Datasets:

kilian-group
/

arxiv-classifier

Dataset card Data Studio Files Files and versions Community

Subset (4)

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

Split (2)

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

paper_id stringlengths 9 16	version stringclasses 26 values	yymm stringclasses 311 values	created timestamp[s]	title stringlengths 6 335	secondary_subfield sequencelengths 1 8	abstract stringlengths 25 3.93k	primary_subfield stringclasses 124 values	field stringclasses 20 values	fulltext stringlengths 0 2.84M
1610.06516	1	1610	2016-10-20T17:42:01	Perfect and semiperfect restricted enveloping algebras	[ "math.RA" ]	For a restricted Lie algebra $L$, the conditions under which its restricted enveloping algebra $u(L)$ is semiperfect are investigated. Moreover, it is proved that $u(L)$ is left (or right) perfect if and only if $L$ is finite-dimensional.	math.RA	math	PERFECT AND SEMIPERFECT RESTRICTED ENVELOPING ALGEBRAS SALVATORE SICILIANO AND HAMID USEFI Abstract. For a restricted Lie algebra L, the conditions under which its restricted enveloping algebra u(L) is semiperfect are investigated. Moreover, it is proved that u(L) is left (or right) perfect if and only if L is finite-dimensional. 1. Introduction Let R be a ring with unity and denote by J (R) the Jacobson radical of R. We recall that R is said to be semiperfect if R/J (R) is Artinian and idempotents of R/J (R) can be lifted to R. Semiperfect rings, introduced by H. Bass in [1], turn out to be a significant class of rings from the viewpoint of homological algebra and representation theory, since they are precisely the rings R for which all finitely generated left or right R-modules have a projective cover (see e.g. [11], Chapter 8, §24). Clearly, one sided Artinian rings and local rings are semiperfect. Recall that R is called left perfect if all left R-modules have projective covers. Right perfect rings are defined in an analogous way. The pioneering work on perfect rings was carried out by H. Bass in 1960 and most of the main characterizations of these rings are contained in his celebrated paper [1]. In particular, it follows from Bass' results that the following conditions are equivalent: R is left perfect; every flat left R-module is projective; R/J (R) is Artinian and for every sequence {ai} in J (R) there exists an integer n such that a1a2 · · · an = 0; R satisfies the descending chain condition on principal right ideals. It should be mentioned that, while semiperfectness is a left-right symmetric property, there exist rings which are perfect on one side but not on the other (see [1], Example 5 on page 476). However, right and left perfectness are clearly equivalent conditions provided R has a nontrivial involution. For instance, this is the case when R is a group algebra or an (ordinary or restricted) enveloping algebra. Left perfect group rings were characterized by G. Renault in [17] and, independently, by S. M. Woods in [23]. It turns out that, for a group G and a field F, the group algebra FG is left perfect if and only if G is finite. Subsequently, a generalization of these results to semigroup rings has been carried out by J. Okni´nski in [14]. On the other hand, although semiperfect group algebras have been also investigated in several papers (see e.g. [3, 6, 22, 24]), a full characterization is not available yet. The best partial result in this direction was obtained by J.M. Goursaud in [6] where, under the assumption that the group G is locally finite, it is proved that FG is semiperfect if and only if either char F = 0 and G is finite or char F = p > 0 and G has a normal p-group of finite index. Date: July 21, 2021. 2010 Mathematics Subject Classification. 17B35, 16P70, 17B50. The research of the second author was supported by NSERC of Canada under grant # RGPIN 418201. 1 2 SALVATORE SICILIANO AND HAMID USEFI In this paper, we consider these problems in the setting of enveloping algebras. For a restricted Lie algebra L over a field F of characteristic p > 0, we denote by u(L) the restricted enveloping algebra of L and by L′ the derived subalgebra of L. Recall that a subset S of L is said to be p-nil if every element of S is p-nilpotent. We first provide some necessary and sufficient conditions on L such that u(L) is semiperfect. We show that if u(L) is semiperfect, then every element of L is p-algebraic and L does not contain any infinite-dimensional torus. Under the assumptions that F is perfect and L′ is p-nil we prove that for a locally finite-dimensional restricted Lie algebra L, u(L) is semiperfect if and only if L contains a p-nil restricted ideal of finite codimension. We construct an infinite-dimensional abelian restricted Lie algebra L over an imperfect field K such that u(L) is semiperfect and L has no nonzero p-nil restricted ideal. Hence, the perfectness assumption on the ground field is necessary. It turns out that the structure of semiperfect ordinary enveloping algebras U (L) of arbitrary Lie algebras L is trivial and also quickly discussed. In the last section, we prove that u(L) is left (or right) perfect if and only if L is finite- dimensional, which represents the Lie-theoretic analogue of the aforementioned result of Renault and Woods. 2. Semiperfectness Throughout the paper, F denotes a field. Of course, every field is a left and right perfect ring. To avoid any possible confusion, we recall that F is called a perfect field if every finite extension of F is separable. It would be an interesting problem to characterize perfect and semiperfect Hopf alge- bras. Our first result, which we actually use later in the setting of restricted enveloping algebras, is a step towards this goal. Recall that an element Λ in a Hopf algebra H with comultiplication ∆ and counit ǫ is called a left integral if hΛ = ǫ(h)Λ, for every h ∈ H. Right integrals are defined analogously. Moreover, an element x of H is called group-like if x 6= 0 and ∆(x) = x ⊗ x and primitive if ∆(x) = 1 ⊗ x + x ⊗ 1. Theorem 2.1. Suppose that a Hopf algebra H is semiperfect. Then H contains no infinite chain of semisimple Hopf subalgebras. Moreover, if H is generated as an algebra by its group-like and primitive elements, then all such elements are algebraic. Proof. For every subspace V of H, we write ¯V for the image of V in H/J (H). By contradiction, suppose that H contains an infinite chain of semisimple Hopf subalgebras 0 ( H1 ( H2 ⊆ · · · ( Hn ( · · · . By Lemma 5.3.1 of [5], each Hopf algebra Hi is finite-dimensional. Moreover, by Theorem 2.2.1 and Corollary 2.2.4 of [13], left and right integrals of Hi coincide and form a 1- dimensional ideal of Hi which is not contained in ker ǫi, where ǫi denotes the counit of Hi. Therefore we can find hi ∈ ker ǫi such that (1 + hi)x = 0 = x(1 + hi), for every x ∈ ker ǫi. Note that for every i ≤ j one has (1 + hj)(1 + hi) = 1 + hj = (1 + hi)(1 + hj). As a consequence, we have a chain H(1 + h1) ⊇ H(1 + h2) ⊇ · · · ⊇ H(1 + hn) ⊇ · · · of left ideals of ¯H. Now, since H is semiperfect, the algebra ¯H is left Artinian and so we must have H(1 + hn) = H(1 + hn+1) PERFECT AND SEMIPERFECT RESTRICTED ENVELOPING ALGEBRAS 3 for some n. Therefore, as the images of 1 + hn and 1 + hn+1 in ¯H are commuting idempotents generating the same left ideal, they must coincide. Thus we have hn+1−hn ∈ J (H). But we have (hn − hn+1)2 = (hn + 1)2 + (hn+1 + 1)2 − 2(hn + 1)(hn+1 + 1) = hn − hn+1. We deduce that hn+1 = hn, because J (H) is free of nonzero idempotents. In particular, 1 + hn+1 ∈ Hn, a contradiction, yielding the first part of the theorem. Now, if H is generated as an algebra by its group-like and primitive elements, then H is clearly pointed. Let x be a group-like or a primitive element of H and suppose, by contradiction, that x is trascendental. Set x0 = x and define recursively xi+1 = xi − x2 i for every n ≥ 0. Since H/J (H) is (right) Artinian, we deduce from Lemma 3.1 in [24] that there exists a positive integer m such that 1 − xm has a right inverse y in H. Let B denote the Hopf subalgebra of H generated by x, so that we have 1 − xm ∈ B. Then B is either the polynomial algebra F[x] (when x is primitive) or the Laurent polynomial algebra F[[x]] (when x is group-like). In particular, B has no nonzero zero-divisors. Therefore, as H is pointed, it follows from the main theorem in [16] that H is a free left B-module, say H = ⊕h∈SBh. Note that by the proof of main theorem in [16], we can assume that 1 ∈ S. Now we can deduce that y is contained in B. But the units of F[x] are non-zero elements of F, and units of F[[x]] are of the form αxk, where 0 6= α ∈ F and k is an integer. This yields a contradiction, completing the proof. ✷ As group algebras are obviously generated by group-like elements, we remark that Theorem 2.1 generalizes Theorem 3.2 in [24]. Let L be a restricted Lie algebra over a field of characteristic p > 0. For S ⊆ L we denote by hSip the restricted subalgebra generated by S. An element x of L is said to be p-algebraic if dim hxip < ∞ and p-transcendental, otherwise. As ordinary and restricted enveloping algebras are generated by primitive elements, Theorem 2.1 yields the following: Corollary 2.2. Let L be restricted Lie algebra over a field of characteristic p > 0 such that u(L) is semiperfect. Then L is p-algebraic and contains no infinite-dimensional torus. Proof. In view of Theorem 2.1, L is clearly p-algebraic. Now suppose, by contradiction, that L contains an infinite dimensional torus T . Clearly, T gives rise to an infinite chain 0 ( T1 ( T2 ⊆ · · · ( Tn ( · · · of finite-dimensional tori of L. Then, by [7], we see that u(Ti) is a semisimple Hopf subalgebra of u(L) for every i, which contradicts Theorem 2.1. ✷ Also, as the non-zero elements of an arbitrary Lie algebra L are transcendental in the ordinary enveloping algebra U (L) of L, from Theorem 2.1 we deduce that U (L) is semiperfect only in the trivial case: Corollary 2.3. Let L be a Lie algebra. Then the ordinary enveloping algebra U (L) is semiperfect if and only if L = 0. Remark 2.4. As the trivial module of an ordinary enveloping algebra U (L) is not projective unless L = 0, one can also prove Corollary 2.3 by using the fact that U (L) is semiprimitive (see Theorem 8.3.14 in [19]) and so only projective U (L)-modules have projective covers (see e.g. [19], Exercise 20 on page 326). 4 SALVATORE SICILIANO AND HAMID USEFI Let X be a basis of L and u ∈ u(L). Then, by the Poincar´e-Birkhoff-Witt (PBW) Theorem for restricted Lie algebras (see [21], Theorem 2.5.1), there exists a finite subset of X, denoted by Supp(u), such that u can be written as a linear combination of PBW monomials in the elements of Supp(u) only. We recall that a restricted Lie algebra is said to be locally finite-dimensional if all of its finitely generated restricted subalgebras are finite-dimensional. Proposition 2.5. Let L be a locally finite-dimensional restricted Lie algebra over a field of characteristic p > 0. Then J (u(L)) is nil. In particular, u(L) is semiperfect if and only if u(L)/J (u(L)) is Artinian. Proof. Let u ∈ J (u(L)). Since Supp(u) is finite and L is locally finite-dimensional, we have that H = hSupp(u)ip is finite-dimensional. Since u ∈ H, we deduce that u(L) is algebraic and the assertion follows from [9, §1.10]. The second part is clear as idempotents can be always lifted modulo a nil ideal. ✷ We need the following result for later use, however we were unable to find a reference for it and as such we include a proof for completeness. Proposition 2.6. Let R be a ring and N a nil ideal of R. If R/N is semiperfect, then so is R. Proof. Note that J (R/N ) = J (R)/N . Hence, we have (R/N )/J (R/N ) ∼= (R/N )/(J (R)/N ) ∼= R/J (R). Since R/N is Artinian, it follows that R/J (R) is Artinian. Now, let e ∈ R such that the image of e modulo J (R) is an idempotent in R/J (R). Since R/N is semiperfect, there exists f ∈ R such that the image of f modulo N is an idempotent in R/N and e − f ∈ J (R). But N is nil, and it is well known that f can be lifted to an idempotent of R, that is, there exists g ∈ R such that f − g ∈ N ⊆J (R). Notice that e − g ∈ J (R) and we are done. ✷ Theorem 2.7. Let L be a locally finite-dimensional restricted Lie algebra over a field of characteristic p > 0. If L has a p-nil restricted ideal of finite codimension, then u(L) is semiperfect. I = P u(L) is nil. To do so, let u ∈ I. Then u = Pn Proof. Let P be a p-nil restricted ideal of finite codimension. We first prove that the ideal i piui, where each pi ∈ P and each ui ∈ u(L). Let B be a basis of L containing a basis of P . Note that there exists a finite subset S of B such that all the elements pi, ui are in u(H), where H is the restricted subalgebra of L generated by S. Now, let Q be the restricted ideal of H generated by all the pi's. Then Q ⊆ P ∩ H, and so we see that Q is a finite-dimensional p-nilpotent restricted ideal of H. Note that Qu(H) is associative nilpotent. Since u ∈ Qu(H), u is nilpotent and as such I is nil. Now, note that u(L/P ) ∼= u(L)/I. Since L/P is finite-dimensional, clearly u(L/P ) is semiperfect. Hence, u(L)/I is semiper- fect and it follows from Proposition 2.6 that u(L) is semiperfect. ✷ PERFECT AND SEMIPERFECT RESTRICTED ENVELOPING ALGEBRAS 5 We remark that an argument similar to the one used in Theorem 2.7 yields an alter- native and shorter proof of the analogous result for group rings obtained in [24]. For a group algebra FG of a locally finite group G over a field F of characteristic p > 0, it is shown by J.M. Goursaud (see [6], Th´eor`eme 8) that FG is semiperfect if and only [G : Op(G)] < ∞, where Op(G) is the largest normal p-subgroup of G. In analogy, we let Op(L) be the sum of all p-nil restricted ideals of L. Since the sum of every two p-nil restricted ideals of L is again p-nil, it follows that Op(L) is also a p-nil restricted ideal. Clearly, if L is finite-dimensional then Op(L) is just radp(L), the p-radical of L. Therefore one might expect that the analogous result holds for restricted Lie algebras, that is, u(L) is semiperfect if and only dim L/Op(L) < ∞. In Proposition 2.8 we provide a partial converse of Theorem 2.7 and, on the other hand, in Example 2.11 we present an infinite-dimensional abelian restricted Lie algebra L over an imperfect field such that u(L) is semiperfect and yet Op(L) = 0. Thus, even in the abelian case, the Lie-theoretic analogue of Goursaud's Theorem does not hold in general. Proposition 2.8. Let L be a locally finite-dimensional restricted Lie algebra over a perfect field F of characteristic p > 0. If u(L) is semiperfect and L′ is p-nil, then Op(L) has finite codimension in L. Proof. First we show that the associative ideal I = Op(L)u(L) is nil. Let u ∈ I. Then u = P xiui, where xi ∈ Op(L) and ui ∈ u(L). Now let H be the restricted subalgebra generated by all the xi's and all the Supp(ui)'s. Note that H is finite-dimensional. Let N be the restricted ideal of H generated by the xi's. Since N ⊆Op(L), we deduce that N is finite-dimensional and p-nilpotent. It follows that N u(H) is nilpotent. In particular, as u ∈ N u(H), we deduce that u is nilpotent. Therefore, I is a nil ideal and as such I⊆J (u(L)). We now claim that I = J (u(L)). Let L = L/Op(L). Since L′⊆Op(L), L is free of nonzero p-nilpotent elements. We claim that u(L ) ∼= u(L)/I is reduced. Let u be a nilpotent element of u(L ). There exist elements x1, . . . , xn in L that are linearly independent modulo Op(L) such that u = X αxi1 1 · · · xin n modulo I. Thus, upr = 0 for some positive integer r. Hence, by the PBW Theorem, the elements x[p]r are linearly dependent modulo Op(L). Let β1, β2, . . . , βn ∈ F, not all zero, 1 such that , . . . , x[p]r n Since F is a perfect field, we get ∈ Op(L). k ∈ Op(L). X βkx[p]r k xk(cid:19)[p]r 1 pr (cid:18)X β But L has no nonzero p-nilpotent elements. We deduce that P β k xk ∈ Op(L), which contradicts the fact that x1, . . . , xn are linearly independent modulo Op(L). Hence, u(L ) is reduced. On the other hand, by Proposition 2.5, J (u(L )) is nil. We deduce that J (u(L )) = 0. Thus, J (u(L)) ⊆ I, as claimed. At this stage, as u(L ) ∼= u(L)/I = u(L)/J (u(L)) is Artinian, we deduce from [12] that L is indeed finite-dimensional. Therefore Op(L) has finite codimension in L, as required. ✷ 1 pr 6 SALVATORE SICILIANO AND HAMID USEFI As an immediate consequence of Theorem 2.7 and Proposition 2.8 we have the following characterization of semiperfect commutative restricted enveloping algebras over perfect fields: Corollary 2.9. Let L be an abelian restricted Lie algebra over a perfect field of charac- teristic p > 0. Then u(L) is semiperfect if and only if Op(L) has finite codimension in L. It is worth mentioning that, in general, the properties of perfectness or semiperfectness of an algebra are not preserved under extensions of the ground field. For instance, consider the following: 1 4 , . . .) be the Example 2.10. Let F be an algebraically closed field and let E = F(t, t extension of the field F(t) of rational functions in the variable t obtained by adjoining all the 2-power roots of t. Consider the automorphism α : E −→ E, f (t) 7→ f (t2) and form the division algebra D = E((x; α)) of skew Laurent series (for details, see e.g., Section 2.3 of [4]). Obviously, D is left perfect. Now, let K = F((x; α)). Then K is a maximal subfield of D and, clearly, the algebra D is not algebraic over K. Moreover, as F is algebraically closed, K is a transcendental field extension of F. Therefore, Theorem 2.4 of [10] assures that J (D ⊗F K) = 0. Finally, by Exercise 6 in Section IX.6 of [8], we see that D ⊗F K is not left Artinian and, indeed, not semiperfect. 1 2 , t We remark that the analogue of the aforementioned result of Goursaud for restricted Lie algebras fails without the perfectness assumption on the field. We show as follows that the restriction on the ground field is required. For simplicity, we construct our example over a field F of characteristic 2 but this example could be extended to any positive characteristic. In the following we denote by K the field of rational functions in infinitely many indeterminates t1, t2, . . . over F, that is, K = F(t1, t2, · · · ). Theorem 2.11. There exists an infinite-dimensional abelian restricted Lie algebra L over K such that u(L) is semiperfect and Op(L) = 0. Proof. Let L be the abelian restricted Lie algebra over K with basis x, y1, y2, . . . and power mapping defined by x[2] = x, y[2] We have that Op(L) = 0. Indeed, let z ∈ L such that z[2] = 0. Then z = αx +P βiyi for some α, βi ∈ K. We have i = tix for i ≥ 1. z[2] = (α2 +X β2 i ti)x = 0. Hence, α2 +P β2 i ti = 0. It is easy to deduce that α = βi = 0 for all i. Hence, z = 0 and Op(L) = 0. We denote by ¯u the image of an element u modulo J (u(L)). Note that L is locally finite-dimensional. Hence, to show that u(L) is semiperfect, by the second part of Proposition 2.5 it is enough to prove that ¯R = u(L)/J (u(L)) is Artinian. For this, we will first show that u(L)/J (u(L)) satisfies the descending chain condition on principal ideals by proceeding in the following steps. Claim 1: Let u ∈ u(L). Then ¯u is invertible in u(L)/J (u(L)) if and only if u is invertible in u(L). Suppose ¯u¯v = ¯1, for some v ∈ u(L). Then 1 − uv is nilpotent. Thus, uv is invertible which implies that u is invertible. PERFECT AND SEMIPERFECT RESTRICTED ENVELOPING ALGEBRAS 7 Claim 2: Let u = a + bx ∈ u(L), where a, b ∈ K. Then u is invertible if and only if a 6= 0 and a 6= b. Suppose that a 6= 0 and a 6= b. Then we have The converse follows from the fact that x(x + 1) = 0. (a + bx)(a−1 + b(a2 + ab)−1x) = 1. Claim 3: Let 0 6= d ∈ K 2r (t2(r−1) such that w2r = d−1x. Suppose that 1 , t2(r−1) 2 , . . .), where r ≥ 1. Then there exists w ∈ u(L) d = β2r +Xn≥1 γ2r i1,...,in (ti1 · · · tin)2(r−1) for some β and γi1,...,in ∈ K. Put w = βx +Pn≥1 γi1,...,inyi1 · · · yin β2 +Pn≥1 γ2 ti1 · · · tin i1,...,in . We observe that w2r = d−1x. Claim 4: For any chain of principal ideals 0 ( (¯u2)⊆(¯u1) ( ¯R, we have that (¯u2) = (¯u1). Let v ∈ u(L) such that ¯u2 = ¯u1¯v. We may assume that ¯v is not invertible. We show that there exists w ∈ u(L) such that u1 − u1vw is nilpotent. Note that (yi(x − 1))2 = 0, for every yi. It follows that ω(L)(x − 1)⊆J (u(L)). In particular, by the PBW Theorem, every element u ∈ u(L) can be written in the form u = α + βx +X γi1,...,inyi1 · · · yin modulo J (u(L)). Thus, for large r, we have u2r = α2r + β2r x +X γ2r i1,...,in (ti1 · · · tin)2(r−1) x. Let u2r We deduce that u2r = a + bx, where a ∈ K 2r and b ∈ K 2r (t2(r−1) , . . .). 1 = a + bx and v2r = c + dx, where a, c ∈ K 2r and b, d ∈ K 2r (t2(r−1) , . . .). Since ¯u1 is not invertible, it follows from Claim 2 that either a = 0 or a = b. Similarly, since ¯v is not invertible, either c = 0 or c = d. , t2(r−1) , t2(r−1) 1 2 1 2 Suppose first that a = 0. Then, if c = d, we get ¯u2r 1 ¯v2r = bcx(1 + x) = 0. Hence, u2 ∈ J (u(L)) and so ¯u2 = 0 which is a contradiction to the assumptions. We deduce that, if a = 0, then c = 0 and c 6= d. Now, by Claim 3, there exists w ∈ u(L) such that 1 − (u1vw)2r = 0. Now, assume that a 6= 0 and so w2r = d−1x. It is easy to see that u2r a = b. In a similar way as above, we can see that c 6= 0 and c = d. Since c ∈ K 2r , there 1 − (u1vw)2r = 0. This completes exists w ∈ u(L) such that w2r = c−1. It is clear that u2r the proof of Claim 4. 2 = ¯u2r It now follows from Claim 4 that u(L)/J (u(L)) satisfies the descending chain con- dition on principal ideals. Since u(L)/J (u(L)) is semiprime, we deduce from Theorem 10.24 in Chapter 4 of [11] that u(L)/J (u(L)) is Artinian, as required. ✷ 3. Perfectness In this section, we characterize left perfect restricted enveloping algebras u(L). As the antipode of any cocommutative Hopf algebra is an involution, left and right perfectness are, in fact, equivalent conditions for u(L). 8 SALVATORE SICILIANO AND HAMID USEFI Lemma 3.1. Let L be a restricted Lie algebra over a field of characteristic p > 0. If u(L) is left perfect, then L is finitely generated. Proof. Let R = u(L). Suppose, by contradiction, that L is not finitely generated and let x1, x2, . . . ∈ L so that xk+1 /∈ hx1, . . . , xkip. Now consider the principal right ideals x1R ⊇ x1x2R ⊇ · · · ⊇ x1x2 · · · xkR ⊇ · · · . Since, according to Bass' Theorem (see [1], Theorem P), R satisfies the descending chain condition on principal right ideals, there exists an integer i such that Therefore one has x1 · · · xiR = x1 · · · xixi+1R. x1 · · · xi − x1 · · · xixi+1v = 0, (3.1) for some v ∈ u(L). Let H be the restricted subalgebra of L generated by x1, . . . , xi. We extend x1, . . . , xi to a basis X of H and extend X to a basis X ∪ Y of L. Since xi+1 /∈ H, we can assume that xi+1 ∈ Y . It is well known that u(L) is a free left u(H)-module, that is, u(L) = ⊕iu(H)ui, where the ui's are distinct PBW monomials in the elements of Y . Now, we observe that Equation (3.1) is not possible because some monomials in the PBW representation of xi+1v involve xi+1 and this contradicts u(L) being a free u(H)-module. Hence, L must be finitely generated. ✷ Lemma 3.2. Let L be a restricted Lie algebra over a field of characteristic p > 0. If u(L) is left perfect, then so is u(H) for every restricted subalgebra H of L. Proof. Let Y be a basis for a complementary vector subspace of H in L. Since u(L) is a free left u(H)-module, we have where the elements uk are distinct PBW monomials in the elements of Y . Let h ∈ u(H) and let I = hu(H) be the right ideal of u(H) generated by h. Then (3.2) implies that u(L) = ⊕ku(H)uk, (3.2) J = ⊕kIuk is a principal right ideal of u(L). As a consequence, every descending chain {Ik}k of principal right ideals of u(H) gives rise to a descending chain of principal right ideals of u(L) which must stabilize, as u(L) is left perfect. Therefore, by (3.2), the chain {Ik}k stabilizes, which finishes the proof. ✷ For a restricted Lie algebra L, we define ∆(L) = {x ∈ L dim [L, x] < ∞}. Then ∆(L) is clearly a restricted ideal of L which is the Lie algebra analogue of the FC-center of a group. Proposition 3.3. Let L be a restricted Lie algebra over a field of characteristic p > 0. If L = ∆(L) and every element of L is p-algebraic, then L is locally finite-dimensional. PERFECT AND SEMIPERFECT RESTRICTED ENVELOPING ALGEBRAS 9 Proof. Let x1, . . . , xn ∈ L and let H be the (ordinary) subalgebra of L generated by these elements. By Lemma 1.3 of [2] we have that dim [H, H] < ∞. Since H is spanned as a vector subspace by x1, . . . , xn and their Lie commutators, we see that H is indeed finite-dimensional. Since every element of H is p-algebraic, it follows from Proposition 1.3(1) in Chapter 2 of [21] that hx1, . . . , xnip is finite-dimensional, as required. ✷ The following result is an immediate consequence of Corollary 6.4 in [2]: Lemma 3.4. Let L be a restricted Lie algebra over a field of characteristic p > 0. Then u(L) is semiprime if and only if u(∆(L)) is semiprime. We can now prove the main result of this section: Theorem 3.5. Let L be a restricted Lie algebra over a field of characteristic p > 0. Then u(L) is left perfect if and only if L is finite-dimensional. Proof. One implication is clear. Suppose now that u(L) is left perfect. By Lemma 3.2, u(∆(L)) is also left perfect and so, by Lemma 3.1, ∆(L) is finitely generated. Note that, by Corollary 2.2, every element of ∆(L) is p-algebraic. Hence, by Proposition 3.3, we con- clude that ∆(L) is finite-dimensional. Thus, it is enough to prove that L/∆(L) is finite- dimensional. Note that, by Corollary 24.19 of [11], the ring u(L/∆(L)) ∼= u(L)/∆(L)u(L) is left perfect, as well. Hence, we can replace L by L/∆(L). By [18, Lemma 2.7(i)], we have that ∆(L/∆(L)) = 0. Therefore we can assume that ∆(L) = 0. Now, Lemma 3.4 entails that u(L) is semiprime and, by Bass' Theorem (see [1], Theorem P), u(L) satisfies the descending chain condition on principal right ideals. But then, by Theorem 10.24 in [11], u(L) is right Artinian. Since, by [12], right Artinian Hopf algebras are finite-dimensional, we conclude that L is finite-dimensional. This completes the proof. ✷ Some immediate consequences of Theorem 3.5 are now in order. As the category of (left) u(L)-modules is equivalent to the category of restricted (left) L-modules, a combination of Theorem 3.5 and Theorem 24.25 in [11] yields the following: Corollary 3.6. Let L be a restricted Lie algebra. The following conditions are equivalent: (1) every restricted flat L-module is projective; (2) L is finite-dimensional. Let R be a ring and θ : G → Aut(R) be a group homomorphism. Then J.K. Park in [15] proved that the smash product R#FG is left perfect if and only if R is left perfect and G is finite. In particular, in view of Theorem 3.5, if R = u(L) for a restricted Lie algebra L, we deduce the following: Corollary 3.7. Let A = u(L)#FG. Then A is left perfect if and only if A is finite- dimensional. Note also that Theorem 3.5 generalizes Corollary 6.6 in [2]. Finally, the celebrated structure theorem of Cartier-Kostant-Milnor-Moore (see e.g. [13, §5.6]) implies that every cocommutative Hopf algebra over an algebraically closed field of characteristic zero can be presented as a smash product of a group algebra and an enveloping algebra. So, Corollary 2.3 along with Park's result in [15] yields: 10 SALVATORE SICILIANO AND HAMID USEFI Corollary 3.8. Let H be a cocommutative Hopf algebra over an algebraically closed field F of characteristic zero. Then H is left perfect if and only if H is the group algebra of a finite group G over F. References [1] H. Bass: Finitistic dimension and a homological generalization of semi-primary rings, Trans. Amer. Math. Soc. 95 (1960), 466 -- 488. [2] J. Bergen -- D. Passman: Delta methods in enveloping rings, J. Algebra 133 (1990), 277-312. [3] W.D. Burgess: On semi-perfect group rings, Can. J. Math. 12 (1969), 645 -- 652. [4] P.M. Cohn: Skew fields: Theory of general division rings. Cambridge University Press, 1995. [5] S. Dascalescu -- C. Nastasescu -- S. Raianu: Hopf algebras. An introduction, Marcel Dekker, Inc., New York, 2001. [6] J.-M. Goursaud: Sur les anneaux de groupes semi-parfaits, Can. J. Math. 25 (1973), 922 -- 928. [7] G. Hochschild: Representations of restricted Lie algebras of characteristic p, Proc. Amer. Math. Soc. 5 (1954), 603 -- 605. [8] T.W. Hungerford: Algebra. Springer-Verlag, New York, 1974. [9] N. Jacobson: Structure of rings. American Mathematical Society, Providence, 1956. [10] E. Jespers - E. Puczilowsky: On ideals of tensor products, J. Algebra 140 (1991), 124 -- 130. [11] T.Y. Lam: A first course in noncommutative rings. Springer-Verlag, New York, 1991. [12] C.H. Liu -- J.J. Zhang: Artinian Hopf algebras are finite dimensional, Proc. Amer. Math. Soc. 135 (2007), 1679 -- 1680. [13] S. Montgomery: Hopf algebras and their actions on rings, CMBS Regional Conference Series in Mathematics, 82, 1993 [14] J. Okni´nski: When is the semigroup ring perfect?, Proc. Amer. Math. Soc., 89 (1983), 49 -- 51. [15] J.K. Park: Artinian skew group rings, Proc. Amer. Math. Soc. 75 (1979), 1 -- 7. [16] D.E. Radford: Pointed Hopf algebras are free over Hopf subalgebras, J. Algebra 45 (1977), 266 -- 273. [17] G. Renault: Sur les anneaux de groupes, C. R. Acad. Sci. Paris 273 (1971), 84 -- 87. [18] D.M. Riley -- A. Shalev: The Lie structure of enveloping algebras, J. Algebra 162(1) (1993), 46 -- 61. [19] L.H. Rowen: Ring Theory. Volume I. Academic Press, Inc., San Diego, 1988. [20] L.H. Rowen: Ring Theory. Volume II. Academic Press, Inc., San Diego, 1988. [21] H. Strade -- R. Farnsteiner: Modular Lie algebras and their representations. Marcel Dekker, New York, 1988. [22] J. Valette: Anneaux de groupes semi-parfaits, C. R. Acad. Sci. Paris, Ser. A 275 (1972), 1219 -- 1222. [23] S.M. Woods: On perfect group rings, Proc. Amer. Math. Soc. 27 (1971), 49 -- 52. [24] S.M. Woods: Some results on semi-perfect group rings, Can. J. Math. 26 (1974), 121 -- 129. Dipartimento di Matematica e Fisica "Ennio De Giorgi", Universit`a del Salento, Via Provinciale Lecce -- Arnesano, 73100 -- Lecce, Italy E-mail address: [email protected] Department of Mathematics and Statistics, Memorial University of Newfoundland, St. John's, NL, Canada, A1C 5S7 E-mail address: [email protected]
1201.6050	2	1201	2012-05-25T12:03:53	Spherical Hecke algebras for Kac-Moody groups over local fields	[ "math.RA", "math.RT" ]	We define the spherical Hecke algebra H for an almost split Kac-Moody group G over a local non-archimedean field. We use the hovel I associated to this situation, which is the analogue of the Bruhat-Tits building for a reductive group. The stabilizer K of a special point on the standard apartment plays the role of a maximal open compact subgroup. We can define H as the algebra of K-bi-invariant functions on G with almost finite support. As two points in the hovel are not always in a same apartment, this support has to be in some large subsemigroup G+ of G. We prove that the structure constants of H are polynomials in the cardinality of the residue field, with integer coefficients depending on the geometry of the standard apartment. We also prove the Satake isomorphism between H and the algebra of Weyl invariant elements in some completion of a Laurent polynomial algebra. In particular, H is always commutative. Actually, our results apply to abstract "locally finite" hovels, so that we can define the spherical algebra with unequal parameters.	math.RA	math	Spherical Hecke algebras for Kac-Moody groups over local fields Stéphane Gaussent and Guy Rousseau May 22, 2012 Abstract We define the spherical Hecke algebra H for an almost split Kac-Moody group G over a local non-archimedean field. We use the hovel I associated to this situation, which is the analogue of the Bruhat-Tits building for a reductive group. The stabilizer K of a special point on the standard apartment plays the role of a maximal open compact subgroup. We can define H as the algebra of K−bi-invariant functions on G with almost finite support. As two points in the hovel are not always in a same apartment, this support has to be in some large subsemigroup G+ of G. We prove that the structure constants of H are polynomials in the cardinality of the residue field, with integer coefficients depending on the geometry of the standard apartment. We also prove the Satake isomorphism between H and the algebra of Weyl invariant elements in some completion of a Laurent polynomial algebra. In particular, H is always commutative. Actually, our results apply to abstract "locally finite" hovels, so that we can define the spherical algebra with unequal parameters. 2 1 0 2 y a M 5 2 ] . A R h t a m [ 2 v 0 5 0 6 . 1 0 2 1 : v i X r a Contents 1 General framework 2 Convolution algebras 3 The split Kac-Moody case 4 Structure constants 5 Satake isomorphism Introduction 3 8 13 15 22 Let G be a connected reductive group over a local non-archimedean field K and let K be an open compact subgroup. The space H of complex functions on G, bi-invariant by K and with compact support is an algebra for the natural convolution product. Ichiro Satake [Sa63] studied this algebra H to define the spherical functions and proved, in particular, that H is commutative for good choices of K. We know now that one of the good choices for K is the fixator of some special vertex for the action of G on its Bruhat-Tits building I , whose structure is explained in [BrT72]. Moreover H, now called the spherical Hecke algebra, may be entirely defined using I , see e.g. [P06]. 1 2 Stéphane Gaussent & Guy Rousseau Kac-Moody groups are interesting generalizations of reductive groups and it is natural to try to generalize the spherical Hecke algebra to the case of a Kac-Moody group. But there is now no good topology on G and no good compact subgroup, so the "convolution product" has to be defined only with algebraic means. Alexander Braverman and David Kazhdan [BrK10] succeeded in defining such a spherical Hecke algebra, when G is split and untwisted affine. For a well chosen subgroup K, they define H as an algebra of K−bi-invariant complex functions with "almost finite" support. There are two new features: the support has to be in a subsemigroup G+ of G and it is an infinite union of double classes. Hence, H is naturally a module over the ring of complex formal power series. Our idea is to define this spherical Hecke algebra using the hovel associated to the almost split Kac-Moody group G that we built in [GR08], [Ro12] and [Ro13]. This hovel I is a set with an action of G and a covering by subsets called apartments. They are in one-to- one correspondence with the maximal split subtori, hence permuted transitively by G. Each apartment A is a finite dimensional real affine space and its stabilizer N in G acts on it via a generalized affine Weyl group W = W v ⋉ Y (where Y ⊂ −→A is a discrete subgroup of translations) which stabilizes a set M of affine hyperplanes called walls. So, I looks much like the Bruhat-Tits building of a reductive group, but M is not a locally finite system of hyperplanes (as the root system Φ is infinite) and two points in I are not always in a same apartment (this is why I is called a hovel). There is on I a G−invariant preorder ≤ which induces on each apartment A the preorder given by the Tits cone T ⊂ −→A . Now, we consider the fixator K in G of a special point 0 in a chosen standard apartment A. The spherical Hecke algebra HR is a space of K−bi-invariant functions on G with values in a ring R. In other words, it is the space HI R of G−invariant functions on I0 × I0 where I0 = G/K is the orbit of 0 in I . The convolution product is easy to guess from this point of view: (ϕ ∗ ψ)(x, y) =Pz∈I0 ϕ(x, z)ψ(z, y) (if this sum means something). As two points x, y in I are not always in a same apartment (i.e. the Cartan decomposition fails: G 6= KN K), we have to consider pairs (x, y) ∈ I0 × I0, with x ≤ y (this implies that x, y are in a same apartment). For HR, this means that the support of ϕ ∈ HR has to be in K\G+/K where G+ = {g ∈ G 0 ≤ g.0} is a semigroup. In addition, K\G+/K is in one-to-one correspondence with the subsemigroup Y ++ = Y ∩ C v f is the fundamental Weyl chamber). Now, to get a well defined convolution product, we have to ask (as in [BrK10]) the support of a ϕ ∈ HR to be almost finite: supp(ϕ) ⊂Sn + is +) ∩ Y ++ is the subsemigroup of Y generated by the fundamental coroots. Note that (λ − Q∨ infinite except when G is reductive. With this definition we are able to prove that HR is really an algebra, which generalizes the known spherical Hecke algebras in the finite or affine split case (§2). In the split case, we describe the hovel I and give a direct proof that HR is commutative (§3). +) ∩ Y ++, where λi ∈ Y ++ and Q∨ The structure constants of HR are the non-negative integers mλ,µ(ν) (for λ, µ, ν ∈ Y ++) such that cλ ∗ cµ =Pν∈Y ++ mλ,µ(ν)cν, where cλ is the characteristic function of KλK. Each chamber (= alcove) in I has only a finite number of adjacent chambers along a given panel. These numbers are called parameters of I and they form a finite set Q. In the split case, there is only one parameter q: the number of elements of the residue field κ of K. In §4 we show that the structure constants are polynomials in these parameters with integral coefficients depending only on the geometry of an apartment. f of Y (where C v i=1 (λi − Q∨ In §5 we build an action of HR on the module of functions from A ∩ I0 to R. This gives an injective homomorphism from HR into a suitable completion R[[Y ]] of the group algebra Spherical Hecke algebras for Kac-Moody groups over local fields 3 R[Y ]; hence HR is abelian (5.3). After modification by a character this homomorphism gives the Satake isomorphism from HR onto the subalgebra R[[Y ]]W v of W v−invariant elements in R[[Y ]]. The proof involves a parabolic retraction of I onto an extended tree inside it. Actually, this article is written in a more general framework (explained in §1): we ask I to be an abstract ordered hovel (as defined in [Ro11]) and G a strongly transitive group of (positive, type-preserving) automorphisms. The general definition and study of Hecke algebras for split Kac-Moody groups over local fields was also undertaken by Alexander Braverman, David Kazhdan and Manish Patnaik (as we knew from [P10]). A preliminary draft appeared recently [BrKP12]. Their arguments are algebraic without use of a geometric object as a hovel, and the proofs seem complete (temporarily?) only for the untwisted affine case. In addition to the construction of the spherical Hecke algebra and the Satake isomorphism (as here), they give a formula for spherical functions and they build the Iwahori-Hecke algebra. We hope to generalize, in a near future, these results to our general framework. One should notice that these authors use, instead of our group K, a smaller K1, a priori slightly different, see Remark in Section 3.4. 1 General framework 1.1 Vectorial data We consider a quadruple (V, W v, (αi)i∈I , (α∨ i )i∈I ) where V is a finite dimensional real vector space, W v a subgroup of GL(V ) (the vectorial Weyl group), I a finite set, (α∨ i )i∈I a family in V and (αi)i∈I a free family in the dual V ∗. We ask these data to verify the conditions of [Ro11, 1.1]. In particular, the formula ri(v) = v − αi(v)α∨ i defines a linear involution in V which is an element in W v and (W v,{ri i ∈ I}) is a Coxeter system. (αj(α∨ Kac-Moody Lie algebra gM and the associated real root system is To be more concrete we consider the Kac-Moody case of [l.c. ; 1.2]: the matrix M = i ))i,j∈I is a generalized Cartan matrix. Then W v is the Weyl group of the corresponding Φ = {w(αi) w ∈ W v, i ∈ I} ⊂ Q =Mi∈I Z.αi. Z.α∨ i ), Q∨ im ∪ ∆− α∨ = w(α∨ im = ∆+ i ) depend only on α, and rα(v) = v − α(v)α∨. We set Φ± = Φ ∩ Q± where Q± = ±(Li∈I (Z≥0).αi) and Q∨ = (Li∈I ±(Li∈I (Z≥0).α∨ or [Ba96]. We shall sometimes also use the set ∆ = Φ ∪ ∆+ −∆− in the sense of [Ba96]. ± = i ). We have Φ = Φ+ ∪ Φ− and, for α = w(αi) ∈ Φ, rα = w.ri.w−1 and The set Φ is an (abstract reduced) real root system in the sense of [MoP89], [MoP95] im of all roots (with im ⊂ Q+, W v−stable) defined in [Ka90]. It is an (abstract reduced) root system f is the disjoint union of the vectorial faces F v(J) = {v ∈ V αi(v) = 0,∀i ∈ J, αi(v) > 0,∀i ∈ I \ J} for J ⊂ I. The positive (resp. negative) vectorial faces are the sets w.F v(J) (resp. −w.F v(J)) for w ∈ W v and J ⊂ I. The set J or the face w.F v(J) is called spherical if the group W v(J) generated by {ri i ∈ J} is finite. The Tits cone T is the (disjoint) union of the positive vectorial faces. It is a W v−stable convex cone in V . f = {v ∈ V αi(v) > 0,∀i ∈ I}. Its closure C v The fundamental positive chamber is C v 4 Stéphane Gaussent & Guy Rousseau 1.2 The model apartment As in [Ro11, 1.4] the model apartment A is V considered as an affine space and endowed with a family M of walls. These walls are affine hyperplanes directed by Ker(α) for α ∈ Φ. that these walls are the hyperplanes defined as follows: We ask this apartment to be semi-discrete and the origin 0 to be special. This means M (α, k) = {v ∈ V α(v) + k = 0} for α ∈ Φ and k ∈ Λα (with Λα = kα.Z a non trivial discrete subgroup of R). Using the following lemma (i.e. replacing Φ by eΦ) we shall assume that Λα = Z,∀α ∈ Φ. For α = w(αi) ∈ Φ, k ∈ Λα(= Z) and M = M (α, k), the reflection rα,k = rM with respect to M is the affine involution of A with fixed point set the wall M and associated linear involution rα. The affine Weyl group W a is the group generated by the reflections rM for M ∈ M; we assume that W a stabilizes M. half-apartment if k ∈ Λα (= Z). define two W v−invariant preorder relations on A: x For α ∈ Φ and k ∈ R, D(α, k) = {v ∈ V α(v) + k ≥ 0} is an half-space, it is called an The Tits cone T and its interior T o are convex and W v−stable cones, therefore, we can o x ≤ y ⇔ y − x ∈ T ; ≤ y ⇔ y − x ∈ T o. If W v has no fixed point in V \ {0} and no finite factor, then they are orders; but they are not in general. αj .αj(kαi.α∨ αi .αi)i∈I , (kαi .α∨ [Ba96]) associated to (V, W v, (k−1 Lemma 1.3. For all α ∈ Φ we choose kα > 0 and define eα = α/kα, eα∨ = kα.α∨. Then eΦ = {eα α ∈ Φ} is the (abstract reduced) real root system (in the sense of [MoP89], [MoP95] or i ))i,j∈I. Moreover with eΦ, the walls are described using the subgroups eM = (k−1 eΛα = Z. Proof. For α, β ∈ Φ, the group W a contains the translation τ by kα.α∨ and τ (M (β, 0)) = β .kα.β(α∨) ∈ Z. Hence eM = So kα.β(α∨) ∈ Λβ i.e. eβ(eα∨) = k−1 M (β,−β(kα.α∨)). (k−1 αj .αj(kαi.α∨ i ))i,j∈I is a generalized Cartan matrix and the lemma is clear, as kwα = kα. i )i∈I ) hence to the generalized Cartan matrix 1.4 Faces, sectors, chimneys... The faces in A are associated to the above systems of walls and halfapartments (i.e. D(α, k) = {v ∈ A α(v) + k ≥ 0}). As in [BrT72], they are no longer subsets of A, but filters of subsets of A. For the definition of that notion and its properties, we refer to [BrT72] or [GR08]. If F is a subset of A containing an element x in its closure, the germ of F in x is the filter germx(F ) consisting of all subsets of A which are intersections of F and neighbourhoods of x. In particular, if x 6= y ∈ E, we denote the germ in x of the segment [x, y] (resp. of the interval ]x, y]) by [x, y) (resp. ]x, y)). The enclosure clA(F ) of a filter F of subsets of A is the filter made of the subsets of A containing an element of F of the shape ∩α∈∆D(α, kα), where kα ∈ Z ∪ {∞} (here, D(α,∞) = A). A face F in the apartment A is associated to a point x ∈ A and a vectorial face F v in V ; it is called spherical according to the nature of F v. More precisely, a subset S of A is an Spherical Hecke algebras for Kac-Moody groups over local fields 5 element of the face F (x, F v) if and only if it contains an intersection of half-spaces D(α, k) or open halfspaces D◦(α, k) (for α ∈ ∆ and k ∈ Z ⊔ {∞}) which contains Ω ∩ (x + F v), where Ω is an open neighborhood of x in A. The enclosure of a face F = F (x, F v) is its closure: the closed-face F . It is the enclosure of the local-face in x, germx(x + F v). There is an order on the faces: the assertions "F is a face of F ′ ", "F ′ covers F " and "F ≤ F ′ " are by definition equivalent to F ⊂ F ′. The dimension of a face F is the smallest dimension of an affine space generated by some S ∈ F . The (unique) such affine space E of minimal dimension is the support of F . Any S ∈ F contains a non empty open subset of E. A face F is spherical if the direction of its support meets the open Tits cone, then its fixator WF in W is finite. Any point x ∈ A is contained in a unique face F (x, V0) which is minimal (but seldom A chamber (or alcove) is a maximal face, or, equivalently, a face such that all its elements spherical); x is a vertex if, and only if, F (x, V0) = {x}. contain a nonempty open subset of A. A panel is a spherical face maximal among faces which are not chambers, or, equivalently, a spherical face of dimension n − 1. Its support is a wall. So, the set of spherical faces of A and the Tits cone completely determine the set M of walls. A sector in A is a V −translate s = x + C v of a vectorial chamber C v = ±w.C v f (w ∈ W v), x is its base point and C v its direction. Two sectors have the same direction if, and only if, they are conjugate by V −translation, and if, and only if, their intersection contains another sector. The sector-germ of a sector s = x + C v in A is the filter S of subsets of A consisting of the sets containing a V −translate of s, it is well determined by the direction C v. So the set of translation classes of sectors in A, the set of vectorial chambers in V and the set of sector-germs in A are in canonical bijection. We write S−∞ the sector-germ associated to the negative fundamental vectorial chamber −C v f . A sector-face in A is a V −translate f = x + F v of a vectorial face F v = ±wF v(J). The sector-face-germ of f is the filter F of subsets containing a translate f′ of f by an element of F v (i.e. f′ ⊂ f). If F v is spherical, then f and F are also called spherical. The sign of f and F is the sign of F v. A chimney in A is associated to a face F = F (x, F v 0 ), its basis, and to a vectorial face F v, its direction, it is the filter r(F, F v) = clA(F + F v). A chimney r = r(F, F v) is splayed if F v is spherical, it is solid if its support (as a filter, i.e. the smallest affine subspace containing r) has a finite fixator in W v. A splayed chimney is therefore solid. The enclosure of a sector-face f = x + F v is a chimney. A halfline δ with origin in x and containing y 6= x (or the interval ]x, y], the segment [x, y]) is called preordered if x ≤ y or y ≤ x and generic if x ≤ x. With these new notions, a chimney can be defined as the enclosure of a preordered halfline and a preordered segment-germ sharing the same origin. The chimney is splayed if, and only if, the halfline is generic. ≤ y or y o o 1.5 The hovel In this section, we recall the definition of an ordered affine hovel given by Guy Rousseau in [Ro11]. 6 Stéphane Gaussent & Guy Rousseau An apartment of type A is a set A endowed with a set Isom(A, A) of bijections (called isomorphisms) such that if f0 ∈ Isom(A, A), then f ∈ Isom(A, A) if, and only if, there exists w ∈ W a satisfying f = f0 ◦ w. An isomorphism between two apartments φ : A → A′ is a bijection such that f ∈ Isom(A, A) if, and only if, φ ◦ f ∈ Isom(A, A′). As the filters in A defined in 1.4 above (e.g. faces, sectors, walls,..) are permuted by W a, they are well defined in any apartment of type A. Definition. An ordered affine hovel of type A is a set I endowed with a covering A of subsets called apartments such that: (MA1) any A ∈ A admits a structure of an apartment of type A; (MA2) if F is a point, a germ of a preordered interval, a generic halfline or a solid chimney in an apartment A and if A′ is another apartment containing F , then A ∩ A′ contains the enclosure clA(F ) of F and there exists an isomorphism from A onto A′ fixing clA(F ); (MA3) if R is a germ of a splayed chimney and if F is a face or a germ of a solid chimney, then there exists an apartment that contains R and F ; (MA4) if two apartments A, A′ contain R and F as in (MA3), then their intersection contains clA(R ∪ F ) and there exists an isomorphism from A onto A′ fixing clA(R ∪ F ); (MAO) if x, y are two points contained in two apartments A and A′, and if x ≤A y then the two segments [x, y]A and [x, y]A′ are equal. We ask here I to be thick of finite thickness: the number of chambers (=alcoves) containing a given panel has to be finite ≥ 3. This number is the same for any panel in a given wall M [Ro11, 2.9]; we denote it by 1 + qM . We assume that I has a strongly transitive group of automorphisms G (i.e. all isomor- phisms involved in the above axioms are induced by elements of G, cf. [Ro13, 4.10]). We choose in I a fundamental apartment which we identify with A. As G is strongly transi- tive, the apartments of I are the sets g.A for g ∈ G. The stabilizer N of A in G induces a group ν(N ) of affine automorphisms of A which permutes the walls, sectors, sector-faces... and contains the affine Weyl group W a [Ro13, 4.13.1]. We denote the fixator of 0 ∈ A in G by K. We ask ν(N ) to be positive and type-preserving for its action on the vectorial faces. This means that the associated linear map −→w of any w ∈ ν(N ) is in W v. As ν(N ) contains W a and stabilizes M, we have ν(N ) = W v ⋉ Y , where W v fixes the origin 0 of A and Y is a group of translations such that: We ask Y to be discrete in V . This is clearly satisfied if Φ generates V ∗ i.e. (αi)i∈I is a Q∨ ⊂ Y ⊂ P ∨ = {v ∈ V α(v) ∈ Z,∀α ∈ Φ}. basis of V ∗. Examples. The main examples of all the above situation are provided by the hovels of almost split Kac-Moody groups over fields complete for a discrete valuation and with a finite residue field, see [Ro12], [Ch10], [Ch11] or [Ro13]. Some details in the split case can be found in Section 3. Remarks. a) In the following, we often refer to [GR08] which deals with split Kac-Moody groups and residue fields containing C. But the results cited are easily generalized to our present framework, using the above references. Spherical Hecke algebras for Kac-Moody groups over local fields 7 b) For an almost split Kac-Moody group over a local field K, the set of roots Φ is KΦred = {Kα ∈ KΦ 1 2 .Kα 6∈ KΦ} where the relative root system KΦ describes well the commuting relations between the root subgroups. Unfortunately eΦ gives a worst description of these relations. 1.6 Type 0 vertices The elements of Y considered as the subset Y = N.0 of V = A are called vertices of type 0 in A; they are special vertices. We note Y + = Y ∩ T and Y ++ = Y ∩ C v f . The type 0 vertices in I are the points on the orbit I0 of 0 by G. This set I0 is often called the affine Grassmannian as it is equal to G/K. on A [Ro11, 5.9]. We set I + = {x ∈ I 0 ≤ x} , I + so I + In general, G is not equal to KY K = KN K [GR08, 6.10] i.e. I0 6= K.Y . We know that I is endowed with a G−invariant preorder ≤ which induces the known one 0 = I0∩I + and G+ = {g ∈ G 0 ≤ g.0}; 0 = G+.0 = G+/K. As ≤ is a G−invariant preorder, G+ is a semigroup. If x ∈ I + there is an apartment A containing 0 and x (by definition of ≤ ) and all apartments containing 0 are conjugated to A by K (axiom (MA2)); so x ∈ K.Y + as I + 0 ∩ A = Y +. But ν(N ∩ K) = W v and Y + = W v.Y ++ (with uniqueness of the element in Y ++); so I + 0 = G+/K is the disjoint union of the KyK/K for y ∈ Y ++. 0 = K.Y ++, more precisely I + Hence, we have proved that the map Y ++ → K\G+/K is one-to-one and onto. 0 f conjugated by W v to x. 1.7 Vectorial distance and Q∨−order For x ∈ T , we note x++ the unique element in C v Let I ×≤ I = {(x, y) ∈ I × I x ≤ y} be the set of increasing pairs in I . Such a pair (x, y) is always in a same apartment g.A; so g−1y − g−1x ∈ T and we define the vectorial f by dv(x, y) = (g−1y − g−1x)++. It does not depend on the choices we distance dv(x, y) ∈ C v made. For (x, y) ∈ I0 ×≤ I0 = {(x, y) ∈ I0 × I0 x ≤ y}, the vectorial distance dv(x, y) takes values in Y ++. Actually, as I0 = G.0, K is the fixator of 0 and I + 0 = K.Y ++ (with uniqueness of the element in Y ++), the map dv induces a bijection between the set I0×≤I0/G of orbits of G in I0 ×≤ I0 and Y ++. Any g ∈ G+ is in K.dv(0, g0).K. For x, y ∈ A, we say that x ≤ Q∨ y (resp. x ≤ Q∨ R+ =Pi∈I or R+−free (i.e.P aiα∨ i ). We get thus a preorder which is an order at least when (α∨ i = 0, ai ≥ 0 ⇒ ai = 0,∀i). + (resp. y − x ∈ i )i∈I is free y) when y − x ∈ Q∨ 1.8 Paths We consider piecewise linear continuous paths π : [0, 1] → A such that each (existing) tangent f under the vectorial Weyl group W v. Such a vector π′(t) is in an orbit W v.λ of some λ ∈ C v path is called a λ−path; it is increasing with respect to the preorder relation ≤ on A. t 6= 1), we let π′ +(t)) denote the derivative of For any t 6= 0 (resp. from the right). Further, we define w±(t) ∈ W v to be the π at t from the left (resp. ±(t) = w±(t).λ (where (W v)λ is the fixator smallest element in its (W v)λ−class such that π′ in W v of λ). Moreover, we denote by π−(t) = π(t) − [0, 1)π′ −(t) = [π(t), π(t − ε) ) (resp. −(t) (resp. π′ Q∨ R≥0.α∨ R 8 Stéphane Gaussent & Guy Rousseau π+(t) = π(t) + [0, 1)π′ segment-germ of π at t. +(t) = [π(t), π(t + ε) ) (for ε > 0 small) the positive (resp. negative) The reverse path π defined by π = π(1 − t) has symmetric properties, it is a (−λ)−path. For any choices of λ ∈ C v f , π0 ∈ A, r ∈ N \ {0} and sequences τ = (τ1, τ2, . . . , τr) of elements in W v/(W v)λ and a = (a0 = 0 < a1 < a2 < ··· < ar = 1) of elements in R, we define a λ−path π = π(λ, π0, τ , a) by the formula: π(t) = π0 + j−1Xi=1 (ai − ai−1)τi(λ) + (t − aj−1)τj(λ) for aj−1 ≤ t ≤ aj. Any λ−path may be defined in this way (and we may assume τj 6= τj+1). Definition. [KM08, 3.27] A Hecke path of shape λ with respect to −C v that, for all t ∈ [0, 1]\{0, 1}, π′ from π′ (β1, . . . , βs) of real roots such that, for all i = 1, . . . , s: +(t), i.e. finite sequences (ξ0 = π′ −(t), ξ1, . . . , ξs = π′ +(t) ≤W v −(t) to π′ f is a λ−path such π(t)−chain +(t)) of vectors in V and π′ −(t), which means that there exists a W v π(t) i) rβi(ξi−1) = ξi, π(t) i.e. βi(π(t)) ∈ Z: π(t) is in a wall of direction Ker(βi). ii) βi(ξi−1) < 0, iii) rβi ∈ W v iv) each βi is positive with respect to −C v Remarks. 1) The path is folded at π(t) by applying successive reflections along the walls M (βi,−βi(π(t)) ). Moreover conditions ii) and iv) tell us that the path is "positively folded" (cf. [GL05]) i.e. centrifugally folded with respect to the sector germ S−∞ = germ∞(−C v f ). f ) be the negative fundamental chamber (= alcove). A Hecke path of shape λ with respect to c− [BCGR11] is a λ−path in the Tits cone T satisfying the above conditions except that we replace iv) by : 2) Let c− = germ0(−C v f i.e. βi(C v f ) > 0. iv') each βi is positive with respect to c− i.e. βi(π(t) − c−) > 0. Then ii) and iv') tell us that the path is centrifugally folded with respect to the center c−. 2 Convolution algebras 2.1 Wanted We consider the space of G−invariant functions on I0 ×≤ I0 with values in a ring R (essentially C or Z). We want R = bHR(I , G) = {ϕI : I0 ×≤ I0 → R ϕI(gx, gy) = ϕI(x, y),∀g ∈ G} bHI to make bHI R (or some large subspace) an algebra for the following convolution product: ϕI(x, z)ψI(z, y). (ϕI ∗ ψI)(x, y) = Xx≤z≤y It is clear that this product is associative and R−bilinear if it exists. Spherical Hecke algebras for Kac-Moody groups over local fields 9 ϕG(g) = ϕI(0, g.0) and ϕI(x, y) = ϕG(dv(x, y)). R and bHR is given by: Via dv, bHI R is linearly isomorphic to the space bHR = {ϕG : Y ++ = K\G+/K → R}, which can be interpreted as the space of K−bi-invariant functions on G+. The correspondence ϕI ↔ ϕG between bHI In this setting, the convolution product should be: (ϕG∗ ψG)(g) =Ph∈G+/K ϕG(h)ψG(h−1g), where we consider ϕG and ψG as trivial on G \ G+. In the following we shall often make no difference between ϕI or ϕG and forget the exponents I and G. We consider the subspace Hf R of functions with finite support in Y ++ = K\G+/K; its natural basis is (cλ)λ∈Y ++ where cλ sends λ to 1 and µ 6= λ to 0. Clearly c0 is a unit for ∗. In bHI R , (cλ ∗ cµ)I(x, y) is the number of triangles [x, z, y] with dv(x, z) = λ and dv(z, y) = µ. As suggested by [BrK10] and lemma 2.4, we consider also the subspace HR of bHR of +) ∩ Y ++ where λi ∈ Y ++. functions ϕ with almost finite support i.e. supp(ϕ) ⊂ ∪n 2.2 Retractions onto Y + For all x ∈ I + there is an apartment containing x and c− [Ro11, 5.1] and this apartment is conjugated to A by an element of K fixing c− (axiom (MA2) ). So, by the usual arguments and [l.c. , 5.5] we can define a retraction ρc− of I + into A with center c−; its image is ρc−(I +) = T = I + ∩ A and ρc−(I + There is also a retraction ρ−∞ of I onto A with center the sector-germ S−∞ [GR08, 4.4]. For ρ = ρc− or ρ−∞ the image of a segment [x, y] with (x, y) ∈ I ×≤ I and dv(x, y) = i=1 (λi − Q∨ 0 ) = Y +. f is a λ−path [GR08, 4.4]. In particular, ρ(x) ≤ ρ(y). λ ∈ C v 2.3 Convolution product The convolution product in bHR should be defined (for y ∈ Y ++) by (ϕ ∗ ψ)(y) =X ϕ(z)ψ(dv(z, y)) where the sum runs over the z ∈ I + 0 such that 0 ≤ z ≤ y and ϕ(z) = ϕI (0, z) = ϕG(dv(0, z)). 1) Using ρc− we have, for λ, µ, y ∈ Y ++, (cλ ∗ cµ)(y) =Pw∈W v/(W v)λ Nc−(µ, w.λ, y) where Nc−(µ, w.λ, y) is the number of z ∈ I + 0 with dv(z, y) = µ and ρc−(z) = w.λ ∈ Y +. Note that, if Nc−(µ, wλ, y) > 0, there exists a µ−path from wλ to y, hence y ∈ wλ + Y +. So cλ ∗ cµ is the formal sum cλ ∗ cµ = Pν∈Y ++ mλ,µ(ν)cν where the structure constant mλ,µ(ν) =Pw∈W v/(W v)λ Nc−(µ, w.λ, ν) ∈ Z≥0∪{+∞} is also equal to the number of triangles [x, z, y] with dv(x, z) = λ and dv(z, y) = µ, for any fixed pair (x, y) ∈ I0 ×≤ I0 with dv(x, y) = ν (e.g. (x, y) = (0, ν)). 2) Using ρ−∞ we have mλ,µ(ν) = Pz ′ N−∞(µ, z′, ν) where the sum runs over the z′ in Y +(λ) = ρ−∞({z ∈ I + dv(0, z) = λ}) and N−∞(µ, z′, ν) ∈ Z≥0 ∪ {+∞} is the number of 0 with dv(0, z) = λ, dv(z, y) = µ (for any y ∈ I + z ∈ I + 0 with dv(0, y) = ν e.g. y = ν) and ρ−∞(z) = z′. But ρ−∞([0, z]) is a λ−path hence increasing with respect to ≤ , so Y +(λ) ⊂ Y +. Moreover, ρ−∞([z, ν]) is a µ−path, so z′ has to be in ν − Y +. Hence, z′ has to run over the set Y +(λ) ∩ (ν − Y +) ⊂ Y + ∩ (ν − Y +). Actually, the image by ρ−∞ of a segment [x, y] with (x, y) ∈ I ×≤ I and dv(x, y) = λ ∈ Y ++ is a Hecke path of shape λ with respect to −C v f [GR08, th. 6.2]. Hence the following results: 0 10 Stéphane Gaussent & Guy Rousseau b) Let π be a Hecke path of shape λ ∈ Y ++ with respect to −C v Lemma 2.4. a) For λ ∈ Y ++ and w ∈ W v, wλ ∈ λ − Q∨ Then λ = π′(0)++ = π′(1)++, π′(0) ≤ Q∨ λ, π′(0) ≤ Q∨ y1 − y0 ≤ Q∨ λ. +, i.e. wλ ≤ Q∨ λ. (y1 − y0) ≤ Q∨ c) If moreover (α∨ i )i∈I is free, we may replace above ≤ Q∨ d) For λ, µ, ν ∈ Y ++, if mλ,µ(ν) > 0, then ν ∈ λ + µ − Q∨ by ≤ Q∨. + i.e. ν ≤ Q∨ λ + µ. R R f , from y0 ∈ Y to y1 ∈ Y . π′(1) ≤ Q∨ λ and R i )i∈I is free. N.B. By d) above, if x ≤ z ≤ y in I0, then dv(x, y) ≤ Q∨dv(x, z) + dv(z, y). Proof. a) By definition, for λ ∈ Y , wλ ∈ λ + Q∨, hence a) follows from [Ka90, 3.12d] used in a realization where (α∨ b) By definition of Hecke paths in 1.8, λ = π′(0)++ = π′(1)++. Moreover, ∀t ∈ [0, 1], −(t) by successive reflections; R+. By integrating the locally constant function π′(t), we −(t)++ = π′ λ = π′ this proves that π′ get π′(0) ≤ Q∨ It is proved (but not stated) in [GR08, 5.3.3] that any Hecke path of shape λ starting in y0 ∈ Y can be transformed in the path πλ(t) = y0 + λt by applying successively the operators Idem for +(t)++ and we know how to get π′ +(t) ∈ π′ c) By b) y1 − y0 − π′(0) ∈ Q∨ d) If mλ,µ(ν) > 0 we have an Hecke path of shape λ (resp. µ) from 0 to z′ (resp. from z′ eαi oreeαi for i ∈ I; moreover eαi (π)(1) = π(1)+α∨ i andeeαi(π)(1) = π(1), hence y1−y0 ≤ Q∨ λ. +, so π′(0) ≤ Q∨ (y1 − y0). R+ ∩ Q∨ = Q∨ (y1 − y0) ≤ Q∨ π′(1) ≤ Q∨ +(t) from π′ −(t) + Q∨ λ. R y1 − y0 ≤ Q∨ π′(1). to ν). So d) follows from b). R R Proposition 2.5. Suppose (α∨ i )i∈I free in V . Then for all λ, µ, ν ∈ Y ++, mλ,µ(ν) is finite. i )i∈I free by (α∨ N.B. Actually we may replace the condition (α∨ Proof. We have to count the z ∈ I + such that dv(0, z) = λ and dv(z, ν) = µ. We set z′ = ρ−∞(z). By lemma 2.4b, z′ ∈ λ−Q∨ +)∩(ν−µ+Q∨ +) which is finite as (α∨ i )i∈I is free or R+−free. So, we fix now z′. By [GR08, cor. 5.9] there is a finite number of Hecke paths π′ of shape µ from z′ to ν. So, we fix now π′. And by [l.c. th. 6.3] (see also 4.10, 4.11) there is a finite number of segments [z, ν] retracting to π′; hence the number of z is finite. +, hence z′ is in (λ−Q∨ + and ν ∈ z′ +µ−Q∨ i )i∈I R+−free. 0 Theorem 2.6. Suppose (α∨ i )i∈I free or R+−free, then HR is an algebra. Proof. We saw that for λ, µ, ν ∈ Y ++, mλ,µ(ν) is finite; hence cλ∗cµ is well defined (eventually as an infinite formal sum). Let us consider ϕ, ψ ∈ HR: supp(ϕ) ⊂ ∪m +), supp(ψ) ⊂ ∪n j=1 (µj − Q∨ If mλ,µ(ν) > 0 with λ ∈ supp(ϕ), µ ∈ supp(ψ) (hence + for some i, j), we have λ + µ ∈ ν + Q∨ λ ∈ λi − Q∨ + by lemma 2.4d. So +) ⊂ (ν − µj + Q∨ λ ∈ (ν − µ + Q∨ +), a finite set. For the same reasons µ is in a finite set, so ϕ ∗ ψ is well defined. With the above notations ν ∈ (λ + µ − Q∨ +) ∩ (λi − Q∨ +) ⊂ ∪i,j (λi + µj − Q∨ +). Let ν ∈ Y ++. +, µ ∈ µj − Q∨ +) ∩ (λi − Q∨ +), so ϕ ∗ ψ ∈ HR. i=1 (λi − Q∨ Definition 2.7. HR = HR(I , G) is the spherical Hecke algebra (with coefficients in R) associated to the hovel I and its strongly transitive automorphism group G. Remark. We shall now investigate HR and some other possible convolution algebras in bHR by separating the cases: finite, indefinite and affine. Spherical Hecke algebras for Kac-Moody groups over local fields 11 2.8 Finite case In this case Φ and W v are finite, (α∨ hovel I = I + is a locally finite Bruhat-Tits building. i )i∈I is free, T = V and the relation ≤ is trivial. The Let ρ be the half sum of positive roots. As 2ρ ∈ Q and ρ(α∨ i ) = 1, ∀i ∈ I, we see that an almost finite set in Y ++ is always finite. So HR and Hf The algebra HC was already studied by I. Satake in [Sa63]. Its close link with buildings is explained in [P06]. The algebra HZ is the spherical Hecke ring of [KLM08], where the interpretation of mλ,µ(ν) as a number of triangles in I is already given. R are equal. in W v). bHR is not an algebra as e.g. mλ,(−w0)λ(0) 6= 0 ∀λ ∈ Y ++ (where w0 is the greatest element 2.9 Indefinite case im an element δ (of support I) such that δ(α∨ Lemma. Suppose now Φ associated to an indefinite indecomposable generalized Cartan matrix. Then there is in ∆+ i ) < 0, ∀i ∈ I and a basis (δi)i∈I of the real vector space QR spanned by Φ such that δi(T ) ≥ 0, ∀i ∈ I. Proof. Any δ ∈ ∆+ is δ ∈ ∆+ Replacing eventually δ by 3δ [l.c. 5.5], we have (δ + αi)(α∨ The wanted basis is inside {δ} ∪ {δ0 + αi i ∈ I}. im takes positive values on T [Ka90, 5.8]. Now, in the indefinite case, there i ) < 0, ∀i ∈ I [l.c. 4.3], hence δ + αi ∈ ∆+, ∀i ∈ I. j ) < 0, ∀i, j ∈ I, hence δ + αi ∈ ∆+ im. im ∩ (⊕i∈I R>0.αi) such that δ(α∨ + has to be in a finite set. R is in general not a subalgebra. im as in the lemma proves that (α∨ The existence of δ ∈ ∆+ i )i∈I is R+−free. So HR is an algebra. The following example 2.10 proves that Hf If (αi)i∈I generates (i.e. is a basis of) V ∗, bHR is also an algebra (the formal spherical Hecke algebra): Let ν ∈ Y ++, we have to prove that there is only a finite number of pairs (λ, µ) ∈ (Y ++)2 such that mλ,µ(ν) > 0. Let z′ be as in the proof of 2.5. We saw in 2.3 that z′ ∈ Y + ∩ (ν − Y +) = Y ∩ T ∩ (ν − T ). By the lemma, T ∩ (ν − T ) is bounded, hence Y ∩ T ∩ (ν − T ) is finite. So we may fix z′. Now λ ∈ z′ + Q∨ + hence (for δ as in the lemma) δ(λ) ≤ δ(z′); as αi(λ) ∈ Z>0 ∀i ∈ I and δ ∈ ⊕i∈I R>0.αi this gives only a finite number of possibilities for λ. Similarly µ ∈ ν − z′ + Q∨ Actually bHR is often equal to HR when (α∨ i )i∈I is free and (αi)i∈I generates V ∗ (hence the matrix M = (αj(α∨ Let us consider the Kac-Moody matrix M =(cid:18) 2 −3 the matrices of r1, r2, r2r1 and r1r2 are respectively (cid:18)−1 0 and M −1 = (cid:18)−1 −3 2 (cid:19). The basis of Φ and V ∗ is {α1, α2} 2 (cid:19) and −3 −1(cid:19) 8 (cid:19). The eigenvalues of M or M −1 are a± = (7 ± √45)/2. In a basis 1 =(cid:18) 2 2 =(cid:18)−3 −3(cid:19), α∨ 1(cid:19),(cid:18)1 0 −1(cid:19), M =(cid:18) 8 i )) is invertible), see the following example 2.10. 2.10 An indefinite rank 2 example diagonalizing M and M −1 we see easily that (r2r1)n + (r1r2)n = an.IdV where an = an is in N and increasing up to infinity (a0 = 2, a1 = 7, a2 = 47, a3 = 322,...). + + an − and we consider the dual basis (∨ 1 , ∨ 2 ) of V . In this basis α∨ −3 3 3 3 3 12 Stéphane Gaussent & Guy Rousseau Consider now λ = µ = −α∨ 2 . We have (r2r1)n.λ + (r1r2)n.λ = an.λ. This means that mλ,λ(an.λ) ≥ Nc−(λ, (r2r1)nλ, an.λ) ≥ 1, for all positive n (and the same thing for N−∞). So cλ ∗ cλ is an infinite formal sum. 1 ⊕ Z≥0.∨ 1 − α∨ 2 = (cid:18)1 1(cid:19) in Y ++ ⊂ Z≥0.∨ Actually (−Q∨ +) ∩ Y ++ ⊃ Z≥0.5∨ 1 ⊕ Z≥0.5∨ 2 , hence Y ++ itself is almost finite! 2.11 An affine rank 2 example 1 2 1 + α∨ 1 , α∨ 1 , ∨ −1 −2 2 3 3 1 0 Z≥0∨ i=1 −2 0 , ∨ 1 1 3 0 −1 −2 3 0 2 i ∈ Y ++ ⊂ ⊕2 2 = −2∨ 0 ∈ Q∨ −1 2 −2 0 1 0 −1 0 1 0 + is the canonical central element [Ka90, § 6.2] and the above 1 1 0 2 0 −2 −1 2 4n2 4n2 0 0 0 1 2 } is free and with basis of 2 ) is the dual basis of V , we have 2 =  , α∨ ,  1 0 1 0 1 2 0 0 −1 c = α∨ calculations are peculiar cases of [l.c. § 6.5]. i=1 ai∨ 2 (cid:19). The basis of Φ is {α1, α2} but  and the matrices of r1, r2, r1r2 and r2r1 are respectively . A classical , M = . Actually Let us consider the Kac-Moody matrix M = (cid:18) 2 −2 we consider a realization V of dimension 3 for which {α∨ V ∗ {αo = −ρ, α1, α2}. More precisely, if (∨ 1 =  α∨  and M −1 =  calculus using triangulation tells us that (r2r1)n + (r1r2)n =  Let's consider now λ = µ = P2 i . We have (r2r1)n(λ) + (r1r2)n(λ) = λ − 2n2λc with λ = a1 + a2. This means that mλ,λ(λ − 2n2λc) ≥ Nc−(λ, (r2r1)n(λ), λ − 2n2λc) ≥ 1, ∀n ∈ Z (and the same thing for N−∞). So cλ ∗ cλ is an infinite formal sum. Moreover as c is fixed by r1 and r2, (r2r1)n(λ+2n2λc)+(r1r2)n(λ) = λ, so mλ+2n2λc,λ(λ) ≥ 1, ∀n ∈ Z, and bHR is not an algebra. Remark also that, if we consider the essential quotient V e = V /Rc, the above calculus tells that mλ,λ(λ) ≥ Pn∈Z Nc−(λ, (r2r1)n(λ), λ) is infinite if λ > 0. 2.12 Affine indecomposable case We saw in the example 2.11 above that mλ,λ(λ) may be infinite, ∀λ ∈ Y ++ when (α∨ not free. So, in this case, bHR seems to contain no algebra except R.c0. i )i∈I R+−free in the affine indecomposable is c = 0 where c = Pi∈I a∨ i (with An almost finite subset in Y ++ is a finite union of subsets like Yλ = (λ − Q∨ λ. But δ = Pi∈I ai.αi with ai ∈ Z>0 ∀i ∈ I, so the image of Y ′ i )i∈I free is equivalent to (α∨ case as the only possible relation between the α∨ i a∨ i ∈ Z>0 ∀i ∈ I) is the canonical central element. +) ∩ Y ++. Let δ be the smallest positive imaginary root in ∆. Then δ(Q∨ +) = 0 so Yλ ⊂ {y ∈ Y ++ δ(y) = λ in V e = V /Rc δ(λ)} = Y ′ (where Rc = ∩i∈I Ker(αi)) is finite. It is now clear that Yλ is a finite union of sets like µ − Z≥0.c with µ ∈ Y ++. Hence an almost finite subset as defined above is the same as an almost finite union (of double cosets) as defined in [BrK10]. The algebra HC is the one introduced by A. Braverman and D. Kazhdan in [BrK10]. We gave above a combinatorial proof that it is an algebra, without algebraic geometry. Remark also that (α∨ i )i∈I is i .α∨ Spherical Hecke algebras for Kac-Moody groups over local fields 13 3 The split Kac-Moody case 3.1 Situation As in [Ro12] or [Ro13], we consider a split Kac-Moody group G associated to a root generating system (RGS) S = (M, YS , (αi)i∈I , (α∨ i )i∈I ) over a field K endowed with a discrete valuation ω (with value group Λ = Z and ring of integers O = ω−1([0, +∞])) whose residue field κ = Fq is finite . So, M = (ai,j)i,j∈I is a Kac-Moody matrix, YS a free Z−module, (α∨ i )i∈I a family in YS, (αi)i∈I a family in the dual X = Y ∗ S of YS and αj(α∨ i ) = ai,j. If (αi)i∈I is free in X, we consider V = VY = YS⊗Z R and the clear quadruple (V, W v, (αi = αi)i∈I , (α∨ i )i∈I ). In general, we may define Q = ZI with canonical basis (αi)i∈I, then V = VQ = HomZ(Q, R) is also in a quadruple as in 1.1. A third example V xl of choice for V is explained in [Ro13]. We always denote by bar : Q → X the linear map sending αi to αi. Λα = Λ = Z, ∀α ∈ Φ). is G = G(K). By [Ro11, 6.11] or [Ro12, 5.16] we have qM = q for any wall M. When G is a split reductive group, I is its extended Bruhat-Tits building. Now the hovel I in 1.5 is as defined in [Ro12] or [Ro13] and the strongly transitive group With these vectorial data we may define what was considered in 1.1 and 1.2 (we choose 3.2 Generators for G The Kac-Moody group G contains a split maximal torus T with character group X and cochar- acter group YS. We note T = T(K). For each α ∈ Φ ⊂ Q there is a group homomorphism xα : K → G which is one-to-one; its image is the subgroup Uα. Now G is generated by T and the subgroups Uα for α ∈ Φ, submitted to some relations given by Tits [T87], also available in [Re02] or [Ro12]. We set U ± the subgroup generated by the subgroups Uα for α ∈ Φ±. We shall explain now only a few of the relations. For u ∈ K, t ∈ T and α ∈ Φ one has: (KMT4) t.xα(u).t−1 = xα(α(t).u) (where α = bar(α)) For u 6= 0, we noteesα(u) = xα(u).x−α(u−1).xα(u) andesα =esα(1). (KMT5)esα(u).t.esα(u)−1 = rα(t) (W v acts on V , YS, X hence on T ) 3.3 Weyl groups Actually the stabilizer N of A ⊂ I is the normalizer of T in G. The image ν(N ) of N in Aut(A) is a semi-direct product ν(N ) = ν(N0) ⋉ ν(T ) with: N0 is the fixator of 0 in N and ν(N0) is isomorphic to W v acting linearly on A = V . Actually ν(N0) is generated by the elements ν(esα) which act as rα (for α ∈ Φ). t ∈ T acts on A by a translation of vector ν(t) ∈ V such that χ(ν(t)) = −ω(χ(t)) for any χ ∈ X = Y ∗ S and χ ∈ X or Q which are related by χ = χ if V = VY or χ = bar(χ) if V = VQ. So, ν(N ) = W v ⋉ Y where Y is closely related to YS ≃ T /T(O): as Λ = ω(K) = Z, they are equal if V = VY and, if V = VQ, Y = bar∗(YS) is the image of YS by the map bar∗ : YS → HomZ(Q, Z) dual to bar. So, the choice V = VY is more pleasant. The choice V = VQ is made e.g. in [Ch10], [Ch11] or [Re02] and has good properties in the indefinite case, cf. 2.9. They coincide both when (αi)i∈I is a basis of X ⊗ R = V ∗ Y . This assumption generalizes semi-simplicity, in particular the center of G is then finite [Re02, 9.6.2]. 14 Stéphane Gaussent & Guy Rousseau 3.4 The group K The group K = G0 should be equal to G(O) for some integral structure of G over O cf. [GR08, 3.14]. But the appropriate integral structure is difficult to define in general. So we define K by its generators: The group N0 is generated by T0 = T(O) = T ∩ K and the elementsesα for α ∈ Φ (this is 0 = U0 ∩ U ±. In general U ± clear by 3.3). The group U0, generated by the groups Uα,0 = xα(O) for α ∈ Φ, is in K. We note U ± 0 is not generated by the groups Uα,0 for α ∈ Φ± [Ro12, 4.12.3a]. It is likely that K may be greater than the group generated by N0 and U0 (i.e. by U0 0 as follows. In a formal ⊃ U − =Qα∈∆+ Uα,0 of the subgroup positive completion bG+ of G, we can define a subgroup U ma+ U ma+ =Qα∈∆+ Uα of bG+, with U + ⊂ U ma+ (where Uα,0 and Uα are suitably defined for α and T0). We have to define groups U pm+ ∩ U +. The group U nm− is defined similarly 0 and U nm− ⊃ U + = U ma+ 0 0 0 0 0 0 ∩ G = U ma+ .U − 0 .N0 ⊂ U ma− in a formal negative completion bG− of G. [Ro12, 4.14, 5.1] 0 .N0 = U pm+ 0 imaginary). Then U pm+ with ∆− using a group U ma− 0 .U + Now K = G0 = U nm− 0 0 Remark. Let us denote by K1 the group used by A. Braverman, D. Kazhdan and M. Patnaik in their definition of the spherical Hecke algebra. With the notation above, K1 is generated by T0 and U0, i.e. by T0, U + 0 and U − .K1, with U − 0 ⊂ U nm− ⊂ U +. But they prove, at least in the untwisted affine case, that U −∩U +.K1 ⊂ K1 [BrKP12, proof of 6.4.3]; so U nm− ⊂ U −∩K ⊂ U −∩U +.K1 ⊂ K1 and K = K1. This result answers positively a question in [Ro13, 5.4], at least for points of type 0 and in the untwisted affine split case. 0 , hence K = U nm− ⊂ U − and U + 0 ⊂ U pm+ .K1 = U pm+ 0 0 0 0 0 Proposition 3.5. There is an involution θ (called Chevalley involution) of the group G such that θ(t) = t−1 for all t ∈ T and θ(xα(u)) = x−α(u) for all α ∈ Φ and u ∈ K. Moreover K is θ−stable and θ induces the identity on W v = N/T . Proof. This involution is well known on the corresponding complex Lie algebra, see [Ka90, 1.3.4] where one uses for the generators eα a convention different from ours ([eα, e−α] = −α∨ as in [T87] or [Re02]). Hence the proposition follows when κ contains C or is at least of characteristic 0. But here we have to use the definition of G by generators and relations. for (KMT4) and (KMT5) (as rα = r−α). The three other relations are: We see in [Ro12, 1.5, 1.7.5] that esα(−u) = esα(u)−1 and esα(u) = es−α(u−1). So for the wanted involution θ we have θ(esα(u)) = es−α(u) = esα(u−1). We have now to verify the relations between the θ(xα(u)) = x−α(u), θ(t) = t−1 and θ(esα(u)) =esα(u−1). This is clear (KMT3) (xα(u), xβ(v)) = Q xγ(C α,β p,q .upvq) for (α, β) ∈ Φ2 prenilpotent and, for the product, γ = pα+ qβ runs in (Z>0α+ Z>0β)∩ Φ. But the integers C α,β p,q are picked up from the corresponding formula between exponentials in the automorphism group of the corresponding complex Lie algebra. As we know that θ is defined in this Lie algebra, we have C −α,−β = C α,β p,q and (KMT3) is still true for the images by θ. p,q (KMT6)esα(u−1) =esα.α∨(u) for α simple and u ∈ K \ {0}. This is still true after a change by θ as θ(esα(u−1)) =esα(u) and (−α)∨(u) = α∨(u−1). (KMT7) esα.xβ(u).es−1 α = xγ(ε.u) if γ = rα(β) and esα(eβ) = ε.eγ in the Lie algebra (with ε = ±1). This is still true after a change by θ becauseesα(eβ) = ε.eγ ⇒esα(e−β) = ε.e−γ (as rα(β∨) = γ∨). Spherical Hecke algebras for Kac-Moody groups over local fields 15 So, θ is a well defined involution of G, θ(U0) = U0, θ(N0) = N0 and θ(U ± 0 . But the isomorphism θ of U + onto U − can clearly be extended to an isomorphism θ from U ma+ onto U ma− sending U ma+ onto U ma− θ induces the identity on W v = N/T . . So θ(U pm+ 0 ) = U ∓ 0 0 ) = U nm− 0 0 and θ(K) = K. As θ(esα) =esα, Theorem 3.6. The algebra bHR or HR is commutative, when it exists. Notation: To be clearer we shall sometimes write bHR(G,K) or HR(G,K) instead of bHR or HR. Proof. The formula θ#(g) = θ(g−1) defines an anti-involution (θ#(gh) = θ#(h).θ#(g) ) of G which induces the identity on T and stabilizes K. In particular θ#(G+) = θ#(KY ++K) = G+ and θ#(KλK) = KλK, ∀λ ∈ Y ++. For ϕ, ψ ∈ bHR and g ∈ G+, one has: (ϕ∗ ψ)(θ#(g)) =Ph∈G+/K ϕ(h)ψ(h−1θ#(g)). The map h 7→ h′ = θ#(h−1θ#(g)) = gθ#(h−1) is one-to-one from G+/K onto G+/K. So, (ϕ ∗ ψ)(g) =Ph′∈G+/K ϕ(θ#(h′−1g))ψ(θ#(h′)) = Ph′∈G+/K ϕ(h′−1g)ψ(h′) = (ψ ∗ ϕ)(g). Remarks 3.7. 1) This commutativity will be below proved in general as a consequence of the Satake isomorphism. The above proof generalizes well known proofs in the reductive case, e.g. for G = GLn, θ# is the transposition. (ϕ ∗ ψ)(g) = 2) When G is an almost split Kac-Moody group over the field K (supposed complete or henselian) it splits over a finite Galois extension L, the hovel KI over K exists and embeds in the hovel LI over L [Ro13, § 6]. After enlarging eventually L one may suppose that 0 is a special point in KI and LI , more precisely in the fundamental apartments KA ⊂ LA = A associated respectively to a maximal K−split torus KS and a L−split maximal torus T ⊃ KS. If we make a good choice of the homomorphisms xα : L → G(L), the associated involution θ of G(L) should commute with the action of the Galois group Γ = Gal(L/K) hence induce an involution Kθ and an anti-involution Kθ# of G(K) = G(L)Γ such that Kθ(K) = Kθ#(K) = K and Kθ# induces the identity in Y (KS) = KS(K)/KS(O). The commutativity of bHR(G,K) or HR(G,K) would follow. general case. 3) The commutativity of bHR or HR is linked to the choice of a special vertex for the origin 0. Even in the semi-simple case, other choices may give non commutative convolution algebras, see [Sa63] and [KeR07]. This strategy works well when G is quasi split over K; unfortunately it seems to fail in the 4 Structure constants We come back to the general framework of § 1. We shall compute the structure constants of bHR or HR by formulas depending on A and the numbers qM of 1.5. Note that there are only a finite number of them: as qwM = qM , ∀w ∈ ν(N ) and wM (α, k) = M (wα, k),∀w ∈ W v, we may suppose M = M (αi, k) with i ∈ I and k ∈ Z. Now α∨ i ) = 2 the translation by α∨ i permutes the walls M = M (αi, k) (for k ∈ Z) with two orbits. So Y has at most two orbits in the set of the constants qM (αi,k), those of qi = qM (αi,0) and q′ i = qM (αi,±1). Hence the number of (possibly) different parameters is at most 2.I. We denote by Q = {q1,··· , ql, q′ l = q2l} this set of parameters. i ∈ Q∨ ⊂ Y ; as αi(α∨ 1 = ql+1,··· , q′ 16 Stéphane Gaussent & Guy Rousseau 4.1 Centrifugally folded galleries of chambers x (resp. I − We have twinned buildings I + Let x be a point in the standard apartment A. Let Φx be the set of all roots α such that α(x) ∈ Z. It is a closed subsystem of roots. Its associated Weyl group W v x is a Coxeter group. x ) whose elements are segment germs [x, y) = germx([x, y]) for y ∈ I , y 6= x, y ≥ x (resp. y ≤ x). We consider their unrestricted structure, so the associated Weyl group is W v and the chambers (resp. closed chambers) are the local chambers C = germx(x + C v) (resp. local closed chambers C = germx(x + C v)), where C v is a vectorial chamber, cf. [GR08, 4.5] or [Ro11, § 5]. To A is associated a twin system of apartments Ax = (A− We choose in A− x , A+ x a negative (local) chamber C − x its opposite in A+ x and denote C + x ). x . We f is the system f )). We note (αi)i∈I the corresponding basis of Φ x (i.e. Φ+ = wΦ+ f , if Φ+ consider the system of positive roots Φ+ associated to C + Φ+ defined in 1.1 and C + and (ri)i∈I the corresponding generators of W v. x = germx(x + wC v x , C1, ..., Cr) in the apartment A− Fix a reduced decomposition of an element w ∈ W v, w = ri1 ··· rir and let i = (i1, ..., ir) be the type of the decomposition. We consider now galleries of (local) cham- bers c = (C − x and of type i. The set of all these galleries is in bijection with the set Γ(i) = {1, ri1} × ··· × {1, rir} via the map (c1, ..., cr) 7→ (C − x ). Let βj = −c1 ··· cj(αij ), then βj is the root corresponding to the common limit hyperplane Mj = Mβj of Cj−1 = c1 ··· cj−1C − x and Cj = c1 ··· cjC − x and satisfying to βj(Cj) ≥ βj(x) (actually Mj is a wall ⇐⇒ βj ∈ Φx). In the following, we shall identify a sequence (c1, ..., cr) and the corresponding gallery. x , ..., c1 ··· crC − x starting at C − x , c1C − Definition 4.2. Let Q be a chamber in A+ be centrifugally folded with respect to Q if cj = 1 implies βj ∈ Φx and w−1 wQ = w(C + by Γ+ x . A gallery c = (c1, ..., cr) ∈ Γ(i) is said to Q βj < 0, where x ). We denote this set of centrifugally folded galleries x , Q) ∈ W v (i.e. Q = wQC + Q(i). Proposition 4.3. A gallery c = (C − Cj = Cj−1 implies that Mj = Mβj is a wall and separates Q from Cj = Cj−1. Proof. We saw that Mj is a wall ⇐⇒ βj ∈ Φx. We have the following equivalences: (Mj separates Q from Cj = Cj−1) ⇐⇒ (w−1 (w−1 x , C1, ..., Cr) ∈ Γ(i) belongs to Γ+ Q Mj separates C + Q Cj = w−1 x from w−1 Q βj is a negative root). Q(i) if, and only if, Q Cj−1) ⇐⇒ The group Gx = Gx/GIx acts strongly transitively on I + x . For any root α ∈ Φx with α(x) = k ∈ Z, the group U α = Uα,k/Uα,k+1 is a finite subgroup of Gx of cardinality qx,α = qM (α,−α(x)) ∈ Q. We denote by uα the elements of this group. x and I − Next, let ρQ : Ix → Ax be the retraction centered at Q. To a gallery of chambers x , C1, ..., Cr) in Γ(i), one can associate the set of all galleries of type i x that retract onto c, we denote this set by CQ(c). We denote the set of c = (c1, ..., cr) = (C − x in I − starting at C − minimal galleries in CQ(c) by Cm gj =( cj Q (c). Set Q βj > 0 or βj 6∈ Φx Q βj < 0 and βj ∈ Φx. Proposition 4.4. CQ(c) is the non empty set of all galleries (C − ∀j : C ′ C ′ j are in the apartment g1 ··· gj Ax. j = g1 ··· gjC − ucj (αij )cj if w−1 if w−1 r) where x with each gj chosen as in (1) above. For all j the local chambers Q and x = C ′ 1, ..., C ′ 0, C ′ (1) Spherical Hecke algebras for Kac-Moody groups over local fields 17 0, C ′ 1, ..., C ′ x = C ′ Q (c) is empty if, and only if, the gallery c is not centrifugally folded with respect r) is minimal if, and only if, cj 6= 1 for any j with The set Cm to Q. The gallery (C − Q βj > 0 or βj 6∈ Φx and ucj(αij ) 6= 1 for any j with cj = 1 and w−1 w−1 Remark. For gj as in equation (1) we may write gj = ucj(αij )cj (with ucj(αij ) = 1 if w−1 Q βj > 0 or βj 6∈ Φx). Then in the product g1 ··· gj we may gather the ck on the right and, as c1 ··· ck(αik ) = −βk, we may write g1 ··· gj = u−β1 ··· u−βj .c1 ··· cj. Hence C ′ j := g1 ··· gjC − Q βk < 0; so it is clear that ρQ(C ′ x = u−β1 ··· u−βj Cj. When u−βk 6= 1 we have βk ∈ Φx and w−1 j) = Cj. Q βj < 0. j = u−αi1 The gallery (C − 0, C ′ .uri1 (−αi2 ) ··· uri1 ···rij−1 (−αij ).ri1 ··· rij (C − r) (of type i) is minimal if, and only if, we may also write x ) = h1 ··· hj.ri1 ··· rij (C − x ) j ∈ h1 ··· hj Ax. j) = Cj. We know only that, when βk 6∈ Φx (uniquely) C ′ with hk = uri1 ···rik−1 (−αik ) ∈ U ri1 ···rik−1 (−αik ) (which fixes C − But this formula gives no way to know when ρQ(C ′ i.e. ri1 ··· rik−1(−αik ) 6∈ Φx, we have necessarily hk = 1. Proof. As the type i of (C − r) is the type of a minimal decomposition, this gallery is minimal if, and only if, two consecutive chambers are different. So the last assertion is a consequence of the first ones. We prove these properties for (C − j ) by induction on j. We write in the following just Hj for the common limit hyperplane Hβj of Cj−1 and Cj of type ij. x ). In particular, C ′ x = C ′ x = C ′ 1, ..., C ′ 1, ..., C ′ 1, ..., C ′ x = C ′ 0, C ′ 0, C ′ There are five possible relative positions of Q, C − x and C1 with respect to H1 and we seek C ′ 1) = C1 and C ′ 1 with ρQ(C ′ 0) β1 = −c1αi1 6∈ Φx, then H1 is not a wall, each C ′ x ∩ H1. 1 ⊃ C − 1 or C − x and C ′ or ri1C − Q are in g1Ax = Ax with g1 = c1. When C ′ folded. 1 = C − x are contained in the same apartments. So C ′ 1 with C ′ x ∩ H1 is equal to C − x 1 = C1 = c1C − x ; C1 and x , we have c1 = 1 and c is not centrifugally 1 ⊃ C − We suppose now β1 ∈ Φx, so H1 is a wall. 1) C − 1 = g1C − x is on the same side of H1 as Q and C1 not, then c1 = ri1, β1 = αi1, w−1 C ′ x = u−αi1 by H1 containing C − Q β1 < 0, pointwise stabilizes the halfspace bounded 1 are in the apartment g1Ax. C1. But u−αi1 (Q) = Q and C ′ ri1C − x = u−αi1 x , hence u−αi1 x = C1 are separated by H1, then c1 = 1, β1 = −αi1, w−1 pointwise stabilizes the halfspace bounded by H1 not containing C − Q β1 < 0, C ′ 1 = g1C − x = x , hence 2) Q and C − C − x but uαi1 uαi1 Q and C ′ C ′ apartment g1Ax. 4) Q and C − 1 are in the apartment g1Ax. 3) C1 is on the same side of H1 as Q and C − x not, then c1 = ri1, β1 = αi1, w−1 1 has to be C1 so g1 = c1 = ri1, w−1 Q (αi1) > 0, moreover Q and C ′ 1 = ri1C − Q β1 > 0 and x = C1 are in the Q β1 > 0; the gallery c x with g1 = c1 = 1 as in x = C1 are on the same side of H1. Then c1 = 1 and w−1 x = g1C − is not centrifugally folded. So ρQ(C ′ (1). But the gallery (C − 1 = C − j ) cannot be minimal. By induction we assume now that the chambers Q and C ′ 1) = C1 implies C ′ 1, ..., C ′ x = C ′ 0, C ′ x are in the apartment Aj−1 = g1 ··· gj−1Ax. Again, we have five possible relative positions for Q, Cj−1 and Cj with respect to Hj. We seek C ′ . j−1∩ g1 ··· gj−1Hαij j−1 are contained in x and Q are in g1 ··· gj Ax = g1 ··· gj−1Ax with 0) βj = −c1 ··· cjαij 6∈ Φx, then Hj is not a wall, each C ′ is equal to C ′ the same apartments. So C ′ gj = cj. When C ′ j−1, we have cj = 1 and c is not centrifugally folded. j with C ′ x ; moreover C ′ x or g1 ··· gj−1rij C − j−1 = g1 ··· gj−1C − j = g1 ··· gj−1cjC − j−1 ∩ g1 ··· gj−1Hαij j−1 = g1 ··· gj−1C − j ⊃ C ′ j or C ′ j) = Cj and C ′ j with ρQ(C ′ j ⊃ C ′ j = C ′ 18 Stéphane Gaussent & Guy Rousseau We suppose now βj ∈ Φx, so Hj is a wall. 1) Cj−1 is on the same side of Hj = c1 ··· cj−1Hαij Q βj < 0. Moreover Q and C ′ βj = c1 ··· cj−1αij , w−1 in Aj−1, and as Q and Cj not, then cj = rij , j−1 are on the same side of g1 ··· gj−1Hαij C ′ j = g1 ··· gj−1u−αij = g1 ··· gj−1u−αij = g1 ··· gj−1u−αij rij C − x rij (g1 ··· gj−1)−1C ′ (g1 ··· gj−1)−1g1 ··· gj−1rij (g1 ··· gj−1)−1C ′ j−1 j−1, where g1 ··· gj−1rij (g1 ··· gj−1)−1C ′ Aj−1. Moreover, g1 ··· gj−1u−αij g1 ··· gj−1Hαij containing C ′ j−1 is the chamber adjacent to C ′ in (g1 ··· gj−1)−1 pointwise stabilizes the halfspace bounded by j along g1 ··· gj−1Hαij j are in the apartment g1 ··· gj Ax. 2) Cj−1 = Cj and Q are separated by Hj, then cj = 1, βj = −c1 ··· cj−1αij , w−1 j−1 and Q. So Q and C ′ Q βj < 0. in Aj−1, and Q and the chamber Moreover C ′ j−1 and Q are separated by g1 ··· gj−1Hαij g1 ··· gj−1rij (g1 ··· gj−1)−1C ′ j−1 are on the same side of this wall. For uαij 6= 1 C ′ j = g1 ··· gj−1uαij C − x = g1 ··· gj−1uαij (g1 ··· gj−1)−1C ′ j−1 is a chamber adjacent (or equal) to C ′ g1 ··· gj Ax (with gj = uαij ). j−1 along g1 ··· gj−1Hαij = g1 ··· gj−1uαij Hαij in The root-subgroup g1 ··· gj−1Uαij and containing the chamber g1 ··· gj−1rij (g1 ··· gj−1)−1C ′ (g1 ··· gj−1)−1 pointwise stabilizes the halfspace bounded j−1. So Q and C ′ j by g1 ··· gj−1Hαij are in the apartment g1 ··· gj Ax. 3) Cj is on the same side of Hj = c1 ··· cj−1Hαij Q βj > 0. and so C ′ j = g1 ··· gj−1rij C − as Q and Cj−1 not, then cj = rij , j are in the x . Whence Q and C ′ βj = c1 ··· cj−1αij , w−1 apartment g1 ··· gj Ax. 4) Cj−1 = Cj and Q are on the same side of Hj = c1 ··· cj−1Hαij Q βj > 0. The gallery c is not centrifugally folded. So ρQ(C ′ x with gj = cj = 1 as in (1). But the gallery (C − , then cj = 1, βj = j) = Cj x = −c1 ··· cj−1αij and w−1 implies C ′ 0, C ′ C ′ j−1 = g1 ··· gjC − j ) cannot be minimal. j = C ′ 1, ..., C ′ Corollary 4.5. If c ∈ Γ+ Q(i), then the number of elements in Cm Q (c) is: ♯Cm Q (c) = t(c)Yk=1 qjk × r(c)Yl=1 (qjl − 1) where qj = qx,βj = qx,αij ∈ Q, t(c) = ♯{j cj = rij , βj ∈ Φx and w−1 r(c) = ♯{j cj = 1, βj ∈ Φx and w−1 Q βj < 0}. Q βj < 0} and Spherical Hecke algebras for Kac-Moody groups over local fields 19 4.6 Galleries and opposite segment germs Suppose now x ∈ A∩ I +. Let ξ and η be two segment germs in A+ respectively η and ξ in A− where C−ξ is the negative (local) chamber containing −ξ such that w(C − length. Let Q be a chamber of A+ x . Let i be the type of a minimal gallery between C − x . Let −η and −ξ opposite x and C−ξ, x , C−ξ) is of minimal x containing η. We suppose ξ and η conjugated by W v x . Lemma. The following conditions are equivalent: x ). (ζ) = −ξ. x x such that ρAx,C− Q(i) ending in −η. Q (c) for c in the set Γ+ Q(i,−η) of galleries in Γ+ x η (in the sense of 1.8, with Φ+ defined as in 4.1 using C − (i) There exists an opposite ζ to η in I − (ii) There exists a gallery c ∈ Γ+ (iii) ξ ≤ W v Moreover the possible ζ are in one-to-one correspondence with the disjoint union of the Q(i) ending in −η. More precisely, if sets Cm m ∈ CQ(c) is associated to (h1,··· , hr) as in remark 4.4, then ζ = h1 ··· hr(−ξ). Proof. If ζ ∈ I − (ζ) = −ξ, then any minimal gallery m = (C − x and C−ξ. So we can as well assume that m has type i = (i1, ..., ir) and then ζ determines m. Now, if we retract m from Q, we get a gallery c = ρAx,Q(m) in A− x , ending in −η and centrifugally folded with respect to Q. Q(i), such that −η ∈ Cr. According to proposition r) in the set CQ(c), and the x where each hk fixes x , M1, ..., Mr ∋ ζ) retracts onto a minimal gallery between C − and remark 4.4, there exists a minimal gallery m = (C − chambers C ′ C − j can be described by C ′ x = h1 ··· hj.ri1 ··· rij C − x opposites η and if ρAx,C− x , C1, ..., Cr) ∈ Γ+ Reciprocally, let c = (C − j = g1 ··· gjC − x starting at C − x , C ′ 1, ..., C ′ x x restricts on C ′ x , hence ρAx,C− Let ζ ∈ C ′ gallery m = (C − both opposite η up to conjugation by W v x . 1, ..., C ′ r opposite η in any apartment containing those two. The minimality of the x , C ′ (ζ) = −ξ as they are (ζ) ∈ C−ξ; hence ρAx,C− r) ensures that ρAx,C− x x j to the action of (h1 ··· hj)−1. So we proved the equivalence (i) ⇐⇒ (ii) and the last two assertions. Now the equivalence (i) ⇐⇒ (iii) is proved in [GR08, Prop. 6.1 and Th. 6.3]: this reference we speak of Hecke paths with respect to −C v discussion in Ix (using only C − equivalence. x and the twin building structure of I ± in f , but the essential part is a local x ) which gives this 4.7 Liftings of Hecke paths Let π be a λ−path from z′ ∈ Y + to y ∈ Y + entirely contained in the Tits cone T , hence f with w ∈ W v. By [GR08, 5.2.1], for each w ∈ W v in a finite union of closed sectors wC v there is only a finite number of s ∈]0, 1] such that the reverse path ¯π(t) = π(1 − t) leaves, in π(s), a wall positively with respect to −wC v f , i.e. this wall separates π−(s) from −wC v f . Therefore, we are able to define ℓ ∈ N and 0 < t1 < t2 < ··· < tℓ ≤ 1 such that the zk = π(tk), k ∈ {1, ..., ℓ} are the only points in the path where at least one wall containing zk separates π−(tk) and the local chamber c− of 1.8.2. zk (as in 4.1) the germ in zk of the sector of vertex zk containing c−. Let ik be a fixed reduced decomposition of the element w−(tk) ∈ W v and let Qk be a fixed chamber in I + zk containing ηk = π+(tk). We note −ξk = π−(tk). When π is a Hecke path (or a billiard path as in [GR08]), ξk and ηk are conjugated by W v zk. For each k ∈ {1, ..., ℓ} we choose for C − 20 Stéphane Gaussent & Guy Rousseau When π is a Hecke path with respect to c−, {z1,··· , zℓ} includes all points where the (ik,−ηk) are piecewise linear path π is folded and, in the other points, all galleries in Γ+ Qk unfolded. Let Sc−(π, y) be the set of all segments [z, y] such that ρc−([z, y]) = π. Theorem 4.8. Sc−(π, y) is non empty if, and only if, π is a Hecke path with respect to c−. Then, we have a bijection Sc−(π, y) ≃ ℓYk=1 ac∈Γ+ Qk (ik,−ηk) Cm Qk (c) In particular the number of elements in this set is a polynomial in the numbers q ∈ Q with coefficients in Z≥0 depending only on A. N.B. So the image by ρc− of a segment in I + is a Hecke path with respect to c−. Proof. The restriction of ρc− to Izk is clearly equal to ρAzk ,C− is a Hecke path with respect to c− if, and only if, each Γ+ Qk (ik,−ηk) is non empty. zk ; so the lemma 4.6 tells that π 1...hℓ 1 Qk Qk (ik,−ηk) Cm We set t0 = 0 and tℓ+1 = 1. We shall build a bijection from Sc−(π[tn−1,1], y) onto (c) by decreasing induction on n ∈ {1,··· , ℓ + 1}. For n = ℓ + 1 and if tℓ 6= 1, no wall cutting π([tℓ, 1]) separates y = π(1) from c−; so a segment s in I with s(1) = y and ρc− ◦ s = π has to coincide with π on [tℓ, 1]. Suppose now that s ∈ Sc−(π[tn,1], y) is determined, in the following way, by a unique (c): For an element (mn+1, mn+2, ..., mℓ) in this last set, rk ) ∈ rn+1 )π(t) j is a chosen element of U−ri1 ···rij−1 (αij ) whose class in U −ri1 ···rij−1 (αij ) Qℓ k=n`c∈Γ+ element inQℓ each mk = (C − (Gzk )rk, as in remark 4.4 and, for t ∈ [tn, tn+1], we have s(t) = (hℓ where actually each hk is the hk (ik,−ηk) Cm rk ) is a minimal gallery given by a sequence of elements (hk rℓ )··· (hn+1 k=n+1`c∈Γ+ j defined above; in particular each hk 1, ..., hk ...hn+1 Qk 1 , ..., C k j fixes c−. zk , C k 1...hℓ Qk 1 ...hn+1 rℓ )··· (hn+1 (in,−ηn). rn+1 ) ∈ Gc−. Then g−1s(tn) = π(tn) = zn. We note g = (hℓ If s ∈ Sc−(π[tn−1,1], y) and s[tn,1] is as above, then g−1s−(tn) is a segment germ in I − zn opposite g−1s+(tn) = π+(tn) = ηn and retracting to π−(tn) by ρc−. By lemma 4.6 and the above remark, this segment germ determines uniquely a minimal gallery mn ∈ Cm (c) with c ∈ Γ+ Conversely such a minimal gallery mn determines a segment germ ζ ∈ I − zn, opposite rn)π−(tn) for some π+(tn) = ηn such that ρAzn ,C− well defined (hn j by a chosen element of G(zn∪c−) whose class in Gzn is this gn j . As no wall cutting [zn−1, zn] separates zn = π(tn) from c−, any segment retracting by ρc− onto [zn−1, zn] and with [zn, x) = π−(tn) (resp. rn)[zn−1, zn]). We set = ζ, = gζ) is equal to [zn−1, zn] (resp. 1 ...hn s(t) = (hℓ (ζ) = π−(tn). By lemma 4.6, ζ = (hn rn) ∈ (Gzn)rn. As above we replace each gn rn )[zn−1, zn], g(hn 1 ,··· , hn rn+1 )(hn With this inductive definition, s is a λ−path, s(1) = y, ρc− ◦ s = π and s[tk−1,tk] is a segment ∀k ∈ {1, ..., ℓ + 1}. Moreover, for k ∈ {1, ..., ℓ}, the segment germs [s(tk), s(tk+1)) and [s(tk), s(tk−1)) are opposite. By the following lemma this proves that s itself is a segment. Lemma 4.9. Let x, y, z be three points in an ordered hovel I , with x ≤ y ≤ z and suppose the segment germs [y, z) , [y, x) opposite in the twin buildings Iy. Then [x, y] ∪ [y, z] is the segment [x, z]. rn )π(t) for t ∈ [tn−1, tn]. rℓ )··· (hn+1 ...hn+1 1 ...hn 1 ...hn 1...hℓ (hn 1 ...hn Qn Qn zn 1 Spherical Hecke algebras for Kac-Moody groups over local fields 21 Proof. For any u ∈ [y, z], we have x ≤ y ≤ u ≤ z, hence x and [u, y) or [u, z) are in a same apartment [Ro11, 5.1]. As [y, z] is compact we deduce that there are points u0 = y, u1,··· , uℓ = z such that x and [ui−1, ui] are in a same apartment Ai, for 1 ≤ i ≤ ℓ. Now A1 contains x and [y, u1], hence also [x, y] (axiom (MAO) of 1.5). But [y, x) and [y, u1) = [y, z) are opposite, so [x, y] ∪ [y, u1] = [x, u1]. The lemma follows by induction. Remark 4.10. The same things as above may be done for the retraction ρ−∞ instead of ρc−: f ). For a λ−path π in A from z′ to y, [GR08, 5.2.1] for all x we choose C − tells that we have a finite number of points zk = π(tk) where at least a wall is left positively by the path ¯π(t) = π(1 − t). We define as above ik, Qk, ηk and ξk. Now S−∞(π, y) is the set of all segments [z, y] such that ρ−∞([z, y]) = π. In [GR08, Theorems 6.2 and 6.3], we have proven that S−∞(π, y) is nonempty if, and only if, π is a Hecke path with respect to −C v f . Moreover, we have shown that, for I associated to a split Kac-Moody group over C((t)), S−∞(π, y) is isomorphic to a quasi-affine toric complex variety. The arguments above prove that, with our choice for I , S−∞(π, y) is finite, with the following precision (which generalizes to the Kac-Moody case some formulae of [GL11]): x = germx(x − C v Proposition 4.11. Let π be a Hecke path with respect to −C v bijection: f from z′ to y. Then we have a S−∞(π, y) ≃ ℓYk=1 ac∈Γ+ Qk (ik,−ηk) Cm Qk (c) In particular the number of elements in this set is a polynomial in the numbers q ∈ Q with coefficients in Z≥0 depending only on A. a) The number of Hecke paths of shape µ with respect to c− starting in z′ = wλ (for some Theorem 4.12. Let λ, µ, ν ∈ Y ++, c− the negative fundamental alcove and suppose (α∨ R+−free. Then w ∈ W v fixing 0) and ending in y = ν is finite. λ and dv(z, ν) = µ is equal to: b) The structure constant mλ,µ(ν) i.e. the number of triangles [0, z, ν] in I with dv(0, z) = i )i∈I mλ,µ(ν) = Xw∈W v/(W v)λXπ ℓπYk=1 Xc∈Γ+ Qk (ik,−ηk) ♯Cm Qk (c) (2) Qk (c) are defined as above for each such π. where π runs over the set of Hecke paths of shape µ with respect to c− from wλ to ν and ℓπ, Γ+ Qk (ik,−ηk) and Cm c) In particular the structure constants of the Hecke algebra HR are polynomials in the numbers q ∈ Q with coefficients in Z≥0 depending only on A. Proof. We saw in 2.3.1 that mλ,µ(ν) is the number of z ∈ I + such that dv(0, z) = λ and dv(z, ν) = µ. Such a z determines uniquely a Hecke path π = ρc−([z, ν]) of shape µ with respect to c− from z′ = ρc−(z) to ν. But dv(0, z) = λ and 0 ∈ c−, so dv(0, z′) = λ i.e. z′ = wλ with w ∈ W v. So the formula (2) follows from theorem 4.8. We know already that mλ,µ(ν) is finite (2.5) and Sc−(π, y) 6= ∅ (theorem 4.8), hence a) is clear. Now c) follows from corollary 4.5 0 22 Stéphane Gaussent & Guy Rousseau 5 Satake isomorphism In this section, we prove the Satake isomorphism. From now on, we assume that the α∨ free. i 's are We denote by U − the fixator in G of the sector germ S−∞, i.e. any u ∈ U − has to fix f ⊂ A. By definition, for z ∈ I , ρ−∞(z) is the only point of the orbit pointwise a sector x− C v U −.z in A. 5.1 The module of functions on the type 0 vertices in A Let A0 = ν(N ) · 0 = Y · 0 be the set of vertices of type 0 in A. Note that A0 can be identified to the set of horocycles of U − in I0, i.e. to I0/U −, via the retraction ρ−∞. We consider first cF = cFR = F (A0, R), the set of functions on A0 with values in a ring R. Equivalently, cF can be identified with the set of U −−invariant functions on I0. For µ ∈ Y , we define χµ ∈ cF as the characteristic function of U −.µ in I0 (or {µ} in Y ). Then, any χ ∈ cFR may be written χ =Pµ∈Y aµχµ with aµ ∈ R. We set supp(χ) = {µ aµ 6= 0}. Now, let F = FR = {χ ∈ cF supp(χ) ⊂ ∪n be the set of functions on I0 with almost finite support. j=1(µj − Q∨ +) for some µj ∈ A0} We define also the following completion of the group algebra R[Y ]: R[[Y ]] = {f =Xy∈Y ayey supp(f ) = {y ∈ Y ay 6= 0} ⊂ ∪n j=1(µj − Q∨ +) for some µj ∈ A0} it is clearly a commutative algebra (with ey.ez = ey+z). Actually, it is the Looijenga's coweight algebra, see Section 4.1 in [Loo]. The formula (f.χ)(µ) =Py∈Y ayχ(µ − y), for f =P ayey ∈ R[[Y ]], χ ∈ F and µ ∈ Y , defines an element f.χ ∈ F ; in particular ey.χµ = χµ+y. Clearly, the map R[[Y ]] × F → F , (f, χ) 7→ f.χ makes F into a free R[[Y ]]−module of rank 1, with any χµ as basis element. Definition-Proposition 5.2. The map F × H → F (χ, ϕ) 7→ χ ∗ ϕ, where, for x ∈ I0, (χ ∗ ϕ)(x) =Py∈I0 commutes with the actions of Z = {n ∈ N ν(n) ∈ Y } and (more generally) R[[Y ]]. Proof. It is relatively clear that χ∗ ϕ is a function on I0/U − and that the map indeed defines an action. Let us check that this action commutes with the one of Z. Let t ∈ Z and x ∈ I0, then χ(y)ϕI (y, x), defines a right action of H on F that χ(y)ϕI (y, tx) (χ ∗ ϕ)(tx) = Py∈I0 = Py′∈I0 χ(ty′)ϕI (ty′, tx) = Py′∈I0 = ((χ ◦ t) ∗ ϕ)(x). χ(ty′)ϕI (y′, x) (y = ty′) So, (χ ◦ t) ∗ ϕ = (χ ∗ ϕ) ◦ t. For ν(t) = µ ∈ Y and χ ∈ F , we have clearly χ ◦ t = e−µ.χ. As a formal consequence, the right action of H commutes with the left action of R[[Y ]]. Spherical Hecke algebras for Kac-Moody groups over local fields 23 The difficult point is to show that the support condition is satisfied. For any λ ∈ Y ++, and any ν ∈ Y , (χµ ∗ cλ)(ν) = Py∈I0 χµ(y)cI λ (y, ν) = ♯{y ∈ I0 ρ−∞(y) = µ and dv(y, ν) = λ} The latest is also the cardinality of the set of all segments [y, ν] in I (y ≤ ν) of "length" λ such that y ∈ U − · µ. In addition, since the action of H commutes with the one of Z, we set nλ(ν − µ) = (χµ ∗ cλ)(ν). Then nλ(ν − µ) =Pπ ♯S−∞(π, ν) where the sum runs over the set of Hecke λ−paths with respect to −C v f from µ to ν (see 4.10 for the definition of S−∞(π, ν)). Now, Lemma 2.4 b) shows that nλ(ν − µ) 6= 0 implies ν− µ ≤Q+ λ. Moreover, if ν = λ + µ, then nλ(λ) = 1. Therefore, we get χµ ∗ cλ = Xν≤Q∨ λ+µ nλ(ν − µ)χν = χλ+µ + Xν<Q∨ λ+µ nλ(ν − µ)χν. (3) This formula shows that, for any ϕ ∈ H with supp(ϕ) ⊂ ∪n supp(χ) ⊂ ∪n precisely, for any ν ∈ ∪i,j(λi + µj − Q∨ µ ∈ supp(χ) such that ν ≤Q+ λ + µ. Hence, χ ∗ ϕ is well defined. +) and any ξ ∈ F with +). More +) there exists a finite number of λ ∈ supp(ϕ) and +), the support of χ ∗ ϕ is contained in ∪i,j(λi + µj − Q∨ j=1(µj − Q∨ i=1(λi − Q∨ 5.3 The Satake isomorphism 5.3.1 The morphism S∗ As F is a free R[[Y ]]−module of rank one, we have EndR[[Y ]](F ) = R[[Y ]]. So the right action of H on the R[[Y ]]−module F gives an algebra homomorphism S∗ : H → R[[Y ]] such that χ ∗ ϕ = S∗(ϕ).χ for any ϕ ∈ H and any χ ∈ F . As eν.χµ = χµ+ν, equation (3) gives S∗(cλ) = Xν≤Q∨ λ nλ(ν)eν = eλ + Xν<Q∨ λ nλ(ν)eν We shall modify S∗ by some character to get the Satake isomorphism. i 5.3.2 The module δ We define a map δ : Q∨ → R∗ the beginning of Section 4. We extend this homomorphism and its square root to Y (as R∗ uniquely divisible). So, we get homomorphisms δ, δ1/2 : Y → R∗ Z → R∗ +. depend only on δ Q∨. i ∈ Q ⊂ N are as in + is + and δ = δ◦ ν, δ1/2 = δ1/2◦ ν : We made a choice for δ. But we shall see in theorem 5.4 that the expected properties + ,Pi∈I aiα∨ 7→Qi∈I (qiq′ i)ai, where qi, q′ In the classical case, where G is a split semi-simple group and I its Bruhat-Tits building, i , δ1/2(µ) = qP ai = qρ(µ) we have qi = q′ where ρ is the half sum of positive roots. i = q for any i ∈ I. Hence, if we set µ =Pi∈I aiα∨ 24 Stéphane Gaussent & Guy Rousseau 5.3.3 The Satake isomorphism From now on, we suppose that the algebra R contains the image of δ1/2 in R∗ +. We define S(cλ) = Xµ≤Q∨ λ δ1/2(µ)nλ(µ)eµ = δ1/2(λ)eλ + Xµ<Q∨ λ δ1/2(µ)nλ(µ)eµ and extend it to formal combinations of the cλ with almost finite support. it is one to one: We get thus an algebra homomorphism S : H → R[[Y ]] called the Satake isomorphism, as For ϕ = Pλ aλcλ ∈ H, we have S(ϕ) = Pλ aλ(cid:0)δ1/2(λ)eλ +Pµ<Q∨ λ δ1/2(µ)nλ(µ)eµ(cid:1). If ϕ 6= 0 and λ0 is a maximum element in supp(ϕ), then λ0 is also a maximum element in supp(S(ϕ)) and S(ϕ) 6= 0. Remarks. a) So we already know that H is commutative. b) In the classical case where G is a split semi-simple group, S(cλ) is defined as an integral over a maximal unipotent subgroup, we choose here U −. The Haar measure du on U − is chosen to give volume 1 to K ∩ U −, and, for an element t in the torus Z, the formula for changing variables is given by d(tut−1) = δ(t)−1du. So the classical formula for the Satake isomorphism given e.g. in [Ca79, (19) p 146] when ν(t) = µ, is: S(cλ)(t) = δ(t)1/2RU − cG = δ(t)1/2RU − cI = δ(t)1/2Py∈I0 This is the same formula as ours. λ (ut)du = δ(t)1/2RU − cI λ (u−1.0, t.0)du = δ(t)1/2Py∈U −.0 cI λ (y, µ) = δ(t)1/2(χ0 ∗ cλ)(µ) χ0(y).cI λ (0, ut.0)du λ (y, µ) 5.3.4 W v−invariance There is an action of W v on Y , hence on R[Y ] by setting w.eλ = ewλ for w ∈ W v and λ ∈ Y . = {f =P aλeλ ∈ R[[Y ]] aλ = This action does not extend to R[[Y ]], but we define R[[Y ]]W v awλ,∀λ ∈ Y,∀w ∈ W v}. This is a subalgebra of R[[Y ]] and actually the image of the Satake isomorphism (see Theorem 5.4). Remark. Let C ∨ = {π ∈ V ∗ α∨ i (π) ≥ 0,∀i ∈ I} and T ∨ = ∪w∈W v wC ∨ be the fundamental dual chamber and the dual Tits cone in V ∗. By definition, for f ∈ R[[Y ]] and π ∈ C ∨, π(supp(f )) is bounded above. Hence, for f ∈ R[[Y ]]W v , π(supp(f )) is also bounded above ∨ is the closed convex hull Γ of the set for any π ∈ T ∨. We know that the dual cone of T ∆∨im + ∪{0}, where ∆∨im + is the set of positive imaginary roots in the dual system of roots ∆∨, [Ka90, 5.8]. So, the only directions along which points in supp(f ) (for f ∈ R[[Y ]]W v) may go to infinity are the directions in −Γ. Theorem 5.4. The Hecke algebra HR is isomorphic via S to the commutative algebra R[[Y ]]W v of Weyl invariant elements in R[[Y ]]. + ⊂ Q∨ Proof. As S(cλ) = Pµ≤Q∨ λ δ1/2(µ)nλ(µ)eµ we only have to prove that, for w ∈ W v, δ1/2(µ)nλ(µ) = δ1/2(wµ)nλ(wµ) or nλ(wµ) = nλ(µ)δ1/2(µ − wµ). It is sufficient to prove this for w = ri a fundamental reflection, hence to prove that nλ(riµ) = nλ(µ)δ1/2(µ − riµ) = nλ(µ)δ1/2(αi(µ)α∨ i ). By the given definition of δ, the wanted formula is: Spherical Hecke algebras for Kac-Moody groups over local fields nλ(riµ) = nλ(µ)(cid:16)qqiq′ i(cid:17)αi(µ) 25 (4) j=1 (λj − Q∨ j=1 (λj − Q∨ +), we shall build a sequence ϕn in H such that +n)), where +n) and supp(ϕn+1 − ϕn) ⊂ Y ++ ∩ (∪r i. So, in any case(cid:0)pqiq′ The proof of this formula is postponed to the following subsections, starting with 5.5. One can already notice that αi(µ) is an integer. If it is odd, since any t ∈ Z with ν(t) = µ exchanges the walls M (αi, 0) and M (αi, αi(µ)), hence qi = q′ i(cid:1)αi(µ) is an integer. Once the formula (4) is proved we know that S(H) ⊂ R[[Y ]]W v . For f = P aµeµ ∈ R[[Y ]]W v with supp(f ) ⊂ ∪r j=1 (λj − Q∨ supp(f − S(ϕn)) ⊂ ∪r j=1 (λj − Q∨ + P ni ≥ n}. Then, the limit ϕ of this sequence exists in H and +n = {Pi∈I niα∨ Q∨ i ∈ Q∨ S(ϕ) = f . So, S is onto. We build the sequence by induction. We set ϕ0 = 0. If ϕ0,··· , ϕn are given as above, we set {µ1,··· , µs} = supp(f − S(ϕn)) \ ∪r +(n+1)). For any w ∈ W v, wµk ∈ j=1 (λj − Q∨ supp(f − S(ϕn)) ⊂ ∪r +n), so wµk cannot be strictly greater than µk for ≤Q∨; this proves that µk ∈ Y ++. So we define ϕn+1 = ϕn −Ps k=1 aµk (f − S(ϕn))δ(µk)−1/2cµk. As S(cλ) = δ1/2(λ)eλ +Pµ<Q∨ λ δ1/2(µ)nλ(µ)eµ, this ϕn+1 is suitable. Remark. Suppose G is a split Kac-Moody group as in Section 3. And consider the complex Kac-Moody algebra g∨ associated with G∨, the Langlands dual of G. Let h∨ = C ⊗Z Y be the Cartan subalgebra of g∨. Let Rep(g∨) be the category of g∨−modules V such that V is h∨−diagonalizable, the weight spaces Vλ are finite dimensional and the set P(V ) of weights of V satisfies P(V ) ⊂ ∪r +), for some λj. One can check that Rep(g∨) is stable by tensoring, hence, we can consider its Grothendieck ring K(g∨). Now, the map [V ] 7→Pλ(dim Vλ)eλ is an isomorphism from K(g∨) onto C[[Y ]]W v. Therefore, by composing it with S, we get an isomorphism between HC and K(g∨). 5.5 Extended tree associated to (A, αi) We consider the vectorial panel −F v({i}) in −C v f and its support the vectorial wall Ker(αi). Their respective directions are a panel F∞ in a wall M∞, in the twin buildings I ±∞ at infinite of I [Ro11, 3.3, 3.4, 3.7]. j=1 (λj − Q∨ The germs of the sector panels in I of direction F∞ are the points of an (essential) affine building I (F∞), which is of rank 1 i.e. a tree [Ro11, 4.6]. The union I (M∞) of the apartments in I containing a wall of direction M∞ is an inessential affine building whose essential quotient is I (F∞) [Ro11, 4.9]. More precisely I (M∞) may be identified with the product of the tree I (F∞) and an affine space quotient of A. The canonical apartment of I (M∞) is A endowed with a smaller set of walls: uniquely the walls of direction Ker(αi). As we chose I semi-discrete (1.2), this is a locally finite set of hyperplanes; hence I (M∞) is discrete and I (F∞) a discrete tree (not an R−tree). By [Ro11, 2.9] the valences of these walls are the same in I (M∞) and in I , i.e. 1 + qi and 1 + q′ i; hence I (F∞) is a semi-homogeneous tree of valences 1 + qi and 1 + q′ i. By definition, 0 ∈ A is in a wall of valence 1 + qi. We asked that the stabilizer N of A in G is positive and type preserving (1.5) i.e. acts on V = −→A via W v. So, the stabilizer in W v of M∞ is {1, ri} and M∞ determines ∞ = Ker(1 + ri). The ∞ with associated in V a supplementary vectorial subspace of dimension one : M ⊥ affine space A decomposes as the product of the affine space E = A/M ⊥ 26 Stéphane Gaussent & Guy Rousseau vector space Ker(αi) and an affine line (= A/Ker(αi)). This decomposition is canonical i.e. invariant by the stabilizer N (M∞) of M∞ in N. As a consequence we get the decomposition I (M∞) = E × I (F∞) which is canonical i.e. invariant by the stabilizer G(M∞) of M∞ in G. Moreover G(M∞) acts on E by translations only. Remark. Suppose G is an almost split Kac-Moody group over a local field K and I its associated hovel as in [Ro13]. Then the stabilizer G(F∞) of F∞ in G is a parabolic subgroup, endowed with a Levi decomposition G(F∞) = G(M∞) ⋉ U (F∞) (with U (F∞) ⊂ U −) and I (M∞) (resp. I (F∞)) is the extended (resp. essential) Bruhat-Tits building associated to the reductive group of rank one G(M∞), embedded in I [Ro13, 6.12.2]. Any orbit of U (F∞) in I meets I (M∞) in one and only one point. The tree I (F∞) is a piece of the polyhedral "compactification" of I (a true compactifica- tion when G is reductive). With the notation of [Ro13], I (M∞) (resp. I (F∞)) is the façade I (G,K, A)F∞ (resp. I (G,K, A )F∞). e 5.6 Parabolic retraction Let x be a point in I . There is a unique sector-panel x + F∞ of vertex x and direction F∞ [Ro11, 4.7.1]. The germ of this sector-panel is a point in I (F∞), the projection prF∞(x) of x onto I (F∞), cf. [Ch10], [Ch11] or [Ro13, 4.3.5] in the Kac-Moody case. Let Ax be an apartment in I containing x and F∞, hence x + F∞ and germ∞(x + F∞). But this germ is in an apartment Bx of I (M∞) (axiom (MA3) applied to germ∞(x+F∞) and a sector of direction C v f ) and there exists an isomorphism ψx of Ax onto Bx fixing this germ (axiom (MA2)). One writes ρ(x) = ψx(x) ∈ I (M∞). We have thus defined the retraction ρ = ρF∞,M∞ of I onto I (M∞) with center F∞. We shall now verify that ρ(x) does not depend on the choices made. x : A′ x By definition, ρ(x) is in the hyperplane Hx of Bx containing germ∞(x + F∞) and of direction M∞, this Hx does not depend on the choice of Bx. Moreover for two choices ψx : Ax → Bx and ψ′ is the identity on germ∞(x + F∞) hence on Hx. It is now clear that ψx(x) = ψ′ x(x). Actually ρ(x) may also be defined in the following simple way: there exist y, z ∈ (x + F∞) ∩ Bx such that y is the middle of [x, z] in Ax, then ρ(x) is the point of Hx ⊂ Bx such that y is the middle of [ρ(x), z] in Bx. Remark. It is possible to prove that the image by ρ of a preordered segment is a polygonal line and, in some generalized sense, a Hecke path. x → Bx, ψ′ x ◦ ψ−1 5.7 Factorization of ρ−∞ The panel F∞ is in the closure of the chamber C−∞ of I −∞ associated to −C v f . So this chamber or the associated sector-germ S−∞ determines an end of the tree I (F∞) [Ro11, 4.6] i.e. a sector-germ S′ in I (M∞): S′ is one of the two sector-germs in A (considered as an apartment of I (M∞) with its small set of walls), each element in S′ contains an half apartment of equation αi(y) ≤ k with k ∈ Z. We write ρ′ −∞ the retraction of I (M∞) onto A with center S′. Lemma. The retraction ρ−∞ factorizes through ρ : ρ−∞ = ρ′ Proof. For x ∈ I , one chooses an apartment Ax containing x and C−∞, hence the sector its sector-germ S−∞ and its panel x + F∞. One chooses also an apartment x + C−∞, −∞ ◦ ρ. Spherical Hecke algebras for Kac-Moody groups over local fields 27 Bx of I (M∞) containing germ∞(x + F∞) and S−∞. Hence, Ax and Bx contain both germ∞(x + F∞) and S−∞; by axiom (MA4) there exists an isomorphism ψx of Ax onto Bx fixing these two germs. By the definition of the parabolic retraction, in 5.6, ρ(x) = ψx(x). Now the apartments Ax and Bx of I (M∞) contain both S−∞, hence S′. So there is −∞ ◦ ρ(x) = an isomorphism θ : Bx → A fixing S′, hence S−∞. As ρ(x) ∈ Bx, one has ρ′ θ(ρ(x)) = θ ◦ ψx(x) and this is ρ−∞(x) as θ ◦ ψx : Ax → A is an isomorphism fixing S−∞. 5.8 Counting We want to prove equation (4): nλ(riµ) = nλ(µ)(cid:0)pqiq′ i(cid:1)αi(µ) for λ ∈ Y ++ and µ ∈ Y , where nλ(µ) is the number of points y ∈ I0 such that ρ−∞(y) = −µ and dv(y, 0) = λ, cf. 5.2. For z ∈ I (M∞) one writes pλ(z) ∈ Z≥0∪{∞} for the number of points y ∈ I0 such that ρ(y) = z and dv(y, 0) = λ. By lemma 5.7, nλ(µ) is the sum of pλ(z) for z ∈ I (M∞) ∩ I0 such that ρ′ −∞(z) = −µ. Let M0 = 0 + M∞ = Ker(αi) be the wall in A of direction M∞ containing 0. Its fixator G(M0) (⊂ G(M∞)) acts transitively on the apartments of I or I (M∞) containing it (by axiom (MA4), as M0 is the enclosure of two sector panel germs). Moreover G(M∞) fixes F∞, hence ρ is G(M∞)−equivariant. As a consequence, the weight function pλ is constant on the orbits of G(M0) in I (M∞)∩ I0. Hence nλ(µ) =PΩ pλ(Ω)nΩ(−µ), where the sum runs over the orbits Ω of G(M0) in I (M∞)∩ I0 and nΩ(ν) is the number of points in the orbit Ω such that ρ′ To prove formula (4), it is sufficient to prove for any orbit Ω as above and any ν ∈ Y that: −∞(z) = ν. nΩ(riν) = nΩ(ν)(cid:16)qqiq′ i(cid:17)−αi(ν) We saw, in 5.5, that G(M∞) leaves the decomposition I (M∞) = I (F∞) × E invariant and acts on E by translations. But G(M0) fixes M0 ∋ 0, so it acts trivially on E. As G(M0) is transitive on the apartments containing M0, an orbit Ω is a set Sr × {e} where Sr is the sphere of radius r ∈ Z≥0 and center 0 in the tree I (F∞). The apartment A (with its small set of walls) is the product (R, Z) × E, where αi is the projection of A onto the one dimensional apartment R with vertex set Z. So, the above formula, hence Formula (4) and Theorem 5.4 are consequences of the i when m = αi(ν) is odd, was explained in the following proposition. The fact that qi = q′ proof of 5.4. 5.9 The tree case Let T be a (discrete) semi-homogeneous tree. Let A ≃ R be an apartment in T whose vertices are identified with Z. The valence of the vertex s ∈ Z is 1 + q (resp. 1 + q′) if s is even (resp. odd). Let −∞ be the end of A corresponding to integers converging towards −∞. Let ρ′ be the retraction of T onto A with center −∞. For m ∈ Z ⊂ A and r ∈ Z≥0 we write nr(m) the number of vertices in the sphere Sr of center 0 and radius r in T such that ρ′(z) = m. Proposition. One has nr(m) = nr(−m)(√qq′)m. Remark. This formula is equivalent to the W v(T)−invariance of the image of the Satake isomorphism for the Bruhat-Tits tree T. As this invariance is known, the following proof is not necessary; we give it for the convenience of the reader. If m is odd we ask that q = q′. 28 Stéphane Gaussent & Guy Rousseau For a Bruhat-Tits tree I = T, there are two choices for I0 (and Y ): the set of vertices at even distance from 0 or the full set of vertices. In this last case, we have to allow m to be odd and we see below that the hypothesis q = q′ is necessary to get the formula. So, even for classical Bruhat-Tits buildings, to get the good image for the Satake isomorphism, I0 cannot be any G−stable set of special vertices (we chose I0 to be a G−orbit). Proof. For z ∈ Sr, let sz ∈ Z be the vertex of A such that [0, sz] = [0, z] ∩ A. Then ρ′(z) = sz + (r − sz) ∈ Z. We can calculate the number nr(m) of vertices z ∈ Sr such that ρ′(z) = m: First case: sz ≥ 0 ⇐⇒ ρ′(z) = r. So nr(r) = qq′qq′ ··· (r factors). Second case: −r ≤ sz < 0 ⇐⇒ ρ′(z) < r and then ρ′(z) = r + 2sz i.e. sz = (ρ′(z) − r)/2. The number nr(m) is then: 1 (q − 1)q′qq′ ··· (q′ − 1)qq′q ··· (r + sz = (r + m)/2 factors) (r + sz = (r + m)/2 factors) if m = sz = −r if if sz ∈] − r, 0[ is even sz ∈] − r, 0[ is odd It is now easy to compare nr(m) and nr(−m). We get the wanted formula, using that q = q′ when m is odd. References [Ba96] Nicole Bardy-Panse, Systèmes de racines infinis, Mémoire Soc. Math. France (N.S.) 65 (1996). 3, 4 [BCGR11] Nicole Bardy-Panse, Cyril Charignon, Stéphane Gaussent & Guy Rousseau, Une preuve plus immobilière du théorème de saturation de Kapovich-Leeb- Millson, submitted. cf. ArXiv;1007.3803. 8 [BrK10] Alexander Braverman & David Kazhdan, The spherical Hecke algebra for affine Kac-Moody groups I, ArXiv: 0809.1463v3 (2010). 2, 9, 12 [BrKP12] Alexander Braverman, David Kazhdan & Manish Patnaik, Hecke algebras for p−adic loop groups, preliminary draft, January 2012. 3, 14 [BrT72] François Bruhat & Jacques Tits, Groupes réductifs sur un corps local I, Données radicielles valuées, Publ. Math. Inst. Hautes Études Sci. 41 (1972), 5-251. 1, 4 [Ca79] Pierre Cartier, Representations of p−adic groups: a survey, in Automorphic forms, representations and L-functions, Corvallis 1977, Proc. Symp. Pure Math. 33 (1979), 111-155. 24 [Ch10] Cyril Charignon, Structures immobilières pour un groupe de Kac-Moody sur un corps local, preprint Nancy (2010), arXiv [math.GR] 0912.0442v3. 6, 13, 26 [Ch11] Cyril Charignon, Immeubles affines et groupes de Kac-Moody, masures bordées (thèse Nancy, 2 juillet 2010) ISBN 978-613-1-58611-8 (Éditions universitaires eu- ropéennes, Sarrebruck, 2011). 6, 13, 26 Spherical Hecke algebras for Kac-Moody groups over local fields 29 [GL05] Stéphane Gaussent & Peter Littelmann, LS-galleries, the path model, and MV- cycles, Duke Math. J. 127, (2005), 35 -- 88. 8 [GL11] Stéphane Gaussent & Peter Littelmann, One-skeleton galleries, the path model and a generalization of Mac Donald's formula for Hall-Littlewood polynomials, Int. Math. Res. Notices, to appear; preprint, arXiv:1004.0066, 39 pages. 21 [GR08] Stéphane Gaussent & Guy Rousseau, Kac-Moody groups, hovels and Littelmann paths, Annales Inst. Fourier 58 (2008), 2605-2657. 2, 4, 6, 7, 9, 10, 14, 16, 19, 21 [Ka90] Victor G. Kac, Infinite dimensional Lie algebras, third edition, (Cambridge University Press, Cambridge, 1990). 3, 10, 11, 12, 14, 24 [KLM08] Misha Kapovich, Bernhard Leeb & John J. Millson, Polygons in symmetric spaces and buildings with applications to algebra, Memoir Amer. Math. Soc. 896 (2008). 11 [KM08] Misha Kapovich & John J. Millson, A path model for geodesics in euclidean buildings and its applications to representation theory, Geometry, Groups and Dynamics 2, (2008), 405-480. 8 [KeR07] Ferdaous Kellil & Guy Rousseau, Opérateurs invariants sur certains immeubles affines de rang 2, Annales Fac. Sci. Toulouse 16 (2007), 591-610. 15 [Loo] Eduard Looijenga, Invariant theory for generalized root systems, Invent. Math. 61 (1980), 1-32. 22 [MoP89] Robert Moody & Arturo Pianzola, On infinite root systems, Trans. Amer. Math. Soc. 315 (1989) 661-696. 3, 4 [MoP95] Robert Moody & Arturo Pianzola, Lie algebras with triangular decompositions, (Wiley-Interscience, New York, 1995).3, 4 [P06] James Parkinson, Buildings and Hecke algebras, J. of Algebra 297 (2006), 1-49. 1, 11 [P10] Manish Patnaik, The Satake map for p−adic loop groups and the analogue of Mac Donald's formula for spherical functions, lecture Nancy, December 10 2010. 3 [Re02] Bertrand Rémy, Groupes de Kac-Moody déployés et presque déployés, Astérisque 277 (2002). 13, 14 [Ro11] Guy Rousseau, Masures affines, Pure Appl. Math. Quarterly 7 (no 3 in honor of J. Tits) (2011), 859-921. 3, 4, 5, 6, 7, 9, 13, 16, 21, 25, 26 [Ro12] Guy Rousseau, Groupes de Kac-Moody déployés sur un corps local, 2 Masures ordonnées, preprint Nancy, January 2010, ArXiv [math.GR] 1009.0135v2. 2, 6, 13, 14 [Ro13] Guy Rousseau, Almost split Kac-Moody groups over ultrametric fields, preprint nancy February 2012, ArXiv [math.GR] 1202.6232v1. 2, 6, 13, 14, 15, 26 [Sa63] Ichiro Satake, Theory of spherical functions on reductive algebraic groups over p−adic fields, Publ. Math. Inst. Hautes Études Sci. 18 (1963), 5-69. 1, 11, 15 30 Stéphane Gaussent & Guy Rousseau [T87] Jacques Tits, Uniqueness and presentation of Kac-Moody groups over fields, J. of Algebra 105 (1987), 542-573. 13, 14 Institut Élie Cartan Nancy, UMR 7502, Université de Lorraine, CNRS Boulevard des aiguillettes, BP 70239, 54506 Vandoeuvre lès Nancy Cedex (France) [email protected] [email protected] Acknowledgement: The first author acknowledges support of the ANR grants ANR-09- JCJC-0102-01 and ANR-2010-BLAN-110-02.
0910.2825	2	0910	2010-03-19T07:17:53	Compatibility support mappings in effect algebras	[ "math.RA", "math.QA" ]	We give a characterization of subsets of effect algebras, that can be embedded into a range of an observable. To give this characterization, we introduce a new notion of {\em compatibility support mappings.}	math.RA	math	COMPATIBILITY SUPPORT MAPPINGS IN EFFECT ALGEBRAS GEJZA JEN CA Abstract. We give a characterization of subsets of effect algebras, that can be embedded into a range of an observable. To give this characterization, we introduce a new notion of compatibility support mappings. 1. Introduction and motivation Question 1. Let S be a set of effects on a separable Hilbert space H. Is there a measurable space (X, A) and a POV-measure α : (X, A) → E(H) such that S is a subset of the range of α? If S consists only of orthogonal projections (that means, idempotent effects), then the answer is simple: S is a subset of the range of a POV-measure iff the elements of S commute. On the other hand, if there are non-idempotent effects in S, the answer is not known. In the present paper, we examine a related question: Question 2. If S is a subset of an effect algebra E, is there a Boolean algebra B and a morphism of effect algebras α : B → E such that S ⊆ α(B)? This can be considered as a quantum-logical version of Question 1. We prove that, given subset S of an effect algebra E such that 1 ∈ S, there exist a Boolean algebra B and a morphism α : B → E with S ⊆ α(B) if and only if there is a mapping J . , . K : F in(S) × F in(S) → E satisfying certain properties. We call them compatibility support mappings. The proof uses a modification of the limit techniques introduced in [3]. We show that compatibility support mappings, and hence pairs (B, α), exist whenever S is an MV-algebra or S is a pairwise commuting set of effects on a Hilbert space. We prove several properties of strong compatibility support maps, generalizing the properties of the prototype Example 2. The results presented in this paper are more general than the results from an earlier paper [7], where only interval effect algebras were considered. In that paper, a related notion of witness 06 was introduced to characterize coexistent subsets of interval effect algebras. In the last section, we examine connections between compatibility support map- pings and witness mappings. We prove that, for a subset S of an interval effect algebra, every compatibility support map for S gives rise to a witness mapping for S. We do not know whether this relationship is a one-to-one correspondence. 1991 Mathematics Subject Classification. Primary: 03G12, Secondary: 06F20, 81P10. Key words and phrases. effect algebra, observables. This research is supported by grants VEGA G-1/0080/10 of MS SR, Slovakia and by the Slovak Research and Development Agency under the contracts No. APVT-51-032002, APVV-0071-06. 1 2 GEJZA JEN CA 2. Definitions and basic relationships An effect algebra is a partial algebra (E; ⊕, 0, 1) with a binary partial operation ⊕ and two nullary operations 0, 1 satisfying the following conditions. (E1) If a ⊕ b is defined, then b ⊕ a is defined and a ⊕ b = b ⊕ a. (E2) If a ⊕ b and (a ⊕ b) ⊕ c are defined, then b ⊕ c and a ⊕ (b ⊕ c) are defined and (a ⊕ b) ⊕ c = a ⊕ (b ⊕ c). (E3) For every a ∈ E there is a unique a′ ∈ E such that a ⊕ a′ = 1. (E4) If a ⊕ 1 exists, then a = 0 Effect algebras were introduced by Foulis and Bennett in their paper [5]. In- dependently, Kopka and Chovanec introduced an essentially equivalent structure called D-poset (see [8]). Another equivalent structure, called weak orthoalgebras was introduced by Giuntini and Greuling in [6]. For brevity, we denote the effect algebra (E, ⊕, 0, 1) by E. In an effect algebra E, we write a ≤ b iff there is c ∈ E such that a⊕c = b. It is easy to check that every effect algebra is cancellative, thus ≤ is a partial order on E. In this partial order, 0 is the least and 1 is the greatest element of E. Moreover, it is possible to introduce a new partial operation ⊖; b ⊖ a is defined iff a ≤ b and then a ⊕ (b ⊖ a) = b. It can be proved that a ⊕ b is defined iff a ≤ b′ iff b ≤ a′. It is usual to denote the domain of ⊕ by ⊥. If a ⊥ b, we say that a and b are orthogonal. Example 1. The prototype example of an effect algebra is the standard effect algebra E(H). Let H be a Hilbert space. Let S(H) be the set of all bounded self-adjoint operators. on H. Let I be the identity operator H. For A, B ∈ S(H), write A ≤ B if and only if, for all x ∈ H, hAx, xi ≤ hBx, xi. Put E(H) = {X ∈ S(H) : 0 ≤ X ≤ I} and for A, B ∈ E(H) define A ⊕ B iff A ⊕ B ≤ I, A ⊕ B = A + B. Then (E(H), ⊕, 0, I) is an effect algebra. The elements of E(H) are called Hilbert space effects. An effect algebra E is lattice ordered iff (E, ≤) is a lattice. An effect algebra is an orthoalgebra iff a ⊥ a implies a = 0. An orthoalgebra that is lattice ordered is an orthomodular lattice. An MV-effect algebra is a lattice ordered effect algebra M in which, for all a, b ∈ M , (a ∨ b) ⊖ a = b ⊖ (a ∧ b). It is proved in [4] that there is a natural, one-to one correspondence between MV-effect algebras and MV-algebras given by the following rules. Let (M, ⊕, 0, 1) be an MV-effect algebra. Let ⊞ be a total operation given by x ⊞ y = x ⊕ (x′ ∧ y). Then (M, ⊞,′ , 0) is an MV-algebra. Similarly, let (M, ⊞, ¬, 0) be an MV-algebra. Restrict the operation ⊞ to the pairs (x, y) satisfying x ≤ y′ and call the new partial operation ⊕. Then (M, ⊕, 0, ¬0) is an MV-effect algebra. Among lattice ordered effect algebras, MV-effect algebras can be characterized in a variety of ways. Three of them are given in the following proposition. Proposition 1. [1], [4] Let E be a lattice ordered effect algebra. The following are equivalent (a) E is an MV-effect algebra. (b) For all a, b ∈ E, a ∧ b = 0 implies a ≤ b′. (c) For all a, b ∈ E, a ⊖ (a ∧ b) ≤ b′. (d) For all a, b ∈ E, there exist a1, b1, c ∈ E such that a1 ⊕ b1 ⊕ c exists, a1 ⊕ c = a and b1 ⊕ c = b. COMPATIBILITY SUPPORT MAPPINGS IN EFFECT ALGEBRAS 3 Let B be a Boolean algebra and let E be an effect algebra. An observable is a mapping α : B → E such that α(0) = 0, α(1) = 1 and for every x, y ∈ B such that x ∧ y = 0, φ(x ∨ y) = φ(x) ⊕ φ(y). 3. Compatibility support mappings -- definition and examples In this section we introduce (strong) compatibility support mappings and present two examples. Definition 1. Let E be an effect algebra, let S ⊆ E be such that 1 ∈ S. We say that J . , . K : F in(S) × F in(S) → E is a compatibility support mapping for S if and only if the following conditions are satisfied. (a) If V1 ⊆ V2, then JU, V1K ≤ JU, V2K. (b) JU, V K ≤ JU, {1}K. (c) JU, ∅K = 0. (d) J∅, {c}K = c. (e) If c /∈ U ∪ V , then JU ∪ {c}, {1}K ⊖ JU ∪ {c}, VK = JU, V ∪ {c}K ⊖ JU, V K (e) For all c, JU ∪ {c}, {1}K ⊖ JU ∪ {c}, V K = JU, V ∪ {c}K ⊖ JU, V K A compatibility mapping is strong if and only if the following condition is satisfied. Note that (e) implies (e). Example 2. Let M be an MV-effect algebra. Define J . , . K : F in(M )×F in(M ) → E by Then J . , . K is a strong compatibility support mapping. The conditions (a)-(d) are easy to prove. Let us prove (e). JU, V K = (^ U ) ∧ (_ V ). JU, V ∪ {c}K ⊖ JU, VK = (^ U ) ∧ (c ∨ (_ V )) ⊖ ((^ U ) ∧ (_ V )) = = ((^ U ) ∧ c) ∨ ((^ U ) ∧ (_ V )) ⊖ ((^ U ) ∧ (_ V )) = = ((^ U ) ∧ c) ⊖ ((^ U ) ∧ c ∧ (_ V )) = = JU ∪ {c}, {1}K ⊖ JU ∪ {c}, V K Example 3. Let ⊔ be an operation on the set of all operators on a Hilbert space H given by a ⊔ b := a + b − ab. It is easy to check that ⊔ is associative with neutral element 0. If a and b are commuting effects, then a.b is an effect with a.b ≤ a, b. Moreover, a⊔b is an effect. Indeed, since a, b are commuting effects, 1 − a, 1 − b are commuting effects. Since 1 − a, 1 − b are commuting effects, (1 − a).(1 − b) is an effect and 1 − (1 − a).(1 − b) = 1 − (1 − a − b + ab) = a + b − ab is an effect. Let S be a set of commuting effects with 1 ∈ S; there exists a commutative C∗ algebra A with S ⊆ A. The operations ⊔, . are commutative and associative on A ∩ E(H) ⊇ S. Let U, V be a finite subsets of S. Write d U for the product of elements of U . Write F ∅ = 0, F{c} = c and, for V = {v1, . . . , vn} with n > 1, write G V = v1 ⊔ · · · ⊔ vn. 4 GEJZA JEN CA Define J . , . K : F in(S) × F in(S) → E JU, V K = (l U ).(G V ). Let us prove that J . , . K is a compatibility support mapping. JU, V2K. Let us prove that F V1 ≤ F V2. Since V1 ⊆ V2, we may write Proof of condition (a): Suppose that V1 ⊆ V2. We need to prove that JU, V1K ≤ G V2 = (G V1) ⊔ (G(V2 \ V1)) = = (G V1) + (G(V2 \ V1)) − (G V1).(G(V2 \ V1)) Therefore, (G V2) − (G V1) = (G(V2 \ V1)) − (G V1).(G(V2 \ V1)) ≥ 0, so F V1 ≤ F V2. Since F V1 ≤ F V2, JU, V1K = (l U ).(G V1) ≤ (l U ).(G V1) = JU, V2K. The conditions (b)-(d) are trivially satisfied. Proof of condition the (e): JU, V ∪ {c}K − JU, V K = (l U ).(c ⊔ G V ) − (l U ).(G V ) = = (l U ).(c + G V − c.(G V )) − (l U ).(G V ) = = (l U ).c + (l U ).(G V ) − (l U ).c.(G V ) − (l U ).(G V ) = = (l U ).c − (l U ).c.(G V ) = JU ∪ {c}, {1}K ⊖ JU ∪ {c}, V K Note that, if S contains some non-idempotent c, then J . , . K is not strong. To see that (e) is not satisfied, put U = V = {c} and compute JU ∪ {c}, {1}K ⊖ JU ∪ {c}, V K = c ⊖ c.c 6= 0 JU, V ∪ {c}K ⊖ JU, V K = c.c ⊖ c.c = 0 4. Observables from compatibility support mappings The aim of this section is to prove that for every S such that S ∪ {1} admits a compatibility support mapping, then S is coexistent. The direct limit method used here is a dual of the projective limit method introduced in [3]. See also [9] for another application of the projective limit method. Several proofs in this section (Lemma 3 through Theorem 1) are very similar, or even the same, as in [7]. The reason for this is that they are basically an application of Lemma 2, which is the Proposition 4 of [7]. However, the author decided to include them here, to keep the present paper more streamlined. Running assumption 1. In this section, we assume the following. • E is an effect algebra. • S is a subset of E with 1 ∈ S. • J . , . K : F in(S) × F in(S) → S is a compatibility support mapping. Lemma 1. For all c ∈ S, J{c}, {1}K = c. COMPATIBILITY SUPPORT MAPPINGS IN EFFECT ALGEBRAS 5 Proof. Put U = V = ∅ in condition (e) of Definition 1. We see that J{c}, {1}K ⊖ J{c}, ∅K = J∅, {c}K ⊖ J∅, ∅K. By conditions (c) and (d), this implies that J{c}, {1}K = c. Let us write, for A, X ∈ F in(S) such that X ⊆ A, Lemma 2. Let A, X ∈ F in(S), X ⊆ A and let c ∈ S be such that c 6∈ A. Then D(X, A) = JX, {1}K ⊖ JX, A \ XK. D(X, A) = D(X, A ∪ {c}) ⊕ D(X ∪ {c}, A ∪ {c}). Proof. We see that D(X, A ∪ {c}) =JX, {1}K ⊖ JX, {c} ∪ (A \ X)K D(X ∪ {c}, A ∪ {c}) =JX ∪ {c}, {1}K ⊖ JX ∪ {c}, A \ XK and, by condition (e) of Definition 1, we see that JX ∪ {c}, {1}K ⊖ JX ∪ {c}, A \ XK = JX, {c} ∪ (A \ X)K ⊖ JX, A \ XK. Therefore, D(X, A ∪ {c}) ⊕ D(X ∪ {c}, A ∪ {c}) = = (JX, {1}K ⊖ JX, {c} ∪ (A \ X)K) ⊕ (JX, {c} ∪ (A \ X)K ⊖ JX, A \ XK) = = JX, {1}K ⊖ JX, A \ XK = D(X, A). (cid:3) (cid:3) Lemma 3. Let C, A, X ∈ F in(S) be such that X ⊆ A and C ∩ A = ∅. Then (D(X ∪ Y, A ∪ C))Y ⊆C is an orthogonal family and D(X ∪ Y, A ∪ C) = D(X, A). M Y ⊆C Proof. The proof goes by induction with respect to C. For C = ∅, Lemma 3 is trivially true. Assume that Lemma 3 holds for all C with C = n and let c ∈ S, c 6∈ A ∪ C. Let us consider the family (D(X ∪ Z, A ∪ C ∪ {c}))Z⊆C∪{c}. For every Z ⊆ C ∪ {c}, either c ∈ Z or c 6∈ Z, so either Z = Y ∪ {c} or Z = Y , for some Y ⊆ C. Therefore, we can write (D(X ∪ Z, A ∪ C ∪ {c}))Z⊆C∪{c} = (D(X ∪ Y, A ∪ C ∪ {c}), D(X ∪ Y ∪ {c}, A ∪ C ∪ {c}))Y ⊆C . By Lemma 2, D(X ∪ Y, A ∪ C ∪ {c}) ⊕ D(X ∪ Y ∪ {c}, A ∪ C ∪ {c}) = D(X ∪ Y, A ∪ C). It only remains to apply the induction hypothesis to finish the proof. (cid:3) Corollary 1. For every A ∈ F in(S), (D(X, A))X⊆A is a decomposition of unit. 6 GEJZA JEN CA Proof. Obviously, By Lemma 3, D(∅, ∅) = J∅, {1}K ⊖ J∅, ∅K = 1 ⊖ 0 = 1. (D(∅ ∪ X, ∅ ∪ A)) = D(∅, ∅). M X⊆A Corollary 2. For every A ∈ F in(S), the mapping αA : 2(2A) → E given by (cid:3) αA(X) = M X∈X D(X, A) is a simple observable. Proof. The atoms of 2(2A) are of the form {X}, where X ⊆ A. By Corollary 1, (αA({X}) : X ⊆ A) is a decomposition of unit; the remainder of the proof is trivial. (cid:3) For A, B ∈ F in(S) with A ⊆ B, let us define mappings gA B : 2(2A) → 2(2B ) gA B(X) = {X ∪ C0 : X ∈ X and C0 ⊆ (B \ A)} and let us write G for the collection of all such mappings. It is an easy exercise to prove that every gA B ∈ G is an injective homomorphism of Boolean algebras and that ((2(2A) : A ∈ F in(S)), G) is a direct family of Boolean algebras. Let us prove that the mappings gA B behave well with respect to the observables αA and αB. Lemma 4. Let A, B ∈ F in(S) with A ⊆ B. The diagram 2(2A) αA E gA B 2(2B ) αB commutes. Proof. For all X ∈ 2(2A), αB(gA B(X)) = αB({X ∪ C0 : X ∈ X and C0 ⊆ (B \ A)}) = = M(D(X ∪ C0, B) : X ∈ X and C0 ⊆ (B \ A)) = D(X ∪ C0, B)(cid:17) (cid:16) M = M X∈X C0⊆(B\A) Put Y := C0, C := B \ A; by Lemma 3, M C0⊆(B\A) D(X ∪ C0, B) = D(X, A). Therefore, αB(gA B(X)) = M X∈X D(X, A) = αA(X) and the diagram commutes. (cid:3) COMPATIBILITY SUPPORT MAPPINGS IN EFFECT ALGEBRAS 7 Corollary 3. For every B ∈ F in(S), B is a subset of the range of αB. Proof. We need to prove that every a ∈ B is an element of the range of αB. For B = ∅, this is trivial. Suppose that B is nonempty and let a ∈ B. Let A = {a}. and let X = gA B({{a}}). By Lemma 4, αB(X) = αB(gA B({{a}})) = αA({{a}}), and we see that, by (c) of Definition 1 and by Lemma 1 αA({{a}}) = α{a}({{a}}) = D({a}, {a}) = J{a}, {1}K⊖J{a}, {a} \ {a}K = a⊖0 = a. (cid:3) Theorem 1. Let E be an effect algebra, let S ⊆ E. If S ∪{1} admits a compatibility support mapping, then S is coexistent. Proof. Suppose that S ∪ {1} admits a compatibility support mapping. Let us construct FB(S) as the direct limit of the direct family (22A : A ∈ F in(S)), equipped with morphisms of the type gA B. After that, we shall define an observable α : FB(S) → E. Consider the set ΓS = [ A∈F in(S) {(X, A) : X ⊆ 2A} and define on it a binary relation ≡ by (X, A) ≡ (Y, B) if and only if gA gB A∪B(Y), that means A∪B(X) = {X ∪ CA : X ∈ X and CA ⊆ A ∪ B \ A} = {Y ∪ CB : Y ∈ Y and CB ⊆ A ∪ B \ B}. Then FB(S) = ΓS/ ≡ and the operations on FB(S) are defined by [(X, A)]≡ ∨ [(Y, B)]≡ = [(gA A∪B(X) ∪ gB A∪B(Y), A ∪ B)]≡ and similarly for the other operations. Then FB(S) is a direct limit of Boolean algebras, hence a Boolean algebra. Let αS : FB(S) → E be a mapping given by the rule αS([(X, A)]≡) = αA(X). We shall prove that αS is an observable. Let us prove αS is well-defined. Suppose that (X, A) ≡ (Y, B), that means, gA A∪B(X) = gB A∪B(Y). By Lemma 4, and αA(X) = αA∪B(gA A∪B(X)) αB(Y) = αA∪B(gB A∪B(Y)), hence αS is a well-defined mapping. Let us prove that αS is an observable. The bounds of the Boolean algebra FB(S) are [(∅, A)]≡ and [(2A, A)]≡, where A ∈ F in(S). Obviously, by Corollary 2, and αS([(∅, A)]≡) = αA(∅) = 0 αS([(2A, A)]≡) = αA(2A) = 1. 8 GEJZA JEN CA Let [(X, A)]≡ and [(Y, B)≡] be disjoint elements of FB(S), that is, gA gB A∪B(Y) = ∅. Then A∪B(X) ∩ αS([(X, A)]≡ ∨ [(Y, B)]≡) = αS([gA A∪B(X) ∪ gB = αA∪B(gA A∪B(Y), A ∪ B]≡) = A∪B(X) ∪ gB A∪B(Y)). Since αA∪B is an observable, A∪B(X) ∪ gB αA∪B(gA A∪B(Y)) = αA∪B(gA A∪B(X)) ⊕ αA∪B(gB A∪B(Y)). It remains to observe that and that αA∪B(gA A∪B(X)) = αS([(X, A)]≡) αA∪B(gB A∪B(Y)) = αS([(Y, B)]≡). Let us prove that the range of αS includes S. Let a ∈ S. By Corollary 3, the range of α{a} includes a and, by an obvious direct limit argument, the range of α{a} is a subset of the range of αS. (cid:3) 5. Compatibility support mappings from observables The aim of the single theorem of this section is to prove that every subset S of the range of an observable admits a strong compatibility support mapping. Theorem 2. For every coexistent subset S of an effect algebra E, S ∪ {1} admits a strong compatibility support mapping. Proof. Let B be a Boolean algebra and let α : B → E be an observable, let S be a subset of the range of α. For every a ∈ S ∪ {1}, fix an element pa ∈ α−1(a) and define JU, VK = α(( ^ a∈U pa) ∧ ( _ b∈V pb)). Let us check the condition in the definition of a strong compatibility support mapping. Let c 6∈ U, V . Then JU ∪ {c}, {1}K ⊖ JU ∪ {c}, VK = pa) ∧ pc) ⊖ α((( ^ pa) ∧ pc) ∧ ( _ pb)). a∈U b∈V = α(( ^ a∈U To simplify the matters, write mU = ( ^ a∈U pa) jV = ( _ b∈V pb) We can write α(( ^ a∈U pa) ∧ pc) ⊖ α((( ^ a∈U pa) ∧ pc) ∧ ( _ b∈V pb)) = α(mU ∧ pc) ⊖ α(mU ∧ pc ∧ jV ) = = α((mU ∧ pc) ⊖ (mU ∧ pc ∧ jV )) Similarly, JU, V ∪ {c}K ⊖ JU, V K = α((mU ∧ (pc ∨ jV )) ⊖ (mU ∧ jV )). COMPATIBILITY SUPPORT MAPPINGS IN EFFECT ALGEBRAS 9 Since B is a Boolean algebra, (mU ∧ pc) ⊖ (mU ∧ pc ∧ jV ) = (mU ∧ (pc ∨ jV )) ⊖ (mU ∧ jV ) The remaining conditions are trivial to check. (cid:3) Let us note that, if we start with a non-strong compatibility support mapping, apply Theorem 1 to construct an observable and then apply Theorem 2 to con- struct a compatibility support mapping, we cannot obtain the compatibility support mapping we started with, since Theorem 2 always produces a strong compatibility support mapping. 6. Properties of strong compatibility support mappings The aim of this section is to prove that several properties of the Example 2 are valid for all strong compatibility support mappings. It remains open whether and which of these properties are valid for all compatibility support mappings. The main vehicle here is Proposition 2, that is interesting in its own right: it shows that, for a given S, every strong compatibility support mapping on S is determined by its D( . , . ). Running assumption 2. In this section, we assume the following. • E is an effect algebra. • S is a subset of E with 1 ∈ S. • J . , . K : F in(S) × F in(S) → S is a strong compatibility support mapping. Lemma 5. If U, V are not disjoint, then JU, VK = JU, {1}K. Proof. Let c ∈ U ∩ V . This implies that U ∪ {c} = U and V ∪ {c} = V . Therefore, by (e), JU, {1}K ⊖ JU, VK = JU, VK ⊖ JU, V K = 0, hence JU, V K = JU, {1}K. Lemma 6. JU ∪ {c}, {1}K = JU, {c}K. Proof. Put V = ∅ in (e): JU ∪ {c}, {1}K ⊖ JU ∪ {c}, ∅K = JU, {c}K ⊖ JU, ∅K. By condition (c), JU ∪ {c}, ∅K = JU, ∅K = 0, therefore JU ∪ {c}, {1}K = JU, {c}K. Proposition 2. Let U, V ⊆ S. (1) If U ∩ V 6= ∅, then JU, V K = JU, {1}K = D(U, U ). (2) If U ∩ V = ∅, then JU, VK = M ∅6=Y ⊆V D(U ∪ Y, U ∪ V ). Proof. (1) By Proposition 5, JU, V K = JU, {1}K and D(U, U ) = JU, {1}K ⊖ JU, ∅K = JU, {1}K ⊖ 0 = JU, {1}K. (cid:3) (cid:3) 10 GEJZA JEN CA (2) By Lemma 3, D(U, U ) = M Y ⊆V D(U ∪ Y, U ∪ V ). Therefore, Moreover, D(U, U ) ⊖ D(U, U ∪ V ) = M ∅6=Y ⊆V D(U ∪ Y, U ∪ V ). D(U, U ) ⊖ D(U, U ∪ V ) = (JU, {1}K ⊖ JU, ∅K) ⊖ (JU, {1}K ⊖ JU, V K) = = JU, V K ⊖ JU, ∅K = JU, V K ⊖ 0 = JU, V K. (cid:3) Proposition 3. If U1 ⊆ U2, then JU1, V K ≥ JU2, VK. Proof. (Case 1) Suppose that U1 ∩ V 6= ∅. Then U2 ∩ V 6= ∅. By Proposition 2 and Lemma 3, JU1, VK = D(U1, U1) = M D(U1 ∪ Y, U2) ≥ D(U2, U2) = JU2, V K. (Case 2) Suppose that U2 ∩ V = ∅. Then U1 ∩ V = ∅. By Proposition 2, Y ⊆U2\U1 JU1, V K = M ∅6=Y ⊆V D(U1 ∪ Y, U1 ∪ V ). By Lemma 3, for every ∅ 6= Y ⊆ V , D(U1 ∪ Y, U1 ∪ V ) = M W ⊆U2\U1 D(U1 ∪ Y ∪ W, U2 ∪ V ). Obviously (put W = U2 \ U1), this implies that D(U1 ∪ Y, U1 ∪ V ) ≥ D(U2 ∪ Y, U2 ∪ V ), hence we may write JU1, V K = M ∅6=Y ⊆V D(U1 ∪ Y, U1 ∪ V ) ≥ M ∅6=Y ⊆V D(U2 ∪ Y, U2 ∪ V ). It remains to apply Proposition 2 again: M ∅6=Y ⊆V D(U2 ∪ Y, U2 ∪ V ) = JU2, V K. (Case 3) Suppose that U1 ∩ V = ∅ and U2 ∩ V 6= ∅. By Proposition 2, JU1, V K = M ∅6=Y ⊆V D(U1 ∪ Y, U1 ∪ V ). By Lemma 3, D(U1 ∪ Y, U1 ∪ V ) = M W ⊆U2\(U1∪V ) D(U1 ∪ W ∪ Y, U2 ∪ V ) We can put W = U2 \ (U1 ∪ V ), proving that D(U1 ∪ Y, U1 ∪ V ) ≥ D((U2 \ V ) ∪ Y, U2 ∪ V ). COMPATIBILITY SUPPORT MAPPINGS IN EFFECT ALGEBRAS 11 Therefore, JU1, VK = M ∅6=Y ⊆V D(U1 ∪ Y, U1 ∪ V ) ≥ M ∅6=Y ⊆V D((U2 \ V ) ∪ Y, U2 ∪ V ) ≥ M V ∩U2⊆Y ⊆V D((U2 \ V ) ∪ Y, U2 ∪ V ). For every V ∩ U2 ⊆ Y ⊆ V , there is exactly one Z ⊆ V \ U2 such that (U2 \ V ) ∪ Y = U2 ∪ Z. Thus, we can rewrite M V ∩U2⊆Y ⊆V D((U2 \ V ) ∪ Y, U2 ∪ V ) = M Z⊆V \U2 D(U2 ∪ Z, U2 ∪ V ). By Lemma 3 and Proposition 2, M Z⊆V \U2 D(U2 ∪ Z, U2 ∪ V ) = D(U2, U2) = JU2, V K. Proposition 4. JU, {1}K is a lower bound of U . Proof. Any element is a lower bound of ∅. Suppose that the proposition is true for some U and pick c ∈ S \ U . By Propo- sition 3, (cid:3) JU ∪ {c}, {1}K ≤ JU, {1}K. By the induction hypothesis, JU, {1}K is a lower bound of U . It remains to prove that JU ∪ {c}, {1}K ≤ c. By Proposition 6, JU ∪ {c}, {1}K = JU, {c}K. By Proposition 3 and condition (d), JU, {c}K ≤ J∅, {c}K = c. Corollary 4. JU, V K is a lower bound of U . Proof. By Proposition 4, JU, {1}K is a lower bound of U . By condition (b), JU, VK ≤ JU, {1}K. Proposition 5. J∅, VK is an upper bound of V . (cid:3) Proof. Any element is an upper bound of ∅. Suppose that the proposition is true for some V and pick c ∈ S \ V . By condition (a), and by induction hypothesis, J∅, V K is an upper bound of V . It remains to prove that c ≤ J∅, V ∪ {c}K. Put U = ∅ in condition (e*): J∅, V K ≤ J∅, V ∪ {c}K (cid:3) (cid:3) Add J∅, V K to both sides to obtain J{c}, {1}K ⊖ J{c}, VK = J∅, V ∪ {c}K ⊖ J∅, V K. (J{c}, {1}K ⊖ J{c}, VK) ⊕ J∅, VK = J∅, V ∪ {c}K. As J{c}, VK ≤ J∅, VK, By Lemma 1, J{c}, {1}K = c. J{c}, {1}K ≤ (J{c}, {1}K ⊖ J{c}, VK) ⊕ J∅, VK. 12 GEJZA JEN CA 7. Compatibility support mappings and witness mappings Let (G, ≤) be a partially ordered abelian group and u ∈ G be a positive element. For 0 ≤ a, b ≤ u, define a ⊕ b if and only if a + b ≤ u and put a ⊕ b = a + b. With such a partial operation ⊕, the closed interval [0, u]G = {x ∈ G : 0 ≤ x ≤ u} becomes an effect algebra ([0, u]G, ⊕, 0, u). Effect algebras which arise from partially ordered abelian groups in this way are called interval effect algebras, see [2]. Let E be an interval effect algebra in a partially ordered abelian group G. Let S ⊆ E. Let us write F in(S) for the set of all finite subsets of S. We write I(F in(S)) for the set of all comparable elements of the poset (F in(S), ⊆), that means, I(F in(S)) = {(X, Y ) ∈ F in(S) × F in(S) : X ⊆ Y }. For every mapping β : F in(S) → G, we define a mapping Dβ : I(F in(S)) → G. For (X, A) ∈ I(F in(S)), the value Dβ(X, A) ∈ G is given by the rule Dβ(X, A) := X X⊆Z⊆A (−1)X+Zβ(Z). In [7], we introduced and studied the following notion: Definition 2. Let E be an interval effect algebra. We say that a mapping β : F in(S) → E is a witness mapping for S if and only if the following conditions are satisfied. (A1) β(∅) = 1, (A2) for all c ∈ S, β({c}) = c, (A3) for all (X, A) ∈ I(F in(S)), Dβ(X, A) ≥ 0. We proved there, that a subset S of an interval effect algebra E is coexistent if and only if there is a witness 06 β : F in(S) → E. The aim of this section is to explore the connection between the notion of a witness mapping and the notion of compatibility support mappings. Proposition 6. Let E be an interval effect algebra, let S be a subset of E with 1 ∈ S. Suppose there is a compatibility support mapping J . , . K : F in(S) × F in(S) → S. Then β : F in(S) → E, given by β(X) = JX, {1}K is a witness mapping and D(X, A) = Dβ(X, A), for all (X, A) ∈ I(F in(S)). Proof. We see that, by the condition (d) of Definition 1, so the condition (A1) of Definition 2 is satisfied. By Lemma 1, β(∅) = J∅, {1}K = 1, hence (A2) is satisfied. β({c}) = J{c}, {1}K = c, For the proof of (A3), it suffices to prove that D(X, A) = Dβ(X, A), for all (X, A) ∈ I(F in(S)). The positivity of Dβ then follows from the positivity of D. The proof goes by induction with respect to A \ X. If A \ X = 0, then A = X and Dβ(X, A) = β(X) = JX, {1}K = JX, {1}K ⊖ 0 = JX, {1}K ⊖ JX, ∅K = D(X, A). COMPATIBILITY SUPPORT MAPPINGS IN EFFECT ALGEBRAS 13 Suppose that D(X, A) = Dβ(X, A), for all (X, A) ∈ I(F in(S)) such that A \ X = n. Let (Y, B) ∈ I(F in(S)) be such that B \ Y = n + 1. Pick c ∈ B \ Y and put X = Y , A = B \ {c}. By Lemma 1 of [7], for any mapping β : F in(S) → E, for all (X, A) ∈ I(F in(S)) and for all c ∈ S \ A, the following equality is satisfied: Dβ(X, A) = Dβ(X, A ∪ {c}) + Dβ(X ∪ {c}, A ∪ {c}). Therefore, Dβ(Y, B) = Dβ(X, A ∪ {c}) = Dβ(X, A) ⊖ Dβ(X ∪ {c}, A ∪ {c}). By the induction hypothesis, Dβ(X, A) = D(X, A) and Dβ(X ∪ {c}, A ∪ {c}) = Dβ(X ∪ {c}, A ∪ {c}). Thus, Dβ(Y, B) = D(X, A) ⊖ D(X ∪ {c}, A ∪ {c}). By Lemma 2, D(X, A) ⊖ D(X ∪ {c}, A ∪ {c}) = D(X, A ∪ {c}) = Dβ(Y, B). (cid:3) The following problem remains open. Problem 1. Let E be an effect algebra, let S ⊆ E, let β : F in(S) → E be a witness mapping. Is there always a compatibility support mapping J . , . K : F in(S) × F in(S) → S such that β(X) = JX, {1}K? References [1] M.K. Bennett and D.J. Foulis. Phi-symmetric effect algebras. Foundations of Physics, 25:1699 -- 9722, 1995. [2] M.K. Bennett and D.J. Foulis. Interval and scale effect algebras. Advances in Applied Mathe- matics, 19:200 -- 215, 1997. [3] G. Cattaneo, M. L. Dalla Chiara, R. Giuntini, and S. Pulmannov´a. Effect Algebras and Para- Boolean Manifolds. International Journal of Theoretical Physics, 39(3):551 -- 564, 2000. [4] F. Chovanec and F. Kopka. Boolean D-posets. Tatra Mt. Math. Publ, 10:183 -- 197, 1997. [5] D.J. Foulis and M.K. Bennett. Effect algebras and unsharp quantum logics. Found. Phys., 24:1325 -- 1346, 1994. [6] R. Giuntini and H. Greuling. Toward a formal language for unsharp properties. Found. Phys., 19:931 -- 945, 1989. [7] G. Jenca. Coexistence in interval effect algebras. 2009, arXiv:quant-ph/0910.2823. [8] F. Kopka and F. Chovanec. D-posets. Math. Slovaca, 44:21 -- 34, 1994. [9] S. Pulmannov´a. A note on observables on MV-algebras. Soft Computing, 4(1):45 -- 48, 2000. Department of Mathematics and Descriptive Geometry, Faculty of Civil Engineer- ing, Radlinsk´eho 11, Bratislava 813 68, Slovak Republic E-mail address: [email protected]
1105.4398	3	1105	2012-04-05T12:10:36	Semisimple Hopf algebras of dimension $2q^3$	[ "math.RA" ]	Let $q$ be a prime number, $k$ an algebraically closed field of characteristic 0, and $H$ a non-trivial semisimple Hopf algebra of dimension $2q^3$. This paper proves that $H$ can be constructed either from group algebras and their duals by means of extensions, or from Radford's biproduct $H\cong R#kG$, where $kG$ is the group algebra of $G$ of order 2, $R$ is a semisimple Yetter-Drinfeld Hopf algebra in ${}^{kG}_{kG}\mathcal{YD}$ of dimension $q^3$.	math.RA	math	SEMISIMPLE HOPF ALGEBRAS OF DIMENSION 2q3 JINGCHENG DONG AND LI DAI Abstract. Let q be a prime number, k an algebraically closed field of charac- teristic 0, and H a non-trivial semisimple Hopf algebra of dimension 2q3. This paper proves that H can be constructed either from group algebras and their duals by means of extensions, or from Radford's biproduct H ∼= R#kG, where kG is the group algebra of G of order 2, R is a semisimple Yetter-Drinfeld Hopf algebra in kG kGY D of dimension q3. 1. Introduction In the last twenty years, various classification results were obtained for finite- dimensional semisimple Hopf algebras over an algebraically closed field of charac- teristic 0. Up to now, semisimple Hopf algebras of dimension p, p2, p3, pq, pq2 and pqr have been completely classified, where p, q, r are distinct prime numbers. See [3, 4, 8, 9, 22] for details. In the present paper, we shall continue the investigation on the classification of semisimple Hopf algebras. The main purpose of this paper is to investigate semisimple Hopf algebras of dimension 2q3, where q is a prime number. Of course, some other interesting results are also obtained in this paper. The paper is organized as follows. In Section 2, we recall the definitions and basic properties of semisolvability, characters, Radford's biproducts and Drinfeld double, respectively. Some useful lemmas are also contained in this section. In Section 3, we study the structure of non-trivial semisimple Hopf algebras of dimension pq3, where p, q are prime numbers with p2 < q. In Section 4, we study the structure of non-trivial semisimple Hopf algebras of dimension 2q3, where q is a prime number. Throughout this paper, all modules and comodules are left modules and left comodules, and moreover they are finite-dimensional over an algebraically closed field k of characteristic 0. ⊗, dim mean ⊗k, dimk, respectively. For two positive integers m and n, gcd(m, n) denotes the greatest common divisor of m, n. Our references for the theory of Hopf algebras are [11] or [21]. The notation for Hopf algebras is standard. For example, the group of group-like elements in H is denoted by G(H). 2. Preliminaries 2.1. Characters. Throughout this subsection, H will be a semisimple Hopf alge- bra over k. Let V be an H-module. The character of V is the element χ = χV ∈ H ∗ defined by hχ, hi = TrV (h) for all h ∈ H. The degree of χ is defined to be the 2000 Mathematics Subject Classification. 16W30. Key words and phrases. semisimple Hopf algebra, semisolvability, Radford's biproduct, char- acter, Drinfeld double. 1 2 J. DONG AND L. DAI integer degχ = χ(1) = dimV . We shall use Xt to denote the set of all irreducible characters of H of degree t. All irreducible characters of H span a subalgebra R(H) of H ∗, which is called the character algebra of H. The antipode S induces an anti- algebra involution ∗ : R(H) → R(H), given by χ 7→ χ∗ := S(χ). The character of the trivial H-module is the counit ε. Let χU , χV ∈ R(H) be the characters of the H-modules U and V , respectively. The integer m(χU , χV ) = dimHomH (U, V ) is defined to be the multiplicity of U in V . Let bH denote the set of irreducible characters of H. Then bH is a basis of R(H). If χ ∈ R(H), we may write χ = Pα∈ bH m(α, χ)α. For any group-like element g in G(H ∗), m(g, χχ∗) > 0 if and only if m(g, χχ∗) = 1 if and only if gχ = χ. The set of such group-like elements forms a subgroup of G(H ∗), of order at most (degχ)2. See [15, Theorem 10]. Denote this subgroup by G[χ]. In particular, we have χχ∗ = X g + X m(α, χχ∗)α. (2.1) g∈G[χ] α∈ bH,degα>1 A subalgebra A of R(H) is called a standard subalgebra if A is spanned by ir- reducible characters of H. Let X be a subset of bH. Then X spans a standard subalgebra of R(H) if and only if the product of characters in X decomposes as a sum of characters in X. There is a bijection between ∗-invariant standard subalge- bras of R(H) and quotient Hopf algebras of H. See [15, Theorem 6]. H is said to be of type (d1, n1; · · · ; ds, ns) as an algebra if d1 = 1, d2, · · · , ds are the dimensions of the simple H-modules and ni is the number of the non-isomorphic simple H-modules of dimension di. That is, as an algebra, H is isomorphic to a direct product of full matrix algebras H ∼= k(n1) × sY i=2 Mdi(k)(ni). If H ∗ is of type (d1, n1; · · · ; ds, ns) as an algebra, then H is said to be of type (d1, n1; · · · ; ds, ns) as a coalgebra. Lemma 2.1. Let χ be an irreducible character of H. Then (1) The order of G[χ] divides (degχ)2. (2) The order of G(H ∗) divides n(degχ)2, where n is the number of non-isomorphic irreducible characters of degree degχ. Proof. It follows from Nichols-Zoeller Theorem [16]. See also [14, Lemma 2.2.2]. (cid:3) 2.2. Semisolvability. Let B be a finite-dimensional Hopf algebra over k. A Hopf subalgebra A ⊆ B is called normal if h1AS(h2) ⊆ A and S(h1)Ah2 ⊆ A, for all h ∈ B. If B does not contain proper normal Hopf subalgebras then it is called simple. The notion of simplicity is self-dual, that is, B is simple if and only if B ∗ is simple. Let π : H → B be a Hopf algebra map and consider the subspaces of coinvariants H coπ = {h ∈ H(id ⊗ π)∆(h) = h ⊗ 1}, and coπH = {h ∈ H(π ⊗ id)∆(h) = 1 ⊗ h}. Then H coπ (respectively, coπH) is a left (respectively, right) coideal subalgebra of H. Moreover, we have dimH = dimH coπdimπ(H) = dimcoπHdimπ(H). SEMISIMPLE HOPF ALGEBRAS 3 The left coideal subalgebra H coπ is stable under the left adjoint action of H. Moreover H coπ = coπH if and only if H coπ is a (normal) Hopf subalgebra of H. If this is the case, we shall say that the map π : H → B is normal. See [19] for more details. The following lemma comes from [13, Section 1.3]. Lemma 2.2. Let π : H → B be a Hopf epimorphism and A a Hopf subalgebra of H such that A ⊆ H coπ. Then dimA divides dimH coπ. The notions of upper and lower semisolvability for finite-dimensional Hopf alge- bras have been introduced in [10], as generalizations of the notion of solvability for finite groups. By definition, H is called lower semisolvable if there exists a chain of Hopf subalgebras Hn+1 = k ⊆ Hn ⊆ · · · ⊆ H1 = H such that Hi+1 is a normal Hopf subalgebra of Hi, for all i, and all quotients Hi/HiH + i+1 are trivial. That is, they are isomorphic to a group algebra or a dual group algebra. Dually, H is called upper semisolvable if there exists a chain of quotient Hopf algebras H(0) = H π1−→ H(1) π2−→ · · · πn−−→ H(n) = k such that H coπi (i−1) are trivial. In analogy with the situations for finite groups, it is enough for many applications (i−1) is a normal Hopf subalgebra of H(i−1), and all H coπi to know that a Hopf algebra is semisolvable. By [10, Corollary 3.3], we have that H is upper semisolvable if and only if H ∗ is lower semisolvable. If this is the case, then H can be obtained from group algebras and their duals by means of (a finite number of) extensions. Proposition 2.3. Let H be a semisimple Hopf algebra of dimension pq3, where p, q are distinct prime numbers. If H is not simple as a Hopf algebra then it is semisolvable. Proof. By assumption, H has a proper normal Hopf subalgebra K. Moreover, by Nichols-Zoeller Theorem [16], dimK divides dimH = pq3. We shall examine every possible dimK. If dimK = q2 or pq then k ⊆ K ⊆ H is a chain such that K and H/HK + are both trivial (see [3, 8]). Hence, H is lower semisolvable. If dimK = q3 then [8] shows that K has a non-trivial central group-like element g. Let L = khgi be the group algebra of the cyclic group hgi generated by g. Then k ⊆ L ⊆ K ⊆ H is a chain such that L, K/KL+ and H/HK + are all trivial (see [22]). Hence, H is lower semisolvable. If dimK = pq2 then [2, Lemma 2.2] and [12, Theorem 5.4.1] show that K has a proper normal Hopf subalgebra L of dimension p, q, pq or q2. Then k ⊆ L ⊆ K ⊆ H is a chain such that L, K/KL+ and H/HK + are all trivial. Hence, H is lower semisolvable. Finally, we consider the case that dimK = p or q. Let L be a proper normal Hopf subalgebra of H/HK + (Notice that H/HK + is not simple). Write K = H/HK + π2−→ L −→ k is a chain such that every map is and L = K/KL+. Then H normal and H coπ1, (K)coπ2 are trivial. Hence, H is upper semisolvable. (cid:3) π1−→ K 4 J. DONG AND L. DAI 2.3. Radford's biproduct. Let A be a semisimple Hopf algebra and let A AYD de- note the braided category of Yetter-Drinfeld modules over A. Let R be a semisimple Yetter-Drinfeld Hopf algebra in A AYD. Denote by ρ : R → A ⊗ R, ρ(a) = a−1 ⊗ a0, and · : A ⊗ R → R, the coaction and action of A on R, respectively. We shall use the notation ∆(a) = a1 ⊗ a2 and SR for the comultiplication and the antipode of R, respectively. Since R is in particular a module algebra over A, we can form the smash product (see [11, Definition 4.1.3]). This is an algebra with underlying vector space R ⊗ A, multiplication is given by (a ⊗ g)(b ⊗ h) = a(g1 · b) ⊗ g2h, for all g, h ∈ A, a, b ∈ R, and unit 1 = 1R ⊗ 1A. Since R is also a comodule coalgebra over A, we can dually form the smash coproduct. This is a coalgebra with underlying vector space R⊗A, comultiplication is given by ∆(a ⊗ g) = a1 ⊗ (a2)−1g1 ⊗ (a2)0 ⊗ g2, for all h ∈ A, a ∈ R, and counit εR ⊗ εA. As observed by D. E. Radford (see [17, Theorem 1]), the Yetter-Drinfeld condi- tion assures that R ⊗ A becomes a Hopf algebra with these structures. This Hopf algebra is called the Radford's biproduct of R and A. We denote this Hopf algebra by R#A and write a#g = a ⊗ g for all g ∈ A, a ∈ R. Its antipode is given by S(a#g) = (1#S(a−1g))(SR(a0)#1), for all g ∈ A, a ∈ R. A biproduct R#A as described above is characterized by the following prop- erty(see [17, Theorem 3]): suppose that H is a finite-dimensional Hopf algebra endowed with Hopf algebra maps ι : A → H and π : H → A such that πι : A → A is an isomorphism. Then the subalgebra R = H coπ has a natural structure of Yetter-Drinfeld Hopf algebra over A such that the multiplication map R#A → H induces an isomorphism of Hopf algebras. Following [20, Proposition 1.6], H ∼= R#A is a biproduct if and only if H ∗ ∼= R∗#A∗ is a biproduct. The following lemma is a special case of [13, Lemma 4.1.9]. Lemma 2.4. Let H be a semisimple Hopf algebra of dimension pq3, where p, q are distinct prime numbers. If gcd(G(H), G(H ∗)) = p, then H ∼= R#kG is a biproduct, where kG is the group algebra of group G of order p, R is a semisimple Yetter-Drinfeld Hopf algebra in kG kGYD of dimension q3. 2.4. Drinfeld double. For a finite-dimensional Hopf algebra H, D(H) = H ∗cop ⊲⊳ H will denote the Drinfeld double of H. D(H) is a Hopf algebra with underlying vector space H ∗cop ⊗ H. More details on D(H) can be found in [11, Section 10.3]. The following theorem follows directly from [18, Proposition 9,10]. Theorem 2.5. Suppose that H is a finite-dimensional Hopf algebra. (1) The map G(H ∗) × G(H) → G(D(H)), given by (η, g) 7→ η ⊲⊳ g, is a group isomorphism. (2) Every group-like element of D(H)∗ is of the form g ⊗ η, where g ∈ G(H) and η ∈ G(H ∗). Moreover, g ⊗ η ∈ G(D(H)∗) if and only if η ⊲⊳ g is in the center of D(H). SEMISIMPLE HOPF ALGEBRAS 5 Corollary 2.6. Suppose that H is a finite-dimensional Hopf algebra such that G(D(H)∗) is non-trivial. If gcd(G(H), G(H ∗)) = 1 then H or H ∗ has a non- trivial central group-like element. Proof. Let 1 6= g ⊗ η ∈ G(D(H)∗). We may assume that 1 6= g ∈ G(H), since otherwise η ∈ G(H ∗) would be a non-trivial central group-like element, and sim- ilarly we may assume that ε 6= η ∈ G(H ∗). Since gcd(G(H), G(H ∗)) = 1, the order of g and η are different. Assume that the order of g is n. Then (g ⊗ η)n = gn ⊗ ηn = 1 ⊗ ηn 6= 1 ⊗ ε implies that ηn ⊲⊳ 1 is in the center of D(H). Hence, ηn is a non-trivial central group-like element in G(H ∗). Similarly, we can prove that G(H) also has a non-trivial central group-like element. (cid:3) 3. Semisimple Hopf algebras of dimension pq3 Lemma 3.1. Let H be a semisimple Hopf algebra of dimension pq3, where p < q are prime numbers. If H has a Hopf subalgebra K of dimension q3 then H is lower semisolvable. Proof. Since the index [H : K] = p is the smallest prime number dividing dimH, the result in [7] shows that K is a normal Hopf algebra of H. The lemma then follows from Proposition 2.3. (cid:3) In the rest of this section, p, q will be distinct prime numbers such that p2 < q, and H will be a non-trivial semisimple Hopf algebra of dimension pq3. Recall that a semisimple Hopf algebra A is called of Frobenius type if the di- mensions of the simple A-modules divide the dimension of A. Kaplansky conjec- tured that every finite-dimensional semisimple Hopf algebra is of Frobenius type [5, Appendix 2]. It is still an open problem. However, many examples show that a positive answer to Kaplansky's conjecture would be very helpful in the classification of semisimple Hopf algebras. By [2, Lemma 2.2], H is of Frobenius type and G(H ∗) 6= 1. Therefore, the dimension of a simple H-module can only be 1, p, q or pq. It follows that we have an equation pq3 = G(H ∗) + p2a + q2b + p2q2c, (3.1) where a, b, c are the numbers of non-isomorphic simple H-modules of dimension p, q and pq, respectively. By Nichols-Zoeller Theorem [16], the order of G(H ∗) divides dimH. We shall give some results concerning the order of G(H ∗). Lemma 3.2. The order of G(H ∗) can not be q. Proof. Suppose on the contrary that G(H ∗) = q. Let χ be an irreducible character of degree p. By Lemma 2.1 (1) and the fact that G[χ] is a subgroup of G(H ∗), we It follows that the decomposition of χχ∗ (2.1) know that G[χ] = {ε} is trivial. gives rise to a contradiction, since p2 < q. Therefore, a = 0 and equation (3.1) is pq3 = q + q2b + p2q2c, which is impossible. (cid:3) Lemma 3.3. If G(H ∗) = q2 then a = 0 and b 6= 0. Proof. A similar argument as in Lemma 3.2 shows that a = 0. Therefore, equation (3.1) is pq3 = q2 + q2b + p2q2c. Obviously, b 6= 0, otherwise a contradiction will occur. (cid:3) Lemma 3.4. If G(H ∗) = q3 then H is upper semisolvable. 6 J. DONG AND L. DAI Proof. It follows from Lemma 3.1. (cid:3) Lemma 3.5. If G(H ∗) = q2 and H has a Hopf subalgebra K of dimension pq2 then H is lower semisolvable. Proof. Considering the map π : H ∗ → K ∗ obtained by transposing the inclusion K ⊆ H, we have dim(H ∗)coπ = q (see Section 2.2). By Lemma 3.3, the dimension of every irreducible left coideal of H ∗ is 1, q or pq. Therefore, by Lemma 2.2, as a left coideal of H ∗, (H ∗)coπ decomposes in the form (H ∗)coπ = khgi, where hgi is a subgroup of G(H ∗) generated by g which is of order q. Therefore, (H ∗)coπ is a normal Hopf subalgebra of H ∗. The lemma then follows from Proposition 2.3. (cid:3) Lemma 3.6. If G(H ∗) = p then H is either semisolvable, or isomorphic to a Radford's biproduct H ∼= R#kG, where kG is the group algebra of G of order p, R is a semisimple Yetter-Drinfeld Hopf algebra in kG kGYD of dimension q3. Proof. Notice that G(D(H)∗) is not trivial by [2, Lemma 2.2]. If G(H) = q2 or q3 then H is not simple by Corollary 2.6. Hence, H is semisolvable. If G(H) = p, pq or pq2 then, by Lemma 2.4, H is isomorphic to a Radford's biproduct H ∼= R#kG, where kG is the group algebra of G of order p, R is a semisimple Yetter-Drinfeld Hopf algebra in kG (cid:3) kGYD of dimension q3. 4. Semisimple Hopf algebras of dimension 2q3 In this section, H will be a non-trivial semisimple Hopf algebra of dimension 2q3, where q is a prime number. We shall discuss the structure of H. When q = 2, the structure of H is given in [6]. When q = 3, the structure of H is given in [13, Chapter 12]. Therefore, in the rest of this section, we always assume that q > 3. Using notations from Section 3, we have 2q3 = G(H ∗) + 4a + q2b + 4q2c. (4.1) By the results in Section 3, it suffices to consider the cases that G(H ∗) = 2q, 2q2 and q2. Lemma 4.1. If G(H ∗) = 2q or 2q2 then H is either semisolvable, or isomorphic to a Radford's biproduct H ∼= R#kG, where kG is the group algebra of G of order 2, R is a semisimple Yetter-Drinfeld Hopf algebra in kG kGYD of dimension q3. Proof. By Lemma 3.4 and 3.6, we may assume that G(H) 6= 2 and q3. First, if gcd(G(H), G(H ∗)) = 2 then the lemma follows from Lemma 2.4. Second, we assume that G(H ∗) = 2q and G(H) = q2. Notice that a 6= 0, otherwise equation (4.1) will give rise to a contradiction. Hence, G(H ∗) ∪ X2 spans a standard subalgebra of R(H). It follows that H has a quotient Hopf algebra K of dimension 2q + 4a. By Nichols-Zoeller Theorem [16], dimK = 2q2 or 2q3. If dimK = 2q2 then the duality of Lemma 3.5 proves the lemma. If dimK = 2q3 then H is of type (1, 2q; 2, a) as an algebra. Then [1, Theorem 6.4] shows that either H or H ∗ must contain a non-trivial central group-like element. The lemma then follows from Proposition 2.3. Finally, we assume that G(H ∗) = 2q2 and G(H) = q2. In this case, the duality (cid:3) of Lemma 3.5 proves the lemma. Lemma 4.2. If G(H ∗) = q2 then H is semisolvable. SEMISIMPLE HOPF ALGEBRAS 7 Proof. By the discussion above, we may assume that G(H) = q2. Considering the map π : H → (kG(H ∗))∗ obtained by transposing the inclusion kG(H ∗) ⊆ H ∗, we have dimH coπ = 2q. By Lemma 3.3, the dimension of every irreducible left coideal of H is 1, q or 2q. Therefore, by Lemma 2.2, as a left coideal of H, H coπ decomposes in the form H coπ = khgi ⊕ V , where khgi is the group algebra of a subgroup hgi of G(H) generated by g which is of order q, and V is an irreducible left coideal of H of dimension q. Since gV and V g are irreducible left coideals of H isomorphic to V , and gV, V g are contained in H coπ, we have gV = V = V g. Then [13, Corollary 3.5.2] shows that khgi is a normal Hopf subalgebra of k[C], where C is the simple subcoalgebra of H containing V , and k[C] is a Hopf subalgebra of H generated by C as an algebra. Clearly, dimk[C] ≥ q + q2. Moreover, by Nichols-Zoeller Theorem [16], dimk[C] = 2q3, q3 or 2q2. If dimk[C] = 2q3 then k[C] = H and khgi is a normal Hopf subalgebra of H. The lemma then follows from Proposition 2.3. If dimk[C] = q3 then the lemma follows from Lemma 3.1. If dimk[C] = 2q2 then the lemma follows from Lemma 3.5. (cid:3) From the discussion in Section 3 and 4, we obtain the main theorem. Theorem 4.3. Let H will be a non-trivial semisimple Hopf algebra of dimension 2q3, where q > 3 is a prime number. Then (1) If gcd(G(H), G(H ∗)) = 2 then H is isomorphic to a Radford's biproduct H ∼= R#kG, where kG is the group algebra of G of order 2, R is a semisimple Yetter-Drinfeld Hopf algebra in kG kGYD of dimension q3. (2) In all other cases, H is semisolvable. As an immediate consequence of Theorem 4.3, we have the following result. Corollary 4.4. If H is simple as a Hopf algebra then H is isomorphic to a Rad- ford's biproduct H ∼= R#kG, where kG is the group algebra of G of order 2, R is a semisimple Yetter-Drinfeld Hopf algebra in kG kGYD of dimension q3. References [1] J. Bichon, S. Natale, Hopf algebra deformations of binary polyhedral groups. Trans. Groups, DOI: 10.1007/s00031-011-9133-x. [2] J. Dong, Structure of semisimple Hopf algebras of dimension p2q2. arXiv:1009.3541v2, to appear in Communications in Algebra. [3] P. Etingof and S. Gelaki, Semisimple Hopf algebras of dimension pq are trivial. J. Algebra 210(2), 664 -- 669 (1998). [4] P. Etingof, D. Nikshych and V. Ostrik, Weakly group-theoretical and solvable fusion cate- gories. Adv. Math. 226 (1), 176 -- 505 (2011). [5] I. Kaplansky, Bialgebras. Chicago, University of Chicago Press 1975. [6] Y. Kashina , Classification of semisimple Hopf algebras of dimension 16, J. Algebra 232, 617 -- 663(2000). [7] T. Kobayashi and A. Masuoka, A result extended from groups to Hopf algebras. Tsukuba J. Math. 21(1), 55 -- 58 (1997). [8] A. Masuoka, The pn theorem for semisimple Hopf algebras. Proc. Amer. Math. Soc. 124, 735 -- 737 (1996). [9] A. Masuoka, Self-dual Hopf algebras of dimension p3 obtained by extension. J. Algebra 178, 791-806 (1995) [10] S. Montgomery and S. Whiterspoon, Irreducible representations of crossed products. J. Pure Appl. Algebra 129, 315 -- 326 (1998). [11] S. Montgomery, Hopf algebras and their actions on rings. CBMS Reg. Conf. Ser. Math. 82. Providence. Amer. Math. Soc. 1993. 8 J. DONG AND L. DAI [12] S. Natale, On semisimple Hopf algebras of dimension pqr. Algebras Represent. Theory 7 (2), 173 -- 188 (2004). [13] S. Natale, Semisolvability of semisimple Hopf algebras of low dimension. Mem. Amer. Math. Soc. 186 (2007). [14] S. Natale, On semisimple Hopf algebras of dimension pq2. J. Algebra 221(2), 242 -- 278 (1999). [15] W. D. Nichols and M. B. Richmond, The Grothendieck group of a Hopf algebra. J. Pure Appl. Algebra 106, 297 -- 306(1996). [16] W. D. Nichols and M. B. Zoelle, A Hopf algebra freeness theorem. Amer. J. Math. 111(2), 381 -- 385 (1989). [17] D. Radford, The structure of Hopf algebras with a projection. J. Algebra 92, 322 -- 347 (1985). [18] D. Radford, Minimal Quasitriangular Hopf algebras. J. Algebra, 157, 285 -- 315 (1993). [19] H.-J. Schneider, Normal basis and transitivity of crossed products for Hopf algebras. J. Al- gebra 152, 289 -- 312 (1992). [20] Y. Sommerhauser, Yetter-Drinfeld Hopf algebras over groups of prime order, Lectures Notes in Math. 1789, Springer-Verlag (2002). [21] M. E. Sweedler, Hopf Algebras. New York, Benjamin 1969. [22] Y. Zhu, Hopf algebras of prime dimension. Internat. Math. Res. Notices 1, 53 -- 59 (1994). College of Engineering, Nanjing Agricultural University, Nanjing 210031, Jiangsu, People's Republic of China E-mail address, J. Dong: [email protected] E-mail address, L. Dai: [email protected]
1909.10210	1	1909	2019-09-23T08:21:15	A power Cayley-Hamilton identity for nxn matrices over a Lie nilpotent ring of index k	[ "math.RA" ]	For an nxn matrix A over a Lie nilpotent ring R of index k, we prove that an invariant "power" Cayley-Hamilton identity of degree (n^2)2^{k-2} holds. The right coefficients are not uniquely determined by A, and the cosets lambda_i+D, with D the double commutator ideal R[[R,R],R]R of R, appear in the so-called second right characteristic polynomial of the natural image of A in the nxn matrix ring M_{n}(R/D) over the factor ring R/D.	math.RA	math	A power Cayley-Hamilton identity for n × n matrices over a Lie nilpotent ring of index k Jeno Szigeti, Szilvia Szil´agyi, and Leon van Wyk* Abstract. For an n × n matrix A over a Lie nilpotent ring R of index k, with k ≥ 2, we prove that an invariant "power" Cayley-Hamilton identity −1λ(2) n2 −1 + An2 (cid:16)Inλ(2) 0 + Aλ(2) 1 + · · · + An2 n2 (cid:17)2k−2 λ(2) of degree n22k−2 holds. The right coefficients λ(2) ∈ R, 0 ≤ i ≤ n2 are i not uniquely determined by A, and the cosets λ(2) i + D, with D the double commutator ideal R[[R, R], R]R of R, appear in the so-called second right characteristic polynomial pA,2(x) of the natural image A of A in the n × n matrix ring Mn(R/D) over the factor ring R/D: pA,2(x) = (λ(2) 1 + D)x + · · · + (λ(2) 0 + D) + (λ(2) n2 + D)xn2 −1 + (λ(2) + D)xn2 = 0 n2 −1 . 1. INTRODUCTION The Cayley-Hamilton theorem and the corresponding trace identity play a fun- damental role in proving classical results about the polynomial and trace identities of the n × n matrix algebra Mn(K) over a field K (see, for example, [Dr], [DrF] and [Row]). In case of char(K) = 0, Kemer's pioneering work (see [K]) on the T-ideals of associative algebras revealed the importance of the identities satisfied by the n × n matrices over the Grassmann (exterior) algebra E = K hv1, v2, ..., vi, ... vivj + vjvi = 0 for all 1 ≤ i ≤ ji generated by the infinite sequence of anticommutative indeterminates (vi)i≥1. Accordingly, the importance of matrices over non-commutative rings features prominently in the theory of PI-rings; indeed, this fact has been obvious for a long time in other branches of algebra, for example, in the structure theory of semisimple rings. Thus any Cayley-Hamilton type identity for such matrices seems to be of general interest. 2010 Mathematics Subject Classification. 15A15,15A24,15A33,16S50. Key words and phrases. Lie nilpotent ring, commutator ideals, the Lie nilpotent right Cayley- Hamilton identity. The first named author was partially supported by the National Research, Development and Innovation Office of Hungary (NKFIH) K119934. Corresponding author. 1 2 JEN O SZIGETI, SZILVIA SZIL ´AGYI, AND LEON VAN WYK In the general case (when R is an arbitrary non-commutative ring with 1) Par´e and Schelter proved (see [PaSch]) that a matrix A ∈ Mn(R) satisfies a monic identity in which the leading term is Am for some large integer m, i.e., m ≥ 22n−1 . The other summands in the identity are of the form r0Ar1Ar2 · · · rl−1Arl, with left scalar coefficient r0 ∈ R, right scalar coefficient rl ∈ R and "sandwich" scalar coefficients r2, . . . , rl−1 ∈ R. An explicit monic identity for 2 × 2 matrices arising from the argument of [PaSch] was given by Robson in [Rob]. Further results in this direction can be found in [Pe1] and [Pe2]. Obviously, by imposing extra algebraic conditions on the base ring R, we can expect "stronger" identities in Mn(R). A number of examples show that certain polynomial identities satisfied by R can lead to "canonical" constructions providing invariant Cayley-Hamilton identities for A of degree much lower than 22n−1 . If R satisfies the polynomial identity [[[. . . [[x1, x2], x3], . . .], xk], xk+1] = 0 of Lie nilpotency of index k (with [x, y] = xy − yx), then for a matrix A ∈ Mn(R), a left (and right) Cayley-Hamilton identity of degree nk was constructed in [S1] (see also [MaMeSvW]). Since E is Lie nilpotent of index k = 2, this identity for a matrix A ∈ Mn(E) is of degree n2. In [Do], Domokos considered a slightly modified version of the mentioned iden- tity, in which the left (as well as the right) coefficients are invariant under the conjugate action of GLn(K) on Mn(E). For a 2 × 2 matrix A ∈ M2(E), the left scalar coefficients of this Cayley-Hamilton identity are expressed as polynomials (over K) of the traces tr(A), tr(A2) and tr(A3). If 1 2 ∈ R and R satisfies the so-called weak Lie solvability identity [[x, y], [x, z]] = 0, then for a 2 × 2 matrix A ∈ M2(R), a Cayley-Hamilton trace identity (of degree 4 in A) with sandwich coefficients was exhibited in [MeSvW]. If R satisfies the identity [x1, x2, ..., x2s]solv = 0 of general Lie solvability, then a recursive construction (also in [MeSvW]) gives a similar Cayley-Hamilton trace identity (the degree of which depends on s) for a matrix A ∈ M2(R). In the present paper we consider an n × n matrix A ∈ Mn(R) over a ring R (with 1) satisfying the identity [[x1, y1], z1][[x2, y2], z2] · · · [[xt, yt], zt] = 0, and we prove that an invariant "power" Cayley-Hamilton identity of the form (cid:16)Inλ(2) 0 + Aλ(2) 1 + · · · + An2−1λ(2) n2−1 + An2 λ(2) n2(cid:17)t = 0 holds, with certain right coefficients λ(2) i ∈ R, 0 ≤ i ≤ n2 − 1, and λ(2) n2 = n(cid:8)(n − 1)!(cid:9)1+n , which are only partially determined by A. The cosets λ(2) i + D, with D the double commutator ideal R[[R, R], R]R of R, appear in the second right characteristic poly- nomial pA,2(x) of the natural image A ∈ Mn(R/D) of A over the factor ring R/D: pA,2(x) = (λ(2) 0 + D) + (λ(2) 1 + D)x + · · · + (λ(2) n2−1 + D)xn2 −1 + (λ(2) n2 + D)xn2 . A POWER CAYLEY-HAMILTON IDENTITY 3 We note that [[x1, y1], z1][[x2, y2], z2] · · · [[xt, yt], zt] = 0 is a typical identity of the ring Ut(R) of t × t upper triangular matrices over a ring R satisfying the identity [[x, y], z] = 0 (i.e., Lie nilpotency of index 2). Finally, using a theorem of Jennings (see [J]), we prove that if R is Lie nilpotent of index k, then an identity of the form (cid:16)Inλ(2) 0 + Aλ(2) 1 + · · · + An2−1λ(2) n2−1 + An2 λ(2) n2(cid:17)2k−2 = 0 (∗) holds for A ∈ Mn(R). The total degree of this identity (in A) is n22k−2, a much smaller integer than the degree nk of A in the right Cayley-Hamilton identity Inλ(k) 0 + Aλ(k) 1 + · · · + Ank −1λ(k) nk −1 + Ank λ(k) nk = 0 (∗∗) arising from the k-th right characteristic polynomial pA,k(x) = λ(k) 0 + λ(k) 1 x + · · · + λ(k) nk −1xnk −1 + λ(k) nk xnk ∈ R[x] of A (see [S1] and [SvW1]). The advantage of (∗∗) is that all the coefficients are on the right side of the powers of A, while the expansion of the power in (∗) yields a sum of products of the form Ai1 λi1 Ai2 λi2 · · · Ais λis , with s = 2k−2. In order to provide a self-contained treatment, we present the necessary pre- requisites in sections (2) and (3). 2. SOME RESULTS ON LIE NILPOTENT RINGS Let R be a ring, and let [x, y] = xy − yx denote the additive commutator of the elements x, y ∈ R. It is well-known that (R, +, [ , ]) is a Lie ring, [y, x] = −[x, y] and [[x, y], z] + [[y, z], x] + [[z, x], y] = 0 (the Jacobian identity). For a sequence x1, x2, . . . , xk of elements in R we use the notation [x1, x2, . . . , xk]k for the left normed commutator (Lie-)product: [x1]1 = x1 and [x1, x2, . . . , xk]k = [. . . [[x1, x2], x3], . . . , xk]. Clearly, we have [x1, x2, . . . , xk, xk+1]k+1 = [[x1, x2, . . . , xk]k, xk+1] = [[x1, x2], x3, . . . , xk, xk+1]k. A ring R is called Lie nilpotent of index k (or having property Lk) if [x1, x2, . . . , xk, xk+1]k+1 = 0 is a polynomial identity on R. If R has property Lk, then [x1, x2, . . . , xk]k is central for all x1, x2, . . . , xk ∈ R. A concise proof of Theorem 2.1 due to Jennings can be found in [SvW2]. 2.1. Theorem ([J]). Let k ≥ 3 be an integer and R be a ring with Lk. Then [x1, x2, . . . , xk]k · [y1, y2, . . . , yk]k = 0 for all xi, yi ∈ R, 1 ≤ i ≤ k. Thus the two-sided ideal N = R(cid:8)[x1, x2, . . . , xk]k xi ∈ R, 1 ≤ i ≤ k(cid:9) =(cid:8)[x1, x2, . . . , xk]k xi ∈ R, 1 ≤ i ≤ k(cid:9)R generated by the (central) elements [x1, x2, . . . , xk]k is nilpotent, with N 2 = {0}. 4 JEN O SZIGETI, SZILVIA SZIL ´AGYI, AND LEON VAN WYK* 2.2. Corollary ([J]). If R is a ring with Lk ( k ≥ 2), then the double commutator ideal D = R[[R, R], R]R = {P1≤i≤mri[[ai, bi], ci]si ri, ai, bi, ci, si ∈ R, 1 ≤ i ≤ m} ⊳ R is nilpotent, with D2k−2 = {0}. Proof. This follows from Theorem 2.1 by an easy induction on k. (cid:3) 3. THE LIE NILPOTENT CAYLEY-HAMILTON THEOREM A Lie nilpotent analogue of classical determinant theory was developed in [S1]; further details can be found in [Do], [S2] and [SvW1]. Here we present the basic definitions and results about the sequences of right determinants and right char- acteristic polynomials, including the so-called Lie nilpotent right Cayley-Hamilton identities. Let R be an arbitrary (possibly non-commutative) ring or algebra with 1, and let Sn = Sym({1, . . . , n}) denote the symmetric group of all permutations of the set {1, 2, . . . , n}. If A = [ai,j] is an n × n matrix over R, then the element sdet(A) = Xτ,ρ∈Sn = Xα,β∈Sn sgn(ρ)aτ (1),ρ(τ (1)) · · · aτ (t),ρ(τ (t)) · · · aτ (n),ρ(τ (n)) sgn(α)sgn(β)aα(1),β(1) · · · aα(t),β(t) · · · aα(n),β(n) of R is called the symmetric determinant of A. The (r, s)-entry of the symmetric adjoint matrix A∗ = [a∗ r,s] of A is defined as follows: a∗ r,s =Xτ,ρ =Xα,β sgn(ρ)aτ (1),ρ(τ (1)) · · · aτ (s−1),ρ(τ (s−1))aτ (s+1),ρ(τ (s+1)) · · · aτ (n),ρ(τ (n)) sgn(α)sgn(β)aα(1),β(1) · · · aα(s−1),β(s−1)aα(s+1),β(s+1) · · · aα(n),β(n) , where the first sum is taken over all τ, ρ ∈ Sn with τ (s) = s and ρ(s) = r, while the second sum is taken over all α, β ∈ Sn with α(s) = s and β(s) = r. We note that the (r, s) entry of A∗ is exactly the signed symmetric determinant (−1)r+ssdet(As,r) of the (n − 1) × (n − 1) minor As,r of A arising from the deletion of the s-th row and the r-th column of A. The trace tr(M ) of a matrix M ∈ Mn(R) is the sum of the diagonal entries of M . In spite of the fact that the well known identity tr(AB) = tr(BA) is no longer valid for matrices A, B ∈ Mn(R) over a non-commutative R, we still have (see [SvW1]) sdet(A) = tr(AA∗) = tr(A∗A). If R is commutative, then sdet(A) = n!det(A) and A∗ = (n − 1)!adj(A), where det(A) and adj(A) denote the ordinary determinant and adjoint, respectively, of A. The right adjoint sequence (Pk)k≥1 of a matrix A ∈ Mn(R) is defined by the following recursion: P1 = A∗ and Pk+1 = (AP1 · · · Pk)∗ A POWER CAYLEY-HAMILTON IDENTITY 5 for k ≥ 1. The k-th right adjoint of A is defined as radj(k)(A) = nP1 · · · Pk. The k-th right determinant of A is the trace of AP1 · · · Pk: rdet(k)(A) = tr(AP1 · · · Pk). The following theorem shows that radj(k)(A) and rdet(k)(A) can play a role sim- ilar to that played by the ordinary adjoint and determinant, respectively, in the commutative case. 3.1. Theorem ([S1], [SvW1]). If index k, then for a matrix A ∈ Mn(R) we have 1 n ∈ R and the ring R is Lie nilpotent of Aradj(k)(A) = nAP1 · · · Pk = rdet(k)(A)In. The above Theorem 3.1 is not used explicitly in the sequel, however it helps our understanding and serves as a starting point in the proof of Theorem 3.3. Let R[x] denote the ring of polynomials in the single commuting indetermi- nate x, with coefficients in R. The k-th right characteristic polynomial of A is the k-th right determinant of the n × n matrix xIn − A in Mn(R[x]): pA,k(x) = rdet(k)(xIn − A). 3.2. Proposition ([S1], [SvW1]). The k-th right characteristic polynomial pA,k(x) ∈ R[x] of A ∈ Mn(R) is of the form pA,k(x) = λ(k) 0 + λ(k) 1 x + · · · + λ(k) nk −1xnk −1 + λ(k) nk xnk , where λ(k) 0 , λ(k) 1 , . . . , λ(k) nk −1, λ(k) nk ∈ R and λ(k) nk = n(cid:8)(n − 1)!(cid:9)1+n+n2+···+nk−1 . 3.3. Theorem ([S1], [SvW1]). If index k, then a right Cayley-Hamilton identity 1 n ∈ R and the ring R is Lie nilpotent of (A)pA,k = Inλ(k) 0 + Aλ(k) 1 + · · · + Ank−1λ(k) nk −1 + Ank λ(k) nk = 0 with right scalar coefficients holds for A ∈ Mn(R). We also have (A)u = 0, where u(x) = pA,k(x)h(x) and h(x) ∈ R[x] is arbitrary. 3.4. Theorem ([Do]). If 1 2 ∈ R and the ring R is Lie nilpotent of index 2, then for a 2 × 2 matrix A ∈ M2(R) the right Cayley-Hamilton identity in the above 3.3 can be written in the following trace form: (A)pA,2 =I2(cid:18)1 tr(A2)tr2(A)+(cid:2)tr(A3), tr(A)(cid:3)(cid:19)+ A(cid:16)tr(A)tr(A2)+tr(A2)tr(A)−2tr3(A)(cid:17)+A2(cid:16)4tr2(A)−2tr(A2)(cid:17)−A3(cid:16)4tr(A)(cid:17)+2A4 = 0. tr2(A)tr(A2)− tr2(A2)+ tr4(A)+ 1 4 5 4 1 2 2 3.5. Corollary ([Do]). If 1 for every A ∈ M2(R), 2 ∈ R and the ring R is Lie nilpotent of index 2, then, tr(A) = tr(A2) = 0 implies that A4 = 0. 6 JEN O SZIGETI, SZILVIA SZIL ´AGYI, AND LEON VAN WYK* 4. MATRICES OVER A RING WITH [[x1, y1], z1][[x2, y2], z2] · · · [[xt, yt], zt] = 0 We shall make use of the following well-known fact. 4.1. Proposition. If identity on a ring R, then Dt = {0}, with D the ideal R[[R, R], R]R of R. [[x1, y1], z1][[x2, y2], z2] · · · [[xt, yt], zt] = 0 is a polynomial 4.2. Theorem. If 1 2 ∈ R and A ∈ Mn(R) is a matrix over a ring R satisfying the polynomial identity [[x1, y1], z1][[x2, y2], z2] · · · [[xt, yt], zt] = 0, then an invariant "power" Cayley-Hamilton identity of the form (cid:16)Inλ(2) 0 + Aλ(2) 1 + · · · + An2−1λ(2) n2−1 + An2 holds, with certain right coefficients λ(2) i ∈ R, 0 ≤ i ≤ n2 − 1, and λ(2) λ(2) n2(cid:17)t = 0 n2 = n(cid:8)(n − 1)!(cid:9)1+n (only partially determined by A). The cosets λ(2) appear in the second right characteristic polynomial pA,2(x) of the natural image A ∈ Mn(R/D) of A over the factor ring R/D: pA,2(x) = (λ(2) i + D with D = R[[R, R], R]R ⊳ R 1 +D)x+· · ·+(λ(2) 0 +D)+(λ(2) n2−1+D)xn2 n2 +D)xn2 −1+(λ(2) ∈ (R/D)[x]. Proof. Consider the factor ring R/D, where D = R[[R, R], R]R ⊳ R is the double commutator ideal. If A = [ai,j] ∈ Mn(R), then we use the notation A = [ai,j + D] for the image of A in Mn(R/D). Since R/D is Lie nilpotent of index 2, Theorem 3.3 implies that, in Mn(R/D), (A)pA,2 = In(λ(2) where 1 +D)+· · ·+(A)n2−1(λ(2) n2 −1+D)+(A)n2 0 +D)+A(λ(2) n2 +D) = 0, (λ(2) pA,2(x) = rdet(k)(xIn − A) = (λ(2) 0 + D) + (λ(2) 1 + D)x + · · · + (λ(2) n2−1 + D)xn2−1 + (λ(2) n2 + D)xn2 is the second right characteristic polynomial of A in (R/D)[x]. Clearly, Inλ(2) 0 + D) + A(λ(2) 0 + Aλ(2) 1 + · · · + An2−1λ(2) 1 + D) + · · · + (A)n2−1(λ(2) n2−1 + An2 λ(2) n2 −1 + D) + (A)n2 n2 = In(λ(2) implies that (λ(2) n2 + D) = 0 Inλ(2) 0 + Aλ(2) 1 + · · · + An2−1λ(2) n2−1 + An2 λ(2) n2 ∈ Mn(D). Now Dt = {0} is a consequence of Proposition 4.1, whence (Mn(D))t = {0} and (cid:16)Inλ(2) 0 + Aλ(2) 1 + · · · + An2−1λ(2) n2−1 + An2 λ(2) n2(cid:17)t = 0 follows. (cid:3) 4.3. Remark. If [x1, y1][x2, y2] · · · [xt, yt] = 0 is a polynomial identity on R and A ∈ Mn(R), then using the commutator ideal T = R[R, R]R and the natural image eA ∈ Mn(R/T ) of A over the commutative ring R/T , a similar argument as in the proof of Theorem 4.2 gives that (cid:16)Inλ(1) 0 + Aλ(1) 1 + · · · + An−1λ(1) n−1 + Anλ(1) = 0 n (cid:17)t A POWER CAYLEY-HAMILTON IDENTITY 7 holds, where p eA,1(x) = (λ(1) 0 + T )+ (λ(1) 1 + T )x+ · · ·+ (λ(1) n−1 + T )xn−1 + (λ(1) n + T )xn is the n! times scalar multiple of the classical characteristic polynomial of eA in (R/T )[x] with λ(1) n = n!. 4.4. Theorem. If 1 of index k, then an invariant "power" Cayley-Hamilton identity of the form 2 ∈ R and A ∈ Mn(R) is a matrix over a Lie nilpotent ring R (cid:16)Inλ(2) 0 + Aλ(2) 1 + · · · + An2−1λ(2) n2−1 + An2 λ(2) n2(cid:17)2k−2 = 0 n2 = n(cid:8)(n − 1)!(cid:9)1+n holds, with certain right coefficients λ(2) i ∈ R, 0 ≤ i ≤ n2 − 1, and λ(2) (only partially determined by A). The cosets λ(2) appear in the second right characteristic polynomial pA,2(x) of the natural image A ∈ Mn(R/D) of A over the factor ring R/D: pA,2(x) = (λ(2) i + D with D = R[[R, R], R]R ⊳ R n2−1+D)xn2−1+(λ(2) 1 +D)x+· · ·+(λ(2) 0 +D)+(λ(2) n2 +D)xn2 ∈ (R/D)[x]. Proof. According to Jennings's result (Corollary 2.2), the double commutator ideal D = R[[R, R], R]R =(cid:8)P1≤i≤mri[[ai, bi], ci]si ri, ai, bi, ci, si ∈ R, 1 ≤ i ≤ m(cid:9) ⊳ R = {0}. Thus the application of Theorem 4.2 gives our is nilpotent, with D2k−2 identity. (cid:3) If k = 2, then R[[R, R], R]R = {0}, and the identity in Theo- 4.5. Remark. rem 4.4 remains the same as the Lie nilpotent right Cayley-Hamilton identity in Theorem 3.3. 4.6. Remark. The Grassmann algebra E = K hv1, v2, ..., vi, ... vivj + vjvi = 0 for all 1 ≤ i ≤ ji over a field K (with 2 6= 0) has property L2, and [v1, v2] · [v3, v4] · · · · · [v2t−1, v2t] = 2tv1v2 · · · v2t 6= 0 shows that L2 does not imply the identity [x1, y1][x2, y2] · · · [xt, yt] = 0 for any t. Thus the identity mentioned in Remark 4.3 cannot be used directly to derive new identities for matrices over a Lie nilpotent ring of index k ≥ 2. However, as the referee pointed out, the following (weak) version of Latyshev's theorem provides a possibility to use Remark 4.3 in order to obtain another remarkable "power" Cayley-Hamilton identity. Theorem ([L]). If S is a Lie nilpotent algebra (over an infinite field) of index k, generated by m elements, then there exists an integer d = d(k, m) such that S satisfies the polynomial identity [x1, y1][x2, y2] · · · [xd, yd] = 0. (In the original version S satisfies a so-called nonmatrix polynomial identity.) If A ∈ Mn(R) is a matrix over a Lie nilpotent algebra (over an infinite field) R of index k, then A ∈ Mn(S), where S is the (unitary) subalgebra generated by the n2 8 JEN O SZIGETI, SZILVIA SZIL ´AGYI, AND LEON VAN WYK* entries of A. Thus [x1, y1][x2, y2] · · · [xd, yd] = 0 is a polynomial identity on S with d = d(k, n2) and Remark 4.3 gives that an identity (cid:16)Inλ(1) 0 + Aλ(1) 1 + · · · + An−1λ(1) n−1 + Anλ(1) n (cid:17)d(k,n2) = 0 of degree nd(k, n2) holds. Unfortunately, our knowledge about d(k, n2) is very limited, the fact that d(2, 4) = 3 was mentioned by the referee. 4.7. Remark. If R is an algebra over a field K of characteristic zero, then the invariance of the identities in 4.2 and 4.4 means that pT −1AT ,2(x) = pA,2(x) holds for any T ∈ GLn(K) (see [Do]). 4.8. Corollary. If 1 every A ∈ M2(R), 2 ∈ R and the ring R is Lie nilpotent of index k, then, for tr(A) = tr(A2) = 0 implies that A2k = 0. Proof. Using D = R[[R, R], R]R ⊳ R, A ∈ M2(R/D) and tr(A) = tr(A) + D = 0, tr((A)2) = tr(A2) = tr(A2) + D = 0, the application of Corollary 3.5 ensures that A4 =(cid:0)A(cid:1)4 = {0}) gives that A2k of D (D2k−2 = 0. (cid:3) =(cid:0)A4(cid:1)2k−2 = 0. Thus the nilpotency 4.9. Remark. According to the following important observation of the referee, the use of Latyshev's theorem gives an n × n variant of Corollary 4.8. If A ∈ Mn(R) is a matrix over a Lie nilpotent algebra (over a field K of characteristic zero) R of index k, then we prove that tr(A) = tr(A2) = · · · = tr(An) = 0 implies that And(k,n2) = 0. Indeed, A ∈ Mn(S), where S ⊆ R is the (unitary) subalgebra of R generated by the entries of A. Now consider the natural image eA ∈ Mn(S/S[S, S]S) of A. The application of the well-known fact that implies that fAn = (eA)n =e0 (it is a consequence of the Newton trace formulae for the coefficients of the characteristic polynomial p eA,1(x) ∈ (S/S[S, S]S)[x], where the factor S/S[S, S]S is a commutative algebra over K). Since (S[S, S]S)d(k,n2) = {0} by Latyshev's theorem and An ∈ Mn(S[S, S]S), we obtain the desired equality. tr(eA) = tr((eA)2) = · · · = tr((eA)n) =e0 ACKNOWLEDGMENT The authors thank the referee for the valuable report and the important contribu- tions mentioned in Remarks 4.6 and 4.9. REFERENCES [Do] M. Domokos, Cayley-Hamilton theorem for 2 × 2 matrices over the Grassmann algebra, J. Pure Appl. Algebra 133 (1998), 69-81. [Dr] V. Drensky, Free algebras and PI-algebras. Graduate course in algebra. Springer- Verlag Singapore, Singapore, 2000. A POWER CAYLEY-HAMILTON IDENTITY 9 [DrF] V. Drensky and E. Formanek, Polynomial identity rings. Advanced Courses in Mathematics. CRM Barcelona. Birkhauser-Verlag, Basel, 2004. [J] S. A. Jennings, On rings whose associated Lie rings are nilpotent, Bull. Amer. Math. Soc. 53 (1947), 593-597. [K] A. R. Kemer, Ideals of identities of associative algebras. Translated from the Russian by C. W. Kohls. Translations of Math. Monographs, 87. American Math- ematical Society, Providence, RI, 1991. [L] V. N. Latyshev, Generalization of the Hilbert theorem on the finiteness of bases (Russian), Sib. Mat. Zhurn. 7 (1966), 1422-1424. Translation: Sib. Math. J. 7 (1966), 1112-1113. [MaMeSvW] L. M´arki, J. Meyer, J. Szigeti and L. van Wyk, Matrix representations of finitely generated Grassmann algebras and some consequences, Israel J. Math. 208 (2015), 373-384. [MeSvW] J. Meyer, J. Szigeti and L. van Wyk, A Cayley-Hamilton trace identity for 2 × 2 matrices over Lie-solvable rings, Linear Algebra Appl. 436 (2012), 2578-2582. [PaSch] R. Par´e and W. Schelter, Finite extensions are integral, J. Algebra 53 (1978), 477-479. [Pe1] K. R. Pearson, A lower bound for the degree of polynomials satisfied by ma- trices, J. Aust. Math. Soc. Ser. A 27 (1979), 430-436. [Pe2] K. R. Pearson, Degree 7 monic polynomials satisfied by a 3 × 3 matrix over a noncommutative ring, Comm. Algebra 10 (1982), 2043-2073. [Rob] J. C. Robson, Polynomials satisfied by matrices, J. Algebra 55 (1978), 509- 520. [Row] L. H. Rowen, Polynomial identities in ring theory. Pure and Applied Math- ematics, 84. Academic Press, New York-London, 1980. [S1] J. Szigeti, New determinants and the Cayley-Hamilton theorem for matrices over Lie nilpotent rings, Proc. Amer. Math. Soc. 125 (1997), 2245-2254. [S2] J. Szigeti, On the characteristic polynomial of supermatrices, Israel J. Math. 107 (1998), 229-235. [SvW1] J. Szigeti and L. van Wyk, Determinants for n × n matrices and the sym- metric Newton formula in the 3 × 3 case, Linear Multilinear Algebra 62 (2014), 1076-1090. [SvW2] J. Szigeti and L. van Wyk, On Lie nilpotent rings and Cohen's theorem, Comm. Algebra 43 (2015), 4783 -- 4796. Institute of Mathematics, University of Miskolc, 3515 Miskolc-Egyetemv´aros, Hun- gary E-mail address: [email protected] Institute of Mathematics, University of Miskolc, 3515 Miskolc-Egyetemv´aros, Hun- gary E-mail address: [email protected] Department of Mathematical Sciences, Stellenbosch University, Private Bag X1, Matieland 7602, Stellenbosch, South Africa E-mail address: [email protected]
1805.02573	1	1805	2018-05-07T15:30:41	Diophantine problems in rings and algebras: undecidability and reductions to rings of algebraic integers	[ "math.RA", "math.LO", "math.NT" ]	We study systems of equations in different families of rings and algebras. In each such structure $R$ we interpret by systems of equations (e-interpret) a ring of integers $O$ of a global field. The long standing conjecture that $\mathbb{Z}$ is always e-interpretable in $O$ then carries over to $R$, and if true it implies that the Diophantine problem in $R$ is undecidable. The conjecture is known to be true if $O$ has positive characteristic, i.e. if $O$ is not a ring of algebraic integers. As a corollary we describe families of structures where the Diophantine problem is undecidable, and in other cases we conjecture that it is so. In passing we obtain that the first order theory with constants of all the aforementioned structures $R$ is undecidable.	math.RA	math	Diophantine problems in rings and algebras: undecidability and reductions to rings of algebraic integers Albert Garreta∗, Alexei Miasnikov† and Denis Ovchinnikov‡ Abstract We study systems of equations in different families of rings and algebras. In each such structure R we interpret by systems of equations (e-interpret) a ring of integers O of a global field. The long standing conjecture that Z is always e-interpretable in O then carries over to R, and if true it implies that the Diophantine problem in R is undecidable. The conjecture is known to be true if O has positive characteristic, i.e. if O is not a ring of algebraic integers. As a corollary we describe families of structures where the Diophantine problem is undecidable, and in other cases we conjecture that it is so. In passing we obtain that the first order theory with constants of all the aforementioned structures R is undecidable. Contents 1 Introduction 2 Preliminaries 2.1 Model theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1 Multi-sorted structures . . . . . . . . . . . . . . . . . . . . . . . . 2.1.2 Diophantine problems and reductions. . . . . . . . . . . . . . . . Interpretations by equations . . . . . . . . . . . . . . . . . . . . . 2.1.3 2.2 Rings of integers of number and function fields 2.3 Notation and conventions . . . . . . . . . . . . . . . . . . . . . . . . . . 2 8 8 8 8 9 . . . . . . . . . . . . . . . 11 13 3 From bilinear maps to commutative rings and algebras 3.1 Ring of scalars of a full non-degenerate bilinear map . . . . . . . . . . . 3.2 E-interpreting Z(Sym(f )) and the largest ring of scalars . . . . . . . . . 3.3 Arbitrary bilinear maps 14 14 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 4 From rings of scalars to rings of integers of global fields 4.1 Subrings of global fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Rings of scalars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 24 26 ∗University of the Basque Country UPV/EHU, [email protected] †Stevens Institute of Technology, [email protected] ‡Stevens Institute of Technology, [email protected] 1 5 Rings and algebras over finitely generated rings of scalars 27 5.1 Module-finite rings and algebras . . . . . . . . . . . . . . . . . . . . . . 28 5.2 Finitely generated rings and algebras satisfying an infiniteness condition 30 5.3 Finitely generated associative commutative non-unitary rings and algebras 32 5.4 Undecidability of first order theories . . . . . . . . . . . . . . . . . . . . 34 6 Finitely generated algebras over infinite fields 7 References 1 Introduction 35 38 In this paper we study systems of equations in different families of rings and algebras. For each R in one of these families we interpret by systems of equations a ring of integers O of a global field (i.e. O is the integral closure of Z or Fp[t] in a finite extension of Q or Fp(t), respectively). In particular this reduces the Diophantine problem (decidability of systems of equations) in O to the same problem in R. It is known that D(O) is unde- cidable if O has positive characteristic [41], and it is conjectured to be also undecidable if otherwise O is a ring of algebraic integers [31, 6]. This leads us to conjecture that the Diophantine problem in the above structures is also undecidable. In many cases we verify that this is so. A field K is a global field if it is either a number field (a finite extension of Q) or a global function field (a finite extension of Fp(t), for some prime p). A ring of integers of a number field is called a ring of algebraic integers. The Diophantine problem in a structure R (also known as Hilbert's 10th problem in R), denoted D(R), asks whether there exists an algorithm that, given a system of equations S with coefficients in R, determines if S has a solution in R or not. The original version of this problem was posed by Hilbert for the ring of integers Z. This was solved in the negative in 1970 by Matyasevich [24] building on the work of Davis, Putnam, and Robinson [3]. Subsequently the same problem has been studied in a wide variety of rings, for instance in rings of algebraic integers, where it remains open. It is conjectured that the ring Z is e-definable in all such rings [6, 31], and indeed this has been shown to be true in some partial cases, for example for rings of integers of abelian number fields [38] (see [43] for further partial results). The conjecture is known to hold provided that for each number field K there exists an elliptic curve E over K with rkE(K) = rkE(Q) = 1 [32]. Later this was shown to be true assuming the Safarevich-Tate conjecture [25]. The scenario is much clearer for rings of integers of global function fields, indeed Shlapentokh [41] showed that Z is e-interpretable in any such ring O, and consequently that D(O) is undecidable. Other results in this direction include undecidability of D(K) for any global function field K [8, 39], and of D(R[t]), for R any integral domain [4, 5]. Some rings where the Diophantine problem remains open are most remarkably Q (it is known however that this problem is undecidable in Z[S−1], for S an infinite set of primes of Dirichlet density 1 [33]); the rational functions C(t) (even though D(C(t1, t2)) 2 is undecidable [21]); and the field of Laurent series Fp((t)). The decidability of the first order theory of C(t) and of Fp((t)) remains an open problem as well, whereas a classic result of Julia Robinson states that the first order theory of Q is undecidable. We refer to [34, 31, 43, 22] for further information and surveys of results in this direction. Regarding non-commutative rings, Romankov [37] showed that D(F ) is undecidable in several types of free rings F , which include free Lie rings, free associative or non- associative rings, and free nilpotent rings. In the same direction the second author, in collaboration with Kharlampovich, recently proved undecidability of D(A) in the language of rings, for A any of the following: a free associative k-algebra, a free Lie k-algebra (of rank at least 3), and many group k-algebras [15, 17]. In all these cases k is an arbitrary field. In Corollary 1.8 we obtain further results of this type. Many results regarding the first order theory of free algebras can be found in [16, 18, 19]. In this paper all rings and algebras are possibly non-associative, non-commutative, and non-unitary, unless stated otherwise. A ring (or algebra) of scalars is an associative, commutative, unitary ring (or algebra). We will always consider algebras over rings of scalars, and we fix Λ to denote such ring. Given a Λ-algebra L, we let L2 be the Λ-module generated by all products of two elements of L. A ring is the same as a Z-algebra. One of the main results of the paper is the following: Theorem 1.1. Let R be a ring that is finitely generated as an abelian group. Suppose that R2 is infinite. Then there exists a ring of algebraic integers O that is interpretable in R by systems of equations in the language of rings. Consequently, D(O) is reducible to D(R). On the other hand, if R2 is finite, then D(R) is decidable. In number theoretic terms, an interpretation by systems of equations (in short, an e-interpretation), is roughly a Diophantine definition up to a Diophantine equivalence relation. Here Diophantine definitions are considered by means of systems of equations, as opposed to single equations. We also convene that all systems of equations and all e-interpretations allow the use of any constant elements of the structures at hand, not necessarily in the signature. See Subsections 2.1.3, 2.1.2 and 2.3 for further comments on these matters. Theorem 1.1 is further generalized to algebras. A Λ-algebra is called module-finite if it is finitely generated as a Λ-module. The language of Λ-modules Lmod , or of Λ-algebras Lalg, consists in the usual language of groups Lgroup or of rings Lring, respectively, together with unary functions {·λ λ ∈ Λ} representing multiplication by elements of Λ (see Subsection 2.3). We write (R; L) to indicate that a structure R is considered with a language L. Theorem 1.2. Let A be a module-finite algebra over a finitely generated ring of scalars Λ. Suppose that A2 is infinite. Then there exists a ring of integers O of a global field such that (O; Lring) is e-interpretable in (A; Lalg ), and D(O; Lring) is reducible to D(A; Lalg). If additionally Λ has positive characteristic p, then (Fp[t]; Lring) is e-interpretable in If otherwise A2 is finite and D(A; Lmod ) is (A; Lalg ), and D(A; Lalg ) is undecidable. decidable, then D(A; Lalg) is decidable. 3 If Λ is a finite field, then all the above holds after replacing (A; Lalg) by (A; Lring), and (A; Lmod ) by (A; Lgroup). In this case, if A2 is finite, then D(A; Lring) is decidable. The simultaneous appearance of number fields and global function fields in our results should come as no surprise: there is a well-known analogy between these two classes of fields (see for example the preface of [45]). Theorems 1.1 and 1.2 are further extended to a large class of finitely generated non-module-finite rings and algebras (Theorems 1.6 and 1.7). Similar statements are obtained for finite-dimensional algebras over arbitrary fields (Corollary 1.11 and Theo- rem 1.12). All of these will be discussed later in this introduction. Our main results rely on the combination of two techniques: in the first we move from non-commutative algebra to commutative algebra, and in the second we move from the latter to number theory. The first reduction is achieved through the study of rings of scalars of bilinear maps between Λ-modules. This is relevant for us because much of the structure of a Λ-algebra can be "seen" in its ring multiplication operation, which is indeed a Λ-bilinear map between Λ-modules. In fact, bilinear maps also arise in a natural way in other structures, and in some cases it is possible to apply the methods presented in this paper to these, for example, in some classes of groups. This line of work will be explored further in an upcoming paper. Non-commutative algebra. Other structures Bilinear maps Commutative algebra Number theory The approach regarding bilinear maps is described further at the end of this introduc- tion. The reduction from commutative algebra to number theory is summarized in the following result, and also in Theorem 1.10, where algebras over arbitrary fields are considered. Theorem 1.3. Let R be an infinite finitely generated ring of scalars. Then there exists a ring of integers O of a global field such that (O; Lring) is e-interpretable in (R; Lring), and D(O; Lring) is reducible to D(R; Lring). Moreover: 1. If R is finitely generated as an abelian group, then O is a ring of algebraic integers. 2. If R has positive characteristic p, then O is the ring of integers of a global function field, (Fp[t]; Lring) is e-interpretable in (R; Lring), and D(R; Lring) is undecidable. Some of the main results of this paper have been presented above. We now proceed to state the remaining ones, starting with the following non-unitary version of Theorem 1.3. See also Theorem 1.10 for a similar statement involving algebras. Theorem 1.4. Let R be an infinite finitely generated associative commutative non- unitary ring. Then the conclusions of Theorem 1.3 hold for (R; Lring). We now make an important observation. 4 Remark 1.5. All subsequent results are stated for Λ-algebras. These hold in particular for rings by taking Λ = Z, in which case algebras become rings, modules become groups, and one has Lalg = Lring, Lmod = Lgroup. The reader interested solely in rings may choose to read the whole paper with these considerations in mind (note that this does not apply for the results concerning algebras over infinite fields, which are stated at the end of this introduction and in the last section of the paper). Theorems 1.1 and 1.2 are extended to some families of finitely generated non-module- finite rings and algebras. Let A be a Λ-algebra (or a ring) generated by a set S. If A is unitary let S = S\{λ · 1 λ ∈ Λ}, and otherwise let S = S. In both cases let In be the Λ-ideal generated by all Λ-multiples of products of n elements of S. A non-associative Λ-algebra will be called right-normed-generated if each In is generated as a Λ-module by a (possibly infinite) set of elements of the form (s1(s2(. . . (sk−1sk) . . . ))), with k ≥ n and si ∈ S for all i. Theorem 1.6. Let A be a finitely generated algebra over a finitely generated ring of scalars Λ. Suppose that A is associative or right-normed-generated, and that (A/In)2 is infinite for some n ≥ 1. Then there exists a ring of integers of a global field O such that (O; Lring) is e-interpetable in (A; Lalg ), and D(O; Lring) is reducible to D(A; Lalg). Moreover, if Λ has positive characteristic then D(A; Lalg ) is undecidable. If Λ = Z (so A is just a ring), then O is a ring of algebraic integers. If Λ is either Z or a finite field, then the whole theorem holds after replacing (A; Lalg ) by (A; Lring). This theorem is proved by showing that A/In is e-interpretable in L. Then, since A/In is module-finite, it suffices to apply Theorem 1.2 and transitivity of e-interpretations. A gradation L = ⊕i≥1Li of a non-unitary algebra L will be called simple if L1 generates L as an algebra. We will see that all simply graded Lie algebras are right- normed-generated. Thus Theorem 1.6 immediately yields the following: Corollary 1.7. Let L be a finitely generated simply graded Lie Λ-algebra (for Λ as in Theorem 1.6). Suppose that [L/In, L/In] is infinite for some n ≥ 1. Then the conclusions of Theorem 1.6 hold for L. In a free algebra F one has that (F/In)2 is infinite for all n ≥ 2. Hence the next result. Below Λ is as in Theorem 1.6. Corollary 1.8. Let F be a finitely generated free associative Λ-algebra (commutative or non-commutative, and unitary or non-unitary) or a free Lie Λ-algebra of rank at least 2. Then the conclusions of Theorem 1.6 hold for F . This complements the aforementioned results of Romankov [37] and of Kharlam- povich and Miasnikov [15, 17] regarding free algebras. We remark that in [37] it is proved (among others) that the algebras of Corollary 5.10 actually have undecidable Diophantine problem if Λ = Z. Noskov [30] proved that all finitely generated infinite rings of scalars have undecidable first order theory. Since this applies to the ring of integers of any global field, the next result follows. 5 Theorem 1.9. Suppose that A satisfies the hypotheses of any of the theorems and corollaries above. Then the first order theory of A in the corresponding language with constants is undecidable. We next consider finitely generated algebras over a field. Note that an infinite field cannot be finitely generated, and so such an algebra may not fall into any of the cases studied previously. From now on k denotes an arbitrary field. The following is analogous to Theorem 1.3. As before it relies on some results due to Shlapentokh [39, 40]. The proof is very similar to that of Theorem 1.3, with the main divergence being due to different behavior when the Krull dimension is 0. Note that only the language of rings (as opposed to the language of k-algebras) is used here. Theorem 1.10. Let R be a nonzero finitely generated k-algebra of scalars. Suppose that R has Krull dimension at least one. Then (Z; Lring) is e-interpretable in (R; Lring), and D(R; Lring) is undecidable. If otherwise R has Krull dimension zero, then there exists a finite field extension K of k such that (K; Lring) is e-interpretable in (R; Lring). A similar result holds if R is non-unitary. In this case however one must consider (R; Lalg) instead of (R; Lring). Corollary 1.11. Let R be a finitely generated associative commutative non-unitary k- algebra such that R2 6= 0. Then the the first or the second conclusion of Theorem 1.10 hold for (R; Lalg). All previous results regarding module-finite algebras over a f.g. ring of scalars have their analogue for module-finite algebras over k. As in Theorem 1.10, the e-interpretations below use only the language of rings. This contrasts with Theorems 1.1 and 1.2. Theorem 1.12. Let R be a finitely generated k-algebra. Suppose that R satisfies one of the following: 1. R has finite dimension over k and R2 6= 0. 2. R is associative or right-normed-generated and (R/In)2 6= 0 for some n ≥ 1. 3. R is a simply graded Lie algebra and [R/In, R/In] 6= 0 for some n ≥ 1. Then there exists a finite field extension K of k such that (K; Lring) is e-interpretable in (R; Lring), and D(K; Lring) is reducible to D(R; Lring). So for example if k is a global function field, then D(R; Lring) is undecidable for any R as above. The same is true if k is a field all of whose finite extensions have undecidable Diophantine problem. 6 From bilinear maps to commutative algebra Next, we briefly explain the ap- proach taken in order to pass from possibly non-associative, non-commutative, and non- unitary algebras to algebras of scalars (and similarly for rings). The ideas we present here were introduced by the second author in [26], and they have been used successfully to study different first order theoretic aspects of different types of structures, includ- ing rings whose additive group is finitely generated [27], free algebras [16, 18, 19], and nilpotent groups [28, 29]. Observe that ring multiplication · of a Λ-algebra R is, by definition, a Λ-bilinear map between Λ-modules. One can try to replace Λ by a "larger" ring of scalars ∆. To do so, one needs to find a ring of scalars ∆ that acts on R by Λ-module endomorphisms (thus making the additive group of R into a ∆-module), in a way that · becomes a ∆-bilinear map between ∆-modules. In this case we say that ∆ is a ring of scalars of the map ·. These considerations apply in the exact same way if one starts with an arbitrary Λ-bilinear map f : N × N → M between Λ-modules N and M . If f is full and non- degenerate (see Subsection 3.1) then one can define the largest ring of scalars of f , denoted R(f ). This ring constitutes an important feature of f , and in some sense it provides an "approximation" to interpreting (in (N, M ; f ; Lmod )) multiplication of constant elements from N and M by integer variables, or alternatively by variables taking values in Λ. Another important property of R(f ) is that it is interpretable in (N, M ; f ; Lmod ) by first order formulas without constants [26]. In this paper we prove that this is still true if one uses systems of equations instead (with constants). Theorem 1.13. If f is full and non-degenerate, and if N and M are finitely generated, then both Z(Sym(f )) and the largest ring of scalars of f are e-interpretable in the two sorted structure (N, M ; f, Lmod ). Here Sym(f ) is defined as Sym(f ) = {α ∈ EndΛ(A) f (αx, y) = f (x, αy) ∀ x, y ∈ A}, and Z(Sym(f )) denotes the center of Sym(f ). The interest we have for Z(Sym(f )) is mostly technical. This is explained in Remark 3.11. Idea of the proof of Theorem 1.13. There are two main observations. The first goes as follows: Both Z(Sym(f )) and R(f ) can be seen as subalgebras of the algebra of Λ- endomorphisms of N , denoted End Λ(N ). Let a1, . . . , ak be a module generating set of N . Then each α ∈ EndΛ(N ) can be identified with the tuple (αa1, . . . , αak) ∈ N k, and so we can think of Z(Sym(f )) and R(f ) as Λ-submodules of N k with an extra ring multiplication operation. Using that N k is a Notherian Λ-module, we will see that Z(Sym(f )) and R(f ) as Λ-modules are e-definable in N . Moreover: since αai is the i-th component of α seen as a tuple, it is possible to e-interpret in N the action of any endomorphism α on any generator ai (and thus on any constant element of N ). The second idea is to use the properties of f in order to "express" statements about endomorphisms from Z(Sym(f )) in terms of their actions on a1, . . . , ak. For example, given α, β, γ ∈ Z(Sym(f )), one has that γ = αβ if and only if f (γai, aj) = f (βai, αaj) for all 1 ≤ i, j ≤ k (this is proved using bilinearity of f and the fact that f (αβx, y) = f (βx, αy) for all x and y). This and the considerations in the previous paragraph can be 7 combined to show (after some work) that multiplication in Z(Sym(f )) is e-interpretable in (N, M ; f, Lmod ). The rest of the proof follows similarly. In Subsection 3.3 we generalize Theorem 1.13 to the following result. Theorem 1.14. Let f : A × B → C be a Λ-bilinear map between finitely generated Λ-modules. Then there exists a ring of scalars Θ that is a module-finite Λ-algebra, such that (Θ; Lalg) is e-interpretable in F = (A, B, C; f, Lmod ). If Λ is the ring Z or a field, then Lalg and Lmod can be replaced by Lring and Lgroup, respectively. As mentioned above, the ring multiplication of any module-finite Λ-algebra R is a Λ- bilinear map · : R × R → R between finitely generated Λ-modules, and (R, R, R; ·, Lmod ) is e-interpretable in (R; Lalg). Applying Theorem 1.14 and transitivity of e-interpretations we manage to move from the possibly non-associative, non-commutative, and non- unitary R to an algebra of scalars. 2 Preliminaries 2.1 Model theory 2.1.1 Multi-sorted structures A multi-sorted structure A is a tuple A = (Ai; fj, rk, cℓ i, j, k, ℓ), where the Ai are pairwise disjoint sets called sorts; the fj are functions of the form fj : Aℓ1 × · · · × Aℓm → Aℓm+1 for some ℓi; the rk are relations of the form rk : As1 × · · · × Asp → {0, 1}; and the cℓ are constants, each one belonging to some sort. The tuple (fj, rk, cℓ j, k, ℓ) is called the signature or the language of A. We always assume that A contains the relations "equality in Ai", for all sorts Ai, but we do not write them in the signature. If A has only one sort then A is a structure in the usual sense. One can construct terms in a multi-sorted structure in an analogous way as in uniquely-sorted structures. In this case, when introducing a variable x one must specify a sort where it takes values, which we denote Ax. A set S of generators of A is a collection of elements from different sorts such that any element from any sort can be written as a term using only constants from S and from the signature of A (and using function symbols). Let A1, . . . , An be a collection of multi-sorted structures. We denote by (A1, . . . , An) the multi-sorted structure that is formed by all the sorts, functions, relations, and constants of each Ai. Given a function f or a relation r we use the notation (A, f ) or (A, r) with analogous meaning. If two different Ai's have the same sort, then we view one of them as a formal disjoint copy of the other. 2.1.2 Diophantine problems and reductions. Let A be a multi-sorted structure. An equation in A is an expression of the form r(τ1, . . . , τk), where r is a signature relation of A (typically, the equality relation), and each τi is a term in A (taking values in an appropriate sort) where some of its variables 8 may have been substituted by elements of A. Such elements are called the coefficients (or the constants) of the equation. These may not be signature constants. A system of equations is a finite conjunction of equations. A solution to a system of equations ∧iΣi(x1, . . . , xn) on variables x1, . . . , xn is a tuple (a1, . . . , an) ∈ Ax1 × · · · × Axn such that all equations Σi(a1, . . . , an) are true in A. The Diophantine problem in A, denoted D(A), refers to the algorithmic problem of determining if each given system of equations in A (with coefficients in a fixed com- putable set) has a solution. Sometimes this is also called Hilbert's tenth problem in A. An algorithm L is a solution to D(A) if, given a system of equations S in A, deter- mines whether S has a solution or not. If such an algorithm exists, then D(A) is called decidable, and, if it does not, undecidable. An algorithmic problem P1 is said to be reducible to another problem P2 if a solution to P2 (if it existed) could be used as a subroutine of a solution to P1. For example, D(Z) is undecidable, and hence D(A) is undecidable for any structure A such that D(Z) is reducible to D(A). In some cases one restricts the set of coefficients C that can be used in the input equations of the Diophantine problem of a structure. For instance, one typically takes C = Z when studying D(Q) (equivalently one can take C = {0, 1}). In this paper we will always need that C contains certain coefficients, namely those used in a certain e-interpretation, and maybe also the preimage of some constants of the structure that is being e-interpreted. For this reason, and to simplify the exposition, we agree that C is always the whole structure, or a suitable computable subset if the structure is not countable. All structures considered in this paper are finitely generated (and thus computable), except in Subsection 6. 2.1.3 Interpretations by equations Interpretability by equations (e-interpretability) is the analogue of the classic model- theoretic notion of interpretability by first order formulas (see [12, 23]). In e-interpretability one requires that only systems of equations with coefficients are used, instead of first order formulas. From a number theoretic viewpoint, e-interpretability is roughly Dio- phantine definability (by systems of equations) up to a Diophantine definable equivalence relation. In this paper -in e-interpretations and in Diophantine problems- we consider systems of equations and not just single equations. This may contrast with some number- theoretic settings, where systems of equations are equivalent to single equations, and both notions are treated interchangeably (for example in integral domains whose field of fractions is not algebraically closed). Let A be a structure with sorts {Ai i ∈ I}. A basic set of A is a set of the form Ai1 × · · · × Aim for some m and ij's. Definition 2.1. Let M be a basic set of a multi-sorted structure M. A subset A ⊆ M is called definable by equations (or e-definable) in M if there exists a system of equations ΣA(x1, . . . , xm, y1, . . . , yk) on variables (x1, . . . , xm, y1, . . . , yk) = (x, y) such that x takes values in M , and such that for any tuple a ∈ M , one has that a ∈ A if and only if the 9 system ΣA(a, y) on variables y has a solution in M. In this case ΣA is said to define A in M. From an algebraic geometric viewpoint, an e-definable set is a projection onto some coordinates of an affine algebraic set. Definition 2.2. Let A = (A1, . . . ; f, . . . , r . . . , c, . . . ) and M be two multi-sorted struc- tures. A is said to be interpretable by equations (or e-interpretable) in M if for each sort Ai there exists a basic set M (Ai) of M, a subset Xi ⊆ M (Ai), and a surjective map φi : Xi → Ai such that: 1. Xi is e-definable in M, for all i. 2. For each function f and each relation r in the signature of A (including the equality relation of each sort), the preimage by φ = (φ1, . . . ) of the graph of f (and of r) is e-definable in M, in which case we say that f (or r) is e-interpretable in M. The same terminology applies to functions and relations that are not necessarily in the signature of A. The tuple of maps φ = (φ1, . . . ) is called an e-interpretation of A in M. The coefficients appearing in the equations above are called the coefficients of the e-interpretation. The next lemma illustrates a key application of e-interpretability. Lemma 2.3. Let R be a ring, not necessarily commutative or associative. Suppose I ≤ R is an ideal that is e-definable as a set in R. Then (R/I; Lring) is e-interpretable in (R; Lring). Proof. Let ΣI(x, y) be a system of equations that e-defines I as a set in R, so that a ∈ R belongs to I if and only if ΣI (a, y) has a solution y. It suffices to check that the natural epimorphism π : R → R/I is an e-interpretation of R/I in R. First observe that the preimage of π is the whole R, which is e-definable in R by an empty system of equations. Regarding equality in R/I, π(a1) = π(a2) in R/I if and only if a1 − a2 ∈ I, i.e. if and only if ΣI (a1 − a2, y) has a solution. From this it follows that the preimage of equality in R/I, {a1, a2 ∈ R π(a1) = π(a2)} , is e-definable in R by the system of equations Σ′ I(x1, x2, y) obtained from ΣI(x, y) after substituting each occurrence of x by x1 − x2, where x1 and x2 are new variables. By similar arguments the preimages of the addition and multiplication operations in R/I are e-definable in R: indeed, π(a1) + π(a2) = π(a3) if and only if a1 + a2 − a3 ∈ I, and π(a1)π(a2) = π(a3) if and only if a1a2 − a3 ∈ I. Remark 2.4. It is clear from the proof that an analogue of Lemma 2.3 holds for other structures, such as groups with e-definable normal subgroups, modules with e-definable submodules, etc. The next two results are fundamental. They follow from Lemma 2.7, which we present at the end of this subsection. Proposition 2.5 (E-interpretability is transitive). If A is e-interpretable in B and B is e-interpretable in M, then A is e-interpretable in M. 10 Proposition 2.6 (Reduction of Diophantine problems). Let A and M be (possibly multi-sorted) structures such that A is e-interpretable in M. Then D(A) is reducible to D(M). As a consequence, if D(A) is undecidable, then so is D(M). Similarly, the first order theory of A is reducible to the first order theory (with constants1) of M, and the second is undecidable if the first is. Both Propositions 2.5 and 2.6 are a consequence of the following lemma, which states in technical terms that if one structure is e-interpretable in the other, then one may "express" equations in the first as systems of equations in the second. Lemma 2.7. Let φ = (φ1, . . . ) be an e-interpretation of a multi-sorted structure A = (A1, . . . ; f, . . . , r, . . . , c, . . . ) in another multi-sorted structure M, with φi : Xi ⊆ M (Ai) → Ai (see Definition 2.2). Let σ(x) = σ(x1, . . . , xn) be an arbitrary system of equations in A with each variable xi taking vaules in Aji. Then there exists a system of equations Σσ(y1, . . . , yn) in M, such that each tuple of variables yi takes values in M (Aji), and i=1 M (Aji) is a solution to Σσ(y1, . . . , yn) if and only i=1 Xji and (φj1(b1), . . . , φjn(bn)) is a solution to σ. such that a tuple (b1, . . . , bn) ∈Qn if (b1, . . . , bn) ∈ Qn Proof. It suffices to follow step by step the proof of Theorem 5.3.2 from [12], which states that the above holds for the classical notion of interpretability by first order formulas of uniquely-sorted structures. Indeed, the result follows from the definitions after observing that σ can be rewritten as a system of equations in which all its atomic subformulas are unnested. 2.2 Rings of integers of number and function fields An (algebraic) number field is a finite field extension of Q. An (algebraic) function field F over a field k is a finite field extension of k(t). If k is a finite field then F is called a global function field. A global field is either a number field or a global function field. Let A and B be two rings of scalars such that A ≤ B. An element b ∈ B is integral over A if there exists a monic polynomial from A[x] that has b as a root. The integral closure of A in B is the set C of all elements of B that are integral over A. Such set is a subring of B. An integral domain D is called integrally closed if the integral closure of D in its field of fractions is D itself. Let K be a number field or a field of functions over k. The ring of integers of K, denoted OK , is the integral closure of Z or k[t] in K, respectively. In the first case OK is called a ring of algebraic integers. Rings of S-integers or holomorphy rings Our main references for this subsection are [44, 43, 10]. Throughout the end of this subsection let K be either a number field or a function field over a field k. A (non-archimidean) valuation of K is a map v : K → Z ∪ {∞} that satisfies all of the following: 1The considerations made regarding the use of constants in systems of equations and Diophantine problems are made as well for first order formulas and their decidability problems (see Paragraph 3 of Subsection 2.3 or Subsection 2.1.2). 11 1. v induces a homomorphism v : K ∗ → Z from the multiplicative group of K into the additive group of Z. 2. v(x + y) ≥ min{v(x), v(y)} for all x, y ∈ K. 3. v(x) = ∞ if and only if x = 0. 4. v(x) 6= 0 for some x ∈ K. 5. If K is a function field, then v(c) = 0 for all c in the algebraic closure of k in K. We will refer to a non-archimedean valuation simply as a valuation (archimidean valu- ations are never considered in this paper). Two valuations v1 and v2 are equivalent if {x ∈ K v1(x) ≥ 0} = {x ∈ K v2(x) ≥ 0}. An equivalence class of a valuation v will be called a prime. We identify v with its equivalence class, and we write ord v(x) instead of v(x), for any x ∈ K. Let S be a (finite or infinite) set of primes of K. The ring of S-integers of K is defined as OK,S = {x ∈ K ord p(x) ≥ 0 ∀p /∈ S} . In many texts OK,S is called a holomorphy ring when K is a function field. Remark 2.8. Rings of S-integers include rings of integers of number and function fields. Indeed, if K is a number field, then OK,∅ is precisely the ring of algebraic integers OK of K (this follows, for example, from B.1.15 and B.1.22 of [43]). Suppose otherwise that K is a function field over k. The ring of integers of k(t) is k[t] = Ok(t),{v∞}, where v∞ is the prime of k(t) defined by v∞(f /g) = deg(g) − deg(f ) (see 3.2 of [44]). Furthermore, the ring of integers of K is OK,S∞, where S∞ is the set of all primes of K that lie above v∞, i.e. those primes that, when restricted to k(t), coincide with v∞ (see 1.2.2 and 3.2.6 of [44]). The set S∞ is finite (see B.1.11 of [43]), and each p ∈ S∞ is called a pole of t. Notice that if S is an arbitrary set of primes of K and OK,S contains k[t], then S∞ ⊆ S, by definition of OK,S and S∞. It follows from the above remark that if K is a number field, then the field of fractions F of OK,S is K, for any set of primes S. Similarly, if K is a function field, then the field of fractions of OK,S is again K, for any set of primes S containing S∞. Proposition 2.9 (Propositions B.1.21 and B.1.27 of [43], and Corollary 3.2.8 of [44]). Let R be a subring of K. Then R is integrally closed if and only if R = OF,S, where F is the field of fractions of R, and S is some set of primes of F . We will also need the following result. Lemma 2.10. Let S be a set of primes of K. Suppose that one of the following hold: 1. K is a number field, and OK,S is finitely generated as a ring. 2. K is a function field over k, and OK,S is finitely generated as a k-algebra. 12 Then there exists a finite subset S0 of S such that OK,S = OK,S0. Proof. Let r be either Z or k, depending on whether Item 1 or 2 holds. Suppose that OK,S is generated as a r-algebra by a finite set A = {a1, . . . , an}. Let S0 be the subset of S consisting in all primes p such that ord p(a) 6= 0 for some a ∈ A. It is well known that for any x ∈ OK,S there are only finitely many primes p of K such that ord p(x) 6= 0 (see B.1.18 of [43] and 1.3.4 of [44]). Therefore S0 is finite. By assumption, for all x ∈ OK,S there exists a polynomial q ∈ r[a1, . . . , an] such that x = q. Using the axioms of non-archimedean valuations and the fact that ord p(a) = 0 for all p ∈ S\S0 and all a ∈ A, we obtain that ord p(x) ≥ 0 for all x ∈ OK,S and all p ∈ S\S0 , we have ord p(x) ≥ min{ord p(ri mi ) i} = min{ord p(ri ) i} ≥ 0 ). Thus by definition OK,S ≤ OK,S0. The opposite inclusion follows immediately, since S0 ⊆ S. (indeed, writing q = Pi rimi for some ri ∈ r and some products mi = ai1 . . . aiji 2.3 Notation and conventions We would like to emphasize some relevant aspects and to fix some notation. 1. By ring or algebra of scalars we mean an associative commutative unitary ring or algebra, respectively. Given a Λ-algebra R and a subset S ⊆ R, we let hSiΛ be the Λ-submodule of R generated by a set S. We also let R2 = hxy x, y ∈ RiΛ. All modules are assumed to be left modules. Similarly, the underlying module of an algebra is assumed to be a left module. All arguments work in the same way if we replace left for right, or left module for bimodule. 2. There is no restriction on the coefficients used in an e-interpretation. In particular, the coefficients may not belong to the signature. The same consideration applies for the Diophantine problem of a computable (for example countable) structure. If the structure is not computable, then the input systems of equations in the Diophantine problem must have coefficients in a fixed computable set. See Subsection 2.1.2. 3. The language of groups is Lgroup = (+, 0). The language of Λ-modules is Lmod = (Lgroup, ·Λ), where the ·Λ = {·λ λ ∈ Λ} are unary functions representing multiplication by scalars: ·λ(x) = λx. The language of rings Lring is either (+, ·, 0) or (+, ·, 0, 1). The language of Λ-algebras is Lalg = (Lring, ·Λ). The presence or lack of 1 will be clear If Λ admits a finite generating set S, then one can replace ·Λ by from the context. ·S = {·λ λ ∈ S}. Hence, in an equation (or in a formula) over a Λ-module or Λ-algebra R, one is allowed to multiply any element of R by any constant element of Λ, but this is as far as one can involve Λ: no variable can take values in Λ, no quantification over Λ can be made, etc. 13 4. The notion of Z-module or Z-algebra with the languages above is equivalent to the notion of abelian group or ring, respectively. Indeed, multiplication by scalars from Z is e-interpretable using only the group addition +, since nx = x + n. . . + x for all all n ∈ Z and all element x. 5. Sometimes we will want to look at a Λ-algebra L as a Λ-module, or as a ring, or as a group, forgetting about the corresponding additional operations of L. We will use the notation (L; Lmod ), (L; Lring), (L; Lgroup) when this is done, respectively. We will also write (L; Lalg) to emphasize that L is considered with all its Λ-algebra operations. A similar terminology will be used for other structures such rings and modules. This notation will be used extensively in expressions of the type (L; L1) is e-interpretable in (K; L2). This means that L with the operations of the language L1 is e-interpretable in K considered with the operations of L2 (and with constants possibly not in the sig- nature of L2 –see Paragraph 3). In the particular case that L1 = L2 we will also say that L is e-interpretable in K in the language L1 (with constants). Given some modules and algebras L1, . . . , Ln, we write (L1, . . . , Ln; Lmod ) instead of ((L1; Lmod ), . . . , (Ln; Lmod )). The expression (L1, . . . , Ln; Lgroup) has an analogous meaning. 3 From bilinear maps to commutative rings and algebras A brief description of the arguments used in this section can be found at the last part of the introduction. 3.1 Ring of scalars of a full non-degenerate bilinear map Throughout this subsection, Λ denotes a (possibly infinitely generated) ring of scalars, i.e., an associative, commutative, unitary ring. A map f : N × N → M between Λ-modules N and M is Λ-bilinear if, for all a ∈ N , the maps f (a, ·) and f (·, a) from N to M are homomorphisms of Λ-modules. We call f non-degenerate if whenever f (a, x) = 0 for all x ∈ N , we have a = 0, and similarly for f (x, a). The map f is full if the Λ-submodule generated by the image of f is the whole M . The set of module endomorphisms of a Λ-module N , denoted End Λ(N ), forms an associative unitary Λ-algebra once we equip it with the operations of addition and composition (henceforth called multiplication). We simply write αx instead of α(x), for α ∈ EndΛ(N ) and x ∈ N . An action of a ring ∆ on N is a ring homomorphism φ : ∆ → EndΛ(N ). Any such action φ endows N with a structure of ∆-module. Furthermore the action is called faithful if φ is an embedding. Definition 3.1. Let f : N × N → M be a Λ-bilinear map between Λ-modules. A ring ∆ is called a ring of scalars of f if it is associative, commutative, and unitary, and there exist faithful actions of ∆ on M and N such that f (αx, y) = f (x, αy) = αf (x, y) for all α ∈ ∆ and all x, y ∈ N . 14 Since the actions of a ring of scalars ∆ of f on M and N are faithful, there exist ring embeddings ∆ ֒→ EndΛ(M ) and ∆ ֒→ EndΛ(N ). For this reason and for convenience we always assume that a ring of scalars of f is a subring of EndΛ(N ). Definition 3.2. We say that ∆ is the largest ring of scalars of f , and we denote it R(f ), if for any other ring of scalars ∆′ of f , one has ∆′ ≤ ∆ as subrings of End Λ(N ). The next result was proved by the second author in [26]. We recover its proof since we will need to examine it in the next subsection. Theorem 3.3 ([26]). Let f : N × N → M be a full non-degenerate Λ-bilinear map between Λ-modules. Then the largest ring of scalars R(f ) of f exists and is unique. Proof. Consider the following subsets of EndΛ(N ): Sym(f ) = {α ∈ EndΛ(N ) f (αx, y) = f (x, αy) for all x, y ∈ N } , Z(Sym(f )) = {α ∈ Sym(f ) αβ = βα for all β ∈ Sym(f )} . (1) (2) It is straightforward to check that both Sym(f ) and Z(Sym(f )) are Λ-modules. More- over, for all α1, α2 ∈ Z(Sym(f )) and all x, y ∈ N , f (α1α2x, y) = f (α2x, α1y) = f (x, α2α1y) = f (x, α1α2y), and thus α1α2 ∈ Sym(f ). Since both α1 and α2 commute with any element from Sym(f ), so does α1α2. Hence, α1α2 ∈ Z(Sym(f )), and so Z(Sym(f )) is a subalgebra of EndΛ(N ). Next, we show that any ring of scalars ∆ of f is a subring of Z(Sym(f )). Indeed, by definition, ∆ ⊆ Sym(f ). To see that ∆ ⊆ Z(Sym(f )), let α ∈ ∆ and β ∈ Sym(f ). Then, for all x, y ∈ N , f (αβx, y) = αf (βx, y) = αf (x, βy) = f (αx, βy) = f (βαx, y). Hence f ((αβ −βα)x, y) = 0 for all x, y ∈ N . Since f is non-degenerate and y is arbitrary, (αβ − βα)x = 0 for all x ∈ N . It follows that αβ = βα, and thus ∆ ⊆ Z(Sym(f )). By what we have seen so far, Z(Sym(f )) is an associative commutative unitary algebra that acts faithfully on N . We now wish to define an action of a subring ∆ of Z(Sym(f )) on M . Since f is full, for all z ∈ M we have z = Pi f (xi, yi) for some xi, yi ∈ N . Hence, one may try to define the following action: αz = X f (αxi, yi) for α ∈ ∆. (3) However, this is not necessarily well-defined, because the same z ∈ M may have different expressions as sums of elements f (xi, yi). With this in mind, we let ∆ be the set of all α ∈ Z(Sym(f )) such that X f (αxi, yi) = X f (αx′ i, y′ i) whenever X f (xi, yi) = X f (x′ i, y′ i). (4) 15 Clearly, ∆ is closed under addition and multiplication, and therefore it is a subring of Z(Sym(f )) with a well-defined action on M given by (3). Since the action of ∆ on N is faithful, and f is a non-degenerate map, the action of ∆ on M is faithful as well. It follows that ∆ is a ring of scalars of f . Moreover, any ring of scalars ∆′ of f satisfies (4), and thus, since ∆′ ⊆ Z(Sym(f )), we have ∆′ ≤ ∆. We conclude that ∆ is the unique largest ring of scalars of f . Remark 3.4. It is clear from the proof above that R(f ) is closed under multiplication by Λ. Hence, R(f ) admits the structure of a Λ-algebra. It is possible to show that the ring structure of R(f ) is independent of the actions of Λ on N and M . In other words, R(f ) is the same ring whether we look at f as a bilinear map between Λ-modules, or as a bilinear map between abelian groups (we do not prove this, since it will not be used or commented further in the paper). However, the second point of view may be inconvenient for e-interpretability purposes, since N and M may be finitely generated as Λ-modules, but infinitely generated as abelian groups. 3.2 E-interpreting Z(Sym(f )) and the largest ring of scalars Throughout this subsection Λ denotes a possibly infinitely generated Noetherian ring of scalars. In this case any finitely generated Λ-module is Noetherian and finitely presented (see [7] or [11]). We refer to Subsection 2.3 for important notation and terminology conventions. The goal of this subsection is to prove the following result. Theorem 3.5. Let f : N × N → M be a full non-degenerate bilinear map between finitely generated Λ-modules. Then both Z(Sym(f )) and the largest ring of scalars R(f ) of f are module-finite Λ-algebras, and they are e-interpretable as Λ-algebras in F = (N, M ; f, Lmod ). Moreover, 1. If Λ is either a field or the ring Z, then (Z(Sym(f )); Lring) is e-interpretable in (N, M ; f, Lgroup) (i.e. multiplication by scalars is not required). 2. Z(Sym(f )), R(f ), N , and M are all simultaneously zero, finite, or infinite. We state some lemmas and observations before proving Theorem 3.5, starting with a useful description of End Λ(N ). Remark 3.6. Let N be a Λ-module with finite module presentation ha1, . . . , am for all j. Conversely, any m-tuple from N m with this property determines an el- ement from End Λ(N ). Thus EndΛ(N ) can be identified with the set of m-tuples Pi xj,iai, j = 1, . . . , T iΛ, where xj,i ∈ Λ for all i, j. Each element α of EndΛ(N ) is uniquely determined by the m-tuple (αa1, . . . , αam) ∈ N m, and one hasPi xj,i(αai) = 0 (α1, . . . , αm) ∈ N m that satisfy Pi xj,iαi = 0 for all j. In the particular case that Λ is a field we have that N is a vector space. In particular, N is a free Λ-module, and so it admits a finite presentation with an empty set of relations. In this case, EndΛ(N ) = N m. 16 The identification of EndΛ(N ) with a subset of N m is used to prove the following result. Lemma 3.7. Let N be a finitely generated Λ-module. Then the following hold: 1. (End Λ(N ); Lmod ) is e-interpretable in (N ; Lmod ). 2. Let SN = {a1, . . . , am} be a generating set of N , and define maps ·ai : EndΛ(N ) → N by α 7→ αai ∈ N . Denote ·SN = {·a1, . . . , ·am}. Then the two-sorted structure ENDΛ(N ) = (EndΛ(N ), N ; ·SN , Lmod ) is e-interpretable in (N ; Lmod ). 3. In the particular case that Λ is a field or the ring of integers Z, the previous statements are still valid after replacing Lmod by Lgroup in all structures. Proof. As mentioned above, since Λ is a Noetherian ring of scalars, any finitely generated Λ-module is finitely presented with respect to any finite generating set. Let P xj,iai, j = 1, . . . , T be a finite set of relations of N , with xj,i ∈ Λ for all i, j. Following Remark 3.6, identify each element α of EndΛ(N ) with the m-tuple (α1, . . . , αm) = (αa1, . . . , αam) ∈ N m. By this same remark, any m-tuple α = (α1, . . . , αm) belongs to EndΛ(N ) if and only if P xj,iαi = 0 for all j. This is a finite system of equations in N with variables αi, and so End Λ(N ) as a set is e-definable in (N ; Lmod ). As observed in Remark 3.6, if Λ is a field then End Λ(N ) = N n, and so the e-definition consists in an empty equation. In particular, it does not use multiplication by scalars. The group addition of two tuples from the Λ-module EndΛ(N ) is obtained by component-wise addition. Hence the graph of the addition operation of EndΛ(N ) (which is a subset of N 3m) is e-definable in N . By similar reasons, so are the graphs of the equality relation of EndΛ(N ) and of multiplication by fixed elements of Λ (i.e. multi- plication by scalars). This proves that (End Λ(N ); Lmod ) is e-interpretable in (N ; Lmod ). In the case that Λ is a field, (End Λ(N ); Lgroup) is e-interpretable in (N ; Lgroup). It follows, of course, that the two-sorted structure (EndΛ(N ), N ; Lmod ) is e-interpretable in (N ; Lmod ). Finally, notice that, for α ∈ EndΛ(N ) and x ∈ N , the tuple (α, x) = (α1, . . . , αm, x) ∈ N m+1 belongs to the graph of ·ai if and only if x = αi. In other words, (y1, . . . , ym+1) ∈ Graph(·ai) ⊆ N m+1 if and only if yi = ym+1, and so the graph of ·ai is e-definable in (N ; Lgroup) by the equation yi = ym+1. This completes the proof that ENDΛ(N ) is e-interpretable in (N ; Lmod ). If Λ is a field then multiplication by scalars was not used in any equation other than when e-interpreting the scalar multiplication of End Λ(N ), and if Λ = Z then a Λ-module is just a group, because nx = x + n. . . + x for all n ∈ Z. Hence, Item 3 holds. Remark 3.8. It follows from Lemma 3.7 that there exists an e-interpretation φ of the three-sorted structure F1 = (EndΛ(N ), N, M ; f, ·SN , Lmod ) 17 in F = (N, M ; f, Lmod ). If Λ is a field or Z, then one can replace Lmod by Lgroup. Thus by transitivity of e-interpretations (Proposition 2.5), in order to prove that (R(f ); Lalg) or (Z(Sym(f )); Lalg) is e-interpretable in F it suffices to show that it is so in F1. For this one must keep in mind that an equation in F1 can involve constants and variables from N , M , and End Λ(N ); the map f ; actions of endomorphisms on the ai's given by ·SN ; and the operations of (N ; Lmod ), (M ; Lmod ), and (EndΛ(N ); Lmod ) without its ring multiplication. For example, the equation f (αai, aj) = f (ai, αaj ) on the variable α is valid in F1, whereas α1α2ai = α2α1ai or αx = ai is not (for variables α1, α2, α ∈ EndΛ(N ), x ∈ N ). One can think of examples in which Λ = N = Z such that, if the ring multiplication of (End Λ(N ); Lalg ) was e-interpretable in F1, then multiplication of integers would be e-interpretable in the free abelian group (Z; Lgroup), contradicting the negative answer to the original Hilbert's 10th Problem. We next prove the main result of this subsection. Proof of Theorem 3.5. First observe that End Λ(N ) is Λ-module-finite, because N m is a Noetherian module and EndΛ(N ) embeds as a Λ-module into N m, by Remark 3.6. By the same reason both R(f ) and Z(Sym(f )) are Λ-module-finite as well. Denote F = (N, M ; f, Lmod ). We proceed to prove that (Z(Sym(f )); Lalg ) is e- interpretable in F . By the previous Remark 3.8, it suffices to show that (Z(Sym(f )); Lalg) is e-interpretable in F1 for some generating set SN = {a1, . . . , an} of N . We start by proving that Sym(f ) can be e-defined as a set in F1. Indeed, take any x, y ∈ N and write x = P xiai and y = P yiai for some xi, yi ∈ Λ. Since αx = P xiαai for all α ∈ EndΛ(N ), we have f (αx, y) = P xiyjf (αai, aj), and similarly for f (x, αy). It follows that Sym(f ) = {α ∈ EndΛ(N ) f (αai, aj) = f (ai, αaj ) for all 1 ≤ i, j ≤ n} . We conclude that Sym(f ) is e-definable as a set in F1 by the system ^1≤i,j≤n(cid:2)f (αai, aj) = f (ai, αaj )(cid:3) (5) on the single variable α ∈ EndΛ(N ). Here the expressions αat stands for ·at(α) (t = 1, 2), which are terms in the structure F1. Observe that (5) does not use multiplication by scalars Λ. Since the signature of F1 contains all operations of (EndΛ(N ); Lmod ), we have that Sym(f ) ≤ EndΛ(N ) as a Λ-module is e-interpretable in F1. Next, we show that Z(Sym(f )) is e-definable as a set in F1. As before, this im- mediately implies that the Λ-module (Z(Sym(f )); Lmod ) is e-interpretable in F1. Let β1, . . . , βk be a finite generating set of (Sym(f ); Lmod ). Then, α ∈ Z(Sym(f )) if and only if α ∈ Sym(f ) and αβt = βtα for all t = 1, . . . , k. This implies that f (αai, βtaj) = f (ai, αβtaj) = f (ai, βtαaj) = f (βtai, αaj ) for all i, j, t. (6) 18 We claim that (6) is a sufficient condition for an endomorphism α ∈ Sym(f ) to belong to Z(Sym(f )). As a consequence one has that Z(Sym(f )) is definable as a set in F1 by means of the following system of equations on the variable α: ^t=1,...,k, 1≤i,j≤n(cid:2)f (αai, βtaj) = f (βtai, αaj)(cid:3), (7) together with the system (5), which ensures that α ∈ Sym(f ). Again, we have written αai and βtaj instead of ·ai(α) and ·aj(βt), and similarly for the other entries of f . As before, (7) does not use multiplication by scalars Λ. To prove the claim suppose (7) holds. Then f (βtαai, aj) = f (αβtai, aj) for all i, j, t, and thus, for fixed i and t, f ([βt, α]ai, aj) = 0 for all j, where [βt, α] = βtα − αβt. By bilinearity of f and the fact that a1, . . . , an generate N , we have that f ([βt, α]ai, x) = 0 for all x ∈ N and for all i, t. Since f is non-degenerate, [βt, α]ai = 0 for all i, t. Similar arguments as before yield [βt, α]x = 0 for all x ∈ N , and thus [βt, α] = 0 for all t. This completes the proof of the claim. We have seen that the Λ-module (Z(Sym(f )); Lmod ) is e-interpretable in F1. By analogous reasons as above, for any triple γ1, γ2, γ3 ∈ Z(Sym(f )) the equality γ3 = γ1γ2 holds if and only if f (γ3ai, aj) = f (γ2ai, γ1aj) for all 1 ≤ i, j ≤ n. (8) Hence the ring multiplication of Z(Sym(f )) is e-interpretable in F1 by means of (8) and appropriate systems of the form (5) and (7) (which ensure that γi ∈ Z(Sym(f ))). We conclude that Z(Sym(f )) as a Λ-algebra is e-interpretable in F1. We now prove Item 1. As observed in the arguments above, multiplication by scalars of F1 was only used in order to e-interpret multiplication by scalars of Z(Sym(f )). Hence (Z(Sym(f )); Lring) is e-interpretable in F1 = (EndΛ(N ), N, M ; f, ·SN , Lgroup). By Lemma 3.7 and Remark 3.8, the latter structure is e-interpretable in (M, M ; f, Lgroup), and so (Z(Sym(f )); Lring) is e-interpretable as a ring in (M, M ; f, Lgroup). This con- cludes the proof of Item 1. Next we show that (R(f ); Lalg) is e-interpretable in (N, M ; f, Lgroup). By the previ- ous arguments and by transitivity of e-interpretations, it suffices to prove that (R(f ); Lalg) is e-interpretable in (Z(Sym(f )); Lalg). First recall from the proof of Theorem 3.3 that R(f ) is the set of all α ∈ Z(Sym(f )) such that if Xi f (xi, yi) = Xi This condition is equivalent to f (x′ i, y′ i), then Xi f (αxi, yi) = Xi f (αx′ i, y′ i). if Xi f (xi, yi) = 0, then Xi f (αxi, yi) = 0. (9) We claim that α ∈ Z(Sym(f )) satisfies (9) if and only if it satisfies the following condi- tion: if Xj,k zj,kf (aj, ak) = 0 for some zj,k ∈ Λ, then Xj,k zj,kf (αaj, ak) = 0. (10) 19 Indeed, the direct implication is immediate. Conversely, suppose that α satisfies (10), and let {xi, yi} be such that Pi f (xi, yi) = 0. Write each xi and yi in terms of the generators a1, . . . , an, xi,jaj, xi = Xj and yi = Xk yi,kak, xi,j, yi,k ∈ Λ. Since f is bilinear, f (xi, yi) = Xj,k Xi Xi xi,jyi,k! f (aj, ak) = 0. Thus 0 = Pj,kPi xi,jyi,kf (αaj, ak) = P f (αxi, yj). This completes the proof of the claim. The set S of all tuples (zi,j) ∈ Λn2 i,j=1 zi,jf (ai, aj) = 0 forms a submod- , and so it admits a finite generating set, say X = {si i = 1, . . . , T }. Write ule of Λn2 si = (si,j,k 1 ≤ j, k ≤ n). Then α ∈ Z(Sym(f )) belongs to R(f ) if and only if such that Pn Xj,k Xi tisi,j,k! f (αaj, ak) = 0, for all t1, . . . , tr ∈ Λ. (11) si,j,kf (αaj, ak)  = 0, for all t1, . . . , tr ∈ Λ. (12) Equivalently, Xi ti Xj,k By making appropriate choices for the ti's, one sees that (12) holds if and only if each one of the expressions inside the parenthesis is 0. It follows that R(f ) is e-definable as a set in (Z(Sym(f )); Lalg ). Consequently, R(f ) is e-interpretable as a Λ-algebra in (Z(Sym(f )); Lalg), since all the operations of (R(f ); Lalg) are already present in the signature of the latter. Finally we prove Item 2, i.e. that Z(Sym(f )), R(f ), N , and M are all simultaneously either zero, finite, or infinite. We claim that if Θ is a ring of scalars, then any finitely generated faithful Θ-module K is zero, finite, or infinite if and only if Θ is zero, finite, or infinite, respectively. Indeed, if K is finite (or zero), then End Θ(K) is finite (zero) as well, because EndΘ(K) embeds as a Θ-module into K n, for some n (see Remark 3.6). Since K is a faithful Θ-module, there exists an embedding Θ ֒→ EndΘ(K), and hence Θ is finite (zero) as well. On the other hand, if K is infinite, then, since K is finitely generated, there must exist k ∈ K such that the set {θk θ ∈ Θ} is infinite. The claim follows. Observe that both N and M are faithful R(f )-modules, and that N is also a faithful Z(Sym(f ))-module. We claim that all these modules are finitely generated. Indeed, let ΛN = {λ ∈ Λ λn = 0 for all n ∈ N }, and define ΛM similarly. Then N (resp. M ) is a finitely generated faithful Λ/ΛN -module (Λ/ΛM -module). Using that f is full and non-degenerate, one can see that N is also a faithful Λ/ΛM -module under the action 20 (λ + ΛM )x = λx. With this action, Λ/ΛM becomes a ring of scalars of f , and so by maximality of R(f ) we have Λ/ΛM ≤ R(f ) ≤ Z(Sym(f )) as subrings of End Λ(N ). Similar arguments yield Λ/ΛN ≤ R(f ) ≤ Z(Sym(f )). Hence N is finitely generated as a R(f )-module and a Z(Sym(f ))-module. Similarly, M is finitely generated as a R(f )- module. This completes the proof of the claim. Item 2 follows now from such claim and from the observation in the paragraph above. 3.3 Arbitrary bilinear maps In this subsection we keep the assumption that Λ is a (possibly infinitely generated) Noetherian ring of scalars. Our next goal is to generalize Theorem 3.5 to arbitrary bilinear maps. Theorem 3.9. Let f : A × B → C be a Λ-bilinear map between finitely generated Λ- modules. Then there exists a module-finite Λ-algebra of scalars Θ such that (Θ; Lalg) is e-interpretable in F = (A, B, C; f, Lmod ). Moreover, 1. If Λ is a field or the ring Z, then (Θ; Lring) is e-interpretable in (A, B, C; f, Lgroup). 2. Θ is zero, finite or infinite if and only if both hf (A, B)iΛ and A/Annl(f ) × B/Annr(f ) are zero, finite or infinite, respectively. The proof of this result relies in constructing from f a suitable full non-degenerate bilinear map of the form f : X × X → Y , so that we can apply Theorem 3.5 to it. To this end, let the left and right annihilators of f be, respectively, Annl(f ) = {a ∈ A f (a, y) = 0 for all y ∈ B} , Annr(f ) = {b ∈ B f (x, b) = 0 for all x ∈ A}. Observe that f induces a full non-degenerate Λ-bilinear map f1 : A/Annl(f ) × B/Annr(f ) → hf (A, B)iΛ, (13) where by hSiΛ we mean the Λ-submodule generated by S. Write F = (A, B, C; f, Lmod ), A1 = A/Annl(f ), B1 = B/Annr(f ), C1 = hf (A, B)iΛ, and F1 = (A1, B1, C1; f1, Lmod ). Note that A1, B1 and C1 are finitely generated since A, B and C are Noetherian modules. If A1 = B1, then f1 is of the desired form, and Theorem 3.5 can be applied to it. Otherwise consider the map f2 :(cid:0)A1 × B1(cid:1) ×(cid:0)A1 × B1(cid:1) → C1 × C1 (cid:0)(a, b), (a′, b′)(cid:1) 7→ (cid:0)f1(a, b′), f1(a′, b)(cid:1). (14) One can easily check that f2 is a full non-degenerate Λ-bilinear map between finitely generated Λ-modules. Denote F2 = (A1 × B1, C1 × C1; f1, Lmod ). Either f1 or f2 are of the desired form, and thus Theorem 3.5 can be applied to at least one of them. Moreover: Lemma 3.10. Both F1 and F2 are e-interpretable in F . The same is true if one replaces Lmod with Lgroup in both structures. 21 Proof. Let SA = {a1, . . . , an} and SB = {b1, . . . , bm} be generating sets of A and B, respectively. The submodules Annl(f ) and Annr(f ) are e-definable as sets in F by the systems of equations ∧if (x, bi) = 0 and ∧if (ai, y) = 0, respectively. Here x and y are variables taking values in A and in B, respectively. Any c ∈ C1 = hf (A, B)iΛ can be written as c = Xi,j λi,jf (ai, bj) = Xj f Xi λi,jai, bj! for some λi,j ∈ Λ. It follows that C1 is e-definable as a set in F by the equation z = Pj f (xj, bj) on variables z and {xj j = 1, . . . , n}. These variables take values in C and in A, respectively. The operations of Annl(f ), Annr(f ) and C1 are e-interpretable in F because they are already present in the signature of F . Hence (A1; Lmod ) and (B1; Lmod ) are e- interpretable as Λ-modules in (A; Lmod ) and (B; Lmod ), respectively (see Lemma 2.3 and Remark 2.4). Moreover, from the proof of Lemma 2.3 and the fact that the e-definitions of Annl(f ) and Annr(f ) do not use multiplication by scalars, we have that (A1; Lgroup) and (B1; Lgroup) are e-interpretable in (A; Lgroup) and (B; Lgroup), respectively. The preimage in F of the graph of f1 is e-definable in F by the system of equations z = f (x, y), z = Pj f (xj, bj) (the second equation ensures that z takes values in C1). Again this equation does not use multiplication by scalars. We conclude that F1 = (A1, B1, C1; f1, Lmod ) is e-interpretable in (A, B, C; f, Lmod ), and that the same holds if one drops multiplication by scalars in both structures. Finally, note that (A1 × B1; Lmod ) is e-interpretable in (A1, B1; Lmod ), and (C1 × C1; Lmod ) is e-interpretable in (C1; Lmod ) (they are basic sets of (A1, B1) and C1, and so they are defined as sets by empty systems of equations). The equations z = f1(x, y′) and z′ = f1(x′, y) on variables x, y, x′, y′, z, z′ e-define the graph of f2 in F1. It follows that the whole two-sorted structure F2 is e-interpretable in F1, and also in F by transitivity. Moreover, this is still valid if one replaces Lmod by Lgroup in the three structures F2, F1 and F . Proof of Theorem 3.9. The result follows immediately after using Theorem 3.5 in order to e-interpet (Z(Sym(f2)); Lalg ) or (Z(Sym(f1)); Lalg ) in F2 or F1, depending on whether or not A1 = B1, respectively. Items 1 and 2 are a direct consequence of Items 1 and 2 of Theorem 3.5. Remark 3.11. In Theorem 3.9 we have e-interpreted Z(Sym(f2)) in F2 (or Z(Sym(f1)) in F1 if A1 = B1). Alternatively one can also e-interpret the largest ring of scalars R(f2) of f2 in F2 (similarly for f1). This may have some advantages if one seeks to study the structure of A, B, C and f , because R(f2) is determined by "more properties" of these than Z(Sym(f2)). However, when it comes to the Diophantine problem, Z(Sym(f2)) is a more practical choice than R(f2), because it uses a simpler e-interpretation. For instance, as we have seen, if Λ is a field then one can drop multiplication by scalars in the e-interpretation of (Z(Sym(f2)); Lring), whereas there is no apparent way to do the same with (R(f2); Lring). 22 4 From rings of scalars to rings of integers of global fields In the previous section we established a method for passing from bilinear maps to commutative structures. In this section we provide a reduction from the latter to rings of number theoretic flavour. We refer to Subsection 2.3 for relevant conventions, notation, and terminology used throughout the paper. We start with two useful results. Lemma 4.1. Let R be a module-finite algebra over a Noetherian ring of scalars A. Then (R; Lalg) (in particular, (R; Lring)) is e-interpretable in (A; Lring). Proof. Let r1, . . . , rn be a finite set of generators of R as an A-module, and let An be the direct product of n copies of A, i.e. the free A-module of rank n. Consider the natural projection φ : An → R defined by φ(a1, . . . , an) = P airi. We claim that φ is an e-interpretation of (R; Lalg) in (A; Lring). Indeed, it is clear that φ is surjective and that φ−1(R) = An is e-definable in (A; Lalg ). From here on, we look at r1, . . . , rn indistinctly as a set of generators of R and as a natural base of An. We now check that the preimage of the equality relation on R is e-definable in A. Since A is Noetherian, R is finitely presented as an A-module. Let this presentation be hr1, . . . , rn w1, . . . , wkiA, where for all i = 1, . . . , k we have wi = P ai,jrj for some ai,j ∈ A (j = 1, . . . , n). If xi, yi ∈ A (i = 1, . . . , n), then P xiri = P yiri holds in R if and only if P xiri = P yiri + P zjwj holds in An. Expanding wj = P ai,jri, and collecting summands with the same ri, we conclude that Xi xiri = Xi yiri in R if and only if ^i  xi − yi = Xj zjai,j  in A. (15) The right-hand side is a finite system of equations on variables {xi, yi, zi i}. This defines in A the preimage of equality in R. Next we prove that the preimage of the graph of the ring multiplication in R is e- Take any three elements x, y, z of R, and let xj, yk, zi be elements of A such that x = definable in (A; Lalg). For each 1 ≤ i, j, k ≤ n let bj,k,i ∈ A be such that rjrk = P bj,k,iri. P xjrj, y = P ykrk, and z = P ziri. Then the equality in R xy = (cid:16)X xjrj(cid:17)(cid:16)X ykrk(cid:17) = X ziri = z is equivalent to xjykbj,k,i Xi  Xj,k  ri = Xi ziri. (16) Using (15) we see that (16) can be written as a system of equations on variables {xj, yk, zi} and some new variables {wℓ}. Finally, e-interpretability of the remaining operations of R (addition and mulitplica- tion by scalars) follows in a similar way using (15) and the expressions P xiri +P yiri = P(xi + yi)ri, and a (P xiri) = P(axi)ri, for a, xi, yi ∈ A. 23 We will also need the following: Lemma 4.2. Let I be a finitely generated ideal of a ring of scalars A. Then I as a set is e-definable in A, and (A/I; Lring) is e-interpretable in (A; Lring). Proof. Let a1, . . . , an be a generating set of I. Then the equation x = P xiri on variables {x} ∪ {xi i} e-defines I in A as a set. Lemma 2.3 now implies that (A/I; Lring) is e- interpretable in (A; Lring). 4.1 Subrings of global fields The goal of this subsection is to prove the result below. See Subsection 2.2 for an explanation of the terminology used here. Theorem 4.3. Let R be a finitely generated integral domain whose field of fractions K is a global field. Then the ring of integers of K is e-interpretable in R in the language of rings (with constants). Furthermore, if K has positive characteristic p, then (Fp[t]; Lring) is e-interpretable in (R; Lring), and D(R; Lring) is undecidable. The proof will be obtained essentially by putting together some already known re- sults, which we state below. Most of them can be found in Shlapentokh's book [43]. We will follow its notation and terminology closely. The next statement was originally proved by Shlapentokh in [41] (here we follow a different reference of the same author). Theorem 4.4 (10.6.2 and 2.1.2 and 2.1.10 of [43]). Suppose that K is a global function field of characteristic p, and let OK be its ring of integers. Then there exists a tran- scendental element t ∈ K such that (Fp[t]; Lring) is e-interpretable in (O; Lring), and D(O; Lring) is undecidable. Proof. Following the terminology of [43], Proposition 10.6.2 states that Fp[t] is Diophantine- generated over OK, while 2.1.2 and 2.1.10 of the same reference yield that in this case this is equivalent to Fp[t] being e-definable in (OK ; Lring) as a set (in the proof of Corol- lary 4.6 we will comment further on this equivalence). Now D(O; Lring) is undecidable because D(Fp[t]; Lring) is reducible to D(O; Lring), and because D(Fp[t]; Lring) is unde- cidable by [5]. We shall also use the fact, stated below, that integrality at a prime (and thus at finitely many) is e-definable in a global field. For global function fields this was proved by Shlapentokh in [42]. For number fields this is implicit in the work of Robinson [35, 36], and it was written down explicitly by Kim and Roush in [20]. See also [9] or Theorems 4.2.4 and 4.3.4 of [43] for a unified account of these results. Theorem 4.5 (E-definability of integrality at a prime for global fields). Let K be a global field and let p be a prime of K. Then the ring of S-integers OK,S = {x ∈ K ord p(x) ≥ 0 } is e-definable as a set in K. Here S is the set of all primes of K but p. Since OK,S is a subring of K, it follows that OK,S is e-interpretable as a ring in K (following the notation of the above Theorem 4.5). As a consequence we have: 24 Corollary 4.6. Let OK be the ring of integers of a global field K. Let S be a finite set of primes of K. In the case that K is a finite extension of Fp(t), assume further that S contains all the poles of t, i.e. that S∞ ⊆ S (see Remark 2.8). Then (OK ; Lring) is e-interpretable in (OK,S; Lring). This result is stated in Subsections 2.3.2 and 2.3.3 of [43] for number fields and global function fields, respectively, though we could not find full direct proofs. In any case, it follows quickly from several results of the cited reference, as we see now. Proof. We simply connect some results from Shlapentokh's book [43], while following its terminology and numeration. Unless stated otherwise all results that have 3 digits in the numeration are from this reference. The author uses the notion of Diophantine- generation (see Definition B.1.20) instead of e-definition or e-interpretation. In 2.1.2 and 2.1.10 it is proved that if R1 and R2 are two subrings of a global field such that R1 ≤ R2, then R1 is Dioph-generated over R2 if and only if R1 as a set is is e-definable in (R2; Lring), in which case (R1; Lring) is e-interpretable in (R2; Lring). During this proof we will always have the case R1 ≤ R2, and thus for our purposes there is no difference between Diophantine-generation and e-definability. Moreover, both of them imply e-interpretability as a ring. Denote the complement of a set of primes W by W c (in the set of all non-archimedean primes). Observe that OK,Sc = ∩p∈SOK,pc. Since S is finite, this intersection is finite, and by the previous Theorem 4.5 we obtain that OK,Sc is e-definable (in particular, Dioph-generated) in K. Lemma 2.2.2 states that the field of fractions of a domain D is Dioph-generated over D, provided that the set of non-zero elements of D is e-definable in D. In Proposition 2.2.4 it is proved that OK,W satisfies this last condition for any set of primes W . Therefore K is Dioph-generated over OK,S, and by transitivity of Dioph-generation (Theorem 2.1.15), OK,Sc is Dioph-generated over OK,S. In 2.1.19 it is seen that the intersection of finitely many subrings of an integral domain R, each Dioph-generated over R, is still Dioph-generated over R. Hence, if K is a number field, then OK = OK,∅ = OK,Sc ∩ OK,S is Dioph-generated over OK,S. Since OK ≤ OK,S, we have then that (OK ; Lring) is e-interpretable in (OK,S; Lring). Suppose otherwise that K is a global function field. Recall that in this case OK = OK,S∞ (see Remark 2.8). By the same argument as before, OK = OK,S∞ = OK,(S\S∞)c ∩ OK,S, and so OK is Dioph-generated over OK,S (recall that S∞ is finite and thus S\S∞ is finite as well). Since OK ≤ OK,S because S∞ ⊆ S, we obtain again that (OK ; Lring) is e-interpretable in (OK,S; Lring). We will also need the following auxiliary result. Lemma 4.7 (Corollary 4.6.5 of [13]). Let D be a finitely generated integral domain. Then the integral closure D of D in a finite extension of the field of fractions of D is module-finite over D (i.e. D is finitely generated as a D-module). We can now prove the main theorem of this subsection. 25 Proof of Theorem 4.3. By Lemma 4.7, the integral closure R of R in K is module-finite over R. Then Lemma 4.1 implies that (R; Lring) is e-interpretable in (R; Lring). Since R is module-finite over R and R is f.g., R is f.g. as well. Then Proposition 2.9 and Lemma 2.10 yield that R = OK,S for some global field K and some finite set S of primes of K. By Theorem 4.6, the ring of integers OK of K is e-interpretable in OK,S in the language of rings, and so by transitivity OK is e-interpetable in R in the same language. The last part of the theorem is a consequence of Theorem 4.4. Remark 4.8. Say that the rank of a ring is the maximum number of Z-linearly inde- pendent elements of the ring. Following the notation of Theorem 4.3, suppose that K is a number field and that R has finite rank n. We claim that R and OK have the same rank. Indeed, since R has characterstic zero, R is freely generated as a Z-module by n elements. It follows that K : Q = n, because K is the field of fractions of R. On the other hand, it is well known that the rank of OK coincides with the degree K : Q, and hence the claim is proved. 4.2 Rings of scalars Recall that by a ring (algebra) of scalars we mean an associative, commutative, unitary ring (algebra). In this subsection we prove the following: Theorem 4.9. Let A be an infinite finitely generated ring of scalars. Then there exists a ring of integers O of a global field such that (O; Lring) is e-interpretable in (A; Lring), and D(O; Lring) is reducible to D(A; Lring). Moreover: 1. If A is finitely generated as an abelian group, then O is a ring of algebraic integers. 2. If A has positive characteristic, then O is the ring of integers of a global function field, (Fp[t]; Lring) is e-interpretable in (A; Lring) for some prime p, and D(A; Lring) is undecidable. This theorem is proved by reducing it to the case when A is a subring of a global field, and then applying the main theorem of the previous subsection. To make such a reduction we closely follow Noskov's arguments from [30], where he proved that any infinite finitely generated ring of scalars has undecidable first order theory. Proof. Throughout the proof we will use implicitly the facts that e-interpretability is transitive, and that the quotient by any ideal of a Noetherian ring R is e-interpretable in R, by Lemma 4.2. Suppose first that A is not an integral domain, and let N be the nilradical of A, i.e. the ideal formed by all nilpotent elements of N . Equivalently, N is the intersection of all minimal prime nonzero ideals of A. There are finitely many such ideals q1, . . . , qn in a Noetherian ring (see Theorem 87 of [14]). Since A is not an integral domain, we have n ≥ 1, because A contains at least one nonzero maximal ideal. As argued in 3.2 of [30] there exists i such that A/qi is infinite. Moreover A/qi is a finitely generated integral domain. Note also that if A is finitely generated as an abelian group, then so is A/qi. Similarly, if A has positive characteristic, then so does A/qi. 26 In views of the previous paragraph, assume that A is an infinite finitely generated integral domain. The Krull dimension of A is the largest integer k for which there exists a proper ascending chain of prime ideals p0 < p1 < . . . < pk < A. Such k is finite under our assumptions (see Section 8.2.1 of [7]). It is not possible that k = 0, for in this case A would be a finitely generated Artinian domain, and thus a finitely generated field (see Proposition 8.30 of [2]), a contradiction because such a field is necessarily finite. Hence k ≥ 1. We may assume that k = 1, since if k ≥ 2 then A/pk−1 is a finitely generated integral domain of Krull dimension 1. Moreover A/pk−1 is finitely generated as an abelian group if A is, and it has positive characteristic if A does. In 1.3, 1.4, and 2.2 of [30] it is proved that if an infinite finitely generated integral domain R has Krull dimension 1, then one of the following hold: 1. R is a subring of a number field (if R has characteristic 0). 2. There exists a prime number p and a transcendental element t ∈ R such that Fp[t] ≤ R and R is integral over Fp[t]. This follows by Noether normalization assuming that R has positive characteristic p, in which case R is a finitely generated Fp-algebra. In particular, R is a subring of a global function field. It follows that the field of fractions K of A1 is a global field. Since A1 is finitely generated, Theorem 4.3 implies that the ring of integers OK of K is e-interpretable in A1 in the language of rings (with constants). By transitivity, (OK ; Lring) is e-interpretable in (A; Lring), and D(OK ; Lring) is reducible to D(A; Lring) by Proposition 2.6. If A is finitely generated as an abelian group, then it has characterstic 0, because it is infinite. Hence, A1 is a subring of a number field, and OK is a ring of algebraic integers. If otherwise A has characteristic p > 0, then Theorem 4.3 and transitiv- ity of e-interpretations yield that (Fp[t]; Lring) is e-interpretable in (A; Lring), and that D(A; Lring) is undecidable. Remark 4.10. The ring OK of Theorem 4.9 has rank at most the rank of R (recall that the rank of a ring is its maximum number of Z-linearly independent elements). This follows by Remark 4.8 and by the proof of the previous Theorem 4.9, since the rank of a ring does not increase after taking a quotient. 5 Rings and algebras over finitely generated rings of scalars The following lemma is a combination of the results obtained so far. It constitutes the main "general tool" presented in this paper (see also Lemma 6.4). Throuhgout the rest of the section we will explore its consequences. In a subsequent paper we plan to apply it further to the area of group theory. We refer again to Subsection 2.3 for important notation and terminology conventions. Lemma 5.1. Let Λ be a ring of scalars, let f : A × B → C be a Λ-bilinear map between finitely generated Λ-modules, and write C1 = hIm(f )iΛ. Suppose that (A, B, C; f, Lmod ) is e-interpretable in some structure M. Then there exists a module-finite Λ-algebra of 27 scalars R such that (R; Lalg) is e-interpretable in M. The ring R is 0, finite, or infinite if and only if C1 is 0, finite, or infinite, respectively. Furthermore, if C1 is infinite and Λ is finitely generated, then there exists a ring of integers O of a global field such that (O; Lring) is e-interpretable in (R; Lring) (hence in M), and additionally in this case: 1. If Λ has positive characteristic p, then (Fp[t]; Lring ) is e-interpretable in M, and D(M) is undeciable. 2. If Λ = Z then O is a ring of algebraic integers. If Λ is Z or a field, then the whole lemma holds after replacing (A, B, C; f, Lmod ) by (A, B, C; f, Lgroup) and (R; Lalg) by (R; Lring), i.e. multiplication by scalars is not re- quired. Proof. By Theorem 3.9, there exists a module-finite Λ-algebra of scalars R such that (R; Lalg) is e-interpretable in (A, B, C; f, Lmod ), and so in (R; Lalg) by transitivity of e-interpretations. The statement regarding the cardinality of R follows from Item 2 of Theorem 3.9. Suppose that C1 is infinite and that Λ is finitely generated. Then R is infinite and finitely generated as a ring. Hence, by Theorem 4.9 there exists a ring of integers O of a global field such that (O; Lring) is e-interpretable in (R; Lring), and thus in M. If Λ has positive characteristic, then so does R, because it is a unitary algebra over Λ. Hence (Fp[t]; Lring) is e-interpretable in M, by Item 2 of Theorem 4.9 and by transitivity. If Λ = Z, then R is finitely generated as an abelian group, and so O is a ring of algebraic integers by Item 1 of Theorem 4.9. If Λ is Z or a field, then (R; Lring) is e-interpretable in (A, B, C; f, Lgroup), by Item 1 of Theorem 3.9. Hence if the latter is e-interpretable in M, then the lemma holds after replacing Lalg by Lring and Lmod by Lgroup, due to the previous considerations and due to transitivity of e-interpretations. 5.1 Module-finite rings and algebras Throughout this subsection Λ denotes a finitely generated ring of scalars. The following is one of the main results of the paper. A similar theorem will be obtained in Subsection 6, where we study the case when Λ is an arbitrary field (possibly infinitely generated). The case Λ = Z will be considered separately afterwards. Theorem 5.2. Let R be a (possibly non-associative, non-commutative, and non-unitary) module-finite algebra over a finitely generated ring of scalars Λ. Suppose that R2 is infinite. Then there exists a ring of integers O of a global field such that (O; Lring) is e-interpretable in (R; Lalg), and D(O; Lring) is reducible to D(R; Lalg). Moreover: 1. If Λ has positive characteristic p, then (Fp[t]; Lring) is e-interpretable in (R; Lalg), and D(R; Lalg) is undecidable. 2. If R2 is finite and D(R; Lmod ) is decidable, then D(R; Lalg) is decidable. 28 If Λ is a finite field, then all the above holds after replacing (R; Lalg) by (R; Lring). Proof. The ring multiplication operation · of R induces a Λ-bilinear map between finitely generated Λ-modules · : R × R → R, with hIm(·)iΛ = R2. Since the three-sorted structure (R, R, R; ·, Lmod ) is e-interpretable in (R; Lalg), the theorem (except Item 2) follows from Lemma 5.1. We now prove Item 2. Let Σ be a system of equations in the Λ-algebra R. By adding new variables and equations, we may assume that Σ consists of 1) equations in the Λ-module (R; Lmod ) (i.e. Λ-linear equations), together with 2) equations of the form x1 = x2 · x3, where each xi is a variable that appears exactly once among all the equations of the second type. We say that x2 is a variable of type left, and that x3 is a variable of type right. Note that Annl(·) and Annr(·) are finite index submodules of R, by Item 2 of The- orem 3.9. Take a1, . . . , as and b1, . . . , bt to be full systems of coset representatives of R/Annl(·) and A/Annr(·), respectively. Let also SR be a finite generating set of R. For each variable x of type left, choose a coset representative ai(x) ∈ {a1, . . . , as}, and intro- duce a new variable x′. Then replace each occurrence of x in Σ by ai(x) + x′. We wish x′ to take values in Annl(·). Clearly, the system of equations ∧r∈SRzr = 0 on the variable z e-defines Annl(·) in R. Hence, for each new variable x′, add ∧r∈SRx′r = 0 to Σ in order to ensure that x′ takes values in Annl(·). Notice that this is a system of equations in the Λ-module (R; Lmod ). Proceed in an analogous way with each variable y of type right. Let Σ′ be the resulting system of equations. Since there are finitely many coset representatives, the number of all possible resulting systems Σ′ is finite. Let Σ1, . . . , Σm be all of them. It is clear that Σ has a solution if and only if Σi has a solution for some i. We now prove that it is possible to decide algorithmically if each one of the Σi has a solution or not, in which case our proof is concluded. Indeed, each Σi consists in some equations in the Λ-module (R; Lmod ), together with some equations of the form x = (ai + x′)(bj + y′), where x′ and y′ are bound to take values in Annl(·) and in Annr(·), respectively. Hence each equation x = (ai + x′)(bj + y′) can be replaced with x = aibj, which is an equation in (R; Lmod ). Thus Σi is equivalent to a system of Λ-linear equations, which is decidable by hypothesis. The following is essentially a particular case of the previous Theorem 5.2. It is stated separately due to its independent interest. Theorem 5.3. Let A be a ring (possibly non-associative, non-commutative, and non- unitary). Assume that A is finitely generated as an abelian group, and that A2 is infinite. Then there exists a ring of algebraic integers O such that (O; Lring) is e-interpretable in (A; Lring), and D(O; Lring) is reducible to D(A, Lring). If otherwise A2 is finite, then D(R; Lring) is decidable. Proof. The first part of the theorem follows in the same way as Theorem 5.2, taking Λ = Z and observing that here O is a ring of algebraic integers by Item 2 of Lemma 5.1. The last part follows by Item 2 of Theorem 5.2 (note that (A; Lmod ) is just a finitely generated abelian group, and so D(A; Lmod ) is decidable). 29 5.2 Finitely generated rings and algebras satisfying an infiniteness con- dition In this subsection Λ denotes a possibly infinitely generated ring of scalars. We next apply the previous Theorems 5.2 and 5.3 to certain classes of possibly non-module-finite finitely generated rings and algebras R. The approach consists in e-defining an ideal In in R that contains "enough" products of at least n elements of R (for example, the ideal generated by all such products), so that the quotient R/In is infinite and module-finite. Then it suffices to apply the results from Subsection 5.1, together with transitivity of e-interpretations. This approach has two limitations: 1. In can be difficult to e-define if R is non-associative (hence Definition 5.5). 2. R/In may be finite. For instance, if R is unitary one cannot simply take In to be the ideal generated by all products of n elements of R, since then In = R (hence the next definition). Definition 5.4. Any unitary Λ-algebra R admits a generating set of the form {1} ∪ T for some set T such that T ∩ {λ · 1 λ ∈ Λ} = ∅. We let In(T ), or simply In, be the ideal generated by all all products of n elements of T (n ≥ 1). If R is non-unitary, then In denotes the ideal generated by all Λ-multiples of all products of n elements of R. In both cases In is a Λ-subalgebra of R. Throughout the rest of the section R denotes a finitely generated Λ-algebra, possibly non-module finite, non-associative, non-commutative, and non-unitary. We fix a finite set T = {a1, . . . , am} as in Definition 5.4, and we define the ideals In accordingly. Definition 5.5. Suppose that R is non-associative. Then R is called right-norm- generated if, for all n ≥ 1, In is generated as a Λ-module by a (possibly infinite) subset of {(ai1(ai2(. . . (aik−1aik) . . . ))) k ≥ n, 1 ≤ i1, . . . , ik ≤ m}. Later we will see that many graded Lie algebras are right-normed-generated. Lemma 5.6. Suppose that R is associative or right-normed-generated, and let n ≥ 1. Then (R/In; Lring) and (R/In; Lalg) are e-interpretable in (R; Lring) and (R; Lalg), respectively. Moreover, R/In is a module-finite Λ-algebra. Proof. Let T = {a1, . . . , am}, and assume first that R is right-normed-generated. Then each element of In is a finite sum of elements of the form λ(ai1 (ai2(. . . (aik−1aik ) . . . ))), λ ∈ Λ, k ≥ n. (17) Hence each element as in (17) can be written in the form (ai1(. . . (ain−1y) . . . )) for some y ∈ R in the non-unitary case, and in the form (ai1(. . . (ain−1 (ainy)) . . . )) in the unitary case. Consequently, if R is non-unitary, then In is e-definable as a set in (R; Lring) by the equation x = X1≤i1,...,in−1≤m (ai1(. . . (ain−1yi1,...,in−1) . . . )) (18) 30 on variables x and {yi1,...,in−1} (the e-definition is in (R; Lring) because it makes no use of multiplication by scalars Λ). Observe however that if R is unitary then a solution to (18) may yield x ∈ In−1, for example if each yi1,...,in−1 is a Λ-multiple of 1. This can be solved by taking the equation x = X1≤i1,...,in≤m (ai1 (. . . (ain−1 (ainyi1,...,in)) . . . )) (19) By our previous considerations, (19) e-defines In as a set in (R; Lring) when R is unitary. Thus in both cases (R/In; Lring) and (R/In; Lalg) are e-interpretable in (R; Lring) and (R; Lalg) by Lemma 4.2. If R is associative, then any element x ∈ In is also a sum of elements of the form (17), and the proof follows as in the non-associative case. Finally, note that R/In is generated as a Λ-module by the projection of all products of less than n elements of T , together with 1 + In if R is unitary. It follows that R/In is module-finite. We now state the main result of this subsection. The ideals In are defined with respect to any set T satisfying the condition of Definition 5.4. Theorem 5.7. Let R be a finitely generated Λ-algebra (possibly non-module-finite, non- associative, non-commutative and non-unitary). Suppose that Λ is finitely generated, that R is associative or right-normed-generated, and that (R/In)2 is infinite for some n ≥ 1. Then there exists a ring of integers O of a global field such that (O; Lring) is e-interpretable in (R; Lalg), and D(O; Lring) is reducible to D(R; Lalg). Moreover: 1. If Λ has positive characteristic p, then (Fp[t]; Lring) is e-interpretable in (R; Lalg), and D(R; Lalg) is undecidable. 2. If R is a ring (i.e. Λ = Z) then O is a ring of algebraic integers. If Λ is Z or a finite field then all the above holds after replacing (R; Lalg) by (R; Lring). Proof. By Lemma 5.6, R/In is a module-finite Λ-algebra that is e-interpretable as an algebra in (R; Lalg). The same result states that R/In is e-interpretable as a ring in (R; Lring). By hypothesis, we can take n so that (R/In)2 is infinite. Now the result follows by Theorems 5.2 and 5.3 applied to R/In, and by transitivity of e-interpretations. We next apply the previous theorem to a class of non-associative algebras. Recall that a Λ-algebra L is called graded (over a semigroup (I, +)) if there exists a Λ-module direct-sum decomposition L = Li∈I Li such that LiLj ⊆ Li+j for all i, j ∈ I. In this case, each Li is called an homogeneous component of L, and the elements from Ln are said to be homogeneous of degree n. We say that L is simply graded if L1 generates A as a Λ-algebra, assuming that L is graded over N = {1, 2, . . . }. The following result is a generalization of Lemma 3.3 of [1]. The proof is essentially the same. 31 Lemma 5.8. Any simply graded Lie Λ-algebra is right-normed-generated. Proof. Let L = ⊕n≥1Ln be a gradation of L over N = {1, 2, . . . } (note that L is non- unitary), and let T = {a1, . . . } be a generating set of L1 as a Λ-module. Then T generates L as a Λ-algebra, and moreover the degree of a product of k elements from T is k. We prove that the set S = {[ai1 , [. . . [aik−1, aik ] . . . ]] aij ∈ T, k ≥ 1} generates L as a Λ-module. It suffices to see that any homogeneous element w ∈ L is a Λ-linear combination of elements from S. This follows by induction on the degree of w and by the identity [[x, y], z] = [x, [y, z]] − [y, [x, z]] (see Lemma 3.3 of [1]). Now let n ≥ 1 and x ∈ In(T ) ⊆ ⊕i≥nLi (see Definition 5.4 for an explanation of the terminology In(T )). By the previous claim, x is a Λ-linear combination of elements from S, and so x = Ps∈S λss for some λs ∈ Λ, almost all 0. Using the gradation of L one obtains that λs = 0 for all s of degree less than n, i.e. for all s = [ai1, [. . . [aik−1, aik ] . . . ]] with k < n. In conclusion, In is generated as a L-module by all elements of S of degree at least n. The next two corollaries follow immediately from Theorem 5.7 and Lemma 5.8. As usual, [R/In, R/In] denotes the Λ-module generated by {[x, y] x, y ∈ R/In}. Corollary 5.9. Let L be a finitely generated simply graded Lie Λ-algebra. Assume that [R/In, R/In] is infinite for some n ≥ 1, and that Λ is finitely generated. Then the conclusions of Theorem 5.7 hold for L. Corollary 5.10. Let F be a free associative Λ-algebra (possibly non-commutative and non-unitary) or a free Lie algebra of rank at least 2, with Λ finitely generated. Then the conclusions of Theorem 5.7 hold for F . Proof. In both cases (F/In)2 is infinite for all n ≥ 3. Moreover, any free Lie algebra is simply graded. Thus the result follows from Theorem 5.7 and the previous Corollary 5.9. Corollary 5.10 complements Romankov's [37], and Kharlampovich and Miasnikov's [15, 17] papers on free algebras. In the first reference it is proved that D(F ; Lring) is undecidable for many types of free rings F . In particular it is proved that the algebras of Corollary 5.10 actually have undecidable Diophantine problem if Λ = Z. In the references [15, 17] it is proved that D(F ; Lring) is undecidable if Λ is an arbitrary field, and F is a free associative non-commutative unitary algebra, or a free Lie algebra of rank at least 3. Note that an infinite field is necessarily infinitely generated, and so our Corollary 5.10 does not cover many of the cases considered in [15, 17]. 5.3 Finitely generated associative commutative non-unitary rings and algebras In this subsection we study non-unitary rings and algebras of scalars. Lemma 5.11. Let A be a finitely generated associative commutative non-unitary algebra over a (possibly infinitely generated) ring of scalars Θ. Then the following exist: 32 1. A ring of scalars Λ and a module-finite Λ-algebra of scalars B such that (B; Lalg) is e-interpretable in (A; Lalg ). 2. A finitely generated Θ-algebra of scalars C such that (C; Lalg) is e-interpetable in (A; Lalg). Additionally, C is module-finite if A is module-finite, and if A2 is zero, nonzero or infinite, then both B and C are zero, nonzero or infinite as well, respectively. Proof. The set Λ = Θ + A = {θ + a θ ∈ Θ, a ∈ A} is a ring of scalars under the obvious operations of addition and multiplication. Moreover, Λ acts naturally by endomorphisms on A, and with this action A is a Λ-algebra. During this proof we write AΛ and AΘ to refer to A seen as a Λ-algebra or as a Θ-algebra, respectively (we will proceed similarly with other algebras below). The operation of multiplication by a given scalar θ + a ∈ Λ is e-interpreted in (AΘ; Lalg) by the equation y = θx + ax. Thus, (AΛ; Lalg) is e-interpretable in (AΘ; Lalg). Suppose that AΘ is generated as a Θ-algebra by n elements a1, . . . , an. Then AΛ is generated as a Λ-module by these same elements, since for all x ∈ A there exists y1, . . . , yn ∈ A ≤ Λ such that x = Pi yiai. In particular, A is a module-finite Λ-algebra. The ring multiplication of AΛ is a Λ-bilinear map between finiely generated Λ- modules · : A × A → A. Moreover, (AΛ, AΛ, AΛ; ·, Lmod ) is e-interpretable in (AΛ; Lalg), which in turn is e-interpretable in (AΘ; Lalg). Hence, by the first part of Lemma 5.1, there exists a module-finite Λ-algebra of scalars DΛ such that (DΛ; Lalg) is e- interpretable in (AΘ; Lalg). Since Θ ≤ Λ, also D is a Θ-algebra. Clearly, (DΘ; Lalg) is e-interpretable in (DΛ; Lalg), and so DΘ is e-interpetable in AΘ in the language of Θ-algebras (with constants). Let SD be a finite set of generators of DΛ as a Λ-module. Identify each ai with the element ai · 1 of D. Write SA = {a1, . . . , an}. Then DΘ is generated as a Θ-algebra by the finite set SD ∪ SA. Now taking B = DΛ and C = DΘ completes the proof of Items 1 and 2. A similar argument as above shows that if AΘ is generated as a Θ-module by a finite set, say SA, then DΘ is generated as a Θ-module by SD ∪ SA, thus in this case both AΘ and DΘ are module-finite. Finally, if (AΘ)2 is zero, nonzero or infinite, then (AΛ)2 = hIm(·)iΛ is nonzero or infinite as well, respectively, and hence the three B, C and D are also respectively zero, nonzero or infinite, by Lemma 5.1. Below we obtain a non-unitary version of Theorem 4.9. We convene that the charac- teristic of a non-unitary ring A is defined as the maximum positive integer n such that nx = 0 for all x ∈ R. Theorem 5.12. Let A be a finitely generated associative commutative non-unitary ring, with A2 infinite. Then the conclusions of Theorem 4.9 hold for (A; Lring). Proof. Let n be the characteristic of A, and write Zn = Z/nZ (if n = 0 let Zn = Z). Then A is a faithful finitely generated Zn-algebra. Let C be the infinite finitely generated Zn-algebra given by Lemma 5.11. By this same result, C as a Zn-algebra (and thus as 33 a ring) is e-interpretable in A as a Zn-algebra. The latter is e-interpretable in (A; Lring) because multiplication by any scalar s ∈ Zn is e-interpretable in (A; Lgroup) by the equation y = x + s. . . + x. Hence (C; Lring) is e-interpretable in (A; Lring). Note that C is finitely generated as a ring because it is finitely generated as a Zn-algebra. So by Theorem 4.9 applied to C there exists a ring of integers O of a global field such that (O; Lring) is e-interpretable in (C; Lring), and also in (A; Lring) by transitivity. By Lemma 5.11 again, if A is Zn-module-finite then so is C. It follows then that if A is finitely generated as an abelian group then so is C. In this case Item 1 of Theorem 4.9 yields that O is a ring of algebraic integers. Finally, if A has positive characteristic (i.e. n > 0) then C has positive characteristic as well, being a Zn-algebra. Hence by Item 2 of Theorem 4.9 there exists a prime p such that (Fp[t]; Lring) is e-interpretable in (C; Lring), and so in (A; Lring). This makes D(A; Lring) undecidable in this case. Lemma 5.11 can be used further to study non-unitary Θ-algebras of scalars, for certain Θ. We note that, unlike in several previous results, here we cannot replace Lalg by Lring in case Θ is a field. Note also that the case Θ = Z has already been treated in the previous Theorem 5.12. Theorem 5.13. Let L be a finitely generated associative commutative non-unitary al- gebra over a finitely generated ring of scalars Θ, with L2 infinite. Then there exists a ring of integers O of a global field such that (O; Lring) is e-interpretable in (L; Lalg), and D(O; Lring) is reducible to D(L; Lalg). Moreover, if Θ has positive characteristic p, then (Fp[t]; Lring) is e-interpretable in (L; Lalg), and D(L; Lalg) is undecidable. Proof. Let B be the module-finite Λ-algebra of scalars given by Item 1 of Lemma 5.11, where Λ = Θ + B. Suppose that L2 is infinite. By this same lemma, B is infinite, and since B is unitary we have that B2 = B is infinite as well. Note further that if Θ has positive characteristic then so does Λ. The result now follows by transitivity of e-interpretations and by Theorem 5.2 applied to B. 5.4 Undecidability of first order theories The first order theory T (or elementary theory) of a structure M over a language L is the set of all first order sentences over L that are true in M . One says that T is decidable if there exists an algorithm that, given a sentence φ over L, determines if φ is true in M or not, i.e. if φ belongs to T . If such an algorithm does not exist then T is said to be undecidable. Noskov proved in [30] that the first order theory of an infinite finitely generated ring of scalars is undecidable in the language of rings with constants. In particular this is true for the ring of integers of any global field. Thus using transitivity of e-interpretations and Proposition 2.6 we immediately obtain the following: Theorem 5.14. Let R be a ring or an algebra satisfying the hypotheses of one of the Theorems 4.9, 5.2, 5.3, 5.7, 5.12, or Corollary 5.9. Let L denote Lring if R is a ring or an algebra over a field. Otherwise let L = Lalg. Then the first order theory of R in the language L (with constants) is undecidable. 34 The same statement holds if R satisfies the hypotheses of Theorem 5.13, i.e. if R is a finitely generated associative commutative non-unitary algebra over a finitely generated ring of scalars Λ, with R2 infinite. In this case however we need to take L to be the language of Λ-algebras, regardless of whether Λ is a field or not. In particular, Theorem 5.14 holds for finitely generated associative commutative non-unitary rings R with R2 infinite (by Theorem 5.12). This result is close to Noskov's original one, but to our knowledge there is no apparent way to prove it with "direct" methods from commutative algebra. 6 Finitely generated algebras over infinite fields Throughout this subsection k denotes an arbitrary field. We next study finitely generated algebras over k. Say R is one such algebra. A separate treatment is required, because an infinite field is necessarily infinitely generated as a ring. Hence the general tool (Lemma 5.1) from the previous section cannot be applied to R. More precisely, the k-algebra of scalars L obtained from Theorem 3.9 by considering the k-bilinear map · : R × R → R is not finitely generated as a ring, and so the results of Section 5 cannot be applied to L. Except in Corollary 6.3, all e-interpretations and reductions of this section are done in the language of rings with constants: the full expressiveness of the language of algebras is never required. This contrasts with our previous results obtained for algebras over arbitrary f.g. rings of scalars. Note also that the algebras in this section are no longer countable, and thus Diophantine problems must be considered using coefficients in a fixed computable subset (see Subsections 2.3 or 2.1.2). We next state the analogue of Theorem 4.9 for algebras over fields. Recall that an algebra of scalars is an associative commutative unitary algebra, and that by function field over k we refer to a finite field extension of k(t) (see Subsection 2.2). The Krull dimension of a ring R is the largest integer n for which there exists a strictly ascending chain of prime ideals p0 ≤ · · · ≤ pn in R. We refer to Section 8 of [7] for further information on this notion. Theorem 6.1. Let R be a nonzero finitely generated k-algebra of scalars. Suppose that R has Krull dimension at least one. Then (Z; Lring) is e-interpretable in (R; Lring), and D(R; Lring) is undecidable. If otherwise R has Krull dimension zero, then there exists a finite field extension K of k such that (K; Lring) is e-interpretable in (R; Lring). To prove this theorem we shall need the following result due to Shlapentokh. Its proof (besides a trivial last step that we prove below) can be found in Theorems 3.1 and 4.1 of [39], and in Theorem 3.1 and Remark 3.1 of [40]. Theorem 6.2 ([39, 40]). Let S be a nonempty finite set of primes of a function field K. Then (Z; Lring) is e-interpretable in (OK,S; Lring). Proof. In Theorem 3.1 and Remark 3.1 of [40] it is shown that if K has characteristic 0 then (Z; Lring) is e-interpretable in (OK,S; Lring). In Theorems 3.1 and 4.1 of [39] it is 35 proved that if otherwise K has positive characteristic p, then (Z; +, p, ) is e-interpretable in (OK,S; Lring)2. The former is the so called arithmetic with addition and localized divisibility. Here is the relation of divisibility (xy iff ∃z y = xz), and p is the relation of localized divisibility: xpy iff there exists n ∈ N such that y = ±xpn. In Section 4 of [5], Denef shows that (Z; +, p, ) is e-interpretable in (Z; Lring) (using only 0 and 1 as coefficients), and so the result follows by transitivity of e-interpretations. We next prove Theorem 6.1. Several ideas are parallel to arguments used in Sub- sections 4.1 and 4.2, with the main difference being that the structures at hand behave differently when the Krull dimension is 0. Indeed, a f.g. integral domain R of Krull dimension 0 has to be necessarily a finite field, while if R is a nontrivial f.g. k-algebra of Krull dimension 0, then R is a finite field extension of k. Proof of Theorem 5.7. Note that any f.g. k-algebra A is Noetherian, because k is a Noetherian ring. Thus (A/I; Lalg ) is e-interpretable in (A; Lalg) for any ideal I, by Lemma 4.2. Throughout the proof we will use implicitly this fact, as well as transitivity of e-interpretations. Suppose first that R is not a domain, and let p0 ≤ · · · ≤ pd be a maximal ascending chain of prime ideals of R. Then R/p0 is a domain with the same Krull dimension as R. Moreover R/p0 is nonzero (since by definition a prime ideal is proper). In views of the previous paragraph, asssume for the rest of the proof that R is an integral domain, and let F be the field of fractions of R. Since R is a finitely generated algebra over a field k, the Krull dimension d of R coincides with the transcendence degree of F over k (see Section 8.2.1 of [7]), and by Noether normalization there exist d algebraically independent (over k) elements t1, . . . , td ∈ R such that k[t1, . . . , td] ≤ R and R is module-finite over k[t1, . . . , td]. Note as well that F is a finite extension of k[t1, . . . , td]. Suppose that d = 0. Since R is Noetherian, this is equivalent to saying that R is Artinian (see Proposition 9.1 of [7]). All Artinian domains are fields, and thus R = F is a finite field extension of k (see Proposition 8.30 of [2]). This completes the proof of the theorem when R is an integral domain with Krull dimension 0, and so the proof of the last part of the theorem is complete. Assume now that d ≥ 1. As done in the proof of Theorem 4.9, we may assume that d = 1 by factoring out the second last ideal in a maximal ascending chain of prime ideals of R. Then, as noted above, F is a finite extension of k(t1), and thus it is a function field over k. Let R be the integral closure of R in F . By Theorem 4.6.3 of [13], R is module-finite over R. Hence (R; Lring) is e-interpretable in (R; Lring) by Lemma 4.1. The ring R is integrally closed and its field of fractions is F . Moreover, R is module- finite over R, and so it is finitely generated as a k-algebra. Hence by Proposition 2.9 and Lemma 2.10 we have that R = OF,S for some finite set S of primes of F . Since k[t1] ≤ R ≤ R, it follows that S must contain all the poles of t (see Remark 2.8). In particular, S is nonempty. This and Theorem 6.2 complete the proof of the theorem. 2We remark that the condition of S being nonempty is implicit in [39, 40] (otherwise OK,∅ coincides with the field of constants of K). 36 Below we obtain a similar result for the case when R is non-unitary, though in this case one must consider R with the language of k-algebras, and not just with the language of rings. Notice also that one does not have such a transparent classification in terms of Krull dimension. Corollary 6.3. Let R be a finitely generated associative commutative non-unitary k- algebra. Assume that R2 6= 0. Then either the first or the second conclusion of Theorem 5.7 hold for (R; Lalg). Proof. By Item 2 of Lemma 5.11 there exists a nonzero finitely generated k-algebra of scalars L such that (L; Lalg) (in particular (L; Lring)) is e-interpretable in (R; Lalg). This same lemma states that if R2 is nonzero then L is nonzero as well. Now the result follows by Theorem 6.1 and transitivity of e-interpretations. Similarly as done in Section 5, a combination of Theorem 6.1 and our main result for bilinear maps (Theorem 3.9) yields the following analogue of Lemma 5.1. Note that the full power of the previous Theorem 6.1 is not exploited here, since the aforemen- tioned result for bilinear maps always gives finite-dimensional k-algebras, hence of Krull dimension 0 by Noether normalization. Lemma 6.4. Let f : A × B → C be a k-bilinear map between finite-dimensional k- modules, with Im(f ) 6= 0 . Suppose that (A, B, C; f, Lgroup) is e-interpretable in a struc- ture M. Then there exists a finite field extension F of k that is e-interpretable in M in the language of rings (with constants), and D(F ; Lring) is reducible to M. Proof. By Theorem 3.9, there exists a nonzero finite-dimensional k-algebra of scalars R such that (R; Lring) is e-interpretable in (A, B, C; f, Lgroup), and thus in M. Since R is finite-dimensional, the Krull dimension of R is necessarily 0 (by Noether normalization), and the lemma now follows by Theorem 6.1. The following can be proved in the exact same way as Theorems 5.2, 5.7 and Corol- lary 5.9, using Lemma 6.4 instead of Lemma 5.1. Below, R may be non-associative, non-commutative, and non-unitary. See Subsection 5.2 for an explanation of the termi- nology used here. Theorem 6.5. Let R be a finitely generated algebra over a field k. Suppose that R satisfies one of the following: 1. R has finite dimension over k, and R2 6= 0. 2. R is associative or right-normed-generated, and (R/In)2 6= 0 for some n ≥ 1. 3. R is a simply graded Lie algebra and [R/In, R/In] 6= 0 for some n ≥ 1. Then there exists a finite field extension K of k such that (K; Lring) is e-interpretable in (R; Lring), and D(K; Lring) is reducible to D(R; Lring). 37 Proof. The three-sorted structure (R, R, R; Lgroup, ·) is e-interpretable in (R; Lring), where as usual · denotes the ring multiplication of R. Hence Item 1 follows from Lemma 6.4. If R is associative or right-normed-generated, then (R/In; Lring) is e-interpretable in (R; Lring) by Lemma 5.6, and so Item 2 follows by such lemma and by Item 1 (and by transitivity of e-interpretations). Item 3 is a particular case of Item 2, given that any simply graded Lie algebra is right-normed-generated due to Lemma 5.8. Hence if k is a field all of whose finite extensions have undecidable Diophantine problem, then D(R; Lring) is undecidable. This occurs, for example, if k is a global function field [8, 39]. 7 References [1] E. S. Chibrikov. A right normed basis for free Lie algebras and Lyndon-Shirshov words. J. Algebra, 302(2):593–612, 2006. [2] P. L. Clark. Commutative algebra, 2015. Available at http://math.uga.edu/~pete/integral.pdf. [3] M. Davis, H. Putnam, and J. Robinson. The decision problem for exponential Diophantine equations. Annals of Mathematics, 74(3):425–436, 1961. [4] J. Denef. The Diophantine problem for polynomial rings and fields of rational functions. Trans. Amer. Math. Soc. 242, pages 391–399, 1978. [5] J. Denef. The Diophantine problem for polynomial rings of positive characteristic. In Maurice Boffa, Dirkvan Dalen, and Kenneth Mcaloon, editors, Logic Colloquium '78, volume 97 of Studies in Logic and the Foundations of Mathematics, pages 131 – 145. Elsevier, 1979. [6] J. Denef and L. Lipshitz. Diophantine sets over some rings of algebraic integers. Journal of the London Mathematical Society, s2-18(3):385–391, 1978. [7] D. Eisenbud. Commutative Algebra: With a View Toward Algebraic Geometry. Graduate Texts in Mathematics. Springer, 1995. [8] K. Eisenträger. Hilbert's Tenth Problem for algebraic function fields of character- istic 2. Pacific Journal of Mathematics, 210, 08 2002. [9] K. Eisenträger. Integrality at a prime for global fields and the perfect closure of global fields of characteristic p>2. Journal of Number Theory, 114(1):170 – 181, 2005. [10] M. D. Fried and M. Jarden. Field arithmetic. Ergebnisse der Mathematik und ihrer Grenzgebiete. Springer-Verlag, 1986. [11] K. R. Goodearl and R. B. Warfield. An Introduction to Noncommutative Noetherian Rings. London Mathematical Society St. Cambridge University Press, 2004. 38 [12] W. Hodges. Model theory, volume 42 of Encyclopedia of Mathematics and its Ap- plications. Cambridge University Press, Cambridge, 1993. [13] C. Huneke and I. Swanson. Integral Closure of Ideals, Rings, and Modules. Number v. 13 in Integral closure of ideals, rings, and modules. Cambridge University Press, 2006. [14] I. Kaplansky. Commutative Rings. Chicago lectures in mathematics. University of Chicago Press, 1974. [15] O. Kharlampovich and A. G. Miasnikov. Equations in algebras. ArXiv e-prints, June 2016. [16] O. Kharlampovich and A. G. Miasnikov. What does a group algebra of a free group know about the group? ArXiv e-prints, July 2016. [17] O. Kharlampovich and A. G. Miasnikov. Undecidability of equations in free Lie algebras. ArXiv e-prints, August 2017. [18] O. Kharlampovich and A. G. Miasnikov. Undecidability of the first order theories of free non-commutative Lie algebras. ArXiv e-prints, April 2017. [19] O. Kharlampovich and A. G. Miasnikov. Tarski-type problems for free associative algebras. Journal of Algebra, 500:589 – 643, 2018. Special Issue dedicated to Efim Zelmanov. [20] K. H. Kim and F. W. Roush. An approach to rational Diophantine undecidability. In Proceedings of Asian Mathematical Conference, page 242–248, River Edge, NJ, 1990. World Sci. Publ. [21] K. H. Kim and F. W. Roush. Diophantine undecidability of c(t1, t2). Journal of Algebra, 150(1):35 – 44, 1992. [22] J. Koenigsmann. Undecidability in Number Theory, pages 159–195. Springer Berlin Heidelberg, Berlin, Heidelberg, 2014. [23] D. Marker. Model Theory : An Introduction. Graduate Texts in Mathematics. Springer New York, 2002. [24] Y. V. Matijasevič. The Diophantineness of enumerable sets. Dokl. Akad. Nauk SSSR, 191:279–282, 1970. [25] B. Mazur and K. Rubin. Ranks of twists of elliptic curves and Hilbert's Tenth Problem. Inventiones mathematicae, 181(3):541–575, Sep 2010. [26] A. G. Miasnikov. Definable invariants of bilinear mappings. Siberian Mathematical Journal, 31(1):89–99, Jan 1990. [27] A. G. Miasnikov, F. Oger, and S. Sohrabi. Elementary equivalence of rings with finitely generated additive groups. Annals of Pure and Applied Logic, 2018. 39 [28] A. G. Miasnikov and M. Sohrabi. Groups elementarily equivalent to a free nilpotent group of finite rank. Annals of Pure and Applied Logic, 162(11):916 – 933, 2011. [29] A. G. Miasnikov and M. Sohrabi. Elementary coordinatization of finitely generated nilpotent groups. ArXiv e-prints, November 2013. [30] G. A. Noskov. Elementary theory of a finitely generated commutative ring. Math- ematical notes of the Academy of Sciences of the USSR, 33(1):12–15, Jan 1983. [31] T. Pheidas and K. Zahidi. Undecidability of existential theories of rings and fields: A survey. Contemporary Mathematics, 270, 49-106, 2000. [32] B. Poonen. Using elliptic curves of rank one towards the undecidability of Hilbert's Tenth Problem over rings of algebraic integers. In C. Fieker and D.R. Kohel, editors, Algorithmic Number Theory, pages 33–42, Berlin, Heidelberg, 2002. Springer Berlin Heidelberg. [33] B. Poonen. Hilbert's tenth problem and mazur's conjecture for large subrings of Q. Journal of the American Mathematical Society, 16(4):981–990, 2003. [34] B. Poonen. Hilbert's Tenth Problem over rings of number-theoretic interest. http://math.mit.edu/~poonen/papers/aws2003.pdf, February 2003. Notes for Arizona Winter School on "Number theory and logic". [35] J. Robinson. Definability and decision problems in arithmetic. The Journal of Symbolic Logic, 14(2):98–114, 1949. [36] J. Robinson. The undecidability of algebraic rings and fields. Proceedings of the American Mathematical Society, 10(6):950–957, 1959. [37] V. A. Roman'kov. Unsolvability of the endomorphic reducibility problem in free nilpotent groups and in free rings. Algebra and Logic, 16(4):310–320, Jul 1977. [38] H. N. Shapiro and A. Shlapentokh. Diophantine relationships between algebraic number fields. Communications on Pure and Applied Mathematics, 42(8):1113– 1122, 1989. [39] A. Shlapentokh. Hilbert's Tenth Problem for rings of algebraic functions in one vari- able over fields of constants of positive characteristic. Transactions of the American Mathematical Society, 333(1):275–298, 1992. [40] A. Shlapentokh. Hilbert's Tenth Problem for rings of algebraic functions of char- acteristic 0. Journal of Number Theory, 40(2):218 – 236, 1992. [41] A. Shlapentokh. Diophantine relations between rings of S-integers of fields of alge- braic functions in one variable over constant fields of positive characteristic. The Journal of Symbolic Logic, 58(1):158–192, 1993. 40 [42] A. Shlapentokh. Diophantine classes of holomorphy rings of global fields. Journal of Algebra, 169(1):139 – 175, 1994. [43] A. Shlapentokh. Hilbert's Tenth Problem: Diophantine Classes and Extensions to Global Fields. New Mathematical Monographs. Cambridge University Press, 2007. [44] H. Stichtenoth. Algebraic Function Fields and Codes. Springer Publishing Company, Incorporated, 2nd edition, 2008. [45] G. van der Geer, B. J. J Moonen, and R. Schoof. Number Fields and Function Fields – Two Parallel Worlds. Progress in Mathematics. Birkhäuser, Boston, 2006. 41
1810.06097	1	1810	2018-10-14T20:06:03	On a partially ordered set associated to ring morphisms	[ "math.RA" ]	We associate to any ring $R$ with identity a partially ordered set Hom$(R)$, whose elements are all pairs $(\mathfrak a,M)$, where $\mathfrak a=\ker\varphi$ and $M=\varphi^{-1}(U(S))$ for some ring morphism $\varphi$ of $R$ into an arbitrary ring $S$. Here $U(S)$ denotes the group of units of $S$. The assignment $R\mapsto{}$Hom$(R)$ turns out to be a contravariant functor of the category Ring of associative rings with identity to the category ParOrd of partially ordered sets. The maximal elements of Hom$(R)$ constitute a subset Max$(R)$ which, for commutative rings $R$, can be identified with the Zariski spectrum Spec$(R)$ of $R$. Every pair $(\mathfrak a,M)$ in Hom$(R)$ has a canonical representative, that is, there is a universal ring morphism $\psi\colon R\to S_{(R/\mathfrak a,M/\mathfrak a)} $ corresponding to the pair $(\mathfrak a,M)$, where the ring $S_{(R/\mathfrak a,M/\mathfrak a)} $ is constructed as a universal inverting $R/\mathfrak a$-ring in the sense of Cohn. Several properties of the sets Hom$(R)$ and Max$(R)$ are studied.	math.RA	math	ON A PARTIALLY ORDERED SET ASSOCIATED TO RING MORPHISMS ALBERTO FACCHINI AND LEILA HEIDARI ZADEH Abstract. We associate to any ring R with identity a partially ordered set Hom(R), whose elements are all pairs (a, M ), where a = ker ϕ and M = ϕ−1(U (S)) for some ring morphism ϕ of R into an arbitrary ring S. Here U (S) denotes the group of units of S. The assignment R 7→ Hom(R) turns out to be a contravariant functor of the category Ring of associative rings with identity to the category ParOrd of partially ordered sets. The maximal elements of Hom(R) constitute a subset Max(R) which, for commutative rings R, can be identified with the Zariski spectrum Spec(R) of R. Every pair (a, M ) in Hom(R) has a canonical representative, that is, there is a universal ring morphism ψ : R → S(R/a,M/a) corresponding to the pair (a, M ), where the ring S(R/a,M/a) is constructed as a universal inverting R/a-ring in the sense of Cohn. Several properties of the sets Hom(R) and Max(R) are studied. 1. Introduction In this paper, we study a contravariant functor Hom(−) : Ring → ParOrd from the category Ring of all associative rings with identity to the category ParOrd of partially ordered sets. This functor associates to every ring R the set of all pairs (a, M ), where a = ker ϕ and M = ϕ−1(U (S)) for some ring morphism ϕ : R → S. Here S is any other ring, that is, any object of Ring, and U (S) denotes the group of units (= invertible elements) of S. With respect to a suitable partial order, the set Hom(R) turns out to be a meet-semilattice (Lemma 2.6). The idea is to measure and classify, via the study of the partially ordered set Hom(R), all ring morphisms from the fixed ring R to any other ring S. We have at least five motivations to study our functor Hom(−): (1) We want to generalize the theory developed by Bavula for left Ore localiza- tions [3, 4, 5] to arbitrary ring morphisms. In those papers, Bavula discovered the importance of maximal left denominator sets. Therefore here we want to extend his idea from ring morphisms R → [S−1]R that arise as left Ore localizations to ar- bitrary ring morphisms ϕ : R → S. In view of Bavula's results, we pay a particular attention to the maximal elements of the partially ordered set Hom(R). For every ring R, the subset Max(R) of all maximal elements of Hom(R) is always non-empty (Theorem 5.6). (2) For a commutative ring R, the set Max(R) is in one-to-one correspondence with the Zarisky spectrum Spec(R) of R (Proposition 5.3). Thus Max(R) could Key words and phrases. Ring morphism, Partially ordered set, Contravariant functor, Univer- sal inverting mapping of rings. 2010 Mathematics Subject Classification. Primary 16B50. Secondary 16S85. The first author is partially supported by Dipartimento di Matematica "Tullio Levi-Civita" of Universit`a di Padova (Project BIRD163492/16 "Categorical homological methods in the study of algebraic structures" and Research program DOR1828909 "Anelli e categorie di moduli"). 1 2 ALBERTO FACCHINI AND LEILA HEIDARI ZADEH be used as a good substitute for the spectrum of a possibly non-commutative ring R. Unluckily, the assignment R 7→ Max(R) is not a contravariant functor (Theorem 5.7). This is not quite surprising, because, in the commutative case, the maximal spectrum, i.e., the topological subspace of Spec(R) whose elements are all maximal ideals of the commutative ring R, is not a functor either. All this is related to the paper [21] by Manuel Reyes. Notice that the category ParOrd of partially ordered sets is isomorphic to the category of all Alexandrov T0-spaces, which is a full subcategory of the category Top of topological spaces. Thus our contravariant functor Hom(−) can be also viewed as a functor of Ring into Top. (3) The Hom of a direct limit of rings Ri is the inverse limit of the corresponding partially ordered sets Hom(Ri) (Theorem 4.1). We are motivated to the study of the (good) behavior of our functor Hom(−) with respect to direct limits of rings, because spectra of commutative monoids has a similar behavior [20, Corollary 2.2]. Notice that Reyes' universal contravariant functor p-Spec : Ring → Set can be defined as the inverse limit of the spectra of the commutative subrings of R [21, Proposition 2.14]. (4) An approach similar to ours appears in the paper [24] by Vale. He also considers a contravariant functor from the category Ring, but to the category of ringed spaces. When the ring R is commutative, he also gets a sort of "completion" of Spec(R). (5) Finally, the partially ordered set Hom(R) always has a least element, the pair (0, U (R)), which corresponds to the identity morphism R → R. More generally, like in Bavula's case, the set Hom(R) has a natural partition into subsets Hom(R, a) (Section 2). The least elements of these subsets Hom(R, a), with a contained in the Jacobson radical J(R) of R, correspond to local morphisms (Proposition 7.6), that is, to the ring morphisms ϕ : R → S such that, for every r ∈ R, ϕ(r) invertible in S implies r invertible in R. Thus our interest in the functor Hom(−) is also motivated by the several applications of local morphisms [9, 12]. Notice that the subset Hom(R, 0) classifies all ring extensions ϕ : R ֒→ S. Every pair (a, M ) in Hom(R) has a canonical representative, that is, a universal ring morphism ψ : R → S(R/a,M/a) corresponding to the pair (a, M ) (Theorem 3.3). The ring S(R/a,M/a) is constructed as a universal inverting R/a-ring in the sense of Cohn [10]. Any other ring morphism ϕ : R → S corresponding to (a, M ) has a canonical factorization through ψ (Theorem 7.3). One of the mappings appearing in this factorization of ϕ is a ring epimorphism ϕT : R → T , which still corresponds to the pair (a, M ). Ring epimorphisms, that is, epimorphisms in the category Ring, currently play a predominant role in Homological Algebra [1, 6, 14, 15, 16], in particular left flat morphism, that is, when the codomain is a flat left R-module. The functor Hom(−) is not representable (Section 2). The meet-semilattice Hom(R) has a smallest element (0, U (R)), but does not have a greatest element in general. Hence, for some results, instead of Hom(R), it is more convenient to enlarge the partially ordered set Hom(R) adjoining to it a further element, a new greatest element 1, setting Hom(R) := Hom(R) ∪ {1}. In some sense, this new greatest element 1 corresponds to the zero morphism R → S for any ring S. This enlarged partially ordered set Hom(R) is a bounded lattice (Theorem 6.3). Finally, we specialize some of our results to Bavula's case of left ring of fractions. In Bavula's case, the ring morphism ϕ : R → S is the canonical mapping of R into ON A PARTIALLY ORDERED SET ASSOCIATED TO RING MORPHISMS 3 the right ring of fractions S of R with respect to some right denominator set. Such a ϕ is clearly a ring epimorphism. Throughout, all rings are associative, with identity 1 6= 0, and all ring morphisms send 1 to 1. The group of (right and left) invertible elements of R will be denoted by U (R), and the Jacobson radical of R will be denoted by J(R). 2. The partially ordered set Hom(R) Let R be a ring. We associate to each ring morphism ϕ : R → S into any other ring S the pair (a, M ), where a := ker(ϕ) is the kernel of ϕ and M := ϕ−1(U (S)) is the inverse image of the group of units U (S) of S. In the next lemmas, we collect the basic properties of these pairs (a, M ). Recall that a monoid S is cancellative if, for every x, y, z ∈ S, xz = yz implies x = y and zx = zy implies x = y. An element x of a ring R is regular if, for all r ∈ R, rx = 0 implies r = 0 and xr = 0 implies r = 0. Lemma 2.1. Let ϕ : R → S be a ring morphism and (a, M ) its associated pair. Then: (1) M is a submonoid of the multiplicative monoid R. (2) U (R) ⊆ M . (3) M = M + a = M + a + J(R) and a ∩ M = ∅. (4) M/a := { m + a m ∈ M } consists of regular elements of R/a. In particu- lar, M/a is a cancellative submonoid of the multiplicative monoid R/a. Proof. (1), (2) and (4) are easy. (3) The inclusions M ⊆ M + a ⊆ M + a + J(R) are trivial. In order to prove that M + a + J(R) ⊆ M , notice that M + a ⊆ M and 1 + J(R) ⊆ U (R) ⊆ M . Since M is multiplicatively closed and contains 1R, it follows that M ⊇ (M + a)(1 + J(R)) = M + a + M J(R) + aJ(R) ⊇ M + a + 1RJ(R) = M + a + J(R). (cid:3) Remark 2.2. The monoid M is not cancellative in general. As an example consider R = Z/6Z, S = Z/2Z, ϕ : Z/6Z → Z/2Z the canonical projection, x = 1 + 6Z and y = z = 3 + 6Z. Then x, y, z ∈ M and xz = yz, but x 6= y. Recall that a multiplicatively closed subset M of a ring R is saturated if, for every x, y ∈ R, xy ∈ M implies x ∈ M and y ∈ M . A ring R is directly finite if, for every x, y ∈ R, xy = 1 implies yx = 1. Lemma 2.3. Let ϕ : R → S be a ring morphism and (a, M ) its associated pair. Then: (1) If S is a (not-necessarily commutative) integral domain, then the ideal a is completely prime. (2) If S is a division ring, then R is the disjoint union of a and M , i.e., {a, M } is a partition of the set R. (3) If S is a directly finite ring, e.g., if S is an integral domain, then M is a saturated multiplicatively closed subset of R. Proof. (3) Suppose S directly finite, x, y ∈ R and xy ∈ M = ϕ−1(U (S)). Then ϕ(x)ϕ(y) = ϕ(xy) ∈ U (S). Hence there exists s ∈ S such that ϕ(x)ϕ(y)s = 1 and sϕ(x)ϕ(y) = 1. Thus ϕ(x) is right invertible and ϕ(y) is left invertible. Since S is directly finite, we have that ϕ(x) and ϕ(y) are both invertible in S, so that x ∈ M and y ∈ M . 4 ALBERTO FACCHINI AND LEILA HEIDARI ZADEH Notice that every integral domain is directly finite, because if x, y are element of an integral domain S and xy = 1, then yxy = y, so (yx−1)y = 0, hence yx = 1. (cid:3) We will now deal with preorders on a set X, that is, reflexive and transitive relations on X. Recall that, if X is a set, or more generally a class, and ρ is a preorder on X, then it is possible to associate to ρ an equivalence relation ∼ρ on X and a partial order ≤ρ on the quotient set X/∼ρ. The equivalence relation ∼ρ on X is defined, for every x, y ∈ X, by x ∼ρ y if xρy and yρx. The partial order ≤ρ on the quotient set X/∼ρ:= { [x]∼ρ x ∈ X } is defined by [x]∼ρ ≤ρ [y]∼ρ if xρy. On the class H(R) of all morphisms ϕ : R → S of R into arbitrary rings S, there If ϕ : R → S, ϕ′ : R → S′ are two ring morphisms, are two natural preorders. we have a first preorder ρ on H(R), defined setting ϕρϕ′ if ker(ϕ) ⊆ ker(ϕ′) and ϕ−1(U (S)) ⊆ ϕ′−1(U (S)). A second preorder σ on H(R) is defined setting ϕ σ ϕ′ if there exists a ring morphism ψ : S → S′ such that ψϕ = ϕ′. Correspondingly, there is a first equivalence relation ∼ on the class H(R), defined, for all ring morphisms ϕ : R → S, ϕ′ : R → S′ with associated pairs (a, M ), (a′, M ′) respectively, by ϕ ∼ ϕ′ if (a, M ) = (a′, M ′). That is, ϕ ∼ ϕ′ if and only if ker(ϕ) = ker(ϕ′) and ϕ−1(U (S)) = ϕ′−1(U (S′)). Let Hom(R) := H(R)/∼ denote the set (class) of all equivalence classes [ϕ]∼ modulo ∼, that is, equivalently, the set of all pairs (ker(ϕ), ϕ−1(U (S))). The partial order ≤ on Hom(R) = H(R)/∼ associated to the preorder ρ on H(R) is defined by setting (a, M ) ≤ (a′, M ′) if a ⊆ a′ and M ⊆ M ′. As far as the second natural preorder σ on H(R) is concerned, the equivalence relation ≡ on H(R) associated to σ is defined, for every ϕ : R → S, ϕ′ : R → S′ in H(R), by ϕ ≡ ϕ′ if there exist ring morphisms ψ : S → S′ and ψ′ : S′ → S such that ψϕ = ϕ′ and ψ′ϕ′ = ϕ. The partial order (cid:22) on the quotient class H(R)/ ≡, associated to the preorder σ on H(R), is defined by setting [ϕ]≡ (cid:22) [ϕ′]≡ if ϕ σ ϕ′. Remark 2.4. If there exists a ring morphism ψ : S → S′ such that ψϕ = ϕ′, then (a, M ) ≤ (a′, M ′). Equivalently, for all ϕ : R → S, ϕ′ : R → S′ in H(R), ϕ σ ϕ′ implies ϕ ρ ϕ′. Thus, for all ϕ : R → S, ϕ′ : R → S′ in H(R), ϕ ≡ ϕ′ implies (a, M ) = (a′, M ′), i.e., ϕ ∼ ϕ′. Equivalently, the identity mapping H(R) → H(R) is a preorder morphism of (H(R), σ) onto (H(R), ρ). Similarly, there is an induced surjective morphism of factor classes H(R)/≡ → H(R)/∼ = Hom(R), [ϕ]≡ 7→ [ϕ]∼. The implication ϕ σ ϕ′ implies ϕ ρ ϕ′ cannot be reversed in general, that is, there are morphisms ϕ : R → S and ϕ′ : R → S′ with (a, M ) ≤ (a′, M ′), but for which there does not exist a ring morphism ψ : S → S′ with ψϕ = ϕ′. For instance, let k be a finite field, k its algebraic closure, M2(k) the ring of 2 × 2 matrices with entries in k, and ϕ : k → k and ϕ′ : k → M2(k) the canonical embeddings. Then k and M2(k) are simple rings, so that all ring morphisms ψ : k → M2(k) are injective. But k is finite and k is infinite, so that there is no ring morphism ψ : k → M2(k). The implication ϕ σ ϕ′ implies (a, M ) ≤ (a′, M ′) can be reversed in some special cases, for instance when we restrict our attention to localizations at left denominator sets. See Remark 7.5. ON A PARTIALLY ORDERED SET ASSOCIATED TO RING MORPHISMS 5 Proposition 2.5. Let Ring be the category of rings with identity and ParOrd the category of partially ordered sets. Then Hom(−) : Ring → ParOrd is a con- travariant functor. Proof. The functor Hom assigns to each ring R the set Hom(R) of all pairs (a, M ), where a := ker(ϕ) and M := ϕ−1(U (S)) for some ring morphism ϕ : R → S, partially ordered by ≤, where (a, M ) ≤ (b, N ) if a ⊆ b and M ⊆ N . Moreover, it assigns to each ring morphism f : R → R′ the increasing mapping Hom(f ) : Hom(R′) → Hom(R), (a′, M ′) ∈ Hom(R′) 7→ (f −1(a′), f −1(M ′)). Notice that if ϕ′ : R′ → S is a ring morphism, a′ := ker(ϕ′) and M ′ := ϕ′−1(U (S)), then ϕ′f : R → S is a ring morphism, and f −1(a′) = ker(ϕ′f ) f −1(M ′) = (ϕ′f )−1(U (S)). (cid:3) The functor Hom(−) is not representable. Namely, suppose the contravariant Hom(−) : Ring → Set representable, i.e., that there exists a ring A with Hom(−) naturally isomorphic to the contravariant functor HomRing(−, A) : Ring → Set. Now, for every ring A there always exists a ring R with HomRing(R, A) = ∅ (If A has characteristic 0, take for R any ring of characteristic 6= 0. If A has characteristic n ≥ 2, take for R any ring of characteristic p prime with p 6= n.) Our functor Hom(−) is such that Hom(R) 6= ∅ for every ring R. Hence the functors Hom(−) and HomRing(−, A) can never be isomorphic. For any fixed proper ideal a of R, set Hom(R, a) := { (ker(ϕ), ϕ−1(U (S))) ϕ : R → S, ker(ϕ) = a }. Clearly, Hom(R) is the disjoint union of the sets Hom(R, a): Hom(R) = Hom(R, a). [a⊳R In particular, the partial order ≤ on Hom(R) induces a partial order on each subset Hom(R, a). The following lemma has an easy proof. Lemma 2.6. Let (a, M ), (a′, M ′) be the elements of Hom(R) corresponding to two morphisms ϕ : R → S and ϕ′ : R → S′. Then the element of Hom(R) corresponding to the product morphism ϕ × ϕ′ : R → S × S′ is (a ∩ a′, M ∩ M ′). As a consequence, the partially ordered set Hom(R) turns out to be a meet- semilattice. In particular, with respect to the operation ∧, Hom(R) is a commuta- tive semigroup in which every element is idempotent and which has a zero element (= the least element (0, U (R)) of Hom(R), which corresponds to the identity mor- phism R → R). We will see in Theorem 5.6 that the partially ordered set Hom(R) always has maximal elements, but does not have a greatest element in general, so the semigroup Hom(R) does not have an identity in general. 6 ALBERTO FACCHINI AND LEILA HEIDARI ZADEH 3. A universal construction Let R be any ring and N be any fixed subset of R. Let X := { xn n ∈ N } be a set of non-commuting indeterminates in one-to-one correspondence with the set N . Let R{X} be the free R-ring over X ([7] and [22, Example 1.9.20 on Page 124]). Then there are a canonical ring morphism ϕ : R → R{X} and a mapping ε : X → R{X} such that for every ring S, every ring morphism ψ : R → S and every mapping ζ : X → S there is a unique ring morphism eψ : R{X} → S such that ψ = eψϕ and ζ = eψε. Let I be the two-sided ideal of R{X} generated by the subset { xnn − 1 n ∈ N } ∪ { nxn − 1 n ∈ N } and S(R,N ) := R{X}/I. Clearly, I could be the improper ideal of R{X} and S(R,N ) could be the zero ring. There is a canonical mapping χ(R,N ) : R → S(R,N ), composite mapping of ϕ : R → R{X} and the canonical projection R{X} → R{X}/I. The R-ring R{X}/I is the universal N -inverting R-ring in the sense of [10, Proposition 1.3.1]. Lemma 3.1. If (0, M ) ∈ Hom(R) for a ring R, then the canonical ring morphism χ(R,M) : R → S(R,M) is injective and χ−1 (R,M)(U (S(R,M))) = M . Proof. If (0, M ) ∈ Hom(R), there are a ring S and ring morphism f : R → S such that (0, M ) is associated to f . In particular, f is an injective mapping. The morphism f clearly factors through χ(R,M), that is, there is a ring morphism g : S(R,M) → S with gχ(R,M) = f . As f is injective, χ(R,M) is also injective. Moreover, gχ(R,M) = f implies that M = f −1(U (S)) ⊇ χ−1 (R,M)(U (S(R,M))). Finally, the elements of M are clearly mapped to invertible elements of S(R,M) via χ(R,M), by construction, and so χ−1 (cid:3) (R,M)(U (S(R,M))) = M . The proof of the following lemma is immediate. Lemma 3.2. If (a, M ) ∈ Hom(R), then (a/a, M/a) ∈ Hom(R/a). Theorem 3.3. Let R be a ring and (a, M ) be an element of Hom(R). Then S(R/a,M/a) is a non-zero ring, and if ψ : R → S(R/a,M/a) denotes the composite map- ping of the canonical projection π : R → R/a and χ(R/a,M/a) : R/a → S(R/a,M/a), then ker(ψ) = a and ψ−1(U (S(R/a,M/a))) = M . Moreover, for any ring morphism f : R → S such that ker(f ) ⊇ a and f −1(U (S)) ⊇ M , there is a unique ring morphism g : S(R/a,M/a) → S such that gψ = f . Proof. Since (a, M ) ∈ Hom(R), there are ring morphisms ϕ : R → S such that ker(ϕ) = a and ϕ−1(U (S)) = M . More generally, let f : R → S be any ring morphism with ker(f ) ⊇ a and f −1(U (S)) ⊇ M . Then f factors as the composite mapping of the canonical projection π : R → R/a and a unique morphism f : R/a → S. Now construct the ring S(R/a,M/a) := (R/a){X}/I, where X := { xm m ∈ M/a }. By the universal property of the free R/a-ring (R/a){X}, there is a unique (R/a){X} and ε : X → (R/a){X} are the canonical mapping and ζ : X → S is defined by ζ(xm) = (f (m))−1 for every m ∈ M/a. See the diagram below. From ring morphism ef : (R/a){X} → S such that f = ef ψ′ and ζ = ef ε, where ψ′ : R/a → f = ef ψ′, we get that ef (m) = f (m) = f (m), and, from ζ = ef ε, we have that ef (xm) = ef ε(xm) = ζ(m) = (f (m))−1 = (f (m))−1. Thus the generators xmm − 1 and mxm −1 of the two-sided ideal I of (R/a){X} are mapped to zero via ef , so that ef factors in a unique way through a ring morphism g : S(R/a,M/a) = (R/a){X}/I → ON A PARTIALLY ORDERED SET ASSOCIATED TO RING MORPHISMS 7 the canonical projection. This, applied to any ring morphisms ϕ : R → S such that ker(ϕ) = a and ϕ−1(U (S)) = M , shows that S(R/a,M/a) is a non-zero ring. S, that is, ef = gπ′, where π′ : (R/a){X} → (R/a){X}/I = S(R/a,M/a) denotes Moreover, set ψ = χ(R/a,M/a)π = π′ψ′π and f = f π = ef ψ′π = gπ′ψ′π. Then f = gψ. This proves the existence of g in the last part of the statement of the theorem. f f 9ttttttttttt S ef R π R/a g◆◆◆◆◆◆◆◆◆◆◆◆◆ g (R/a){X} / S(R/a,M/a) π′ ψ′ Now we apply again the previous results to any ring morphism ϕ : R → S. Since (a/a, M/a) ∈ Hom(R/a) by Lemma 3.2, we now have that χ(R/a,M/a) : R/a → S(R/a,M/a) is an injective mapping by Lemma 3.1. Thus the kernel ker(ψ) of ψ = χ(R/a,M/a)π is equal to ker(π) = a. Also, ψ−1(U (S(R/a,M/a))) = (π′ψ′π)−1(U (S(R/a,M/a))) = = (χ(R/a,M/a)π)−1(U (S(R/a,M/a))) = = π−1χ−1 (R/a,M/a)(U (S(R/a,M/a))) = π−1(M/a) = M. It remains to prove the uniqueness of g, that is, if g′ : S(R/a,M/a) → S is another ring morphism such that g′ψ = f , then g = g′. Now S(R/a,M/a) is generated, as a ring, by the image of R via ψ = π′ψ′π and the inverses of the elements of ψ(M ). Since gψ = g′ψ, both mappings g and g′ send each ψ(r) to f (r) and each ψ(m)−1 to f (m)−1. It follows that g = g′, as desired. (cid:3) Theorem 3.3 shows that, for any pair (a, M ) in Hom(R), there is a canonical ring morphism ψ : R → S(R/a,M/a) that realizes that pair. Moreover, the universal property described in the last part of the statement of the theorem shows that the canonical morphism ψ : R → S(R/a,M/a) is one of the least elements in the class H(R, a) of all morphisms f : R → S such that a ⊆ ker(f ) and M ⊆ f −1(U (S)) with respect to the preorder σ, in the sense that ψ σ f for every morphism f : R → S with a ⊆ ker(f ) and M ⊆ f −1(U (S)). 4. Direct limits Now let (Ri)i∈I be a direct system of rings indexed on a directed set (I, ≤). Hence, for every i, j ∈ I, i ≤ j, we have compatible connecting ring morphisms µij : Ri → Rj. Applying our functor Hom(−), we get an inverse system (Hom(Ri))i∈I of partially ordered sets, with connecting partially ordered set morphisms Theorem 4.1. Hom(µij ) : Hom(Rj) → Hom(Ri). Hom(lim −→ Ri) ∼= lim ←− Hom(Ri). / / / / 9 O O / g 8 ALBERTO FACCHINI AND LEILA HEIDARI ZADEH Proof. Let µj : Rj → lim −→ These morphisms induce partially ordered set morphisms Ri be the canonical ring morphisms, for every j ∈ I. Let H := lim ←− Hom(Ri) ⊆ Yj∈I Hom(µj ) : Hom(lim −→ Ri) → Hom(Rj). Hom(Rj ) be the inverse limit of the inverse system i i (Hom(Ri))i∈I of partially ordered sets, and hj : H → Hom(Rj) the canonical map- ping. By the universal property of inverse limit, there exists a unique partially Ri) → H such that hjΨ = Hom(µj) for every order set morphism Ψ : Hom(lim −→ j ∈ I. Thus Ψ(a, M ) = (µ−1 (a), µ−1 (M ))i∈I . j (a′), µ−1 Ri) with Ψ(a, M ) = Ψ(a′, M ′). Then (µ−1 We will show that Ψ is a bijection and Ψ−1 is a morphism of partially order sets. First we prove that the mapping Ψ is injective. Let (a, M ), (a′, M ′) be two j (M )) = elements of Hom(lim −→ (µ−1 j (M ′)) for every j ∈ I. We claim that, for any two subset X and Ri, if µ−1 If we prove Y of lim −→ j (a′), µ−1 the claim, then (µ−1 j (M ′)) for every j ∈ I implies (a, M ) = (a′, M ′), which proves that Ψ is injective. In order to prove the claim, assume X, Y ⊆ lim −→ j (Y ) for all j ∈ I. If x ∈ X, then there exists i ∈ J and ri ∈ Ri with µi(ri) = x. Hence ri ∈ µ−1 (Y ), so that x = µi(ri) ∈ Y . Thus X ⊆ Y . Similarly Y ⊆ X. This concludes the proof of the claim, which shows that Ψ is injective. j (Y ) for every j ∈ I, then X = Y . j (X) = µ−1 j (a), µ−1 j (M )) = (µ−1 Ri and µ−1 j (X) = µ−1 (X) = µ−1 i j (a), µ−1 i Let us prove that Ψ is surjective. Let ((ai, Mi))i∈I be an element of H, so that ij (aj), µ−1 ij (Mj)) = (ai, Mi) for every i ≤ j in I. Here (ai, Mi) is an element of (µ−1 Hom(Ri), so that (ai, Mi) corresponds to the ring morphism ψi : Ri → S(Ri/ai,Mi/ai) (Theorem 3.3). We now show that (cid:0)S(Ri/ai,Mi/ai)(cid:1)i∈I , with suitable canonical connecting maps, form a direct system of rings. Since µ−1 ij (aj) = ai, the morphisms µij induce monomorphisms µij : Ri/ai → Rj/aj, and µij(Mi/ai) ⊆ Mj/aj. Thus µij extends to a ring monomorphism Ri/ai{ xm+ai m ∈ Mi } → Rj/aj{ xm+aj m ∈ Mj } that maps xm+ai to xµij (m)+aj . These canonical ring monomorphisms induce ring morphisms νij : S(Ri/ai,Mi/ai) → S(Rj /aj ,Mj /aj ), for all i ≤ j. The diagrams ψi Ri S(Ri/ai,Mi/ai) µij νij Rj ψj / S(Rj /aj ,Mj /aj ) clearly commute for every i ≤ j. Hence we have a morphism of direct systems of rings, and, taking the direct limit, we get a ring morphism Let (a, M ) ∈ Hom(lim −→ ψ : lim −→ Ri → lim −→ S(Ri/ai,Mi/ai). Ri) be the pair corresponding to this ring morphism ψ. We claim that Ψ(a, M ) = ((ai, Mi))i∈I , that is, the µ−1 (a) = ai and µ−1 i (M ) = i Mi for each i ∈ I. Let us prove that µ−1 (a) = ai. i / / / ON A PARTIALLY ORDERED SET ASSOCIATED TO RING MORPHISMS 9 An element ri ∈ Ri belongs to µ−1 i (a) if and only if µi(ri) ∈ a = ker ψ, that is, if and only if ψµi(ri) = 0. Now we have commutative diagrams Ri µi ψi S(Ri/ai,Mi/ai) νi lim −→ Ri / lim −→ ψ S(Ri/ai,Mi/ai), i i so that ψµi(ri) = 0 if and only if νiψi(ri) = 0, which occurs if and only if there exists j ≥ i such that νijψi(ri) = 0, that is, ψjµij (ri) = 0, i.e., if and only if there exists j ≥ i such that µij(ri) ∈ aj. Equivalently, if and only if ri ∈ ai. This proves that µ−1 (a) = ai for every i. We will now prove that µ−1 (M ) = Mi. Set S := lim −→ S(Ri/ai,Mi/ai). An ele- i ment ri ∈ Ri belongs to µ−1 (M ) if and only if µi(ri) ∈ M , that is, if and only if ψµi(ri) ∈ U (S), i.e., if and only if νiψi(ri) ∈ U (S). This occurs if and only if there exists s ∈ S such that sνiψi(ri) = 1 and νiψi(ri)s = 1. Now any ele- ment s of S is of the form νj(sj) for some j ≥ i and sj ∈ S(Rj /aj ,Mj /aj). Also, νj(sj)νiψi(ri) = 1 and νiψi(ri)νj(sj) = 1 in S if and only if there exists k ≥ i, j such that νjk(sj)νikψi(ri) = 1 and νikψi(ri)νjk(sj) = 1 in S(Rk/ak,Mk/ak). This occurs if and only if νikψi(ri) is invertible in S(Rk/ak,Mk/ak), that is, if and only if ψkµik(ri) is invertible in S(Rk/ak,Mk/ak), i.e., µik(ri) ∈ Mk, that is, ri ∈ µ−1 ik (Mk) = Mi. This shows that µ−1 (M ) = Mi, and concludes the proof of the claim. Thus Ψ is surjective. Finally, let us prove that Ψ−1 is a morphism of partially order sets. Let i))i∈I , i for every i ∈ I. We have direct systems of i), i ∈ I, and canonical projections i/a i, which extend to the free R/ai-ring R/ai{XMi/ai} (to the free }), sending each indeterminate xm+a ∈ XMi/ai to the in- In this way, we get a canonical ring morphism }, which induces, factoring out the corresponding i) and commutative i ))i∈I be two elements in H with ((ai, Mi))i∈I ≤ ((a′ i, M ′ ((ai, Mi))i∈I , ((a′ that is, with ai ⊆ a′ i and Mi ⊆ M ′ rings S(Ri/ai,Mi/ai), i ∈ I, and S(R′ πi : Ri/ai → R′ R/a′ i-ring R/a′ determinate xm+a R/ai{XMi/ai} → R/a′ ideals Ii and I ′ squares i, a ring morphism S(R/ai,M/ai) → S(R/a i /a ∈ XM ′ i /a i{XM ′ i/a′ i{XM ′ i, M ′ ′ i i /a i,M ′ i /a . ′ i,M ′ i /a ′ ′ ′ i ′ i ′ i ′ i Ri ψi Ri ψ′ i S(Ri/ai,Mi/ai) / S(Ri/a πi ′ i,M ′ i /a ′ i). Taking the direct limit, we get a commutative triangle lim −→ Ri ψ }④④④④④④④ S π ψ′ "❊❊❊❊❊❊❊ / S′. Thus we have ψ σ ψ′ with respect to the preorder σ on H(R), so that (a, M ) ≤ (a′, M ′). (cid:3) / / / / } " / 10 ALBERTO FACCHINI AND LEILA HEIDARI ZADEH For the last paragraph of this section, we have been inspired by [24]. Any preordered set (X, ≤) can be viewed as a category whose objects are the elements of X and, for every pair x, y ∈ X of objects of the category, there is exactly one morphism x → y if x ≤ y, and no morphism x → y otherwise. This applies in particular to our partially ordered set Hom(R), for any ring R. There is a covariant functor FR : Hom(R) → Ring. It associates to any object (a, M ) of Hom(R) the ring S(R/a,M/a). Like in the proof of the previous theorem, where we show that Ψ−1 is a partially ordered set morphism, we have that if (a, M ) ≤ (a′, M ′), then there is a canonical morphism S(R/a,M/a) → S(R/a ′). So we have, for every ring R, a covariant functor FR : Hom(R) → Ring. This can be expressed by means of diagrams in the category Ring. Formally, a diagram of shape J in a category C is a functor F from J to C. Here we are considering only the case in which the category J is a partially ordered set. Thus we have, for every ring R, a diagram of shape Hom(R) in the category Ring. ′,M ′/a 5. Maximal elements in Hom(R) We now recall a classification due to Bokut (see [8] and [11, pp. 515-516]). Let D0 be the class of integral domains, D2 the class of invertible rings, that is, rings R such that the universal mapping inverting all non-zero elements of R is injective, and E be the class of rings embeddable in division rings. Then D0 ⊃ D2 ⊃ E. Notice that a ring R ∈ D0 is in D2 if and only if the mapping χ(R,R\{0}) : R → S(R,R\{0}) is injective, if and only if (0, R \ {0}) ∈ Hom(R). Proposition 5.1. Let a be an ideal of a ring R such that (a, R \ a) ∈ Hom(R). Then a is a completely prime ideal of R, the ring R/a is invertible, and (a, R \ a) ∈ Hom(R) is a maximal element of Hom(R). Proof. Since (a, R \ a) ∈ Hom(R), the set R \ a is multiplicatively closed, that is, a is completely prime. Moreover, (a, R \ a) ∈ Hom(R) implies that there exists a morphism ϕ : R → S with kernel a for which all elements of ϕ(R \ a) are invertible. The induced mapping ϕ : R/a → S is an injective morphisms for which the image of every non-zero element of R/a is invertible in S. The injective morphism ϕ factors through the universal inverting mapping ψ : R/a → S(R/a,M/a) by Theorem 3.3. Thus ϕ injective implies ψ injective, i.e., R/a is an invertible ring. Finally, (a, R \ a) is a maximal element of Hom(R), because if (a, R \ a) ≤ (a′, M ′), then a ⊆ a′ and R \ a ⊆ M ′. Hence a′ ∩ M ′ = ∅ implies a′ = a and M ′ = R \ a. (cid:3) In the following example, we show that not all maximal elements of Hom(R) are of the form (a, R \ a) for some completely prime ideal a. Example 5.2. Let R be the ring of n×n matrices with entries in a division ring D, n > 1. Then any homomorphism ϕ : R → S, S any ring, is injective because R is simple. Every element of M := ϕ−1(U (S)) is regular by Lemma 2.1. But regular elements in R are invertible. This proves that Hom(R) has exactly one element, the pair (0, U (R)). Thus, clearly, Hom(R) has a greatest element, which is not of the form (a, R \ a) because R is simple, but not a domain, and R has no completely prime ideals. Proposition 5.3. For any commutative ring R, the maximal elements of Hom(R) are the pairs (P, R \ P ), where P is a prime ideal. ON A PARTIALLY ORDERED SET ASSOCIATED TO RING MORPHISMS 11 Proof. By Proposition 5.1, the pairs (P, R \ P ), where P is any prime ideal of the commutative ring R, are maximal elements of Hom(R). Conversely, let (a, M ) ∈ Hom(R) be a maximal element. The set F of all ideals b of R with a ⊆ b and b ∩ M = ∅ is non-empty because a ∈ F . By Zorn's Lemma, the set F , partially ordered by set inclusion, has a maximal element P . It is very easy to check that P is a prime ideal of R. Then (a, M ) ≤ (P, R \ P ). But (a, M ) is maximal, so (a, M ) = (P, R \ P ). (cid:3) Proposition 5.4. Let R be a commutative ring. Then Hom(R) has a greatest element if and only if R has a unique prime ideal. Proof. If Hom(R) has a greatest element (a, M ), then (a, M ) is the unique maximal element of Hom(R), so that R has a unique prime ideal by Proposition 5.3. Conversely, let R be a commutative ring with a unique prime ideal P . Then R is a local ring with maximal ideal P . Clearly, the pair (P, R \ P ) belongs to Hom(R), because it is associated to the canonical morphism of R onto the field R/P . For any other ring morphism ϕ : R → S, one has ker ϕ ⊆ P because ker ϕ is a proper ideal of R. In order to show that (P, R \ P ) is the greatest element of Hom(R), it suffices to show that ϕ−1(U (S)) ⊆ R \ P . We claim that ϕ−1(U (S)), which is clearly a multiplicatively closed subset of R, is saturated, that is, if r, r′ ∈ R and ϕ(rr′) ∈ U (S), then ϕ(r) ∈ U (S) (this is sufficient, because R is commutative). If r, r′ ∈ R and ϕ(rr′) ∈ U (S), then there exists s ∈ S such that ϕ(r)ϕ(r′)s = 1. Thus ϕ(r) is invertible in S. This proves the claim. The complement of a saturated multiplicatively closed subset of a commutative ring is a union of prime ideals [2, p. 44, exercise 7]. Since R has a unique prime ideal, the saturated multiplicatively closed subsets of R are only R \ P , the improper subset R of R, and the empty set ∅. It follows that ϕ−1(U (S)) = R \ P . This concludes the proof. (cid:3) Example 5.5. As an example, we now describe the structure of the partially ordered set Hom(Z), where Z is the ring of integers. For an arbitrary element (a, M ) of Hom(Z), we have that a = nZ for some non- negative integer n 6= 1. For n = 0, the set M must be a saturated subset of Z. Hence Z \ M is a union of prime ideals [2, p. 44, exercise 7]. Thus there exists a subset P of the set P := { p p is prime number } such that M is the set MP of all z ∈ Z, z 6= 0, with p ∤ z for every p ∈ P . For any such subset P of P, the pair (0, MP ) corresponds to the embedding of Z into its ring of fractions with denominators in the multiplicatively closed subset MP of Z. Now assume that a = nZ for some n ≥ 2 and that (a, M ) corresponds to some ring morphism ϕ : Z → S. Then ϕ induces an injective ring morphism ϕ : Z/nZ → S, and M/nZ is a multiplicatively closed subset of Z/nZ that consists of regular elements and contains U (Z/nZ). Since in a finite ring all regular elements are invertible, it follows that M/nZ = U (Z/nZ), so that M = Mdiv(n), where div(n) := { p ∈ P pn }. Thus Hom(Z) = { (0, MP ) P is a subset of P } ∪(cid:8) (nZ, Mdiv(n)) n ∈ Z, n > 2(cid:9) . Notice that for any P, P ′ ⊆ P and n, n′ ≥ 2: (1) (0, MP ) ≤ (0, MP ′) if and only if MP ⊆ MP ′ , if and only if P ′ ⊆ P . (2) (nZ, Mdiv(n)) ≤ (n′Z, Mdiv(n′)) if and only if nZ ⊆ n′Z, if and only if n′n (because n′n implies div(n′) ⊆ div(n), from which Mdiv(n) ⊆ Mdiv(n′)). (3) (0, MP ) ≤ (nZ, Mdiv(n)) if and only if div(n) ⊆ P . 12 ALBERTO FACCHINI AND LEILA HEIDARI ZADEH (4) (nZ, Mdiv(n)) ≤ (0, MP ) never occurs. In order to better describe the partially ordered set Hom(Z), we will now present an order-reversing injective mapping ρ : Hom(Z) → (N0)P∗ , where N0 denotes the set of non-negative integers with its usual order, N0 := N0 ∪ {+∞} with n ≤ ∞ for all n ∈ N0, P∗ := P∪{0} and the order on the product (N0)P∗ is the component-wise order. Via this ρ, the partially ordered set Hom(Z) can be identified as a partially ordered subset of the opposite partially ordered set of (N0)P∗ . Notice that every positive integer n can be written uniquely as a product of primes, n = pe1 1 . . . pnt for suitable distinct primes pi ∈ P and positive integers ei, so that there is an order- preserving injective mapping ρ′ : N → NP 0, where the set N of positive integers is ordered by the relation (divides). Here t ρ′(n)(p) =(cid:26) ei 0 if p = pi for some i, if p ∈ P \ {p1, . . . , pt} for every n ∈ N and p ∈ P. Similarly, there are characteristic functions of subsets P of P, so that there is an order-preserving bijection χ : P (P) → {0, +∞}P, defined by χ(P ) = χP for every P in the power set P (P) of all subsets of P, where χP : P → {0, +∞} is such that χP (p) =(cid:26) +∞ if p ∈ P, if p ∈ P \ P 0 for every P ⊆ P and p ∈ P. Our mapping ρ will extend both the order-preserving injective mapping ρ′ and the order isomorphism χ. Define ρ : Hom(Z) → (N0)P∗ by if p = pi for some i, if p ∈ P \ {p1, . . . , pt}, if p = 0 +∞ if p ∈ P, 0 1 if p ∈ P \ P, if p = 0 for every n ≥ 2 and p ∈ P∗ := P ∪ {0}, and ei 0 0 ρ(nZ, Mdiv(n))(p) =  ρ(0, MP )(p) =  for every P ⊆ P. In order to show that this mapping ρ is an order-reversing embedding of Hom(Z) , we must prove that ρ satisfies the following four properties, correspond- into (N0)P∗ ing to the four properties (1)-(4) above: (1′) ρ(0, MP ′)(p) ≤ ρ(0, MP )(p) for every p ∈ P∗ if and only if P ′ ⊆ P . (2′) ρ(n′Z, Mdiv(n′))(p) ≤ ρ(nZ, Mdiv(n))(p) for every p ∈ P∗ if and only if n′n. (3′) ρ(nZ, Mdiv(n))(p) ≤ ρ(0, MP )(p) for every p ∈ P∗ if and only if div(n) ⊆ P . (4′) For every P ⊆ P and every n ≥ 2, there exists p ∈ P∗ such that ρ(nZ, Mdiv(n))(p) < ρ(0, MP )(p). Let P, P ′ be subsets of P. In order to prove (1′), notice that ρ(0, MP ′)(p) ≤ ρ(0, MP )(p) for every p ∈ P∗ if and only if, for every p ∈ P, ρ(0, MP ′)(p) = +∞ implies ρ(0, MP )(p) = +∞. This is equivalent to p ∈ P ′ implies p ∈ P for every p ∈ P, that is, if and only if P ′ ⊆ P . Now let n, n′ ≥ 2 be integers. Then ρ(n′Z, Mdiv(n′))(p) ≤ ρ(nZ, Mdiv(n))(p) for every p ∈ P∗ if and only if, for every prime p, pn′ implies pn and the exponent of ON A PARTIALLY ORDERED SET ASSOCIATED TO RING MORPHISMS 13 p in a prime factorization of n′ is less than or equal to the exponent of p in a prime factorization of n. This is equivalent to n′n, which proves (2′). 1 . . . pet For (3′), let n = pe1 t be a prime factorization of the integer n ≥ 2. Then ρ(nZ, Mdiv(n))(p) ≤ ρ(0, MP )(p) for every p ∈ P∗ if and only if ρ(nZ, Mdiv(n))(p) ≤ ρ(0, MP )(p) for every p ∈ P, if and only if ρ(nZ, Mdiv(n))(pi) ≤ ρ(0, MP )(pi) for every i = 1, 2, . . . , t, if and only if ei ≤ ρ(0, MP )(pi) for every i = 1, 2, . . . , t. That is, if and only if div(n) ⊆ P . Finally, (4′) is trivial, because it suffices to take p = 0. It easily follows that: (1) Hom(Z) has (0, MP) = (0, {1, −1}) = (0, U (Z)) as its least element, and (2) the maximal elements of Hom(Z) are the pairs (0, M∅) = (0, Z \ {0}) and the pairs (pZ, Mdiv(p)) = (pZ, Z \ pZ) for every p ∈ P (cf. Proposi- tion 5.3). Propositions 5.1 and 5.3 show that the set Max(R) of all maximal elements of Hom(R) could be used as a good substitute for the spectrum of a non-commutative ring R. Let us show that the set of all maximal elements is never empty. Theorem 5.6. For every ring R, the partially ordered set Hom(R) has maximal elements. Proof. Let R be a ring. It is known that R always has maximal two-sided ideals, that is, maximal elements in the set of all proper two-sided ideals (this is a very standard application of Zorn's Lemma). Let m be a maximal two-sided ideal of R. Set F := { M M = ϕ−1(U (S)), S is any ring and ϕ : R → S is a ring morphism with ker(ϕ) = m }. Then F is non-empty (consider the canonical projection ϕ : R → R/m). Partially order F by set inclusion. Let Mλ (λ ∈ Λ) be a chain in F . By Theorem 3.3, Mλ = ψ−1 λ (U (S(R/m,Mλ/m))), where ψλ : R → S(R/m,Mλ/m) is the canonical mapping. Since the monoids Mλ (λ ∈ Λ) are linearly ordered by set inclusion, the rings S(R/m,Mλ/m) form a direct system of rings over a linearly ordered set, and there is a ring morphism ψ = lim −→ ψλ : R → S = lim −→ S(R/m,Mλ/m). The elements of Mλ are mapped to invertible elements of Sλ via ψλ, so that they are mapped to invertible elements of S via ψ. Thus ψ−1(U (S)) ⊇ Sλ∈Λ Mλ, i.e., ψ−1(U (S)) ∈ F is an upper bound of the chain of the monoids Mλ. Hence we can apply Zorn's Lemma, which concludes the proof of the theorem. (cid:3) From Proposition 2.5, Theorem 5.6 and [21, Theorem 1.1], we get that: Theorem 5.7. There is no contravariant functor from the category of rings to the category of sets that assigns to each ring R the set of all maximal elements of Hom(R). It would be very interesting to determine, for any ring R, if the set Hom(R, 0) has a greatest element. This would correspond with what has been done by Bavula for left localizations at left Ore sets in [4, Theorem 2.1.2]. The greatest element of Hom(R, 0) corresponds to a suitable submonoid M of the multiplicative monoid RegR of all regular elements of R. Notice that: (1) When R is commutative, the greatest element of Hom(R, 0) is clearly (0, RegR). (2) More generally, if RegR is a right Ore set or a left Ore set, then the greatest element of Hom(R, 0) is (0, RegR). 14 ALBERTO FACCHINI AND LEILA HEIDARI ZADEH (3) If the ring R is contained in a division ring, the greatest element of Hom(R, 0) is (0, R \ {0}). (4) More generally, suppose that the canonical morphism χ(R,RegR) : R → S(R,RegR) is injective, or, equivalently, R is contained in a ring in which all regular elements of R are invertible. For example, R could be an invertible ring (see the definition in the first paragraph of this Section). Then the greatest element of Hom(R, 0) is (0, RegR). (5) If M is a submonoid of R, the pair (0, M ) is a maximal element of Hom(R, 0) if and only if the canonical morphism χ(R,M) : R → S(R,M) is injective and, for every regular element x ∈ R, x /∈ M , the canonical mapping χ(R,M∪{x}) : R → S(R,M∪{x}) is not injective. Remark 5.8. For any ring R, consider the three sets Div(R) := { (a, R \ a) a = ker ϕ for a morphism Cpr(R) := { (P, R \ P ) ∈ Hom(R) P is a completely prime ideal of R } Max(R) of all maximal elements of Hom(R). ϕ : R → D into some division ring D } In general, we have Div(R) ⊆ Cpr(R) ⊆ Max(R) ⊆ Hom(R) (Proposition 5.1), and all these inclusions can be proper. When R is commutative, Div(R) = Cpr(R) = Max(R) ∼= Spec(R) ⊆ Hom(R) (Proposition 5.3). Related to this, we can consider Cohn's spectrum X(R) of the ring R, that is, the topological space X(R) of all epic R-fields, up to isomorphism. Recall that a ring morphism f : R → D is an epic R-field in the sense of [10, p. 154] if D is a division ring and there is no division ring different from D between f (R) and D. Notice that there are rings R for which there is no epic R-field R → D. For instance, if R is a ring that is not IBN, there is no ring morphism R → D, for any division ring D. Clearly, there is an onto mapping X(R) → Div(R). 6. The partially ordered set Hom(R) As we have said in Section 2, the partially ordered set Hom(R) is a meet- semilattice, hence a commutative semigroup in which every element is idempotent, has a smallest element (0, U (R)), but does not have a greatest element in general. Hence we now enlarge the partially ordered set Hom(R) adjoining to it a further ele- ment, a new greatest element 1, setting Hom(R) := Hom(R) ∪ {1}. Here (a, M ) ≤ 1 for every element (a, M ) ∈ Hom(R). This new element 1 of Hom(R) represents in some sense the zero morphism R → 0, where 0 is the zero ring with one element. The zero ring is not a ring in our sense strictly, because we have supposed in the Introduction that all our rings have an identity 1 6= 0. This is the reason why the zero morphism R → 0 does not appear in the definition of Hom(R). Moreover, the pair (a, M ) corresponding to the zero morphism R → 0 would clearly have a = R, but it is not clear what M should be. The contravariant functor Hom(−) : Ring → ParOrd extends to a contravari- ant functor Hom(−) : Ring → ParOrd simply extending, for every ring mor- phism ϕ : R → S, the mapping Hom(ϕ) : Hom(S) → Hom(R) to the mapping Hom(ϕ) : Hom(S) → Hom(R), where Hom(ϕ)(1) = 1. First of all, we will now show that Hom(R1 × R2), where R1 × R2 denotes the ring direct product, is canonically isomorphic to the cartesian product Hom(R1) × ON A PARTIALLY ORDERED SET ASSOCIATED TO RING MORPHISMS 15 Hom(R2). We first need an elementary proposition. For every pair R, S of rings, we will denote the set of all ring morphisms R → S, including the zero morphism R → 0, by HomRing(R, S). Proposition 6.1. Let R1, R2, S be rings. Then there is a bijection between HomRing(R1 × R2, S) and the set of all triples (e, ψ1, ψ2), where e ∈ S is an idempotent element and ψ1 : R1 → eSe, ψ2 : R2 → (1 − e)S(1 − e) are ring morphisms (possibly zero, when e = 0 or e = 1). Proof. Let T denote the set of all the triples (e, ψ1, ψ2) in the statement. Let Φ : HomRing(R1 × R2, S) → T be defined by Φ(ϕ) = (ϕ(1R1 , 0R2), ϕR1 , ϕR2 ). Here ϕ : R1 × R2 → S is any ring morphism, so that e := ϕ(1R1 , 0R2) is an idem- potent element of S, and ϕR1 : R1 → eSe, ϕR2 : R2 → (1 − e)S(1 − e) denote the restrictions of ϕ to R1, R2 respectively (or, more precisely, to the subsets R1 × {0} and {0} × R2 of R1 × R2). We leave to the reader to check that Φ is a well-defined surjective mapping. As far as injectivity is concerned, notice that if ϕ : R1 ×R2 → S and ϕ′ : R1 × R2 → S are ring morphisms and Φ(ϕ) = Φ(ϕ′), then ϕ = ϕ′ because R1 × R2 is the direct sum of R1 and R2 as additive abelian groups, and therefore ϕ = ϕ′ are completely determined by their restrictions to the direct summands R1 and R2 of R1 × R2. (cid:3) Proposition 6.2. Let R1 and R2 be rings. Then there is a canonical bijection between Hom(R1 × R2) and the cartesian product Hom(R1) × Hom(R2). Proof. First of all, we show that, for any ring morphism ϕ : R1 × R2 → S, we have (1) ker(ϕ) = ker(ϕR1 ) × ker(ϕR2 ) and (2) ϕ−1(U (S)) = (ϕR1 )−1(U (eSe)) × (ϕR2 )−1(U ((1 − e)S(1 − e))). Here, like in the proof of Proposition 6.1, e is the image via ϕ of the idempotent element (1R1, 0R2) of R1 × R2, and ϕR1 : R1 → eSe, ϕR2 : R2 → (1 − e)S(1 − e) are the restrictions of ϕ to R1, R2. We leave the easy proof of (1) to the reader. As far as (2) is concerned, notice that this formula makes no sense when one of the morphisms ϕ, ϕR1 or ϕR2 is zero. In these three cases, either e = 0 or e = 1, and the morphisms ϕ, ϕR1 , ϕR2 ) correspond to the greatest element 1 of Hom(R1 × R2), Hom(R1) or Hom(R2), respectively. Also remark that if (r1, r2) ∈ R1 × R2, then (r1, r2) ∈ ϕ−1(U (S)) if and only if ϕ(r1, r2) ∈ U (S). Now ϕ(r1, r2) ∈ eSe × (1 − e)S(1 − e) is invertible in S if and only if it is invertible in eSe × (1 − e)S(1 − e), that is, if and only if (ϕR1 )(r1) is invertible in eSe and (ϕR2 )(r2) is invertible in (1 − e)S(1 − e). This concludes the proof of (2). From (1) and (2), it follows that the mapping Hom(R1 × R2) → Hom(R1) × Hom(R2), (ker(ϕ), ϕ−1(U (S))) 7→ ((ker(ϕR1 ), (ϕR1 )−1(U (eSe))), (ker(ϕR2 ), (ϕR2 )−1(U ((1 − e)S(1 − e)))) is a well-defined injective mapping. Its surjectivity is proved considering, for any pair of ring morphisms ϕ1 : R1 → S1, ϕ2 : R2 → S2, the ring morphism ϕ1 × ϕ2 : R1 × R2 → S1 × S2. (cid:3) 16 ALBERTO FACCHINI AND LEILA HEIDARI ZADEH We saw in Lemma 2.6, and the paragraph following it, that the partially ordered set Hom(R) is a meet-semilattice, so that Hom(R), with respect to the operation ∧, is a commutative semigroup in which every element is idempotent. Now Hom(R) is also a meet-semilattice, but with a greatest element 1, so Hom(R), with respect to the operation ∧, turns out to be a commutative monoid in which every element is idempotent. Hence we can view the functor Hom(−) as a functor of Ring into the category CMon of commutative monoids. Now, commutative monoids have a spectrum, set of its prime ideals, i.e., there is a contravariant functor Spec from the category CMon to the category Top of topological spaces [20]. For every commutative monoid A, Spec(A) is a spectral space in the sense of Hochster. Hence the composite functor Spec ◦Hom(−) : Ring → Top associates to every ring a spectral topological space. Theorem 6.3. For every ring R, the partially ordered set Hom(R) is a bounded lattice. ′,M ′/a ′). Let ω : R → P := S(R/a,M/a) ∗R S(R/a Proof. It is clear that Hom(R) is a partially ordered set with a least element (0, U (R)) and a greatest element 1. Since we already know that Hom(R) is a meet-semilattice, we only have to show that any pair of elements (a, M ), (a′, M ′) of Hom(R) has a least upper bound in Hom(R) (it is clear that the least upper bound exists when one of the two elements is the greatest element 1 of Hom(R) ). Suppose (a, M ), (a′, M ′) ∈ Hom(R). Let ψ : R → S(R/a,M/a) be the ring morphism corresponding to the pair (a, M ) as in the statement of Theorem 3.3. Similarly for ψ′ : R → S(R/a ′) be the pushout of ψ and ψ′ in Ring, and (a′′, M ′′) the element of Hom(R) corresponding to ω. We will now prove that (a′′, M ′′) = (a, M ) ∨ (a′, M ′) in the partially ordered set Hom(R). Since ω factors through ψ, we have that (a′′, M ′′) ≥ (a, M ). Similarly, (a′′, M ′′) ≥ (a′, M ′). Conversely, let χ : R → T be any morphism with associated pair (b, N ) and with (b, N ) ≥ (a, M ), (a′, M ′). By the universal property of The- orem 3.3, there is a unique ring morphism g : S(R/a,M/a) → T such that gψ = χ. Similarly, there is a unique ring morphism g′ : S(R/a ′) → T such that g′ψ′ = χ. By the universal property of pushout, there exists a unique morphism h : P → T such that hε = g and hε′ = g′, where ε : S(R/a,M/a) → P and ε′ : S(R/a′,M/a′) → P are the canonical mappings into the pushout. Then hω = hεψ = gψ = χ, so χ factors through ω, hence (b, N ) ≥ (a′′, M ′′). ′,M ′/a ′,M/a ψ ω S(R/a ′,M ′/a ′) R ψ′ '❖❖❖❖❖❖❖❖❖❖❖❖❖❖❖ S(R/a,M/a) ✻✻ ✻✻✻ ε ✻✻ ✻✻✻ ✻✻ ✻✻✻ ✻ $■■■■■■■■■■■ ❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ P g′ ε′ g h T (cid:3) Hence Hom(R), with respect to the operation ∨, turns out to be a commutative monoid with zero (the element 1) and identity the element (0, U (R)), in which all elements are idempotent. / / ' / / $ ON A PARTIALLY ORDERED SET ASSOCIATED TO RING MORPHISMS 17 7. Ring epimorphisms Recall that a ring morphism ϕ : R → S is an epimorphism if, for all ring mor- phisms ψ, ψ′ : S → T , ψϕ = ψ′ϕ implies ψ = ψ′. Proposition 7.1. ([17], [10, Proposition 4.1.1]), [23, Proposition XI.1.2] The fol- lowing conditions on a ring morphism ϕ : R → S are equivalent: (a) ϕ is an epimorphism, (b) s ⊗ 1 = 1 ⊗ s in the S-S-bimodule S ⊗R S for all s ∈ S. (c) The R-R-bimodule S ⊗R S is isomorphic to the R-R-bimodule S via the canonical isomorphism induced by the multiplication · : S × S → S of the ring R. (d) The pushout R → S ∗R S of ϕ with itself is naturally isomorphic to R → S. (e) S ⊗R (S/ϕ(R)) = 0. Proposition 7.2. Let ϕ : R → S be a ring morphism and (a, M ) be its corre- sponding pair in Hom(R). Let T be the subring of S generated by ϕ(R) and all the elements ϕ(m)−1 (m ∈ M ). Then the corestriction ϕT : R → T is a ring epimorphism and its corresponding pair in Hom(R) is (a, M ). Proof. Let T ′ be the subset of T consisting of all elements a ∈ T with a ⊗ 1 = 1 ⊗ a in the S-S-bimodule T ⊗R T . The subset T ′ of T is a subring of T that contains ϕ(R), because T ⊗R T is a T -T -bimodule in which multiplication by elements of T is defined by t(t′ ⊗ t′′) = (tt′)⊗ t′′ and (t′ ⊗ t′′)t = t′ ⊗ (t′′t), so that a⊗ 1 = 1 ⊗ a and b⊗1 = 1⊗b imply (ab)⊗1 = a(b⊗1) = a(1⊗b) = a⊗b = (a⊗1)b = (1⊗a)b = 1⊗(ab). This shows that T ′ is a subring of T . Moreover if m ∈ M , then in T ⊗R T we have that ϕ(m)−1 ⊗1 = ϕ(m)−1 ⊗ϕ(m)ϕ(m)−1 = ϕ(m)−1ϕ(m)⊗ϕ(m)−1 = 1⊗ϕ(m)−1. It follows that T ⊆ T ′, hence T = T ′. It follows that the corestriction ϕT : R → T is an epimorphism. Finally M = ϕ−1(U (S)) and ϕ(m) ∈ U (T ) for every m ∈ M . Thus M ⊆ ϕ−1(U (T )) ⊆ ϕ−1(U (S)) = M . (cid:3) Now consider the universal construction of Theorem 3.3. Let ϕ : R → S be a ring morphism and (a, M ) be its corresponding pair in Hom(R). Via the canonical ring morphism ψ : R → S(R/a,M/a), the subring T of S(R/a,M/a) generated by ψ(R) and the inverses of the images of the elements of M is the whole ring S(R/a,M/a). It follows that the canonical ring morphism ψ : R → S(R/a,M/a) is a ring epimorphism. More generally, for any ring morphism ϕ : R → S, we have the canonical factor- ization described in the next Theorem: Theorem 7.3. Let ϕ : R → S be any ring morphism and (a, M ) its corresponding element in Hom(R). Then ϕ is the composite ring morphism of the mappings R π / R/a χ / S(R/a,M/a) g / T ε / S where T is the subring of S generated by ϕ(R) and the inverses ϕ(m)−1 of the images of the elements of M , g : S(R/a,M/a) → T is a surjective ring epimorphism and ε : T → S is the ring embedding. This theorem shows that any ring morphisms ϕ : R → S can be factorized as: (1) a canonical mapping R → S(R/a,M/a), which only depends on the pair (a, M ) ∈ Hom(R) associated to ϕ. (2) A ring morphism S(R/a,M/a) → T , which is a surjective mapping and is an epimorphism in the category of rings. (3) A ring embedding ε : T → S. / / / / 18 ALBERTO FACCHINI AND LEILA HEIDARI ZADEH Proof. Apply the universal property of Theorem 3.3 to the corestriction ϕT : R → T , getting a factorization R π / T of ϕT . The mapping g is surjective, because T is generated by the images of the elements of R and the inverses of the elements of M , like S(R/a,M/a). Moreover, g is a ring epimorphism, because ϕT : R → T is a ring epimorphism by Proposition 7.2, so ψg = ψ′g implies ψgχπ = ψ′gχπ, i.e., ψϕT = ψ′ϕT , from which ψ = ψ′. (cid:3) / S(R/a,M/a) / R/a χ g For any other ring morphism f : R → S′ such that ker(f ) = a and f −1(U (S′)) = M , there is a unique ring morphism g : S(R/a,M/a) → S′ such that gϕ = f . It follows that the subring T ′ of S′ generated by f (R) and the elements f (m)−1 is the image g(S(R/a,M/a)) of S(R/a,M/a). Hence the corestriction f T ′ : R → T ′ is the composite mapping of ϕ : R → S(R/a,M/a) and the corestriction gT ′ : S(R/a,M/a) → T ′. Recall that a subset T is a left Ore subset of R if it is a submonoid of R such that T r ∩ Rt 6= ∅ for every r ∈ R and t ∈ T . A subset T of the ring R is called a left denominator set if it is a left Ore subset and, for every r ∈ R, t ∈ T , if rt = 0, then there exists t′ ∈ T with t′r = 0. A left ring of fractions ϕ : R → [T −1]R exists if and only if T is a left denominator set in R. Compare Lemma 2.1(4) with the fact that a left quotient ring [T −1]R of a ring R with respect to a multiplicatively closed subset T of R exists if and only if T is a left Ore set and the set T = { t + ass(T ) ∈ R/ ass(T ) t ∈ T } consists of regular elements (see [19, 2.1.12] and [18]). Here ass(T ) denotes the set of all elements r ∈ R for which there exists an element t ∈ T with tr = 0. That is, ass(T ) is the kernel a of the canonical morphism R → [T −1]R. Lemma 7.4. Let T be a left denominator set in R and ϕ : R → S = [T −1]R the canonical mapping into the left ring of fractions. Then M := ϕ−1(U (S)) is a left denominator set in R containing T , a = ass(T ) = ass(M ) and S = [M −1]R. Proof. It is well known that a = ker(ϕ) = ass(T ). Moreover, M ⊇ T . Let us prove that M is a left Ore subset of R. Fix r ∈ R and m ∈ M . We must show that M r ∩ Rm 6= ∅. Now ϕ(r)ϕ(m)−1 ∈ S = [T −1]R, so that there exist r′ ∈ R and t ∈ T such that ϕ(r)ϕ(m)−1 = ϕ(t)−1ϕ(r′). Then ϕ(t)ϕ(r) = ϕ(r′)ϕ(m) in S, so that there exists t′ ∈ T with rtt′ = mr′t′ ∈ rT ∩ mR ⊆ rM ∩ mR. t′tr = t′r′m ∈ T r ∩ Rm ⊆ M r ∩ Rm. This proves that M is a left Ore subset of R. In order to see that M is a left denominator set, notice that if r ∈ R, m ∈ M and rm = 0, then ϕ(r)ϕ(m) = 0, so ϕ(r) = 0. Hence r ∈ ker(ϕ) = ass(T ), so that tr = 0 for some t ∈ T . But T ⊆ M . This proves that M is a left denominator set. It is now clear that S = [T −1]R = [M −1]R, and thus a = ker(ϕ) = ass(M ). (cid:3) Clearly, Lemma 7.4 holds not only for left denominator sets and left rings of fractions, but also for right denominator sets and right rings of fractions, because associating the pair (a, M ) to a ring morphism ϕ is left/right symmetric. Remark 7.5. We have already noticed in Remark 2.4 that if there exists a ring morphism ψ : S → S′ such that ψϕ = ϕ′, then (a, M ) ≤ (a′, M ′). This can be inverted for left localizations, i.e., if S = [T −1]R and S′ = [T ′−1]R for suitable left denominator sets T, T ′, ϕ : R → S, ϕ′ : R → S′ are the canonical mappings, and (a, M ) ≤ (a′, M ′), then there exists a ring morphism ψ : S → S′ such that ψϕ = ϕ′. / / / ON A PARTIALLY ORDERED SET ASSOCIATED TO RING MORPHISMS 19 To prove it, suppose that S = [T −1]R and S′ = [T ′−1]R for left denominator sets T, T ′, that ϕ : R → S, ϕ′ : R → S′ are the canonical mappings and (a, M ) ≤ (a′, M ′). Since M ⊆ M ′, so T ⊆ M ⊆ M ′, the elements of T are mapped to invertible elements of S′ = [M ′−1]R via the canonical mapping ϕ′ : R → S′. By the universal property of the mapping ϕ : R → S = [T −1]R, there exists a unique ring morphism ψ : S → S′ such that ψϕ = ϕ′. Similarly for the equivalence relation σ: If S = [T −1]R and S′ = [T ′−1]R for left denominator sets T, T ′, and ϕ : R → S, ϕ′ : R → S′ are the canonical mappings, then ϕ σ ϕ′ if and only if there exists a ring isomorphism ψ : S → S′ such that ψϕ = ϕ′. Finally, we have already remarked in the Introduction that a ring morphism ϕ : R → S is local if and only if M = U (R). Moreover, ker(ϕ) ⊆ J(R) for every local morphism ϕ : R → S [13, Lemma 3.1]. It follows that local morphisms correspond to the least elements of Hom(R, a) with respect to the partial order ≤. More precisely: Proposition 7.6. Let ϕ : R → S be a ring morphism and (a, M ) its corresponding pair in Hom(R). Then ϕ is a local morphism if and only if (a, M ) is the least element of Hom(R, a) for some ideal a ⊆ J(R). Proof. Suppose that (a, M ) is the least element of Hom(R, a) for some ideal a ⊆ J(R). By Proposition 7.7, the least element of Hom(R, a) is (a, π−1(U (R/a))), where π : R → R/a is the canonical projection. Thus M = π−1(U (R/a)). Let us prove that ϕ is local. If r ∈ R and ϕ(r) is invertible in S, then r ∈ M , so that r ∈ π−1(U (R/a)). Hence r + a is invertible in R/a. Hence r + J(R) is invertible in R/J(R), so r is invertible in R, as desired. This proves that ϕ is a local morphism. The inverse implication is trivial. (cid:3) More generally, for an arbitrary proper ideal a of R, not-necessarily contained in J(R), we have that: Proposition 7.7. For every proper ideal a of a ring R, the partially ordered set Hom(R, a) always has a least element, which is the pair (a, M ) corresponding to the canonical projection π : R → R/a, that is, the pair (a, M ) with M = π−1(U (R/a)). Proof. We must show that, for every ring morphism ϕ : R → S with ker(ϕ) = a, we have π−1(U (R/a)) ⊆ ϕ−1(U (S)). Now, given ϕ : R → S with ker(ϕ) = a, let π : R → R/a denote the canonical projection. By the first isomorphism theorem for rings, there exists a unique injective ring morphism eϕ : R/a → S such that ϕ = eϕπ. It is now easily checked that π−1(U (R/a)) ⊆ ϕ−1(U (S)). (cid:3) We conclude the paper indicating a further possible generalization of our the results in this paper. In Remark 5.8, we have already mentioned Cohn's spectrum X(R) of a ring R, consisting of all epic R-fields, up to isomorphism. P. M. Cohn has shown that any epic R-field R → D is characterized up to isomorphism by the collection of square matrices with entries in R which are carried to singular matrices with entries in the division ring D. He has also given the conditions under which a collection of square matrices over R is of this type, calling such a collection a "prime matrix ideal" of R. The natural ideal is therefore to refine the theory developed in the previous sections, classifying all morphisms ϕ : R → S, not only via our pairs (a, M ), where M is the set of all elements of R mapped to invertible 20 ALBERTO FACCHINI AND LEILA HEIDARI ZADEH elements of R, but also via the collection of all n × m matrices with entries in R which are carried to invertible n × m matrices with entries in the ring S. References [1] L. Angeleri Hugel and J. S´anchez, Tilting modules arising from ring epimorphisms, Algebr. Represent. Theor. 14 (2011), 217 -- 246. [2] M. F. Atiyah and I. G. Macdonald, "Introduction to commutative algebr", Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills, Ont., 1969. [3] V. V. Bavula, Left localizations of left Artinian rings, J. Algebra Appl. 15(9) (2016), 1650165, 38 pp. [4] V. V. Bavula, The largest left quotient ring of a ring, Comm. Algebra 44 (2016), no. 8, 3219 -- 3261. [5] V. V. Bavula, Weakly left localizable rings, Comm. Algebra 45(9) (2017), 3798 -- 3815. [6] S. Bazzoni and L. Positselski, Contramodules over pro-perfect topological rings, the cover- ing property in categorical tilting theory, and homological ring epimorphisms, available in https://arxiv.org/abs/1807.10671 [7] G. M. Bergman, Coproducts and some universal ring constructions, Trans. Amer. Math. Soc. 200 (1974), 33 -- 88. [8] L. A. Bokut, "Associative Rings 1, 2" (Russian), NGU, Novosibirsk, 1981. [9] R. Camps and W. Dicks, On semilocal rings, Israel J. Math. 81 (1993), 203 -- 211. [10] P. M. Cohn, "Skew fields. Theory of general division rings", Encyclopedia of Mathematics and its Applications, 57. Cambridge University Press, Cambridge, 1995. [11] P. M. Cohn, "Free ideal rings and localization in general rings", New Mathematical Mono- graphs, 3. Cambridge University Press, Cambridge, 2006. [12] A. Facchini and D. Herbera, K0 of a semilocal ring, J. Algebra 225 (2000), 47 -- 69. [13] A. Facchini and D. Herbera, Projective modules over semilocal rings, in "Algebra and its applications", D. V. Huynh, S. K. Jain and S. R. L´opez-Permouth Eds., Contemp. Math. 259, Amer. Math. Soc., Providence, RI, 2000, pp. 181 -- 198. [14] A. Facchini and Z. Nazemian, Equivalence of some homological conditions for ring epimor- phisms, J. Pure Appl. Algebra (2018), available online in the web site of the journal. [15] A. Facchini and Z. Nazemian, Covering classes, strongly flat modules, and completions, sub- mitted for publication, 2018. [16] W. Geigle and H. Lenzing, Perpendicular categories with applications to representations and sheaves, J. Algebra 144(2) (1991), 273 -- 343. [17] J. T. Knight, On epimorphisms of non-commutative rings, Proc. Camb. Phil. Soc. 68 (1970), 589-600. [18] T. Y. Lam, "Lectures on modules and rings", Graduate Texts in Math. 189, New York: Springer-Verlag (1999). [19] J. C. Mcconnell and J. C. Robson, "Noncommutative Noetherian Rings", Chichester: Wiley (1987). [20] I. Pirashvili, On the spectrum of monoids and semilattices, J. Pure Appl. Algebra 217(5) (2013), 901 -- 906. [21] M. L. Reyes, Obstructing extensions of the functor Spec to noncommutative rings, Israel J. Math. 192 (2012), 667 -- 698. [22] L. H. Rowen, "Ring theory", Vol. I. Pure and Applied Mathematics 127, Academic Press, Inc., Boston, MA, 1988. [23] B. Stenstrom, "Rings of Quotients", Grundlehren Math. Wiss. 217, Springer-Verlag, New York-Heidelberg, 1975. [24] R. Vale, On the opposite of the category of rings, available in arXiv:0806.1476v2. Dipartimento di Matematica, Universit`a di Padova, 35121 Padova, Italy E-mail address: [email protected] Department of Mathematics, University of Kurdistan, P. O. Box 416, Sanandaj, Iran. E-mail address: [email protected]; [email protected]
1701.09054	1	1701	2017-01-31T14:22:39	On one-sided (B,C)-inverses of arbitrary matrices	[ "math.RA" ]	In this article one-sided (b, c)-inverses of arbitrary matrices as well as one-sided inverses along a (not necessarily square) matrix, will be studied. In adddition, the (b, c)-inverse and the inverse along an element will be also researched in the context of rectangular matrices.	math.RA	math	On one-sided (B, C)-inverses of arbitrary matrices Julio Ben´ıtez, Enrico Boasso, Hongwei Jin Abstract In this article one-sided (b, c)-inverses of arbitrary matrices as well as one-sided inverses along a (not necessarily square) matrix, will be studied. In adddition, the (b, c)-inverse and the inverse along an element will be also researched in the context of rectangular matrices. Keywords: One-sided (b, c)-inverse; One-sided inverse along an element; (b, c)-inverse; Inverse along an element; Matrices AMS classification: 15A09, 15A23, 15A60, 65F99 1 Introduction and notation Several generalized inverses have been studied in the literature. Recently two important outer inverses have been introduced: the inverse along an element (see [14]) and the (b, c)-inverse (see [9]). In fact, these two generalized inverses encompass some of the most important outer inverses such as the group inverse, the Drazin inverse and the Moore-Penrose inverse. Furthermore, in the context of semigroups the left and right inverses along an element were defined in [24]; these notions extend the inverse along an element. Similarly, in the frame of rings, left and right (b, c)-invertible elements were introduced in [13]; these definitions extend both the (b, c)-inverse and the left and right inverses along an element. As it has been said, the aforementioned outer inverses and their extensions were defined in semigroups or rings. However, observe that the set of n × m complex matrices is not a semigroup (unless n = m). The main purpose of this article is to extend the above mentioned (one-sided) inverses as well as the (b, c)-inverse and the inverse along an element to arbitrary matrices and to study their basic properties. Naturally, the results presented also hold for square matrices. In section 3, after having recalled the main notions considered in this article in section 2, the one-sided (b, c)-inverses and the left and right inverses along an element in the context of arbitrary matrices will be thoroughly studied. In sections 4 and 5 the (b, c)-inverse and the inverse along an element will be introduced and studied in the same frame, respectively. In section 6 it will be characterized when the generalized inverses introduced in sections 4 and 5 are inner inverses. In section 7 the relationships among the notions considered in sections 4 and 5 and the outer inverse with prescribed range and null space will be studied. In section 8 the continuity and the differentiability of the notions introduced in sections 4 and 5 will be considered. Finally, in section 9 algorithms to compute the (b, c)-inverse in the matrix frame will be given. Before going on, the definition of several generalized inverses in the context of rings will be given. The corresponding definitions for complex matrices can be obtained making obvious changes. Let R be a unitary ring and a ∈ R. 1 J. Ben´ıtez, E. Boasso, H. Jin 2 (i) The element a is said to be group invertible, if there exists x ∈ R such that axa = a, xax = x, and ax = xa. This x is unique and it is denoted by a#. (ii) The element a is said to be Drazin invertible, if there exists x ∈ R such that xax = x, xa = ax, and an+1x = an, for some n ∈ . This x is unique and it is denoted by ad. Note that when n = 1, the group inverse is obtained (see [8]). (iii) Let R have an involution. The element a is said to be Moore-Penrose invertible, if exists x ∈ R such that axa = a, xax = x, (ax)∗ = ax, and (xa)∗ = xa. This x is unique and it is denoted by a† (see [16]). (iv) Let R have an involution and let m, n ∈ R be invertible Hermitian elements in R. The element a ∈ R is said to be Moore-Penrose invertible with weights m, n, if there exists x ∈ R such that axa = a, xax = x, (max)∗ = max, (nxa)∗ = nxa. This x is unique and it is denoted by a† m,n. In a ring R with an involution, an element u ∈ R is said to be positive, if there exists a Hermitian v ∈ R such that u = v2. (v) Let R have an involution. The element a is said to be core invertible, if there exists x ∈ R such that axa = a, xR = aR and Rx = Ra∗. This x is unique and it is denoted by a #(cid:13) (see [1, 18]). (vi) Let R have an involution. The element a is said to be dual core invertible, if there exists x ∈ R such that axa = a, xR = a∗R and Rx = Ra. This x is unique and it is denoted by a #(cid:13) (see [1, 18]). To end this section, some notation is introduced. Let m, n ∈ and denote by Cm,n the set of m × n complex matrices. The symbol Cn will stand for Cn,n. Any vector of the space n will be identified with Cn,1. n will be considered as a column vector, i.e., C C Moreover, In will mean the identity matrix of order n, rk(A) the rank of A ∈ Cm,n, and when n = m, tr(A) will stand for the trace of A. Related to a matrix A ∈ Cm,n there are two linear subspaces, the column space and the null space, which are defined respectively by R(A) = {Ax : x ∈ C n}, N(A) = {x ∈ C n : Ax = 0}. m, the subsets R(f ) Recall that rk(A) + dim N(A) = n. Given a linear mapping f : C and N(f ) are defined in a similar way. Observe that if A is the matrix associated to f respect with the standard basis, then R(A) = R(f ) and N(A) = N(f ). n → C In addition, the conjugate transpose of the matrix A will be denoted by A∗. Two basic equalities are N(A∗) = R(A)⊥ and R(A∗) = N(A)⊥, for A ∈ Cn,m. If M is a subspace of C on M and PM for the orthogonal projector onto M. When N and M are two subspaces of C PM,N will stand for the idempotent whose range is M and whose null space is N. n, the symbol IM will stand for the identity linear transformation n, Recall that given X ∈ Cn,m, Y ∈ Cm,n is an inner inverse of A, if XY X = X. In addition, Y is said to be an outer inverse of A, if Y XY = Y . Next the outer inverse with prescribed range and null space will be recalled. Let A ∈ Cn,m and consider subspaces T ⊆ C n such that dim T = s ≤ rk(A) and dim S = n − s. Necessary and sufficient for the matrix A to have an outer inverse Z such n, in which case Z is unique and it is that R(Z) = T and N(Z) = S is that A(T) ⊕ S = C denoted by A(2) T,S (see for example [20, Lemma 1.1]). m and S ⊆ C J. Ben´ıtez, E. Boasso, H. Jin 3 2 The definition of the one-sided (D, E) inverses and their re- lationship with other inverses In first place the definition of the (b, c)-inverse will be recalled (see [9, Definition 1.3]). Definition 2.1. Let S be a semigroup and consider a, b, c ∈ S. The element y ∈ S will be said to be the (b, c)-inverse of a, if the following equations hold: (i) y ∈ (bSy) ∩ (ySc). (ii) b = yab, c = cay. According to [9, Theorem 2.1], if the element y in Definition 2.1 exists, then it is unique. In this case, the element under consideration will be denoted by ak(b,c). As it was pointed out in [9], this inverse generalizes among others the standard inverse, the Drazin inverse, and the Moore-Penrose inverse. To learn more on this inverse, see [5, 6, 9, 10, 12]. The inverse along an element was introduced in [14, Definition 4]. Next its definition will be recalled. Definition 2.2. Let S be a semigroup. An element a ∈ S is said to be invertible along d ∈ S if there exists y ∈ S such that (i) yad = d = day. (ii) yS ⊆ dS. (iii) Sy ⊆ Sd. According to [14, Theorem 6], if the element y ∈ S in Definition 2.2 exists, then it is unique. This element is denoted by akd. It is worth noting that according to [9, Proposition 6.1], the inverse along an element is a particular case of the (b, c)-inverse, i.e., the (d, d)-inverse coincides with the inverse along d. The outer inverses recalled in Definition 2.1 and Definition 2.2 encompass several generalized inverses, as the following two theorems show. Theorem 2.3. ([14, Theorem 11]) Let S be a semigroup and let a ∈ S. (i) If S has a unity, then a is invertible if and only if a is invertible along 1. In this case a−1 = ak1. (ii) a is group invertible if and only if a is invertible along a. In this case a# = aka. (iii) a is Drazin invertible if and only if a is invertible along am for some m ∈ . In this case aD = akam . (iv) If S is a ∗-semigroup, a is Moore-Penrose invertible if and only if a is invertible along a∗. In this case a† = aka∗ . Theorem 2.4. Let R be a ring with an involution and a ∈ R. (i) ([18, Theorem 4.3]) If a is Moore-Penrose invertible, then a is core invertible if and only if it is invertible along aa∗. In this case the inverse along aa∗ coincides a #(cid:13). (ii) ([18, Theorem 4.3]) If a is Moore-Penrose invertible, then a is dual core invertible if and only if it is invertible along a∗a. In this case the inverse along a∗a coincides with a #(cid:13). J. Ben´ıtez, E. Boasso, H. Jin 4 (iii) ([4, Theorem 3.2]) If m, n ∈ R are invertible and positive, then a is weighted Moore- Penrose invertible with weights m and n if and only if a is invertible along n−1a∗m. In this case, the inverse along n−1a∗m coincies with a† m,n. Recently the inverse along an element and the (b, c)-inverse were extended by means of one-sided inverses. Next follow the corresponding definitions. See [13, Definition 2.1] and [24, Definition 2.1]. Definition 2.5. Let R be a ring and let b, c ∈ R. An element a ∈ R is said to be left (b, c)-invertible, if there exists y ∈ R such that (i) yab = b. (ii) Ry ⊆ Rc. In this case y is called a left (b, c)-inverse of a. An element a ∈ R is right (b, c)-invertible, if there exists y ∈ R such that (iii) cay = c. (iv) yR ⊆ bR. In this case y is called a right (b, c)-inverse of a. Recall that given a, b, c elements in a ring R, according to [13, Corollary 3.7], a is (b, c)- invertible if and only if it is both left and right (b, c)-invertible. When in Definition 2.5, b = c, the one-sided inverses along an element are obtained. Definition 2.6. Let S be a semigroup. An element a ∈ S is left invertible along d ∈ S, if there exists y ∈ S such that (i) yad = d. (ii) Sy ⊆ Sd. In this case y is called a left inverse of a along d. An element a ∈ S is right invertible along d ∈ S, if there exists y ∈ S such that (iii) day = d. (iv) yS ⊆ dS. In this case y is called a right inverse of a along d. Recall that given a semigroup S and a, d ∈ S, according to [24, Corollary 2.5], a is invertible along d if and only if a is left and right invertible along d. Naturally, since all the inverses that have been considered in this section up to now have been defined in semigroups and rings, they can not be applied to matrices, unless they are square. However, to extend the aforementioned notions to arbitrary matrices, first it is necessary to recall the following facts. Let U , V ∈ Cm,n. There is X ∈ Cm (respectively Y ∈ Cn) such that U = XV (respectively U = V Y ) if and only if N(V ) ⊆ N(U ) (respectively R(U ) ⊆ R(V )). Now with these facts in mind, the notions in Definition 2.5 and Definition 2.6 can be extended to rectangular matrices. J. Ben´ıtez, E. Boasso, H. Jin 5 Definition 2.7. Let A ∈ Cn,m and D, E ∈ Cm,n. (i) The matrix A is said to be left (D, E)-invertible, if there exists C ∈ Cm,n such that CAD = D and N(E) ⊆ N(C). Any matrix C satisfying these conditions is said to be a left (D, E)-inverse of A. (ii) The matrix A is said to be right (D, E)-invertible, if there exists B ∈ Cm,n such that EAB = E and R(B) ⊆ R(D). Any matrix B satisfying these conditions is said to be a right (D, E)-inverse of A. The proofs of the following results are straightforward and they are left to the reader. Remark 2.8. Let A ∈ Cn,m and D, E ∈ Cm,n. The following statements hold (i) The matrix A is left (D, E)-invertible with a left inverse C ∈ Cm,n if and only if A∗ ∈ Cm,n is right (E∗, D∗)-invertible and C ∗ ∈ Cn,m is a right (E∗, D∗)-inverse of A∗. (ii) The matrix A is right (D, E)-invertible with a right inverse B ∈ Cm,n if and only if A∗ ∈ Cm,n is left (E∗, D∗)-invertible and B∗ ∈ Cm,n is a left (E∗, D∗)-inverse of A∗. Consider D′, E′ ∈ Cm,n such that R(D′) = R(D) and N(E′) = N(E). The following statement holds. (iii) The matrix A is left (D, E)-invertible if and only if it is left (D′, E′)-invertible. In addition, in this case C ∈ Cm,n is a left (D, E)-inverse of A if and only if it is a left (D′, E′)-inverse of A. (iv) The matrix A is right (D, E)-invertible if and only if it is right (D′, E′)-invertible. Moreover, in this case B ∈ Cm,n is a right (D, E)-inverse of A if and only if it is a right (D′, E′)-inverse of A. When the matrices D, E ∈ Cm,n in Definition 2.7 coincide, the notions of left and right inverse along a matrix can be introduced. Definition 2.9. Let A ∈ Cn,m and D ∈ Cm,n. (i) The matrix A is said to be left invertible along D, if there exists C ∈ Cm,n such that CAD = D and N(D) ⊆ N(C). Any matrix C satisfying these conditions is said to be a left inverse of A along D. (ii) The matrix A is said to be right invertible along D, if there exists B ∈ Cm,n such that DAB = D and R(B) ⊆ R(D). Any matrix B satisfying these conditions is said to be a right inverse of A along D. Note that similar results to the ones in Remark 2.8 for the case D = E ∈ Cm,n hold for left and right invertible matrices along a matrix. The details are left to the reader. Recall that given a ring R and a, b, c ∈ R, in [13, Definition 2.3] the left and right annihilator (b, c)-inverses of the element a were introduced. However, in the case of matrices, as under the conditions of [13, Proposition 2.5], these notions coincide with the ones in Definition 2.7. J. Ben´ıtez, E. Boasso, H. Jin 6 3 Characterizations of the one-sided (D, E)-invertibility In first place matrices that satisfy Definition 2.7 will be characterized. Theorem 3.1. Let A ∈ Cn,m and D, E ∈ Cm,n. The following statements are equivalent. (i) The matrix A is right (D, E)-invertible. (ii) R(E) = R(EAD). (iii) rk(E) = rk(EAD). (iv) dim N(E) = dim N(EAD). (v) C n = R(AD) + N(E). (vi) C m = R(D) + N(EA) and rk(EA) = rk(E). Proof. In first place, it will be proved that statement (i) implies statement (ii). Assume that there exists a matrix B ∈ Cm,n such that EAB = E and R(B) ⊂ R(D). Recall that the latter condition is equivalent to the fact that there exists M ∈ Cn such that B = DM . Therefore, E = EAB = EADM. In particular, R(E) = R(EAD). Suppose that statement (ii) holds. Thus, there exists X ∈ Cn such that EADX = E. To prove statement (i), it is enough to define B = DX. Statements (ii), (iii), and (iv) are equivalent. In fact, since R(EAD) ⊆ R(E), statements In addition, since dim N(E) + rk(E) = n = dim N(EAD) + (ii) and (iii) are equivalent. rk(EAD), statements (iii) and (iv) are equivalent. Statements (i) and (v) are equivalent. In fact, according to what has been proved, if statement (i) holds, then there is M ∈ Cn such that E = EADM . In particular, R(In − n can be written as x = ADM x + (x − ADM x), statement ADM ) ⊆ N(E). Since any x ∈ C n) = (v) holds. On the other hand, statement (v) implies statement (ii), since R(E) = E(C E[R(AD) + N(E)] = R(EAD). In this paragraph, it will be proved that statement (i) implies statement (vi). Assume that statement (i) holds. Then there exists a matrix B ∈ Cm,n such that E = EAB and m can be written as y = BAy + (y − BAy), the equality R(B) ⊆ R(D). Since any y ∈ C m = R(D) + N(EA) is obtained. To prove the rank equality, according to statement (iii), C rk(E) = rk(EAD) ≤ rk(EA) ≤ rk(E). Finally, it will be proved that statement (vi) implies statement (iii). In fact, if statement (vi) holds, then R(EA) = EA(C m) = EA(R(D) + N(EA)) = R(EAD). However, rk(E) = rk(EA) = rk(EAD). Theorem 3.2. Let A ∈ Cn,m and D, E ∈ Cm,n. The following statements are equivalent. (i) The matrix A is left (D, E)-invertible. J. Ben´ıtez, E. Boasso, H. Jin 7 (ii) N(D) = N(EAD). (iii) dim N(D) = dim N(EAD). (iv) rk(D) = rk(EAD). (v) N(EA) ∩ R(D) = 0. (vi) R(AD) ∩ N(E) = 0 and rk(D) = rk(AD). Proof. Recall that according to Remark 2.8 (i), A is left-(D, E)-invertible if and only if A∗ ∈ Cm,n is right (E∗, D∗)-invertible (E∗, D∗ ∈ Cn,m). In addition, recall that N(D) = R(D∗)⊥, N(EA) = R(A∗E∗)⊥, R(AD) = N(D∗A∗)⊥, rk(D) = rk(D∗), N(EAD) = R(D∗A∗E∗)⊥. R(D) = N(D∗)⊥. N(E) = R(E∗)⊥. rk(AD) = rk(D∗A∗). To conclude the proof, apply Theorem 3.1 to A∗, E∗ and D∗, use the above identities and note that since N(EAD) ⊆ N(D), statememts (ii) and (iii) are equivalent. In addition, note that since dim N(D) + rk(D) = n = dim N(EAD) + rk(EAD), statements (iii) and (iv) are equivalent. Next given D, E ∈ Cm,n, left and right (D, E)-invertible matrices will be characterized using a particular map. Theorem 3.3. Let A ∈ Cn,m and D, E ∈ Cm,n. Let X be any subspace of C n such that n = N(E) ⊕ X. Consider φ : R(D) → X the map defined by φ(x) = PX,N(E)(Ax), for C x ∈ R(D). The following statements hold. (i) The matrix A is left (D, E)-invertible if and only if φ is injective. (ii) The matrix A is right (D, E)-invertible if and only if φ is surjective. Proof. First statement (i) will be proved. Observe that N(φ) = R(D) ∩ N(EA). Thus, according to Theorem 3.2, N(φ) = 0 if and only if A is (D, E)-left invertible. The assertion (ii) will be proved in this paragraph. Note that x ∈ R(EAD) if and only if exists y ∈ C n such that x = E[PX,N(E)(ADy) + PN(E),X(ADy)]. Since E[PX,N(E)(ADy) + PN(E),X(ADy)] = E(φ(Dy)), the equality R(EAD) = E(R(φ)) is m given by f (x) = Ex is injective obtained. In addition, the linear mapping f : R(φ) → C n), therefore, dim R(φ) = dim E(R(φ)) = rk(EAD), and (because R(φ) ⊆ X and X⊕N(E) = C thus, φ is surjective (which is equivalent to dim R(φ) = dim X) if and only if rk(EAD) = rk(E). However, according to Theorem 3.1, this latter condition is equivalent to the fact that A is right (D, E)-invertible. In the following theorem matrices satisfying simultaneously Theorem 3.1 and Theorem 3.2 will be studied. J. Ben´ıtez, E. Boasso, H. Jin 8 Theorem 3.4. Let A ∈ Cn,m and D, E ∈ Cm,n. The following statements are equivalent. (i) A is left and right (D, E)-invertible. (ii) R(EAD) = R(E) and N(EAD) = N(D). (iii) R(AD) ⊕ N(E) = C n and rk(D) = rk(AD). (iv) R(D) ⊕ N(EA) = C m and rk(E) = rk(EA). (v) The map φ : R(D) → X defined in Theorem 3.3 is bijective. Futhermore, in this case, rk(E) = rk(D). Proof. Apply Theorem 3.1, Theorem 3.2 and Theorem 3.3. Note also that rk(E) = rk(D). Actually, this equality can be derived from the fact that n = rk(EAD) + dim N(EAD) = rk(E) + dim N(D). Next the left and right (D, E)-inverses of a matrix A satisfying Theorem 3.4 will be characterized. Proposition 3.5. Let A ∈ Cn,m and D, E ∈ Cm,n be such that A is both left and right (D, E)-invertible. Then, there exist only one left (D, E)-inverse of A and only one right (D, E)-inverse of A. Moreover, these inverses coincide with the unique matrix R ∈ Cm,n satisfying N(R) = N(E), Ry = f −1(y), ∀y ∈ R(AD), where f : R(D) → R(AD) is the isomorphism defined by f (x) = Ax. Proof. Consider C ∈ Cm,n a left (D, E)-inverse of A. Then, CAD = D and N(E) ⊆ N(C). Let f : R(D) → R(AD) and g : R(AD) → R(D) be given by f (x) = Ax and g(y) = Cy. If n. From CAD = D, it is obtained x ∈ R(D), then gf (x) = CAx and x = Du for some u ∈ C that gf (x) = x. In a similar way, f g = IR(AD) can be proved, and therefore, g = f −1. In this paragraph it will be proved that N(C) = N(E). Since N(E) ⊆ N(C) is already known, it is enough to prove the opposite inclusion. Let x ∈ N(C), by Theorem 3.4 (iii), x can n and w ∈ N(E). Now, 0 = Cx = CADy + Cw = be written as x = ADy + w, where y ∈ C Dy because w ∈ N(E) ⊆ N(C). Finally, x = ADy + w = w ∈ N(E). Now consider B ∈ Cm,n a right (D, E)-inverse of A. In particular, EAB = E and R(B) ⊆ R(D). Let x ∈ N(E). Then, B(x) ∈ R(D) ∩ N(EA) = 0 (Theorem 3.4 (iv)). Thus, N(E) ⊆ N(B). The inclusion N(B) ⊆ N(E) is evident from EAB = E. Therefore, N(B) = N(E). Let h : R(AD) → R(D) and k : R(D) → R(E) defined by h(y) = By and k(y) = Ey. The mapping k is an isomorphism because it is simple to prove in view of Theorem 3.4 that N(k) = 0. Furthermore, EAB = E leads to kf h = k, and using that k is an isomorphism, f h = IR(AD), i.e., h = f −1. Now the relationship between the notions introduced in Definitions 2.9 will be studied. To this end, in first place a characterization of left invertibility along a matrix will be given. Theorem 3.6. Let A ∈ Cn,m and D ∈ Cm,n. The following statements are equivalent. (i) A is right invertible along D. J. Ben´ıtez, E. Boasso, H. Jin 9 (ii) R(D) = R(DAD). (iii) rk(D) = rk(DAD). (iv) C n = N(D) ⊕ R(AD). (v) C m = R(D) ⊕ N(DA). Proof. Suppose that statement (i) holds. Then, according to Theorem 3.1 applied to the n = N(D) + R(AD) and case D = E, statements (ii) and (iii) hold, rk(DA) = rk(D), C m = R(D) + N(DA). Now, since C n = rk(AD) + dim N(D) − dim[R(AD) ∩ N(D)] = rk(D) + dim N(D) − dim[R(AD) ∩ N(D)], R(AD) ∩ N(D) = 0 and statement (iv) holds. Similarly, since m = rk(D) + dim N(DA) − dim[R(D) ∩ N(DA)] = rk(DA) + dim N(DA) − dim[R(D) ∩ N(DA)], R(D) ∩ N(DA) = 0 and statement (v) holds. On the other hand, note that statement (ii) (respectively (iii), (iv), (v)) implies statement (ii) (respectively (iii), (iv), (v)) of Theorem 3.1 applied to the case D = E. For statement (v), note also that since C m = R(D) ⊕ N(DA), the equality rk(D) = rk(DA) can be obtained. It is possible to obtain similar statements for right invertible elements along a matrix, however, as the following theorem shows, left and right inverse along a matrix are equivalent notions. Theorem 3.7. Let A ∈ Cn,m and D ∈ Cm,n. The following statements are equivalent. (i) A is right invertible along D. (ii) A is left invertible along D. (iii) N(D) = N(DAD). (iv) dim N(D) = dim N(DAD). (v) AD ∈ Cn is group invertible and dim N(D) = dim N(AD). (vi) DA ∈ Cm is group invertible and rk(DA) = rk(D). (vii) The map φ : R(D) → X defined in Theorem 3.3 for the case D = E is bijective. Proof. According to Theorem 3.2 applied to the case D = E, statements (ii), (iii) and (iv) are equivalent. In addition, note that statement (iv) is equivalent to Theorem 3.6 (iii). In particular, statements (i) and (ii) are equivalent. Suppose that statement (i) holds. Then according to Theorem 3.6 (iv), C n = N(D) ⊕ R(AD). Moreover, according to Theorem 3.2 (v), rk(D) = rk(AD). However, the latter J. Ben´ıtez, E. Boasso, H. Jin 10 identity is equivalent to R(D) = R(AD), which in turn is equivalent to N(D) = N(AD). In n = N(AD) ⊕ R(AD), i.e., AD is group invertible, and dim N(D) = dim N(AD). particular, C n = N(AD)⊕R(AD) and N(D) = N(AD) On the other hand, if statement (v) holds, then C (N(AD) ⊆ N(D)). Consequently, Theorem 3.6 (iv) holds. The equivalence between statements (i) and (vi) can be proved a similar argument, using in particular Theorem 3.6 (v) and Theorem 3.1 (vi). Since statements (i) and (ii) are equivalent, according to Theorem 3.4, statement (i) and (vii) are equivalent. In the following corollary the left and the right inverses of a matrix A ∈ Cn,m that is left or right invertible along D ∈ Cm,n will be presented. Corollary 3.8. Let A ∈ Cn,m and D ∈ Cm,n such that A is left or right invertible along D. Then, there exists only one left inverse of A along D and only one right inverse of A along D. Moreover, these inverses coincide with the matrix R ∈ Cm,n satisfying N(R) = N(D), Ry = f −1(y), ∀y ∈ R(AD), where f : R(D) → R(AD) is given by f (x) = Ax. Proof. Apply Theorem 3.7 and Proposition 3.5. Now the existence of left and right (D, E)-inverses will be studied. To this end the sets of left and right (D, E)-invertible matrices will be characterized. First of all some notation will be given. Consider D, E ∈ Cm,n. Let (Cn,m)kD,E lef t (D, E)-invertible matrices, respectively, i.e., and (Cn,m)kD,E right be the sets of left and right (Cn,m)kD,E (Cn,m)kD,E lef t = {A ∈ Cn,m : A is left (D, E)-invertible}, right = {A ∈ Cn,m : A is right (D, E)-invertible}. When D = E ∈ Cm,n, the sets of left and right invertible matrices along D, (Cn,m)kD lef t and (Cn,m)kD right respectively, are introduced. (Cn,m)kD (Cn,m)kD lef t = (Cn,m)kD,D right = (Cn,m)kD,D lef t = {A ∈ Cn,m : A is left invertible along D}, right = {A ∈ Cn,m : A is right invertible along D}. Next conditions under which the sets (Cn,m)kD,E given. lef t and (Cn,m)kD,E right are non empty will be Theorem 3.9. Let D, E ∈ Cm,n. The following statements are equivalent. (i) (Cn,m)kD,E lef t 6= ∅. (ii) rk(D) ≤ rk(E). J. Ben´ıtez, E. Boasso, H. Jin 11 (iii) dim N(E) ≤ dim N(D). (iv) dim N(E) + rk(D) ≤ n. Proof. Here, It will be proved that statement (i) implies statement (ii). Suppose that there exists A ∈ Cn,m and C ∈ Cm,n such that C is a right (D, E)-inverse of A. In particular, CAD = D and N(E) ⊆ N(C). Thus, rk(D) ≤ rk(C) and dim N(E) ≤ dim N(C). As a result, rk(D) ≤ rk(C) ≤ rk(E). Suppose that statement (iv) holds. Let r = rk(E) and s = rk(D). Let X = [x1 · · · xn] ∈ Cn and Y = [y1 · · · ym] ∈ Cm be two nonsingular matrices such that the last n − r columns of X span N(E) and the first s columns of Y span R(D). Define 0 0 C = Y (cid:20) Is 0 0 (cid:21) X −1 ∈ Cm,n. A = X(cid:20) Is 0 If x ∈ N(E), then x = Pn 0 (cid:21) Y −1 ∈ Cn,m, i=r+1 αixi for some scalars αi. Since s = rk(D) ≤ rk(E) = r, n−s can be defined (the superscript T means the s such 0 ] (because the first s columns of Y span R(D)). Now, it is trivial to prove the vector v = [0 · · · 0 αr+1 · · · αn]T ∈ C transposition). Hence x = X [ 0 that y = Y [ w CAy = y, which implies CAD = D since y ∈ R(D) is arbitrary. Hence (i) holds. v ], and thus, Cx = 0. If y ∈ R(D), then exists w ∈ C The remaining equivalences are clear. Theorem 3.10. Let A ∈ Cn,m and D, E ∈ Cm,n. The following statements are equivalent. (i) (Cn,m)kD,E right 6= ∅. (ii) rk(E) ≤ rk(D). (iii) dim N(D) ≤ dim N(E). (iv) n ≤ dim N(E) + rk(D). Proof. Recall that according to Remark 2.8 (ii), A is right (D, E)-invertible if and only if A∗ ∈ Cm,n is left (E∗, D∗)-invertible (E∗, D∗ ∈ Cn,m). Consequently, statement (i) is equivalent to (Cm,n)kE∗,D∗ 6= ∅, which in turn is equivalent to rk(E∗) ≤ rk(D∗). However, the latter inequality coincides with statement (ii). lef t The remaining equivalences are clear. Next the case of the left and right inverses along a matrix will be considered. Corollary 3.11. Let D ∈ Cm,n. Then, (Cn,m)kD lef t = (Cn,m)kD right 6= ∅. Proof. Apply Theorem 3.7 and Theorem 3.9 or Theorem 3.10 for the case D = E. Next the results in Corollary 3.11 will be extended to the case dim N(E) + rk(D) = n (D, E ∈ Cm,n). Note that this condition is equivalent to rk(D) = rk(E), which in turn is equivalent to dim N(D) = dim N(E), which is also equivalent to dim N(D) + rk(E) = n. Corollary 3.12. Let D, E ∈ Cm,n. The following statements are equivalent. J. Ben´ıtez, E. Boasso, H. Jin 12 (i) rk(E) = rk(D). (ii) (Cn,m)kD,E lef t 6= ∅ and (Cn,m)kD,E right 6= ∅. (iii) (Cn,m)kD,E lef t = (Cn,m)kD,E right 6= ∅. Proof. To prove the equivalence between statements (i) and (ii), apply Theorem 3.9 and Theorem 3.10. Suppose that statment (i) holds and consider A ∈ (Cn,m)kD,E right . According to Theorem 3.1 (iii), rk(D) = rk(EAD). Thus, dim N(D) = dim N(EAD). Consequently, according to Theorem 3.2 (iii), A ∈ (Cn,m)kD,E . A similar argument, using in particular that dim N(D) = lef t dim N(E), proves that (Cn,m)kD,E right . On the other hand, if statement (iii) holds, then consider A ∈ (Cn,m)kD,E right . According to Theorem 3.1 and Theorem 3.2, rk(E) = rk(EAD) and N(D) = N(EAD). Therefore, dim N(D) + rk(E) = n. lef t = (Cn,m)kD,E lef t ⊆ (Cn,m)kD,E Now the case rk(D) 6= rk(E) will be presented. Corollary 3.13. Let D, E ∈ Cm,n such that rk(D) 6= rk(E). Then, the following statements hold. (i) If rk(D) < rk(E), then (Cn,m)kD,E lef t 6= ∅ and (Cn,m)kD,E right = ∅. (ii) If rk(E) < rk D), then (Cn,m)kD,E right 6= ∅ and (Cn,m)kD,E lef t = ∅. Proof. Apply Theorem 3.9 and Theorem 3.10. Next representations of the sets (Cn,m)kD,E first place some notation needs to be introduced. lef t and (Cn,m)kD,E right will be given. However, in Let D, E ∈ Cm,n and consider A ∈ (Cn,m)kD,E lef t of A will be denoted by I(A)kD,E lef t the set of all right (D, E)-inverses of A. . Similarly, when A ∈ (Cn,m)kD,E . Then, the set of all left (D, E)-inverses right will stand for right , I(A)kD,E In the following theorems representations of (Cn,m)kD,E lef t will be given. , (Cn,m)kD,E right , I(A)kD,E lef t and I(A)kD,E right Theorem 3.14. Let D, E ∈ Cm,n be such that (Cn,m)kD,E lef t Then 6= ∅. Let s = rk(D) ≤ rk(E) = r. (i) A ∈ (Cn,m)kD,E lef t such that if and only if there exist two nonsingular matrices X ∈ Cn and Y ∈ Cm s A = X r−s n−r s m−s A1 0 0 ∗ ∗ ∗     Y −1, (1) where the last n − r columns of X are a basis of N(E), the first s columns of Y are a basis of R(D), and A1 is nonsingular. J. Ben´ıtez, E. Boasso, H. Jin (ii) Under the conditions in statement (i), C ∈ I(A)kD,E lef t if and only if C = Y s m−s s (cid:20) A−1 1 0 r−s n−r ∗ 0 ∗ 0 (cid:21)X −1. 13 (2) Proof. Assume that A ∈ (Cn,m)kD,E . Let X ∈ Cn be any nonsingular matrix such that the lef t last n − r columns span N(E). Let Y ∈ Cm be any nonsingular matrix such that the first s columns span R(D). Let us decompose matrix A as follows: A = X r n−r (cid:20) s m−s B1 B2 ∗ ∗ (cid:21)Y −1. (3) Observe that if yi is the i-th column of Y , then Ayi ∈ R(AD) for i = 1, . . . , s, because the first s columns of Y span R(D). From (3), it is obtained that B2 = 0, because R(AD) ∩ N(E) = 0 (Theorem 3.2 (vi)). Let X be the subspace spanned by the first r columns of X (this subspace satisfies X ⊕ n). It is evident that the matrix of the linear mapping φ : R(D) → X defined in N(E) = C Theorem 3.3 respect to the considered basis of R(D) and X is B1. By Theorem 3.3 (i), φ is injective, thus, X can be decomposed as X = φ(R(D)) ⊕ X1, and without loss of generality, the first n − r columns of X can be rearranged so that the first s are a basis of φ(R(D)). In this way, the decomposition written in the statement (i) is obtained. Assume that A ∈ Cn,m is decomposed as in (1). As in the previous paragraph, let X be the subspace spanned by the first r columns of X and consider the mapping φ : R(D) → X defined in Theorem 3.3. The first s columns of Y is a basis of R(D) and the first r columns of X is a basis of X. The matrix of φ respect to the aforementioned basis is (cid:2) A1 A1 is nonsingular, the mapping φ is injective, and according to Theorem 3.3 (i), A is left (D, E)-invertible. 0 (cid:3). Since Now, it will be proved statement (ii). Let yj be the j-th column of Y and xi be the i-th column of X. Assume that C ∈ Cm,n is a left (D, E)-inverse of A. Let us decompose C as follows: s r−s n−r C1 C2 C3 s C = Y C4 C5 C6 (cid:21)X −1. 0 (cid:3), where {e1, . . . , es} is the standard basis of C (cid:20) m−s (4) Since yj = Y (cid:2) ej tion and (1), it is obtained that CAyj = h C1A1 ej is true because C is a left (D, E)-inverse of A), it follows that C1A1ej = ej and C4A1ej = 0, for any j = 1, . . . , s. Therefore C1 = A−1 and C4 = 0. Having in mind that the last n − r 1 columns of X is a basis of N(E) and N(E) ⊆ N(C), the decomposition (4) yields that C3 and C6 are zero matrices. C4A1 ej i, for any j = 1, . . . , s. From CAD = D (this n, from the above decomposi- Assume that C ∈ Cm,n is written as in (2). The following two relations will be proved: CAD = D and N(E) ⊆ N(C). To prove that CAD = D, it is enough to check that CAyj = yj, for j = 1, . . . , s (because R(D) is spanned by y1, . . . , ys), and this trivially follows from (1), s. To prove that N(E) ⊆ N(C), j=1 is the standard basis of C (2), and yj = Y (cid:2) ej 0 (cid:3), where {ej}s J. Ben´ıtez, E. Boasso, H. Jin 14 it is enough to prove that Cxi = 0, for i = r + 1, . . . , n (because N(E) is spanned by the last n − r columns of X). This is trivial in view of (2). Theorem 3.15. Let D, E ∈ Cm,n be such that (Cn,m)kD,E Then right 6= ∅. Let r = rk(E) ≤ rk(D) = s. (i) A ∈ (Cn,m)kD,E right if and only if there exist two nonsingular matrices X ∈ Cn and Y ∈ Cm such that A = X r n−r (cid:20) r m−s s−r 0 A2 ∗ ∗ ∗ ∗ (cid:21)Y −1, (5) where the last n − r columns of X are a basis of N(E), the first s columns of Y are a basis of R(D), and A2 is nonsingular. (ii) Under the conditions in statement (i), B ∈ I(A)kD,E right if and only if s−r B = Y r m−s r ∗ A−1 2 0   n−r ∗ 0 0 X −1.   (6) Proof. Assume that A ∈ (Cn,m)kD,E right . Let X ∈ Cn be any nonsingular matrix such that the last n − r columns span N(E) and let X be the subspace spanned by the first r columns of X. The mapping φ : R(D) → X defined in Theorem 3.3 is surjective, and therefore, s = dim R(D) = dim N(φ) + dim X = dim N(φ) + r. In Theorem 3.3, it was proved that N(φ) = R(D) ∩ N(EA). Let Y ∈ Cm be any nonsigular matrix such that the first s − r columns of Y are a basis of R(D) ∩ N(EA) and the first s columns of Y are a basis of R(D). Decompose A = X s−r r m−s A1 A2 A3 A4 A5 A6 (cid:21)Y −1. r n−r (cid:20) Let yj be the j-th column of Y . For j = 1, . . . , s − r, it is obtained that EAyj = 0, because the first s − r columns of Y belong to N(EA), in other words, Ayj ∈ N(E), and thus, A1 is a zero matrix. The matrix of φ respect the considered basis of R(D) and X is [A1 A2] = [0 A2]. Since φ : R(D) → X is surjective, r = dim X = rk[0 A2] = rk(A2). By recalling that A2 is an r × r matrix, A2 is nonsingular. Assume that A ∈ Cn,m is decomposed as in (5). As in the previous paragraph, let X be the subspace spanned by the first r columns of X. The matrix of the mapping φ : R(D) → X respect the considered basis is s−r r Since the rank of this latter matrix is r (because A2 ∈ Cr is nonsingular) and dim X = r, the mapping φ is surjective. According to Theorem 3.3 (ii), A ∈ (Cn,m)kD,E right . r (cid:2) 0 A2 (cid:3). J. Ben´ıtez, E. Boasso, H. Jin 15 Assume that B ∈ Cm,n is a right (D, E)-inverse of A, i.e., EAB = B and R(B) ⊆ R(D). Decompose s−r B = Y r m−s r n−r B1 B2 B3 B4 B5 B6     X −1. By the condition R(B) ⊆ R(D), it is obtained that Bx ∈ R(D) for any x ∈ C Y   B1 B2 B3 B4 B5 B6  (cid:20) x1 x2 (cid:21) ∈ R(D), ∀ (x1, x2) ∈ C r × C n−r. n. Therefore, Recall that the first s columns of Y span R(D), and thus, B5x1 + B6x2 = 0, for all (x1, x2) ∈ n−r, which implies that B5 and B6 are zero matrices. The equality EAB = E is r × C C equivalent to R(AB − In) ⊆ N(E). But AB − In = X (cid:20) 0 A2 ∗ ∗ ∗ ∗ (cid:21)  B1 B2 B3 B4 0 0   X −1 − XX −1 = X (cid:20) A2B3 − Ir A2B4 ∗ ∗ (cid:21) X −1. Therefore, X [ A2B3−Ir A2B4 n − r columns of X span N(E), which implies that A2B3 = Ir and B4 = 0. x2 ] ∈ N(E), for all (x1, x2) ∈ C r × C ] [ x1 ∗ ∗ n−r. Recall that the last Assume that B ∈ Cm,n is written as in (6). It will be proved now that R(B) ⊆ R(D). If x ∈ C n, then Bx = Y   ∗ A−1 1 0 ∗ 0 0   X −1x ∈ R(D), because the first s columns of Y belong to R(D). Now it will be proved that EAB = B. Since AB − In = X(cid:20) 0 A2 ∗ ∗ ∗ ∗ (cid:21)  ∗ A−1 2 0 ∗ 0 0   X −1 − XX −1 = X(cid:20) 0 0 ∗ ∗ (cid:21) X −1, it is obtained that (AB − In)x ∈ N(E), for any x ∈ C belong to N(E). Thus, R(AB − In) ⊆ N(E), which is equivalent to EAB = E. n, because the last n − r columns of X Next the case rk(E) = rk(D) will be studied, i.e., when (Cn,m)kD,E (D, E ∈ Cm,n). Compare with Proposition 3.5. lef t = (Cn,m)kD,E right 6= ∅ Corollary 3.16. Let D, E ∈ Cm,n be such that rk(E) = rk(D) = r. Let X ∈ Cn be any nonsingular matrix such that its last n−r columns span N(E) and Y ∈ Cm be any nonsingular matrix such that its first r columns span R(D). If A ∈ Cn,m is written as A = X r m−r (cid:20) r A1 ∗ n−r ∗ ∗ (cid:21)Y −1, J. Ben´ıtez, E. Boasso, H. Jin 16 then A ∈ (Cn,m)kD,E above equivalence, the unique left and the unique right (D, E)-inverse of A is right if and only if A1 is nonsingular. Furthermore, under the lef t = (Cn,m)kD,E Y (cid:20) A−1 1 0 0 0 (cid:21) X −1. Proof. Let X be the subspace spanned by the first r columns of X. The matrix of the mapping φ : R(D) → X defined in Theorem 3.3 respect the considered basis is A1. Therefore, according to Theorem 3.4, the matrix A1 is nonsingular if and only if A is left and right (D, E)-invertible. The proof of Theorem 3.14 (ii) shows that the unique left and right (D, E)-inverse of A as the form of the statement of this corollary. Remark 3.17. Let D, E ∈ Cm,n be such that rk(E) = rk(D) and A ∈ (Cn,m)kD,E (Cn,m)kD,E and right (D, E)-inverse of A, say R, satisfies lef t = n = N(E) ⊕ X be any decomposition. From Corollary 3.16, the unique left right . Let C N(R) = N(E), Ry = φ−1(y), ∀y ∈ X. Next the case of left and right inverses along a fixed matrix will be considered. Corollary 3.18. Let D ∈ Cm,n and r = rk(D). Let X ∈ Cn be any nonsingular matrix such that its last columns span N(D) and Y ∈ Cm be any nonsingular matrix such that its first columns span R(D). If A ∈ Cn,m is written as A = X (cid:20) A1 ∗ ∗ (cid:21) Y −1, A1 ∈ Cr, ∗ then A ∈ (Cn,m)kD above equivalence, the unique left and the unique right inverse of A along D is right if and only if A1 is nonsingular. Furthermore, under the lef t = (Cn,m)kD Y (cid:20) A−1 1 0 0 0 (cid:21) X −1. Proof. Apply Corollary 3.11 and Corollary 3.16. In the following theorem the sets I(A)kD,E lef t Moore-Penrose inverse. and I(A)kD,E right will be represented using the Theorem 3.19. Let D, E ∈ Cm,n. The following statements hold. (i) If A ∈ (Cn,m)kD,E right , then right = nDh(EAD)†E +(cid:16)In − (EAD)†EAD(cid:17) Zi : Z ∈ Cno . I(A)kD,E (ii) If A ∈ (Cn,m)kD,E lef t , then lef t = nhD(EAD)† + Z(Im − EAD(EAD)†)i E : Z ∈ Cmo . I(A)kD,E J. Ben´ıtez, E. Boasso, H. Jin 17 Proof. Consider A ∈ (Cn,m)kD,E right . If B ∈ Cm,n satisfies EAB = E and R(B) ⊂ R(D), then according to the proof of Theorem 3.1, there exists a matrix M such that B = DM and EADM = E. Notice that the general solution of the equation EADX = E is X = (EAD)†E + (In − (EAD)†EAD)Z, where Z ∈ Cn,n is arbitrary (see [16, Theorem 2]). Hence M = (EAD)†E+(In−(EAD)†EAD)Z, for some Z ∈ Cn,n. Therefore, B = D(EAD)†E + D(In − (EAD)†EAD)Z. Thus, (Cn,m)kD,E right ⊆ {D(EAD)†E + D(In − (EAD)†EAD)Z : Z ∈ Cn,n}. To prove the opposite inclusion, let Z ∈ Cn,n be arbitrary and consider Y = Dh(EAD)†E + (In − (EAD)†EAD)Zi . It is evident that R(Y ) ⊂ R(D). Furthermore, according to Theorem 3.1 (ii), R(E) = R(EAD). Now EAY = EADh(EAD)†E + (In − (EAD)†EAD)Zi = EAD(EAD)†E = E. In fact, if e is any column of E, then e ∈ R(E) = R(EAD) and since EAD(EAD)† is the orthogonal projection onto R(EAD), EAD(EAD)†e = e. To prove statement (ii), apply Remark 2.8 (i) and what has been proved. Let D, E ∈ Cm,n. If A ∈ Cn,m is left (respectively right) (D, E)-invertible, the case in which I(A)kD,E lef t (respectively I(A)kD,E right ) is a singleton will be studied. Theorem 3.20. Let A ∈ Cn,m and D, E ∈ Cm,n. The following statements are equivalent. (i) The matrix A has a unique left (D, E)-inverse. (ii) The matrix A has a unique right (D, E)-inverse. (iii) rk(D) = rk(E) = rk(EAD). Furthermore, in this case I(A)kD,E lef t = I(A)kD,E right = {D(EAD)†E}. Proof. Note that according to Theorem 3.2 (iv) and Theorem 3.14 (ii), statement (i) and statement (iii) are equivalent. To prove the equivalence between statements (ii) and (iii), apply Theorem 3.1 (iii) and Theorem 3.15 (ii). Now, according to Proposition 3.5, the unique left and the unique right (D, E)-inverse of A coincide. To conclude the proof, notice that according to Theorem 3.19, D(EAD)†E ∈ I(A)kD,E lef t ∩ I(A)kD,E right . Due to Theorem 3.20, another representation of the left and the right inverses along a matrix can be given. Corollary 3.21. Let A ∈ Cn,m and D ∈ Cm,n. The matrix A is left or right invertible along D if and only if rk(D) = rk(DAD). Moreover, in this case the unique left inverse of A along D and the unique right inverse of A along D coincide with the matrix D(DAD)†D. J. Ben´ıtez, E. Boasso, H. Jin 18 Proof. Apply Theorem 3.7, Corollary 3.8 and Theorem 3.20. It is known that a nonzero matrix can be expressed as the product of a matrix of full column rank and a matrix of full row rank. This factorization is known as a full-rank fac- torization and these factorizations turn out to be a powerful tool in the study of generalized inverses. Recall that given H ∈ Cm,n such that rk(H) = r > 0, the matrix H is said to have a full rank factorization, if there exist F ∈ Cm,r and G ∈ Cr,n such that H = F G. Such a factorization always exists but it is not unique ([17, Theorem 2]). Moreover, F †F = Ir = GG† ([17, Theorem 1]). Note in particular that rk(F ) = rk(G) = rk(H). To learn more results on this topic, see [2, 17]. In the following theorem, given D, E ∈ Cm,n, matrices A ∈ Cn,m that are left or right (D, E)-invertible will be characterized using a full rank factorization. In addition, the sets I(A)kD,E right will be represented using a full rank factorization and the Moore-Penrose inverse. lef t and I(A)kD,E Theorem 3.22. Let A ∈ Cn,m and D, E ∈ Cm,n. Consider D = D1D2 and E = E1E2 two full rank factorizations of D and E, respectively. (i) The matrix A has a right (D, E)-inverse if and only if rk(E2) = rk(E2AD1). In addition, I(A)kD,E right = {D1h(E2AD1)†E2 + (Ir − (E2AD1)†E2AD1)Yi : Y ∈ Cr,n}, where r = rk(D). (ii) The matrix A has a left (D, E)-inverse if and only if rk(D1) = rk(E2AD1). Moreover, I(A)kD,E lef t = {hD1(E2AD1)† + Y (Is − (E2AD1)(E2AD1)†)i E2 : Y ∈ Cm,s}, where s = rk(E). Proof. Let r = rk(D) and s = rk(E). According to [17, Theorem 1], D2D† D† 1E1 = Is. In addition, since 1D1 = Ir and E† 2 = Ir, E2E† 2 = Is, rk(EAD) = rk(E1E2AD1D2) ≤ rk(E2AD1) ≤ rk(E† 1E1E2AD1D2D† 2) ≤ rk(EAD), Let A ∈ (Cn,m)kD,E rk(EAD) = rk(E2AD1). According to Theorem 3.1 (iii), A has a right (D, E)-inverse if and only if rk(E2) = rk(E) = rk(EAD) = rk(E2AD1). right and consider B ∈ I(A)kD,E right , i.e., EAB = E and R(B) ⊆ R(D). Since E1E2AB = E1E2, multiplying by E† 1 on the left hand side of this equation, E2AB = E2. In addition, since R(B) ⊆ R(D), there exists M ∈ Cn such that B = DM . Therefore, E2AD1(D2M ) = E2, and according to [16, Theorem 2], there exists Y ∈ Cr,n such that D2M = (E2AD1)†E2 +(cid:2)Ir − (E2AD1)†(E2AD1)(cid:3) Y . Thus, B = DM = D1D2M implies B = D1h(E2AD1)†E2 +(cid:16)Ir − (E2AD1)†E2AD1(cid:17) Yi . Now suppose that B ∈ Cm,n has this form. Observe that B = D1Z for some matrix Z. 2Z = DD† Thus, B = D1D2D† EAB = E1E2AD1h(E2AD1)†E2 +(cid:16)Ir − (E2AD1)†E2AD1(cid:17) Yi = E1E2AD1(E2AD1)†E2. 2Z implies that R(B) ⊆ R(D). In addition, J. Ben´ıtez, E. Boasso, H. Jin 19 Since rk(E2AD1) = rk(E2) and R(E2AD1) ⊆ R(E2), R(E2AD1) = R(E2). Therefore, E2AD1(E2AD1)† = PR(E2AD1) = PR(E2), which implies E2AD1(E2AD1)†E2 = E2. Thus, EAB = E1E2 = E. In particular, B ∈ I(A)kD,E right . To prove statement (ii), apply Remark 2.8 (i) and what has been proved. Next two particular cases will be derived from Theorem 3.22. Corollary 3.23. Let D, E ∈ Cm,n and consider D = D1D2 and E = E1E2 two full rank factorizations of D and E, respectively. The matrix A ∈ Cn,m is left and right (D, E)- invertible if and only if rk(E2) = rk(D1) = rk(E2AD1). Moreover, in this case, I(A)kD,E lef t = I(A)kD,E right = {D1(E2AD1)−1E2}. Proof. The first statement can be derived from Theorem 3.22. Note that rk(D) = rk(D1), rk(E) = rk(E2) and rk(EAD) = rk(E2AD1) (see the proof right is a singleton. In of Theorem 3.22). Then, according to Theorem 3.20, I(A)kD,E addition, according to Theorem 3.22, D1(E2AD1)†E2 ∈ I(A)kD,E lef t = I(A)kD,E lef t ∩ I(A)kD,E right . Thus, I(A)kD,E lef t = I(A)kD,E right = {D1(E2AD1)†E2}. However, since rk(E2) = r = rk(D1) and (E2AD1)† ∈ Cr is such that rk(E2AD1) = r, (E2AD1)† = (E2AD1)−1. To end this section, the case of left and right inverses along a matrix will be presented. Corollary 3.24. Let D ∈ Cm,n and consider D = D1D2 a full rank factorizations of D. The matrix A ∈ Cn,m is left or right invertible along D if and only if rk(D1) = rk(D2AD1). Moreover, in this case the unique left inverse of A along D and the unique right inverse of A along D coincide with D1(D2AD1)−1D2. Proof. Apply Theorem 3.7 and Corollary 3.23. 4 The (D, E)-inverse of arbitrary matrices First of all the (b, c)-inverse will be extended to rectangular matrices. Compare with Defini- tion 2.1 and recall the observation before Definition 2.7. Definition 4.1. Let A ∈ Cn,m and D, E ∈ Cm,n. The matrix A is said to be (D, E)-invertible, if there exist a matrix X ∈ Cm,n such that the following conditions hold. XAD = D, EAX = E, R(X) ⊆ R(D), N(E) ⊆ N(X). Under the same conditions as in Definition 4.1, note that R(X) ⊆ R(D) (respectively N(E) ⊆ N(X)) is equivalent to R(X) = R(D) (respectively N(E) ⊆ N(X)). In the following theorem, it will be proved that the (D, E)-inverse of a matrix A is unique, if it exists. Theorem 4.2. Let A ∈ Cn,m and D, E ∈ Cm,n. The following statements are equivalent. (i) The (D, E)-inverse of the matrix A exists. J. Ben´ıtez, E. Boasso, H. Jin 20 (ii) The matrix A is both left and right (D, E)-invertible. Furthermore, in this case, the (D, E)-inverse of the matrix A is unique. Proof. It is enough to prove that statement (ii) implies statement (i). To this end, apply Proposition 3.5. Proposition 3.5 also proves that there is only one (D, E)-inverse of A, when it exists. According to Theorem 4.2, if the matrix A ∈ Cn,m has a (D, E)-inverse (D, E ∈ Cm,n), then it will be denoted by Ak(D,E). In addition, note that according to Definition 2.5 and [13, Corollary 3.7], when the matrices A, D and E are square, Definition 4.1 reduces to the (b, c)- inverse ([9, Definition 1.3], i.e., Definition 2.1). In the following remark some basic results on this inverse that can be derived from what has been proved in sections 2 and 3 will be collected. Remark 4.3. Let D, E ∈ Cm,n and consider A ∈ Cn,m. (i) The matrix A is (D, E)-invertible if and only if A∗ ∈ Cm,n is (E∗, D∗)-invertible (E∗, D∗ ∈ Cn,m). Moreover, in this case (A∗)k(E∗,D∗) = (Ak(D,E))∗. Apply Theorem 4.2 and Remark 2.8 (i)-(ii). (ii) Let D′, E′ ∈ Cm,n be such that R(D) = R(D′) and N(E) = N(E′). Necessary and suffi- cient for Ak(D,E) to exist is that Ak(D′,E′) exists. Furthermore, in this case, Ak(D′,E′) = Ak(D,E). Apply Theorem 4.2 and Remark 2.8 (iii)-(iv). (iii) Theorem 3.4 and Theorem 3.20 characterize matrices A such that Ak(D,E) exists. (iv) When A is (D, E)-invertible, Ak(D,E) can be represented as in Propostion 3.5, Corol- lary 3.16 and Remark 3.17. Although some results have been presented in connection to left and right (D, E)-inverses (D, E ∈ Cm,n), they deserve to be considered again for the (D, E)-inverse. Recall that according to [9, Remark 2.4] (see also [9, Theorem 2.2]), when the marices A, D, E are square, A is (D, E)-invertible if and only if rk(D) = rk(EAD) = rk(E). In the following theorem this result will be extended to arbitrary matrices. Theorem 4.4. Let A ∈ Cn,m and D, E ∈ Cm,n. The following statements are equivalent. (i) The (D, E)-inverse of A exists. (ii) rk(D) = rk(E) = rk(EAD). Furthermore, in this case Ak(D,E) = D(EAD)†E. Proof. Apply Theorem 4.2, Proposition 3.5 and Theorem 3.20. Now a corollary will be derived from Theorem 4.4. Corollary 4.5. Let A ∈ Cn,m and D, E ∈ Cm,n be such that Ak(D,E) exists. Then rk(AD) = rk(EA) = rk(E) = rk(D). Proof. Apply Theorem 4.4 and Theorem 3.4. J. Ben´ıtez, E. Boasso, H. Jin 21 Remark 4.6. Let A ∈ Cn,m and D, E ∈ Cm,n. Note that the condition in Corollary 4.5 does not imply that Ak(D,E) exists. In fact, consider A = (cid:20) 0 1 1 0 (cid:21) , D = E = (cid:20) 1 0 0 0 (cid:21) . Since EAD = 0 and rk(D) = rk(E) = 1, according to Theorem 4.4, Ak(D,E) does not exist. However AD = (cid:20) 0 0 1 0 (cid:21) , EA = (cid:20) 0 1 0 0 (cid:21) , which lead to rk(AD) = 1 and rk(EA) = 1. In the following theorem the (D, E)-inverse will be characterized using full rank factor- izations. Theorem 4.7. Let D, E ∈ Cm,n and consider D = D1D2 and E = E1E2 two full rank factorizations of D and E, respectively. The matrix A ∈ Cn,m is (D, E)-invertible if and only if rk(E2) = rk(D1) = rk(E2AD1). Moreover, in this case, Ak(D,E) = D1(E2AD1)−1E2. Proof. Apply Theorem 4.2 and Corollary 3.23. In the following corollaries two particular cases will be derived from Theoerm 4.7. Corollary 4.8. Let A ∈ Cn,m and D, E ∈ Cm,n be such that exists Ak(D,E) exists. The following statements hold. (i) If the columns of D are linearly independent, then Ak(D,E) = D(AD)−1. (ii) If the rows of E are linearly independent, then Ak(D,E) = (EA)−1E. Proof. If D has full column rank, then D = DIn is a full rank factorization of D. Since Ak(D,E) exists, according to Theorem 4.4, rk(E) = rk(D). Consequently, E has full column rank and E = EIn is a full rank factorization of E. Then, according to Theorem 4.7, Ak(D,E) = D(AD)−1. Apply a similar argument to prove statement (ii), using in particular the full rank factor- izatons E = ImE and D = ImD and Theorem 4.7. Corollary 4.9. Let A ∈ Cn,m and D, E ∈ Cm,n be such that rk(D) = rk(E) = 1. D = d1d∗ E, respectively, then Ak(D,E) exists if and only if e∗ If 2 (d1, e1 ∈ Cm,1, d2, e2 ∈ Cn,1) are full rank factorizations of D and 2 and E = e1e∗ 2Ad1 6= 0. Moreover, in this case, Ak(D,E) = 1 e∗ 2Ad1 d1e∗ 2. Proof. According to Theorem 4.7, the (D, E)-inverse of the matrix A exists if and only if rk(d1) = rk(e∗ 2) = rk(e∗ 2Ad1) = 1. But observe that e∗ 2Ad1 is a complex number. Thus, the first part of the theorem has been proved. The expression of Ak(D,E) also follows from Theorem 4.7. 2Ad1). Therefore, Ak(D,E) exists if and only if rk(e∗ J. Ben´ıtez, E. Boasso, H. Jin 22 Next an application of Theorem 4.7 will lead to representations of several generalized inverses in terms of a full rank representation. The explicit expression for A† is attributed to C.C. MacDufee by Ben-Israel and Greville in [2]. Ben-Israel and Greville report that around 1959, MacDufee was the first to point out that a full-rank factorization of A leads to the mentioned formula. Corollary 4.10. Let A ∈ Cn and consider a full rank factorization A = F G. Then, the following statements hold. (i) A† = G∗(F ∗AG∗)−1F ∗. (ii) A is group invertible if and only if GF is nonsingular; in this case, A# = F (GF )−2G (see [7]). (iii) A is core invertible if and only if GF is nonsingular; in this case, A #(cid:13) = F (GF )−1F †. (iv) A is dual core invertible if and only if GF is nonsingular; in this case, A #(cid:13) = G†(GF )−1G. (v) A† M,N = N −1G∗(F ∗M AN −1G∗)−1F ∗M , where M and N are nonsingular and positive. Proof. Observe that A∗ = G∗F ∗ is a full rank factorization of A∗. Recall that according to [9, p. 1912], A is Moore-Penrose invertible if and only if A is (A∗, A∗)-invertible and in this case A† = Ak(A∗,A∗). Now apply Theorem 4.7 with D = E = A∗ = G∗F ∗. According to [9, p. 1910], A is group invertible if and only if A is (A, A)-invertible and in this case A# = Ak(A,A). According to Theorem 4.7, A is group invertible if and only if rk(G) = rk(F ) = rk(GF GF ). Thus, if r = rk(A), then GF ∈ Cr is invertible. To prove the formula representing A#, apply Theorem 4.7 with D = E = A = F G. Recall that according to [1, p. 684], A is core invertible if and only if it is group invertible. Thus, according to what has been proved, the first part of statement (iii) holds. In addition, according to [18, Theorem 4.4] (i), necessary and sufficient for A to be core invertible is that A is (A, A∗)-invertible and in this case A #(cid:13) = Ak(A,A∗). Now, apply Theorem 4.7 with D = A = F G and E = A∗ = G∗F ∗. Observe however first that if r = rk(A), then F ∗F ∈ Cr is such that rk(F ∗F ) = rk(F ) = r, and recall that (F ∗F )−1F ∗ = F † ([17, Theorem 1]). Then, A #(cid:13) = Ak(A,A∗) = F (F ∗F GF )−1F ∗ = F (GF )−1(F ∗F )−1F ∗ = F (GF )−1F †. Note that A is dual core dual invertible if and only if A∗ is core invertible and A #(cid:13) = [(A∗) #(cid:13)]∗. Thus, the first part of statement (iv) can be derived from what has been proved and A #(cid:13) = [(G∗(F ∗G∗)−1(G∗)†)]∗ = G†(GF )−1G. According to [4, Theorem 3.2], A is weighted Moore-Penrose invertible with weights M and N if and only if A is invertible along N −1A∗M . In addition, according to [9, Defini- tion 6.1], this is equivalent to the fact that A is (N −1A∗M, N −1A∗M )-invertible. Now apply Theorem 4.7 with D = E = N −1A∗M and consider the full rank factorization N −1A∗M = (N −1G∗)(F ∗M ). To end this section, the set of all matrices A ∈ Cn,m such that they are (D, E)-invertible will be studied (D, E ∈ Cm,n). First some notation need to be introduced. J. Ben´ıtez, E. Boasso, H. Jin 23 Given D, E ∈ Cm,n, let C (D, E)-inverse of A exists, i.e., kD,E n,m denote the set of all matrices A ∈ Cn,m such that the kD,E n,m = {A ∈ Cn,m : Ak(D,E) exists }. C Theorem 4.11. Let D, E ∈ Cm,n. Necessary and sufficient for C lef t = (Cn,m)kD,E rk(D). Moreover, in this case, C right . kD,E n,m = (Cn,m)kD,E kD,E n,m 6= ∅ is that rk(E) = Proof. Apply Theorem 4.2 and Corollary 3.12. Observe that Corollary 3.16 gives an explicit representation of C kD,E n,m . 5 Invertible matrices along a fixed matrix Now the case D = E will be considered. Compare with Definition 2.2 and recall the observa- tion before Definition 2.7. Definition 5.1. Let A ∈ Cn,m and D ∈ Cm,n. The matrix A is said to be invertible along the matrix D, if there exists a matrix X ∈ Cm,n such that the following conditions hold. XAD = D = DAX, R(X) ⊆ R(D), N(D) ⊆ N(X). According to Theorem 4.2, if the inverse of the matrix A ∈ Cn,m along the matrix D ∈ Cm,n exists, then it is unique and it will be denoted by AkD. In addition, when A and D are square matrices, since according to [9, Proposition 6.1], the inverse of A along D coincides with the (D, D)-inverse of A, Definition 5.1 reduces to the notion of the inverse along an element in a ring of square matrices ([14, Definition 4], i.e., Definition 2.2). Moreover, under the same conditions as in Definition 5.1, note that R(X) ⊆ R(D) (respectively N(D) ⊆ N(X)) is equivalent to R(X) = R(D) (respectively N(D) = N(X)). Now characterizations of the inverse along a matrix will be given. Theorem 5.2. Let A ∈ Cn,m and D ∈ Cm,n. The following statements are equivalent. (i) The matrix A is left invertible along D. (ii) The matrix A is right invertible along D. (iii) The matrix A is invertible along D. Proof. Apply Theorem 4.2 and Theorem 3.7. In the following remark several results on this inverse that can be derived from what has been proved in sections 2, 3 and 4 will be collected. Remark 5.3. Let D ∈ Cm,n and consider A ∈ Cn,m. (i) The matrix A is invertible along D if and only if A∗ ∈ Cm,n is invertible along D∗ ∈ Cn,m. Moreover, in this case (A∗)kD∗ = (AkD))∗. Apply Remark 4.3 (i) to the case E = D. (ii) Let D′ ∈ Cm,n be such that R(D) = R(D′) and N(D) = N(D′). Necessary and sufficient = AkD. Apply exists. Furthermore, in this case, AkD′ for AkD to exists is that AkD′ Remark 4.3 (ii) to the case E = D. J. Ben´ıtez, E. Boasso, H. Jin 24 (iii) Theorem 3.6 and Theorem 3.7 characterize matrices A such that AkD exists. Compare Theorem 3.7 (v)-(vi) with [14, Theorem 7] and [15, Theorem 2.1]. (iv) When A is invertible along D, AkD can be represented as in Corollary 3.8 and Corollary 3.18. (v) According to Remark 3.17 applied to the case D = E, another representation of AkD, when it exists, is the following. In the decomposition C n = N(D) ⊕ X, it holds that N(AkD) = N(D), AkDy = φ−1(y), y ∈ X, where φ : R(D) → X is the map of Theorem 3.3. (vi) According to Corollary 4.5 applied to the case D = E, if AkD exists, then rk(AD) = rk(DA) = rk(D). Some results, however, deserve to be presented separately. Corollary 5.4. Let A ∈ Cn,m and D ∈ Cm,n. The following statements are equivalent. (i) A is invertible along D. (ii) rk(D) = rk(DAD). Furthermore, in this case AkD = D(DAD)†D. Proof. Apply Theorem 4.4 for the case D = E. Corollary 5.5. Let D ∈ Cm,n and consider D = D1D2 a full rank factorizations of D. The matrix A ∈ Cn,m is invertible along D if and only if rk(D1) = rk(D2AD1). Moreover, in this case, AkD = D1(D2AD1)−1D2. Proof. Apply Theorem 4.7 for the case D = E. Corollary 5.6. Let A ∈ Cn,m and D ∈ Cm,n. (i) Suppose that the columns of D are linearly independent. Necessary and sufficient for AkD to exists is that AD ∈ Cn is nonsingular. Moreover, in this case, AkD = D(AD)−1. (ii) Suppose that the rows of D are linearly independent. The inverse of A along D exists if and only if DA ∈ Cm is invertible. Moreover, in this case, AkD = (DA)−1D. Proof. Suppose that the columns of D are linearly independent. According to Theorem 5.2 and Theorem 3.7 (v), the characterization of the inverse of A along D holds. To prove the formula that represents AkD, apply Corollary 4.8 (i). To prove statement (ii), apply a similar argument to the one used to prove statement (i), using in particular Theorem 3.7 (vi) and Corollary 4.8 (ii). Corollary 5.7. Let A ∈ Cn,m and D ∈ Cm,n. If rk(D) = 1, then AkD exists if and only if tr(AD) 6= 0. Moreover, in this case, AkD = 1 tr(AD) D. J. Ben´ıtez, E. Boasso, H. Jin 25 Proof. Let D = d1d∗ to Corollary 4.9 applied to the case D = E, d∗ 2 (d1 ∈ Cm,1 and d2 ∈ Cn,1) be a full rank factorization of D, According 2Ad1 = tr(d∗ 2Ad1) = tr(Ad1d∗ 2) = tr(AD). Let C kD n,m stand for the set of all matrices A ∈ Cn,m such that A is invertible along D, i.e., kD n,m = {A ∈ Cn,m : AkD exists }. C The following corollary proves that the set under consideration is nonempty. Corollary 5.8. Let D ∈ Cm,n. Then, C kD n,m 6= ∅. Proof. Apply Theorem 5.2, Corollary 3.11 and Corollary 3.18. Observe that Corollary 3.18 gives an explicit representation of C kD n,m. 6 Outer and inner inverses In the following theorem it will be proved that the notion introduced in Definition 4.1 is an outer inverse. Theorem 6.1. Let A ∈ Cn,m and D, E ∈ Cm,n. If A is (D, E)-invertible, then Ak(D,E) is an outer inverse of A. Proof. According to Definition 4.1, since R(Ak(D,E)) ⊆ R(D), there exists M ∈ Cn such that Ak(D,E) = DM . Thus, Ak(D,E)AAk(D,E) = Ak(D,E)ADM = DM = Ak(D,E). Corollary 6.2. Let A ∈ Cn,m and D ∈ Cm,n. If A is invertible along D, then AkD is an outer inverse of A. Proof. Apply Theorem 6.1 for the case D = E. Given A ∈ Cn,m and D, E ∈ Cm,n such that Ak(D,E) exists, according to Corollary 3.18 and Theorem 4.4, rk(Ak(D,E)) = rk(D) = rk(E) ≤ rk(A). In particular, when A, D and E are square matrices, Ak(D,E) is nonsingular if and only if D or E are nonsingular. In addition, if Ak(D,E) is nonsingular, then A is nonsingular. However, if A is nonsingular and D, E are such that Ak(D,E) exists, then it may be happen that D, E, or Ak(D,E) are singular. For example, take A = I2, D = E = (cid:20) 1 0 0 0 (cid:21) . According to Theorem 4.4, Ak(D,E) exists (Ak(D,E) = D). However, it is possible to charac- terize when rk(A) = rk(D). This characterization is linked with the following observation: Ak(D,E) is always an outer inverse of A (Theorem 6.1), but it is not necessarily an inner inverse of A. To prove this characterization some preparation is needed first. Theorem 6.3. Let A ∈ Cn,m and D, E ∈ Cm,n be such that Ak(D,E) exists. Then, the following statements hold. (i) R(D) ⊕ N(A) = N(AAk(D,E)A − A). J. Ben´ıtez, E. Boasso, H. Jin 26 (ii) rk(A) = rk(D) + rk(AAk(D,E)A − A). (iii) R(A) + N(E) = C n and R(A) ∩ N(E) = R(AAk(D,E)A − A). Proof. The inclusion N(A) ⊆ N(AAk(D,E)A − A) is evident. The equality Ak(D,E)AD = D leads to R(D) ⊆ N(AAk(D,E)A − A). If x ∈ R(D) ∩ N(A), then there exists u ∈ C n such that x = Du, and therefore, x = Du = Ak(D,E)ADu = Ak(D,E)Ax = 0. Thus, R(D) ⊕ N(A) ⊆ N(AAk(D,E)A − A). To prove the opposite inclusion, take y ∈ N(AAk(D,E)A−A). Now, A(Ak(D,E)Ay − y) = 0 and the decomposition y = (y−Ak(D,E)A(y))+Ak(D,E)A(y) proves that N(AAk(D,E)A−A) ⊆ R(D) ⊕ N(A) (because R(Ak(D,E)) = R(D)). Statement (ii) follows from statement (i). According to Remark 4.3 (i), (A∗)k(E∗,D∗) exists, so that, according to statement (i) applied to A∗, D∗ and E∗, N(A∗) ∩ R(E∗) = 0. Then, [R(A) + N(E)]⊥ = R(A)⊥ ∩ N(E)⊥ = N(A∗) ∩ R(E∗) = 0. Therefore, R(A) + N(E) = C EAAk(D,E) = E. According to statement (ii), Theorem 4.4 and n. The inclusion R(AAk(D,E)A − A) ⊆ R(A) ∩ N(E) follows from dim(R(A) ∩ N(E)) = dim R(A) + dim N(E) − dim(R(A) + N(E)) = rk(A) + n − rk(D) − n = rk(AAk(D,E)A − A), R(AAk(D,E)A − A) and R(A) ∩ N(E) have the same dimension. Therefore, both subspaces are equal. The following corollary characterizes when Ak(D,E) is an inner inverse of A. Corollary 6.4. Let A ∈ Cn,m and D, E ∈ Cm,n be such that Ak(D,E) exists. Then, the following statements are equivalent. (i) AAk(D,E)A = A. (ii) rk(A) = rk(D). (iii) R(D) ⊕ N(A) = C m. (iv) R(A) ⊕ N(E) = C n. Proof. Apply Theorem 6.3. In the next corollary the case when the inverse along an element is an inner inverse will be presented. Corollary 6.5. Let A ∈ Cn,m and D ∈ Cm,n be such that AkD exists. Then, the following statements are equivalent. (i) AAkDA = A. (ii) rk(A) = rk(D). (iii) R(D) ⊕ N(A) = C m. J. Ben´ıtez, E. Boasso, H. Jin 27 (iv) R(A) ⊕ N(D) = C n. Proof. Apply Corollary 6.4 to the case D = E. Let A ∈ Cn,m and D, E ∈ Cm,n. Since Ak(D,E) is an outer inverse, if it exists, Ak(D,E)A and AAk(D,E) are idempotents. Now some properties of these idempotents will be studied. Theorem 6.6. Let A ∈ Cn,m and D, E ∈ Cm,n be such that Ak(D,E) exists. (i) Ak(D,E)A and AAk(D,E) are idempotents, R(Ak(D,E)A) = R(D), R(AAk(D,E)) = R(AD) and rk(D) = rk(AAk(D,E)) = rk(Ak(D,E)A) = rk(E). (ii) N(AAk(D,E)) = N(E) and N(Ak(D,E)A) = N(EA). (iii) N(Ak(D,E)A) = N(A) if and only if Ak(D,E) is an inner inverse of A. (iv) R(AAk(D,E)) = R(A) if and only if Ak(D,E) is an inner inverse of A. (v) AAk(D,E) is an orthogonal projector if and only if R(AD) = R(E∗). (vi) Ak(D,E)A is an orthogonal projector if and only if R((EA)∗) = R(D). Proof. Since Ak(D,E) is an outer inverse (Theorem 6.1), Ak(D,E)A and AAk(D,E)are idem- potents and R(Ak(D,E)A) = R(Ak(D,E)). Moreover, since R(Ak(D,E)) = R(D), according to Theorem 4.4, rk(E) = rk(D) = rk(Ak(D,E)) = rk(Ak(D,E)A). In addition, according to Remark 4.3 (i) and what has been proved, rk(E) = rk(E∗) = rk((Ak(D,E))∗A∗) = rk(AAk(D,E)). Moreover, R(AD) = R(AAk(D,E)AD) ⊆ R(AAk(D,E)). However, since according to Corollary 4.5, rk(AD) = rk(D) = rk(AAk(D,E)), R(AAk(D,E)) = R(AD). Since Ak(D,E) is an outer inverse, N(AAk(D,E)) = N(Ak(D,E)) = N(E). Note that since EA = EAAk(D,E)A, N(Ak(D,E)A) ⊆ N(EA). However, since according to Corollary 4.5, rk(EA) = rk(D) = rk(Ak(D,E)A), dim N(Ak(D,E)A) = dim N(EA). Therefore, N(Ak(D,E)A) = N(EA). Naturally, N(A) ⊆ N(Ak(D,E)A). Since dim N(Ak(D,E)A) = m−rk(Ak(D,E)A) = m−rk(D) and dim N(A) = m − rk(A), N(A) = N(Ak(D,E)A) if and only if rk(A) = rk(D). Now apply Corollary 6.4. Note that Ak(D,E) is an inner inverse of A if and only if (Ak(D,E))∗ is an inner inverse of A∗. Now, according to statement (iii), this is equivalent to N((Ak(D,E))∗A∗) = N(A∗), which in turn is equivalent to R(AAk(D,E)) = R(A). Observe that AAk(D,E) is an orthogonal projector if and only if R(AD) = R(AAk(D,E)) = N(AAk(D,E))⊥ = N(E)⊥ = R(E∗). The proof of statement (vi) follows Remark 4.3 (i) and statement (v). In the next corollary the case of the inverse along a fixed matrix will be studied. Corollary 6.7. Let A ∈ Cn,m and D ∈ Cm,n be such that AkD exists. J. Ben´ıtez, E. Boasso, H. Jin 28 (i) AkDA and AAkD are idempotents, R(AkDA) = R(D), R(AAkD) = R(AD) and rk(D) = rk(AAkD) = rk(AkDA). (ii) N(AAkD) = N(D) and N(AkDA) = N(DA). (iii) N(AkDA) = N(A) if and only if AkD is an inner inverse of A. (iv) R(AAkD) = R(A) if and only if AkD is an inner inverse of A. (v) AAkD is an orthogonal projector if and only if R(AD) = R(D∗). (vi) AkDA is an orthogonal projector if and only if R((DA)∗) = R(D). Proof. Apply Theorem 6.6 to the case D = E. Given A ∈ Cn such that A is group invetible, according to [1] and [18], AA #(cid:13) and A #(cid:13)A are orthogonal projectors. In the next corollaries similar properties for several generalized inverses will be characterized using Corollary 6.7. Note that the following identities hold: R(XY ) = R(X) and R(XX ∗) = R(X), where Y is a nonsingular matrix and X is any matrix. In addition, recall that a matrix A ∈ Cn is said to be EP , if AA† = A†A. It is well known that this condition is equivalent to R(A) = R(A∗). Corollary 6.8. Consider A ∈ Cn a group invertible matrix. The following statements holds. (i) A #(cid:13)A is an orthogonal projector. (ii) AA #(cid:13) is an orthogonal projector. (iii) A is EP. Proof. According to Theorem 2.4 (i) and Theorem 6.7 (vi), A #(cid:13)A is an orthogonal projec- tor if and only if R((AA∗A)∗) = R(AA∗). However, R(AA∗) = R(A) and R((AA∗A)∗) = R(A∗AA∗) = R(A∗A) = R(A∗). Similarly, according to Theorem 2.4 (ii) and Theorem 6.7 (v), AA #(cid:13) is an orthogonal projector if and only R(AA∗A) = R(A∗A). However, R(A∗A) = R(A∗) and R(AA∗A) = R(AA∗) = R(A). Corollary 6.9. Let A ∈ Cn and consider M , N ∈ Cn nonsingular and positive. The following statements hold. (i) AA† M,N is an orthogonal projector if and only if R(A) = R(M A). In particular, if M = In, then AA† M,N is an orthogonal projector. (ii) A† M,N A is an orthogonal projector if and only if R(A∗) = R(N −1A∗). In particular, if N = In, then A† M,N A is an orthogonal projector. Proof. According to Theorem 2.4 (iii) and Theorem 6.7 (v), AA† M,N is an orthogonal pro- jector if and only if R(AN −1A∗M ) = R((N −1A∗M )∗). This last condition is equivalent to R(AN −1A∗) = R(M A). Since N is nonsingular and positive, there exists a Hermitian and nonsingular matrix Q such that N −1 = Q2. Define R = AQ. Then, AN −1A∗ = RR∗, and thus, R(AN −1A∗) = R(RR∗) = R(R) = R(AQ) = R(A). J. Ben´ıtez, E. Boasso, H. Jin 29 Similarly, according to Theorem 2.4 (iii) and Theorem 6.7 (vi), A† M,N A is an orthogonal projector if and only if R((N −1A∗M A)∗) = R(N −1A∗M ). This identity is equivalent to R(A∗M A) = R(N −1A∗). Since M is nonsingular and positive, there exists a Hermitian and nonsingular matrix P such that M = P 2. Then A∗M A = A∗P ∗P A = (P A)∗P A, and thus, R(A∗M A) = R((P A)∗P A) = R((P A)∗) = R(A∗P ∗) = R(A∗). 7 The outer inverse with prescribed range and null space In first place the relationship between the outer inverse with prescribed range and null space and the (D, E)-inverse will be considered (D, E ∈ Cm,n). Theorem 7.1. Let A ∈ Cn,m and D, E ∈ Cm,n. The following statements are equivalent. (i) The matrix A is (D, E)-invertible. (ii) The outer inverse A(2) R(D),N(E) exists. Furthermore, in this case Ak(D,E) = A(2) R(D),N(E). Proof. Suppose that statement (i) holds and let H = Ak(D,E). Then, according to Theo- rem 6.1, H = HAH. In addition, according to Definition 4.1, N(H) = N(E) and R(H) = R(D). In particular, A(2) R(D),N(E) exists and A(2) R(D),N(E) = H. Now suppose that statement (ii) holds and let L = A(2) R(D),N(E). In particular, R(L) ⊆ R(D) and N(E) ⊆ N(L). Since L is an outer inverse of A and R(L) = R(D), it is not difficult to n, EAL = E. prove that LAD = D. In addition, since ALx − x ∈ N(L) = N(E), for all x ∈ C Therefore, L = Ak(D,E). Corollary 7.2. Let A ∈ Cn,m and D ∈ Cm,n. The following statements are equivalent. (i) The matrix A is invertible along D. (ii) The outer inverse A(2) R(D),N(D) exists. Furthermore, in this case AkD = A(2) R(D),N(D). Proof. Apply Theorem 7.1 to the case D = E. Due to Theorem 7.1, the properties of the outer inverse with prescribed range and null space can be easily proved for the (D, E)-inverse (D, E ∈ Cm,n). Here only some of the most well known result are considered. Other results and the case of the inverse along a fixed matrix, i.e. when D = E, are left to the reader. Corollary 7.3. Let A ∈ Cn,m and D, E ∈ Cm,n. Ak(D,E) is the unique matrix X that satifies the following equations: XAX = X, AX = PR(AD),N(E), XA = PR(D),N(EA). J. Ben´ıtez, E. Boasso, H. Jin 30 Proof. Apply Theorem 7.1 and [23, Theorem 1]. Note that, if T = R(D) and S = N(E), then A(T) = R(AD) and (A∗(S⊥))⊥ = N(EA). Corollary 7.4. Let A ∈ Cn,m and D, E ∈ Cm,n. Suppose that there exists G ∈ Cm,n such that R(G) = R(D) and N(G) = N(E). If Ak(D,E) exists, then AG ∈ Cm and GA ∈ Cn are group invertible and Ak(D,E) = G(AG)# = (GA)#G = [GA R(G)]−1G. Proof. Apply Theorem 7.1, [20, Theorem 2.1] and [20, Theorem 2.3]. Corollary 7.5. Let A ∈ Cn,m and D, E ∈ Cm,n. Suppose that there exists G ∈ Cm,n such that R(G) = R(D) and N(G) = N(E). If Ak(D,E) exists, then Ak(D,E) = lim ǫ→0 (GA − ǫIm)−1G = lim ǫ→0 G(AG − ǫIn)−1. Proof. Apply Theorem 7.1 and [20, Theorem 2.4]. Corollary 7.6. Let A ∈ Cn,m and D, E ∈ Cm,n. Suppose that there exists G ∈ Cm,n such that R(G) = R(D) and N(G) = N(E). If Ak(D,E) exists, then Ak(D,E) = Z ∞ 0 exp[−G(GAG)∗GAt]G(GAG)∗Gdt. Proof. Apply Theorem 7.1 and [22, Theorem 2.2]. Corollary 7.7. Let A ∈ Cn,m and D, E ∈ Cm,n. Suppose that there exists G ∈ Cm,n such that R(G) = R(D) and N(G) = N(E). If Ak(D,E) exists and the nonzero spectrum of GA lies in the open left half plane, then Ak(D,E) = −Z ∞ 0 exp(GAt)Gdt. Proof. Apply Theorem 7.1 and [21]. 8 Continuity and differentiability First of all note that if A ∈ Cn,m and D, E ∈ Cm,n are such that Ak(D,E) exists and D′, E′ ∈ Cn,m are such that D = DD′D and E = EE′E, then according to Definition 4.1, DD′Ak(D,E) = Ak(D,E) (because R(Ak(D,E)) ⊆ R(D)) and Ak(D,E) = Ak(D,E)E′E. The last idendity can be easily derived from the fact that there exists a matrix N ∈ Cm,m such that Ak(D,E) = N E (because N(E) ⊆ N(Ak(D,E))). In order to characterize the continuity of the (D, E)-inverse, a technical lemma is needed. Lemma 8.1. Let A, B ∈ Cn,m and D, E, F , G ∈ Cm,n be such that Ak(D,E) and Bk(F,G) exist. Let D′, E′, F ′ and G′ ∈ Cn,m be such that D = DD′D, E = EE′E, F = F F ′F and G = GG′G. Then Bk(F,G) − Ak(D,E) = Bk(F,G)(G′G − E′E)(In − AAk(D,E)) + Bk(F,G)(A − B)Ak(D,E) + (Im − Bk(F,G)B)(F F ′ − DD′)Ak(D,E). J. Ben´ıtez, E. Boasso, H. Jin 31 Proof. Since DD′Ak(D,E) = Ak(D,E) and Bk(F,G)BF = F , then Bk(F,G)BAk(D,E) − Ak(D,E) = −(Im − Bk(F,G)B)DD′Ak(D,E) = [(Im − Bk(F,G)B)F F ′ − (Im − Bk(F,G)B)DD′]Ak(D,E) = (Im − Bk(F,G)B)(F F ′ − DD′)Ak(D,E). In addition, since Bk(F,G) = Bk(F,G)G′G and EAAk(D,E) = E, Bk(F,G) − Bk(F,G)AAk(D,E) = Bk(F,G)G′G(In − AAk(D,E)) = Bk(F,G)[G′G(In − AAk(D,E)) − E′E(In − AAk(D,E))] = Bk(F,G)(G′G − E′E)(In − AAk(D,E)). Thus, Bk(F,G) − Ak(D,E) = Bk(F,G)(G′G − E′E)(In − AAk(D,E)) + Bk(F,G)AAk(D,E) + (Im − Bk(F,G)B)(F F ′ − DD′)Ak(D,E) − Bk(F,G)BAk(D,E) = Bk(F,G)(G′G − E′E)(In − AAk(D,E)) + Bk(F,G)(A − B)Ak(D,E) + (Im − Bk(F,G)B)(F F ′ − DD′)Ak(D,E). Next a result regarding the continuity of the (D, E)-inverse will be presented. Theorem 8.2. Let A ∈ Cn,m and D, E ∈ Cm,n be such that Ak(D,E) exists and consider (Ak)k∈ ⊂ Cn,m and (Dk)n∈, (Ek)n∈ ⊂ Cm,n such that Ak(Dn,En) exists for each k ∈ . Let D′, E′ ∈ Cn,m and (D′ n)n∈ ⊂ Cn,m be such that D = DD′D, E = EE′E, Dk = DkD′ k)k∈ and (E′ kEk)k∈ converge to A, DD′ and E′E, respectively. Then, the following statememts are equivalent. kEk, for each k ∈ . Suppose that (Ak)k∈, (DkD′ kDk and Ek = EkE′ k)n∈, (E′ k (i) (Ak(Dk,Ek) k )k∈ converges to Ak(D,E). (ii) The sequence (Ak(Dk,Ek) k )k∈ ⊂ Cn,m is bounded. Proof. Apply Lemma 8.1. If the Moore-Penrose inverse is used, then a more general result can be presented. Theorem 8.3. Let A ∈ Cn,m and D, E ∈ Cm,n be such that Ak(D,E) exists and consider (Ak)k∈ ⊂ Cn,m and (Dk)k∈, (Ek)n∈ ⊂ Cm,n such that Ak(Dk,Ek) exists for each k ∈ . Suppose that (Ak)k∈ converges to A. Then, the following statememts are equivalent. k (i) (Ak(Dk,Ek) k )k∈ converges to Ak(D,E). (ii) The sequences (D† quence (Ak(Dk,Ek) k)k∈ and (E† )k∈ ⊂ Cn,m is bounded. k k)k∈ converge to D† and E†, respectively, and the se- J. Ben´ıtez, E. Boasso, H. Jin 32 Proof. Suppose that (Ak(Dk,Ek) to Ak(D,E)A. Consequently, k )k∈ converges to Ak(D,E). Then, (Ak(Dk,Ek) k Ak)k∈ converges lim k→∞ tr(Ak(Dk,Ek) k Ak) = tr(Ak(D,E)A). Since Ak(D,E) is an outer inverse (Theorem 6.1), Ak(D,E)A is an idempotent. Thus, tr(Ak(D,E)A) = rk(Ak(D,E)A) = rk(Ak(D,E)). A) = rk(Ak(D,E) k ), for k ∈ . As a result, for sufficiently large k ∈ , ) = rk(Ak(D,E)). However, according to Theorem 4.4 Similarly, tr(Ak(D,E) rk(Ak(D,E) k k rk(Dk) = rk(Ek) = rk(Ak(D,E) k ) = rk(Ak(D,E)) = rk(D) = rk(E). Therefore, according to [19], (D† part of statement (ii) is evident. k)k∈ converges to D† and (E† k)k∈ to E†. The remaining If statement (ii) holds, then apply Theorem 8.2 with D′ = D† and E′ = E†, k ∈ . In the following corollary, the case of the inverse along a matrix will be considered. Theorem 8.4. Let A ∈ Cn,m and D ∈ Cm,n be such that AkD exists and consider (Ak)k∈ ⊂ Cn,m and (Dk)k∈ ⊂ Cm,n such that AkDk exists for each k ∈ . Suppose that (Ak)k∈ converges to A. Then, the following statememts are equivalent. k (i) (AkDk k )k∈ converges to AkD. (ii) The sequences (D† k)k∈ converges to D† and the sequence (AkDk k )k∈ ⊂ Cn,m is bounded. Proof. Apply Theorem 8.3 for the case D = E. Now the differentiability will be studied. Theorem 8.5. Let J ⊆ R be an open set and consider t0 ∈ J . Let functions A : J → Cn,m and D, E : J → Cm,n be such that A(t) is (D(t), E(t))-invertible, for any t ∈ J , and A, D and E are differentiable at t0. Suppose that f : J → Cm,n, f (t) = A(t)k(D(t),E(t)) , is a bounded function in J and that the functions D, E have local constant rank in J . Then, the function f is differentiable at t0 and f ′(t0) = A(t0)k(D(t0),E(t0))(cid:2)G′(t0)E(t0) + G(t0)E′(t0)(cid:3)hIn − A(t0)A(t0)k(D(t0),E(t0))i +hIn − A(t0)k(D(t0),E(t0))A(t0)i(cid:2)D′(t0)F(t0) + D(t0)F′(t0)(cid:3) A(t0)k(D(t0),E(t0)) + A(t0)k(D(t0),E(t0))A′(t0)A(t0)k(D(t0),E(t0)), where F, G : J → Cn,m are the functions F(t) = (D(t))† and G(t) = (E(t))†. Proof. Observe that according to Lemma 8.1, for any t ∈ J, f (t) − f (t0) = A(t)k(D(t),E(t)) hE(t)†E(t) − E(t0)†E(t0)ihIn − A(t0)A(t0)k(D(t0),E(t0))i +hIn − A(t)k(D(t),E(t)) A(t)ihD(t)D(t)† − D(t0)D(t0)†i A(t0)k(D(t0),E(t0)) + A(t)k(D(t),E(t)) [A(t0) − A(t)] A(t0)k(D(t0),E(t0)). J. Ben´ıtez, E. Boasso, H. Jin 33 Now, according to [19], the functions F, G : J → Cn,m, F(t) = (D(t))† and G(t) = (E(t))† are continuous. Consequently, according to Theorem 8.4, lim t→t0 A(t)k(D(t),E(t)) [A(t0) − A(t)] t − t0 A(t0)k(D(t0),E(t0)) = A(t0)k(D(t0),E(t0))A′(t0)A(t0)k(D(t0),E(t0)). In addition, according to [11], the functions F, G : J → Cn,m are also differentiable. Thus lim t→t0 A(t)k(D(t),E(t))(cid:2)E(t)†E(t) − E(t0)†E(t0)(cid:3) hIn − A(t0)A(t0)k(D(t0),E(t0))i = A(t0)k(D(t0),E(t0))(cid:2)G′(t0)E(t0) + G(t0)E′(t0)(cid:3)hIn − A(t0)A(t0)k(D(t0),E(t0))i . t − t0 Similarly, lim A(t0)k(D(t0),E(t0)) = t→t0hIn − A(t)k(D(t),E(t)) A(t)i (cid:2)D(t)D(t)† − D(t0)D(t0)†(cid:3) t − t0 hIn − A(t0)k(D(t0),E(t0))A(t0)i(cid:2)D′(t0)F(t0) + D(t0)F′(t0)(cid:3) A(t0)k(D(t0),E(t0)). Now the differentiability of the inverse along a matrix will be studied. Corollary 8.6. Let J ⊆ R be an open set and consider t0 ∈ J . Let functions A : J → Cn,m and D : J → Cm,n be such that A(t) is invertible along D(t) for any t ∈ J , and A and D are differentiable at t0. Suppose that f : J → Cm,n, f (t) = A(t)kD(t), is a bounded function in J and that the function D has local constant rank in J . Then, the function f is differentiable at t0 and f ′(t0) = A(t0)kD(t0)(cid:2)F′(t0)D(t0) + F(t0)D′(t0)(cid:3)hIn − A(t0)A(t0)kD(t0)i +hIn − A(t0)kD(t0)A(t0)i(cid:2)D′(t0)F(t0) + D(t0)F′(t0)(cid:3) A(t0)k(D(t0) + A(t0)kD(t0)A′(t0)A(t0)kD(t0), where F : J → Cn,m is the function F(t) = (D(t))†. Proof. Apply Theorem 8.5 for the case D = E. 9 Explicit computations In this section some explicit ways to compute Ak(D,E) will be given. Theorem 9.1. Let A ∈ Cn,m, D, E ∈ Cm,n, r = rk(D) and s = rk(E). If {v1, . . . , vr} is a basis of R(D) and {w1, . . . , wn−s} is a basis of N(E), then the following affirmations are equivalent: (i) Ak(D,E) exists. (ii) The matrix [Av1 · · · Avr w1 · · · wn−s] is nonsingular. J. Ben´ıtez, E. Boasso, H. Jin 34 In this situation, Ak(D,E) = [v1 · · · vr 0 · · · 0] [Av1 · · · Avr w1 · · · wn−s]−1 . Proof. If statement (i) holds, then according to Theorem 4.4, rk(D) = rk(E). Let X1 = [Av1 · · · Avr] and X2 = [w1 · · · wn−r]. Observe that n − r = rk(X2) because {wi}n−r i=1 is a basis. According to Theorem 4.2 and Theorem 3.4, rk(X) = rk(X1) + rk(X2) (because R(AD) ∩ N(E) = 0). Since {Avi}r i=1 span R(AD) and r = rk(D) = rk(AD) = dim R(AD), the vectors {Avi}r i=1 are linearly independent, and thus, r = rk(X1). Therefore, n = rk(X) and by recalling that X ∈ Cn, the nonsingularity of X is obtained. Suppose that statement (ii) holds. Since the matrix in statement (ii) must be square, rk(D) = r = s = rk(E). In addition, since the matrix in statement (ii) is invertible, rk(AD) = n. Consequently, according to Theorem 3.4 and Theorem 4.2, rk(D) and R(AD) ⊕ N(E) = C Ak(D,E) exists. Now let v be any arbitrary vector in R(D) and let x ∈ C n be such that v = Dx. According to Definition 4.1, Ak(D,E)Av = Ak(D,E)ADx = Dx = v. In addition, Ak(D,E)w = 0 for any w ∈ N(E). Therefore, Ak(D,E) [Av1 · · · Avr w1 · · · wn−r] = [v1 · · · vr 0 · · · 0] . Next m-file that can be executed in Matlab or in Octave shows how Theorem 9.1 can be used to compute Ak(D,E). function J = pseudo(A,D,E) [n m] = size(A); r = rank(D); s = rank(E); E1 = null(E); % An orthonormal basis of N(E) D1 = orth(D); % An orthonormal basis of R(D) aux = [AD1 E1]; if not(r==s) disp('There does not exist the pseudoinverse') disp('because rank(D) is not equal to rank(E)') else if det(aux)==0 disp('There does not exist the pseudoinverse') disp('because the matrix of Th. 4.1 is singular') else J=[D1 zeros(n,n-r)]inv(aux); end end Theorem 9.2. Let A ∈ Cn,m and D, E ∈ Cm,n be such that Ak(D,E) exists. Let r = rk(D) = rk(E) = rk(EAD). Let P ∈ Cm and Q ∈ Cn be two nonsingular matrices such that P EADQ = (cid:20) Ir 0 0 (cid:21) . 0 J. Ben´ıtez, E. Boasso, H. Jin Then P E = (cid:20)X 0(cid:21) and DQ = (cid:2)Y 0(cid:3) , where X ∈ Cr,n, Y ∈ Cm,r. Furthermore, Ak(D,E) = Y X. Proof. Write P and Q as P2(cid:21) P = (cid:20)P1 and Q = (cid:2)Q1 Q2(cid:3) , where P1 ∈ Cr,m, P2 ∈ Cm−r,m, Q1 ∈ Cn,r and Q2 ∈ Cn,n−r. Now (cid:20) Ir 0 0 (cid:21) = P EADQ = (cid:20)P1 0 P2EADQ1 P2EADQ2 (cid:21) , P2(cid:21) EAD(cid:2)Q1 Q2(cid:3) = (cid:20) P1EADQ1 P1EADQ2 35 (7) (8) which implies P1EADQ2 = 0, P2EADQ1 = 0 and P2EADQ2 = 0. Therefore, P2EADQ = P2EAD[Q1 Q2] = [P2EADQ1 P2EADQ2] = 0. The nonsingularity of Q leads to P2EAD = 0. In a similar way, EADQ2 = 0. Since rk(D) = rk(E) = rk(EAD), the equalities R(EAD) = R(E) and N(EAD) = N(D) In addition, since EADQ2 = 0 and N(EAD) = N(D), it can be deduced 2 belongs to N((EAD)∗) = 2 = 0, i.e., P2E = 0. If Y = DQ1 and are obtained. DQ2 = 0. Since (EAD)∗P ∗ R(EAD)⊥ = R(E)⊥ = N(E∗), and therefore, E∗P ∗ X = P1E, then (7) holds. 2 = (P2EAD)∗ = 0, any column of P ∗ Now it will be proved that Y X satisfies Definition 4.1. First, observe that XAY = P1EADQ1 = Ir. Now, by (7) Y XADQ = Y XA[Y 0] = [Y XAY 0] = [Y 0] = DQ, and the nonsingularity of Q leads to Y XAD = D. Similarly, P EAY X = (cid:20)X 0(cid:21) AY X = (cid:20)XAY X 0 (cid:21) = (cid:20)X 0(cid:21) = P E, and thus, EAY X = E. Since Y X = DQ1X, te inclusion R(Y X) ⊆ R(D) can be obtained. In addition, since Y X = Y P1E, it can be deduced N(E) ⊆ N(Y X). Remark 9.3. Observe that it is possible to use either the Gaussian elimination method or the singular value decomposition of EAD to determine P and Q. Let r = rk(EAD). (i) By using the Gauss-Jordan elimination, there exist an elementary row operation matrix P ∈ Cm,m and an elementary column operation matrix Q ∈ Cn,n, such that P EADQ = (cid:20) Ir 0 0 (cid:21). 0 (ii) Let EAD = U SV ∗ be the singular value decomposition of EAD, where S = Σ ⊕ 0, Σ = diag(σ1, . . . , σr). Hence, U ∗EADV = Σ ⊕ 0, which implies (Σ−1/2 ⊕ Im−r)U ∗EADV (Σ−1/2 ⊕ In−r) = (cid:20) Ir 0 0 (cid:21) . 0 Let P = (Σ−1/2 ⊕ Im−r)U ∗ and Q = V (Σ−1/2 ⊕ In−r). It is easy to see that P and Q are nonsingular. J. Ben´ıtez, E. Boasso, H. Jin 36 Theorem 9.2 and Remark 9.3 (i) yield an elimination method to compute Ak(D,E), which is presented as follows. Algorithm 9.1: Compute the (D, E)-inverse. Input: A ∈ Cn,m, D, E ∈ Cm,n with rk(D) = rk(E) = rk(EAD) = r. Output: Ak(D,E). 1. Execute elementary row operations on the first m rows of the block matrix G = (cid:20)EAD E 0(cid:21) D to get (cid:20)W 0 (cid:21) (cid:20)X 0(cid:21) G1 =   D 0 .   2. Execute elementary column operations on the first m columns of the block matrix G1 to get (cid:20)Ir 0 G2 =   (cid:2)Y 0(cid:3) 0 0(cid:21) (cid:20)X 0(cid:21) 0 .   3. Ak(D,E) = Y X. Theorem 9.2 and Remark 9.3 (ii) yield a more stable numerical method based on the SVD to compute Ak(D,E), which is shown as follows. Algorithm 9.2: Compute the (D, E)-inverse. Input: A ∈ Cn,m, D, E ∈ Cm,n with rk(D) = rk(E) = rk(EAD) = r. Output: Ak(D,E). 1. Compute the SVD of EAD, i.e., EAD = U SV ∗. 2. T = S(1 : r, 1 : r), M = T −1/2 ⊕ Im−r, N = T −1/2 ⊕ In−r. 3. P = M U ∗, Q = V N . 4. X = P E, Y = DQ. 5. Ak(D,E) = Y (1 : m, 1 : r) · X(1 : r, 1 : n). Next m-file shows how Theorem 9.2 and the SVD can be used to compute ADE. function J = pseudo(A,D,E) [n m] = size(A); r = rank(D); s = rank(E); t = rank(EAD); if not(r==s) disp('There does not exist the pseudoinverse') disp('because rank(D) is not equal to rank(E)') J. Ben´ıtez, E. Boasso, H. Jin 37 else if not(s==t) disp('There does not exist the pseudoinverse') disp('because rank(D)=rank(E) but not equal to rank(EAD)') else [U S V] = svd(EAD); T = S(1:r,1:r) M = [T^(-1/2) zeros(r,m-r); zeros(m-r,r) eye(m-r)]; N = [T^(-1/2) zeros(r,n-r); zeros(n-r,r) eye(n-r)]; P = MU'; Q = VN; X = PE; Y = DQ; J = Y(1:m,1:r)*X(1:r,1:n); end end References [1] O. M. Baksalary, G. Trenkler, Core inverse of matrices, Linear Multilinear Algebra 58 (2010) 681-697. [2] A. Ben-Israel, T.N.E. Greville, Generalized inverses, Theory and Applications, Springer, 2003. [3] J. Ben´ıtez, E. Boasso, The inverse along an element in rings, Electron. J. Linear Algebra 31 (2016) 572-592. [4] J. Ben´ıtez, E. Boasso, The inverse along an element in rings with an involution, Banach algebras and C ∗-algebras, Linear Multilinear Algebra 65 (2017) 284-299. [5] E. Boasso, G. Kantun-Montiel, The (b, c)-inverse in rings and in the Banach context, submitted, arXiv:1607.02456. [6] N. Castro-Gonz´alez, J. Chen, L. Wang, Further results on generalized inverses in rings with involution, Electron. J. Linear Algebra 30 (2015) 118-134. [7] R.E. Cline, Inverses of rank invariant powers of a matrix, SIAM J. Numer. Anal. 5 (1968) 182-197. [8] M. P. Drazin, Pseudo-Inverses in Associative Rings and Semigroups, Amer. Math. Monthly 65 (1958) 506-514. [9] M. P. Drazin, A class of outer generalized inverses, Linear Algebra Appl. 436 (2012) 1909-1923. [10] M. P. Drazin, Commuting properties of generalized inverses, Linear Multilinear Algebra 61 (2013) 1675-1681. J. Ben´ıtez, E. Boasso, H. Jin 38 [11] G. H. Golub, V. Pereyra, The differentiation of pseudo-inverses and nonlinear least squares problems whose variables separate, SIAM J. Numer. Anal. 10 (1973) 413-432. [12] Y. Ke, J. Chen, The Bott-Duffin (e, f )-inverses and their applications, Linear Algebra Appl. 489 (2016) 61-74. [13] Y. Ke, J. Visnji´c, J. Chen, One sided-inverses in rings, arxiv: 1607.06230v1. [14] X. Mary, On generalized inverses and Green's relations, Linear Algebra Appl. 434 (2011) 1836-1844. [15] X. Mary, P. Patricio, Generalized inverses modulo H in semigroups and rings, Linear Multilinar Algebra 61 (2013) 1130-1135. [16] R. Penrose, A generalized inverse for matrices, Mathematical Proceedings of the Cam- bridge Philosophical Society 3 (1955) 406-413. [17] R. Piziak, P. L. Odell, Full Rank Factorization of Matrices, Math. Mag. 72 (1999) 193- 201. [18] D. S. Raki´c, N. C. Dinci´c, D.S. Djordjevi´c, Group, Moore-Penrose, core and dual core inverse in rings with involution, Linear Algebra Appl. 463 (2014) 115-133. [19] G. W. Stewart, On the continuity of the generalized inverse, SIAM J. Appl. Math. 17 (1969) 33-45. [20] Y. Wei, A characterization and representation of the generalized inverse A(2) T,S and its application, Linear Algebra Appl. 280 (1998) 87-96. [21] Y. Wei, Integral representation of the generalized inverse A(2) T,S and its applications, in Recent Research on Pure and Applied Algebra, Nova Science, Hauppauge, NY, 2003, pp. 59-65. [22] Y. Wei, D. S. Djordjevi´c, On the integral representation of the generalized inverse A(2) T,S, Appl. Math. Comp. 142 (2003), 189-194. [23] Y. Wei, G. Wang, On the continuity of the generalized inverse A(2) T,S, Appl. Math. Comp. 136 (2003), 289-295. [24] H.H. Zhu, J.L. Chen, P. Patr´ıcio, Further results on the inverse along an element in semigroups and rings, Linear Multilinear Algebra 64 (2016) 393-403. Julio Ben´ıtez E-mail address: [email protected] Enrico Boasso E-mail address: enrico [email protected] Hongwei Jin E-mail address: [email protected]
1501.02964	1	1501	2015-01-13T11:29:30	A Note on $*$-Clean Rings	[ "math.RA" ]	A $$-ring $R$ is called (strongly) $$-clean if every element of $R$ is the sum of a projection and a unit (which commute with each other). In this note, some properties of $$-clean rings are considered. In particular, a new class of $$-clean rings which called strongly $\pi$-$$-regular are introduced. It is shown that $R$ is strongly $\pi$-$$-regular if and only if $R$ is $\pi$-regular and every idempotent of $R$ is a projection if and only if $R/J(R)$ is strongly regular with $J(R)$ nil, and every idempotent of $R/J(R)$ is lifted to a central projection of $R.$ In addition, the stable range conditions of $$-clean rings are discussed, and equivalent conditions among $$-rings related to $*$-cleanness are obtained.	math.RA	math	A Note on ∗-Clean Rings Jian Cuia, Zhou Wangb aDepartment of Mathematics, Anhui Normal University, Wuhu 241000, China bDepartment of Mathematics, Southeast University, Nanjing 210096, China Abstract A ∗-ring R is called (strongly) ∗-clean if every element of R is the sum of a projection and a unit (which commute with each other). In this note, some properties of ∗-clean rings are considered. In particular, a new class of ∗- clean rings which called strongly π-∗-regular are introduced. It is shown that R is strongly π-∗-regular if and only if R is π-regular and every idempotent of R is a projection if and only if R/J(R) is strongly regular with J(R) nil, and every idempotent of R/J(R) is lifted to a central projection of R. In addition, the stable range conditions of ∗-clean rings are discussed, and equivalent conditions among ∗-rings related to ∗-cleanness are obtained. Keywords: (Strongly) ∗-clean ring, (strongly) clean ring, strongly π-∗-regular ring, stable range condition. 2010 MSC: 16W10, 16U99 1. Introduction Rings in which every element is the product of a unit and an idempotent are said to be unit regular. Recall that an element of a ring R is clean if it is the sum of an idempotent and a unit, and R is clean if every element of R is clean (see [12]). Clean rings were introduced by Nicholson in relation to exchange rings and have been extensively studied since then. Recently, Wang et al. [16] showed that unit regular rings have idempotent stable range one (i.e., whenever aR + bR = R with a, b ∈ R, there exists e2 = e ∈ R such that a + be ∈ U(R), written isr(R) = 1 for short), and rings with isr(R) = 1 Email addresses: [email protected] (Jian Cui), [email protected] (Zhou Wang) Preprint submitted to Elsevier August 22, 2018 are clean. In 1999, Nicholson [13] called an element of a ring R strongly clean if it is the sum of a unit and an idempotent that commute with each other, and R is strongly clean if each of its elements is strongly clean. Clearly, a strongly clean ring is clean, and the converse holds for an abelian ring (that is, all idempotents in the ring are central). Local rings and strongly π-regular rings are well-known examples of strongly clean rings. A ring R is a ∗-ring (or ring with involution) if there exists an operation ∗ : R → R such that for all x, y ∈ R (x + y)∗ = x∗ + y ∗, (xy)∗ = y ∗x∗, and (x∗)∗ = x. An element p of a ∗-ring is a projection if p2 = p = p∗. Obviously, 0 and 1 are projections of any ∗-ring. A ∗-ring R is ∗-regular [2] if for every x in R there exists a projection p such that xR = pR. Following Vas [15], an element of a ∗-ring R is (strongly) ∗-clean if it can be expressed as the sum of a unit and a projection (that commute), and R is (strongly) ∗-clean if all of its elements are (strongly) ∗-clean. Clearly, ∗-clean rings are clean and strongly ∗-clean rings are strongly clean. It was shown in [7, 11] that there exists a clean ∗-ring but not ∗-clean, and unit regular ∗-regular rings (which called ∗-unit regular rings in [7]) need not be strongly ∗-clean, which answered two questions raised by Vas in [15]. In this note, we continue the study of (strongly) ∗-clean rings. In Sec- tion 2, several basic properties of (strongly) ∗-clean rings are investigated. Motivated by the close relationship between strong π-regularity and strong cleanness, we introduce the concept of strongly π-∗-regular rings in Sec- tion 3. The structure of strongly π-∗-regular rings is considered and some properties of extensions are discussed. As we know, it is still an open ques- tion that whether a strongly clean ring has idempotent stable range one, or even has stable range one (see [13]). In Section 4, we extend isr(R) = 1 to the ∗-version. We call a ∗-ring R have projection stable range one (written psr(R) = 1) if, for any a, b ∈ R, aR + bR = R implies that a + bp is a unit of R for some projection p ∈ R. It is shown that if R is strongly ∗-clean then psr(R) = 1, and if psr(R) = 1 then R is ∗-clean. Furthermore, several equivalent conditions among (strongly) clean rings, (strongly) ∗-clean rings and ∗-rings with projection (idempotent) stable range one are obtained. Throughout this paper, rings are associative with unity. Let R be a ring. The set of all idempotents, all nilpotents and all units of R are denoted by Id(R), Rnil and U(R), respectively. For a ∈ R, the commutant of a is denoted by comm(a) = {x ∈ R : ax = xa}. We write Mn(R) for the ring of all n × n matrices over R whose identity element we write as In. Let Zn be the ring 2 of integers modulo n. For a ∗-ring R, the symbol P (R) stands for the set of all projections of R. 2. ∗-Clean Rings In this section, some basic properties of ∗-clean rings are discussed, and several examples related to ∗-cleanness are given. Example 2.1. (1) Units, elements in J(R) and nilpotents of a ∗-ring R are ∗-clean. (2) Idempotents of a ∗-regular rings are ∗-clean. Proof. (1) It is obvious. (2) Let R be ∗-regular and e ∈ Id(R). Then there exists a projection p such that (1 − e)R = pR. So we have 1 − e = p(1 − e) and p = (1 − e)p, and hence ep = 0. Note that (e − p)(e − p) = e − ep − pe + p = e + p(1 − e) = e + (1 − e) = 1. So e − p ∈ U(R), and e = p + (e − p) is ∗-clean in R. By Example 2.1, every local ring with involution ∗ is ∗-clean. In [15], Vas asked whether there is an example of a ∗-ring that is clean but not ∗-clean. It was answered affirmatively in [7] and [11]. In fact, one can construct some counterexamples based on the following. Example 2.2. Let R be a boolean ∗-ring. Then R is ∗-clean if and only if ∗ = 1R is the identity map of R. In particular, R = Z2 ⊕ Z2 with (a, b)∗ = (b, a) is clean but not ∗-clean. Proof. Note that every boolean ring is clean. Suppose that R is ∗-clean. Given any a ∈ R. Then −a = p + u = p + 1 = p − 1 for some p ∈ P (R). So we have a = 1 − p ∈ P (R). Thus, a∗ = a, which implies ∗ = 1R. Conversely, if ∗ = 1R, then every idempotent of R is a projection. Thus, R is ∗-clean. Lemma 2.3. Let R be a ∗-ring. If 2 ∈ U(R), then for any u2 = 1, u∗ = u ∈ R if and only if every idempotent of R is a projection. Proof. (⇒). Let e ∈ Id(R). Then (1 − 2e)2 = 1. So we have 2e = 2e∗, and thus 2(e − e∗) = 0. Since 2 ∈ U(R), e = e∗. As desired. (⇐). Given u ∈ R with u2 = 1. Then u+1 2 )2 = 2 . Since every idempotent of R is a projection, it follows from 2 ∈ Id(R) since ( u+1 u2+2u+1 4 = u+1 2 )∗ = u+1 2 ( u+1 that u∗ = u. 3 The ∗-ring R = Z2 ⊕ Z2 in Example 2.2 reveals that "2 ∈ U(R)" in Lemma 2.3 cannot be removed. Corollary 2.4. Let R be a ∗-ring with 2 ∈ U(R). The following are equivalent: (1) R is clean and every unit of R is self-adjoint (i.e., u∗ = u for every u ∈ U(R)). (2) R is ∗-clean and ∗ = 1R. Proof. (2) ⇒ (1) is trivial. (1) ⇒ (2). Let a ∈ R. Then a = e + u for some e ∈ Id(R) and u ∈ U(R). Note that (1 − 2e)2 = 1. By Lemma 2.3, e∗ = e. Thus a ∈ R is ∗-clean and a∗ = a, and so ∗ = 1R. Recall that an element t of a ∗-ring R is self-adjoint square root of 1 if t2 = 1 and t∗ = t. Theorem 2.5. Let R be a ∗-ring, the following are equivalent: (1) R is ∗-clean and 2 ∈ U(R). (2) Every element of R is a sum of a unit and a self-adjoint square root of 1. Proof. (1) ⇒ (2). Let a ∈ R. Then 1+a 2 = p + u for some p ∈ P (R) and u ∈ U(R). It follows that a = (2p−1)+2u where (2p−1)∗ = 2p−1, (2p−1)2 = 1 and 2u ∈ U(R). (2) ⇒ (1). We first show that 2 ∈ U(R). By hypothesis, 1 = x + v with x2 = 1 and v ∈ U(R). So we have (1 − v)2 = x2 = 1, which implies that v2 = 2v. Since v is a unit, v = 2 ∈ U(R). Given any a ∈ R, then there exist y, w ∈ R satisfying 2a − 1 = y + w with y ∗ = y, y2 = 1 and w ∈ U(R). 2 )2 = y+1 Thus, a = y+1 and w 2 ∈ U(R). 2 is a ∗-clean expression since ( y+1 2 + w 2 )∗ = y+1 2 , ( y+1 2 Camillo and Yu [5] showed that if R is a ring in which 2 is a unit, then R is clean if and only if every element of R is the sum of a unit and a square root of 1. Indeed, by the proof of Theorem 2.5, the condition 2 ∈ U(R) is also necessary. Proposition 2.6. The following are equivalent for a ∗-ring R : (1) R is ∗-clean and 0, 1 are the only projections. (2) R is clean ring and 0, 1 are the only idempotents. (3) R is a local ring. 4 Proof. (2) ⇒ (3) follows from [14, Lemma 14] and (3) ⇒ (1) follows by Example 2.1. (1) ⇒ (2). It suffices to show that the only idempotents in R are 0 and 1. For e2 = e ∈ R, the hypothesis implies that e = p + u where p ∈ P (R) = {0, 1} and u ∈ U(R). If p = 0 then e = u is a unit, so e = 1. If p = 1 then 1 − e = −u ∈ U(R), and hence e = 0. As required. Let I be an ideal of a ∗-ring R. We call I is ∗-invariant if I ∗ ⊆ I. In this case, the involution ∗ of R can be extended to the factor ring R/I, which is still denoted by ∗. Lemma 2.7. Let R be ∗-clean. If I is a ∗-invariant ideal of R, then R/I is ∗-clean. In particular, R/J(R) is ∗-clean. Proof. Since the homomorphism image of a projection (resp., unit) is also a projection (resp., unit), the result follows. Next we only need to prove that J(R) is ∗-invariant. For any a∗ ∈ (J(R))∗, we show that a∗ ∈ J(R). Note that a ∈ J(R). Take any x ∈ R. Then 1 − x∗a ∈ U(R). Thus 1 − a∗x = (1 − x∗a)∗ is a unit of R, as desired. Let R be a ∗-ring. Then ∗ induces an involution of the power series ring R[[x]], denoted by ∗, where (P∞ i xi. i=0 aixi)∗ = P∞ i=0 a∗ Proposition 2.8. Let R be a ∗-ring. Then R[[x]] is ∗-clean if and only if R is ∗-clean. Proof. Suppose that R[[x]] is ∗-clean. Note that R ∼= R[[x]]/(x) and (x) is a ∗-invariant ideal of R[[x]]. By Lemma 2.7, R is ∗-clean. Conversely, assume that R is ∗-clean. Let f (x) = P∞ i=0 aixi ∈ R[[x]]. Write a0 = p + u with p ∈ P (R) and u ∈ U(R). Then f (x) = p + (u + P∞ i=1 aixi), where p ∈ P (R) ⊆ P (R[[x]]) and u + P∞ i=1 aixi ∈ U(R[[x]]). Hence f (x) is ∗-clean in R[[x]]. According to [14, Proposition 13], the polynomial ring R[x] is never clean. Hence, R[x] is not ∗-clean for any involution ∗. 5 3. Strongly π-∗-Regular Rings Strong π-regularity is closely related to strong cleanness. In this section, we introduce the notion of strongly π-∗-regular rings which can be viewed as ∗-versions of strongly π-regular rings. The structure and properties of strongly π-∗-regular rings are given. Lemma 3.1. [11, Lemma 2.1] Let R be a ∗-ring. If every idempotent of R is a projection, then R is abelian. Due to [7], an element a of a ∗-ring R is strongly ∗-regular if a = pu = up with p ∈ P (R) and u ∈ U(R); R is strongly ∗-regular if each of its elements is strongly ∗-regular. By [7, Proposition 2.8], any strongly ∗-regular element is strongly ∗-clean. Theorem 3.2. Let R be a ∗-ring. Then the following are equivalent for a ∈ R : (1) There exist e ∈ P (R), u ∈ U(R) and an integer m ≥ 1 such that am = eu and a, e, u commute with each other. (2) There exist f ∈ P (R), v ∈ U(R) such that a = f + v, f v = vf and af ∈ Rnil. (3) There exists p ∈ P (R) such that p ∈ comm(a), ap ∈ U(pRp) and a(1 − p) ∈ Rnil. (4) There exists b ∈ comm(a) such that (ab)∗ = ab, b = bab and a−a2b ∈ Rnil. Proof. (1) ⇒ (2). Write f = 1 − e. Clearly, f ∈ P (R) and am − f ∈ U(R) with the inverse u−1e − f. From af = f a, we have a − f := v is a unit of R (since (a − f )(am−1 + am−2f + · · · + af + f ) = am − f ∈ U(R)) and f v = vf . It is clear that (af )m = amf = 0. (2) ⇒ (3). Set p = 1 − f . Then p ∈ P (R), ap = pa = vp ∈ U(pRp) and a(1 − p) = af ∈ Rnil. (3) ⇒ (4). By (3), aw = wa = p for some w ∈ U(pRp). So we obtain [a−(1−p)][w−(1−p)] = 1−a(1−p) ∈ U(R) since a(1−p) is nilpotent, which implies that a − (1 − p) ∈ U(R). Let b = [a − (1 − p)]−1p. Then b ∈ comm(a), bp = b and ab = [a − (1 − p)]b = p ∈ P (R). Thus (ab)∗ = ab, b = bp = bab and a − a2b = a(1 − ab) = a(1 − p) ∈ Rnil. (4) ⇒ (1). Let e = ab. Then (ab)∗ = ab implies e∗ = e, and bab = b yields e2 = e. So e ∈ P (R). As a − a2b ∈ Rnil, am = ame for some integer m ≥ 1. Take u = am + (1 − e) and u′ = bme + (1 − e). Then uu′ = u′u = 1. Hence, u ∈ U(R) and am = ame = ue with a, e, u commuting with each other. 6 Recall that an element a of a ring R is strongly π-regular if an ∈ an+1R ∩ Ran+1 for some n ≥ 1 (equivalently, an = eu with e ∈ Id(R), u ∈ U(R) and a, e, u all commute [13]); R is strongly π-regular if every element of R is strongly π-regular. Based on the above, we introduce the following concept. Definition 3.3. Let R be a ∗-ring. An element a ∈ R is called strongly π-∗-regular if it satisfies the conditions in Theorem 3.2; R is called strongly π-∗-regular if every element of R is strongly π-∗-regular. Corollary 3.4. Any strongly ∗-regular element is strongly π-∗-regular, and any strongly π-∗-regular element is strongly ∗-clean. Example 3.5. (1) Let R = Z4 and ∗ = 1R. Then R is strongly π-∗-regular. However, 2 ∈ R is not strongly ∗-regular. (2) Let R be a local domain with involution ∗ and J(R) 6= 0. Note that P (R) = Id(R) = {0, 1}. So R is strongly ∗-clean by Proposition 2.6, but any power of a nonzero element in J(R) can not expressed as the product of a projection and a unit. Recall that a ring R is π-regular if for any a ∈ R, there exist n ≥ 1 and b ∈ R such that an = anban. Strongly π-regular rings and regular rings are π-regular (see [13]). A ring R is directly finite if ab = 1 implies ba = 1 for all a, b ∈ R. Abelian rings are directly finite. Theorem 3.6. The following are equivalent for a ∗-ring R : (1) R is strongly π-∗-regular. (2) R is π-regular and every idempotent of R is a projection. (3) For any a ∈ R, there exist n ≥ 1 and p ∈ P (R) such that anR = pR, and R is abelian. (4) For any a ∈ R, there exist n ≥ 1 such that an is strongly ∗-regular. (5) For any a ∈ R, there exist p ∈ P (R) and u ∈ U(R) such that a = p + u, ap ∈ Rnil; and v−1qv is a projection for all v ∈ U(R) and all q ∈ P (R). Proof. (1) ⇒ (2). Note that every strongly π-∗-regular ring is strongly π-regular and strongly ∗-clean. Thus R is a π-regular ring. By [11, Theorem 2.2], every idempotent of R is a projection. (2) ⇒ (3). For any a ∈ R, there exists n ≥ 1 such that an = anxan for some x ∈ R. Write anx = p. Then p ∈ P (R) and an = pan. It is clear that anR = pR. In view of Lemma 3.1, R is abelian. 7 (3) ⇒ (4). Let e ∈ Id(R). Then eR = pR for some p ∈ P (R). Since R is abelian, we have e = pe = ep = p. Thus, every idempotent of R is a projection. Given a ∈ R, there exists n ≥ 1 and q ∈ P (R) such that anR = qR. So one gets an = qan and q = anx for some x ∈ R, which implies an = anxan. Next we show that an − (1 − q) is invertible. Note that [an − (1 − q)][xq − (1 − q)] = 1. Then an − (1 − q) := u ∈ U(R) since R is directly finite. Multiplying the equation an − (1 − q) = u by p yields an = anq = uq = qu, which implies that an is strongly ∗-regular. (4) ⇒ (5). For e ∈ Id(R), e = qv = vq for some q ∈ P (R) and v ∈ U(R) by the assumption. Then e = qv = e2 = qv2, and so we obtain q = qv = e, which implies that every idempotent of R is a projection. Clearly, v−1qv is a projection for all v ∈ U(R) and all q ∈ P (R). Given a ∈ R as in (4), an = (1−p)w = w(1−p) for some p ∈ P (R) and w ∈ U(R). Note that R is abelian. So we have anp = (ap)n = 0 and (a − p)[an−1w−1(1 − p) − Pn i=0 aip] = 1, and hence a − p ∈ U(R) as R is directly finite. (5) ⇒ (1). By (5), every element of R is ∗-clean. In view of [11, Theorem 2.2], R is abelian. Thus R is strongly π-∗-regular by Theorem 3.2(2). Corollary 3.7. Let R be a ∗-ring. The following are equivalent: (1) R is strongly π-∗-regular. (2) R/J(R) is strongly π-∗-regular with J(R) nil, every projection of R is central and every projection of R/J(R) is lifted to a projection of R. (3) R/J(R) is strongly ∗-regular with J(R) nil, and every idempotent of R/J(R) is lifted to a central projection of R. Proof. Write R = R/J(R). By Lemma 2.7, R is a ∗-ring. (1) ⇒ (2). Clearly, R is strongly π-∗-regular. As R is strongly π-regular, for any a ∈ J(R), there exist m ≥ 1, e ∈ Id(R) and u ∈ U(R) such that e = amu ∈ J(R). So am = eu−1 = 0, which implies that J(R) is nil. Note that R is strongly ∗-clean. So the rest follows from [11, Corollary 2.11]. nil (2) ⇒ (3). By virtue of [11, Corollar 2.11], R is reduced (i.e., R = 0), and every idempotent of R is lifted to a central projection of R. So we only need to prove that R is strongly ∗-regular. Given any x ∈ R. By Theorem 3.2, there exist p ∈ P (R) and v ∈ U(R) such that a = p + v, vp = pv and ap ∈ R = 0. It follows that a = a(1 − p) = v(1 − p) = (1 − p)v is strongly ∗-regular in R. nil (3) ⇒ (1). Since R is strongly regular, it is reduced clean. By [11, Corollary 2.11], every idempotent of R is a projection. Note that J(R) is nil 8 and R is π-regular. So R is π-regular by [1, Theorem 4]. In view of Theorem 3.6, R is strongly π-∗-regular. Corollary 3.8. Let R be a ∗-ring. Then R is strongly ∗-clean and π-regular if and only if R is strongly π-∗-regular. Proof. If R is strongly ∗-clean and π-regular, by [11, Theorem 2.2], idem- potents of R are projections. So R is strongly π-∗-regular by Theorem 3.6. The other direction is clear. For a ∗-ring R, the matrix ring Mn(R) has a natural involution inherited ij) (i.e., A∗ = ji)). Henceforth we consider Mn(R) as a ∗-ring with respect to from R : if A = (aij) ∈ Mn(R), A∗ is the transpose of (a∗ (a∗ this natural involution. ij)T = (a∗ Corollary 3.9. Let R be a ∗-ring. Then Mn(R) is not strongly π-∗-regular for any n ≥ 2. Let R be a ∗-ring and S = pRp with p ∈ P (R). Then the restriction of ∗ on S will be an involution of S, which is also denoted by ∗. Corollary 3.10. If R is strongly π-∗-regular, then so is eRe for any e ∈ Id(R). Proof. Let S = eRe with e ∈ Id(R). By hypothesis, e is a projection of R. So S is a ∗-ring. It is well known that S is strongly π-regular (see also [4, Lemma 39]). Clearly, every idempotent of S (⊆ R) is a projection. So the result follows by Theorem 3.6. Let RG be the group ring of a group G over a ring R. According to [11, gg−1 is an Lemma 2.12], the map ∗ : RG → RG given by (Pg agg)∗ = Pg a∗ involution of RG, and is denoted by ∗ again. Corollary 3.11. Let R be a ∗-ring with artinian prime factors, 2 ∈ J(R) and G be a locally finite 2-group. Then R is strongly π-∗-regular if and only if RG is strongly π-∗-regular. 9 Proof. Assume that R is strongly π-∗-regular. Then Id(R) = P (R). In particular, R is abelian. So idempotents of R coincide with idempotents in RG by [8, Lemma 11], and hence every idempotent of RG is a projection. Since R is a ring with artinian prime factors and G is a locally finite 2-group, RG is a strongly π-regular ring by [10, Theorem 3.3]. In view of Theorem 3.6, RG is strongly π-∗-regular. Conversely, R is strongly π-regular by [10, Proposition 3.4]. Note that Id(R) ⊆ Id(RG) and all idempotents of RG are projections. By Theorem 3.6, R is strongly π-∗-regular. Let C be the complex filed. It is well known that for any n ≥ 1, the matrix ring Mn(C) is strongly π-regular. However, Mn(C) is not strongly π- ∗-regular whenever n ≥ 2 by Corollary 3.9. So it is interesting to determine when a matrix of Mn(C) is strongly π-∗-regular. The set of all n × 1 matrices over C is denoted by Cn. Example 3.12. Let S = Mn(C) with ∗ the transpose operation. Then A is strongly π-∗-regular if and only if there exist e1, e2, . . . , en ∈ Cn such that 0 N ) P −1 with i ej = 0 for i = 1, . . . , r; j = r + 1, . . . , n, and A = P ( C 0 e∗ P = (e1, e2, . . . , en) ∈ U(S), C ∈ U(Mr(C)) and N ∈ [Mn−r(C)]nil. In particular, any real symmetric matrix is strongly π-∗-regular. Proof. Given A ∈ S. Assume that rank(A) = r. By the Jordan canoni- cal decomposition, there exists P = (e1, e2, . . . , en) ∈ U(S) such that A = 0 N ) P −1, where ei ∈ Cn for all i, C ∈ U(Mr(C)) and N ∈ [Mn−r(C)]nil. P ( C 0 Write B = P (cid:0) C−1 0 0 (cid:1) P −1. Then one easily gets that BA = AB, B = BAB and A−A2B = P ( 0 0 0 N ) P −1 is nilpotent. Note that B satisfies the above con- ditions is unique (see [3]). In view of Theorem 3.2, A is strongly π-∗-regular if and only if (AB)∗ = AB. Notice that AB = P ( Ir 0 0 0 ) P −1 and 0 (AB)∗ = AB 0 0 ) P ∗ = P ( Ir 0 0 0 ) P −1 0 0 ) (P ∗P ) = ( Ir 0 0 0 ) 0 0 ) (P ∗P ) = (P ∗P ) ( Ir 0 0 0 ) ⇔ (P −1)∗ ( Ir 0 ⇔ (P ∗P )−1 ( Ir 0 ⇔ ( Ir 0 ⇔ P ∗P = (cid:0) V1 0 ⇔ e∗ 0 V2 (cid:1) with V1 ∈ U(Mr(C)) and V2 ∈ U(Mn−r(C)) i ej = 0 for all i ∈ {1, 2, . . . , r}, j ∈ {r + 1, r + 2, . . . , n}, where V1 = (e∗ 1, e∗ 2, . . . , e∗ r)T (e1, e2, . . . , er); V2 = (e∗ r+1, e∗ r+2, . . . , e∗ n)T (er+1, er+2, . . . , en). 10 If A ∈ S is a real symmetric matrix, then there exists an orthogonal 0 0 ) P −1. So the result matrix P (i.e., P −1 = P T = P ∗) such that A = P ( Ir 0 follows. In view of [2, Proposition 3], the involution of a ∗-regular ring R is proper (i.e., x∗x = 0 implies that x = 0 for all x ∈ R). Remark 3.13. If R is strongly π-∗-regular, then for any x ∈ R, x∗x = 0 implies x ∈ Rnil. Indeed, by Theorem 3.2, there exist p ∈ P (R) and u ∈ U(R) such that xm = pu = up for some m ≥ 1. Then 0 = (x∗)mxm = (xm)∗xm = u∗pu, and thus p = 0, whence xm = 0. 4. Stable Range Conditions In [13], Nicholson asked whether every strongly clean ring has stable range one, and it is still open. Recall that a ring R is said to have idempotent stable range one (written isr(R)=1) provided that for any a, b ∈ R, aR + bR = R implies that a + be ∈ U(R) for some e ∈ Id(R) (see [6, 16]). If e is an arbitrary element of R (not necessary an idempotent), then R is said to have stable range one. Clearly, if isr(R) = 1, then R is clean and has stable range one. We extend the notion of isr(R) = 1 to ∗-versions. Definition 4.1. A ∗-ring R is said to have projection stable range one (written psr(R) = 1) if for any a, b ∈ R, aR + bR = R implies there exists p ∈ P (R) such that a + bp ∈ U(R). The following result is motivated by [6, Proposition 2]. Proposition 4.2. Let R be a ∗-ring. The following are equivalent: (1) psr(R) = 1. (2) For any a, b ∈ R, aR + bR = R implies there exists p ∈ P (R) such that a + bp is right invertible. (3) For any a, b ∈ R, aR + bR = R implies there exists p ∈ P (R) such that a + bp is left invertible. Proof. The proof is similar to that of [6, Proposition 2]. (1) ⇒ (2) is clear. 11 (2) ⇒ (3). Let a, b ∈ R with aR + bR = R. Then there is a projection p ∈ R such that a + bp = u is right invertible. Assume that uw = 1 for some w ∈ R. Then wR + (1 − wu)R = R. So the hypothesis implies there exists q ∈ P (R) such that w + (1 − wu)q is right invertible. Note that u[w + (1 − wu)q] = 1. Thus w + (1 − wu)q is also left invertible, and hence invertible. This implies that u ∈ U(R). (3) ⇒ (1). Given any a, b ∈ R with aR + bR = R. Then there exists p ∈ P (R) such that a + bp is left invertible. We may let v ∈ R with v(a + bp) = 1. Then vR + 0R = R. By hypothesis, we can find a projection q such that v + 0q = v is left invertible. So v is a unit, which implies that a + bp ∈ U(R). Therefore, psr(R) = 1. For a ∗-ring R, it is clear that if psr(R) = 1, then isr(R) = 1. However, there exists a ∗-ring with isr(R) = 1 but not satisfies psr(R) = 1. Example 4.3. Define the involution of Z2 by ∗ : x 7→ x. Let S = M2(Z2). Then S is a ∗-ring. In view of [16, Corollary 3.4], isr(S) = 1 since S is unit regular. Notice that P (S) = {O, I2, ( 1 0 1 0 ) S = S. However, ( 1 0 1 0 ) P is not invertible for any P ∈ P (S). Hence, psr(S) 6= 1. 0 1 )}, and ( 1 0 0 0 )+ ( 0 0 0 0 ) , ( 0 0 0 0 ) S + ( 0 0 From Example 4.3, one can also find that the projection stable range one property cannot be inherited to the matrix ring. Proposition 4.4. Let R be a ∗-ring. If psr(R) = 1, then R is ∗-clean. Proof. For any a ∈ R, the equation aR + (−1)R = R implies that a + (−1)p = u ∈ U(R) for some p ∈ P (R). So a = p + u, and hence R is ∗-clean. According to [15, Proposition 4], the ring in Example 4.3 is ∗-clean. So we conclude that the converse of Proposition 4.4 is not true. Following Nicholson [12], a ring R is exchange if for every a ∈ R, there exists e2 = e ∈ aR such that 1 − e ∈ (1 − a)R. Clean rings are exchange, the converse holds whenever the rings are abelian. A ∗-ring R is called ∗-abelian if every projection of R is central [15]. Theorem 4.5. Let R be a ∗-ring. The following are equivalent: (1) psr(R) = 1 and R is ∗-abelian. (2) For any a, b ∈ R, aR + bR = R implies there exists a projection p ∈ 12 comm(a) such that a + bp ∈ U(R). (3) isr(R) = 1 and every idempotent of R is a projection. (4) R is clean (or exchange) and every idempotent of R is a projection. (5) R is ∗-clean and ∗-abelian. (6) R is strongly ∗-clean. (7) For every a ∈ R, there exists a projection p ∈ aR such that 1 − p ∈ (1 − a)R. Proof. (1) ⇒ (2) and (3) ⇒ (4) are clear; (4) ⇒ (5) ⇒ (6) ⇒ (7) follows from [11, Theorem 2.2]. (2) ⇒ (3). We only need to show that all idempotents are projections. Let e ∈ Id(R). Then eR + (−1)R = R. So there exists p ∈ P (R) such that ep = pe and e−p ∈ U(R). Note that (e−p)(1−e−p) = (1−e−p)(e−p) = 0. Thus, e = 1 − p ∈ P (R). Therefore, every idempotent of R is a projection. (7) ⇒ (1). Let e ∈ Id(R). Then there exists a projection p ∈ eR such that 1 − p ∈ (1 − e)R. So we obtain p = ep and 1 − p = (1 − e)(1 − p). It follows that e = p, and thus Id(R) = P (R). In view of Lemma 3.1, R is abelian. Note that R is exchange. Then by [6, Theorem 12], isr(R) = 1, and hence psr(R) = 1. It is still unknown that whether strongly clean rings have stable range one ([13]). However, we have an affirmative answer of their ∗-versions. Corollary 4.6. If R is a strongly ∗-clean ring, then psr(R) = 1. The following example will reveal that the converse of Corollary 4.6 does not hold. Example 4.7. Let S = M2(Z3). The involution of S is defined by A → A∗, where A∗ is the transpose of A ∈ S. Then S is not strongly ∗-clean by [7, Theorem 2.3]. Since S is unit regular, isr(S) = 1 by [16, Corollary 3.4]. In view of [9, Lemma 7], we have Id(S) = {O, I2, ( x y z 1−x ) with yz = x − x2}, and P (S) = {O, I2, ( 1 0 0 0 ) , ( 0 0 0 1 ) , ( 2 1 1 2 ) , ( 2 2 2 2 )}. 13 We next prove that psr(S) = 1. Assume on the contrary. Then there exist c d ) and A′ = (cid:0) a′ b′ c′ d′ (cid:1) with AS + A′S = S but A + A′P is not a unit for A = ( a b any P ∈ P (S). That is, det(A + A′P ) = 0. This implies the following system of equations: ad − bc = 0 a′d − bc′ = 0 ac′ − a′c = bd′ − b′d (i), (iii), (v). ad′ − b′c = 0 a′d′ − b′c′ = 0 (ii), (iv), On the other hand, as isr(S) = 1, there exists E ∈ Id(S) \ P (S) such that A + A′E ∈ U(S). Then E must be of the form ( x y z 1−x ) where yz = x − x2. By Eqs. (i) − (iv), we obtain det(A + A′E) = (ac′ − a′c)y − (bd′ − b′d)z. Next we show that ac′ − a′c = bd′ − b′d = 0. Case 1. c 6= 0. Multiplying Eq. (v) by c and by substituting b′c = ad′, we have (ac′ − a′c)c = bd′c − b′dc = (bc − ad)d′ = 0 by Eq. (i). Thus, ac′ − a′c = bd′ − b′d = 0. Case 2. d 6= 0. Multiplying Eq. (v) by d and by substituting a′d = bc′, we have (bd′ − b′d)d = ac′d − a′cd = (ad − bc)c′ = 0 by Eq. (i). So ac′ − a′c = bd′ − b′d = 0. Case 3. c = d = 0. From Eq. (ii) and (iii), we get ad′ = bc′ = 0. If b 6= 0, then c′ = 0, it follows that ac′ − a′c = 0. If a 6= 0, then d′ = 0, and so bd′ − b′d = 0. Thus ac′ − a′c = bd′ − b′d = 0. Therefore, det(A + A′E) = (ac′ − a′c)y − (bd′ − b′d)z = 0 for any case, which contradicts A + A′E ∈ U(S). Hence, psr(R) = 1. By Theorem 4.5, we have the following result immediately. Corollary 4.8. Let R be a ∗-ring. If Id(R) = P (R), then the following are equivalent: (1) R is (strongly) clean. (3) R is (strongly) ∗-clean. (2) R is exchange. (4) isr(R) = 1. (5) psr(R) = 1. ACKNOWLEDGMENTS This work was supported by the NNSF of China (No. 11326062, 11201064). 14 References [1] A. Badawi, On abelian π-regular rings, Comm. Algebra 25 (1997), 1009 -- 1021. [2] S. K. Berberian, Baer ∗-Rings, Grundlehren der Mathematischen Wis- senschaften, vol. 195, Springer-Verlag, BerlinCHeidelbergCNew York, 1972. [3] K. P. S. Bhaskara Rao, The Theory of Generalized Inverses over Com- mutative Rings, Taylor & Francis, London and New York, 2002. [4] G. Borooah, A. J. Diesl, and T. J. Dorsey, Strongly clean matrix rings over commutative local rings, J. Pure Appl. Algebra 212 (2008), no. 1, 281 -- 296. [5] V. P. Camillo and H. P. Yu, Exchange rings, units and idempotents, Comm. Algebra 22 (1994), no. 12, 4737 -- 4749. [6] H. Chen, Rings with many idempotents, Internat. J. Math. 22 (1999), no. 3, 547 -- 558. [7] J. Chen and J. Cui, Two questions of L. Vas on ∗-clean rings, Bull. Aust. Math. Soc. 88 (2013), no. 3, 499 -- 505. [8] J. Chen, W. K. Nicholson, and Y. Zhou, Group rings in which every element is uniquely the sum of a unit and an idempotent, J. Algebra 306 (2006), no. 2, 453 -- 460. [9] J. Chen, X. Yang, and Y. Zhou, When is the 2 × 2 matrix ring over a commutative local ring strongly clean? J. Algebra 301 (2006), no. 1, 280 -- 293. [10] A. Y. M. Chin and H. V. Chen, On strongly π-regular group rings, South- east Asian Bull. Math. 26 (2002), 387 -- 390. [11] C. Li and Y. Zhou, On strongly ∗-clean rings, J. Algebra Appl. 10 (2011), no. 6, 1363 -- 1370. [12] W. K. Nicholson, Lifting idempotents and exchange rings, Trans. Amer. Math. Soc. 229 (1977), 269 -- 278. 15 [13] W. K. Nicholson, Strongly clean rings and Fitting's lemma, Comm. Al- gebra 27 (1999), no. 8, 3583 -- 3592. [14] W. K. Nicholson and Y. Zhou, Rings in which elements are uniquely the sum of an idempotent and a unit, Glasgow Math. J. 46 (2004), no. 2, 227 -- 236. [15] L. Vas, ∗-Clean rings; some clean and almost clean Baer ∗-rings and von Neumann algebras, J. Algebra 324 (2010), no. 12, 3388 -- 3400. [16] Z. Wang, J. Chen, D. Khurana, and T. Y. Lam, Rings of idempotent stable rang one, Algebr. Represent. Theory 15 (2012), no. 1, 195 -- 200. 16
1206.7013	1	1206	2012-06-29T13:10:28	Free symmetric group algebras in division rings generated by poly-orderable groups	[ "math.RA" ]	We show that the canonical involution on a nonabelian poly-orderable group G extends to the Hughes-free division ring of fractions D of the group algebra k[G] of G over a field k and that, with respect to this involution, D contains a pair of symmetric elements freely generating a free group subalgebra of D over k.	math.RA	math	FREE SYMMETRIC GROUP ALGEBRAS IN DIVISION RINGS GENERATED BY POLY-ORDERABLE GROUPS VITOR O. FERREIRA, JAIRO Z. GONC¸ ALVES, AND JAVIER S ´ANCHEZ Abstract. We show that the canonical involution on a nonabelian poly- orderable group G extends to the Hughes-free division ring of fractions D of the group algebra k[G] of G over a field k and that, with respect to this involution, D contains a pair of symmetric elements freely generating a free group subalgebra of D over k. Introduction A long-standing conjecture of Makar-Limanov states that a division ring which is finitely generated (as a division ring) and infinite-dimensional over its center contains a free algebra of rank 2 over the center [13]. This question has been extensively investigated, see e.g. [4], and recently an interesting advance appeared in the context of division rings generated by a group algebra of an ordered group; namely, in [20] S´anchez proved that if K is a division ring, G is a nonabelian ordered group and K(G) is the subdivision ring of the Malcev-Neumann series division ring K((G)) generated by the group ring K[G], then K(G) contains a free group algebra of rank 2 over its center. In the presence of an involution, a natural question would be whether a divi- sion ring satisfying Makar-Limanov's conjecture contains a free algebra of rank 2 generated by symmetric elements. An analogous problem was considered in [4], where the authors looked for a symmetric pair generating a free subgroup in the multiplicative group of a division ring endowed with an involution. In order to state the main contribution of the present paper, we shall introduce some definitions and notation. In what follows, k will denote a field. Given a k-algebra A, by a k-involution on A one understands a k-linear map ⋆ : A → A satisfying (i) (ab)⋆ = b⋆a⋆, for all a, b ∈ A, and (ii) (a⋆)⋆ = a, for all a ∈ A. An element a ∈ A is said to be symmetric if a⋆ = a. Date: 26 June 2012. 2010 Mathematics Subject Classification. Primary 16K40, 16S35, 16W10; Secondary 16S10, 20F60. Key words and phrases. Infinite dimensional division rings, division rings with involution, free associative algebras, ordered groups. The first and second authors were partially supported by Fapesp-Brazil Proc. 2009/52665-0. The second author was partially supported by CNPq-Brazil Grant 300.128/2008-8. The third author was supported by Fapesp-Brazil Proc. 2009/50886-0. 1 2 V. O. FERREIRA, J. Z. GONC¸ ALVES, AND J. S ´ANCHEZ If G is a group and k[G] denotes the group algebra of G over k, then the map k[G] −→ k[G] where for all x ∈ G, ax are elements of k all but a finite number of which nonzero, is a k-involution k[G], henceforth called the canonical involution of k[G]. Px∈G xax 7−→ Px∈G x−1ax, If G is ordered, we let k((G)) denote the division ring of Malcev-Neumann series of G over k, and let k(G) be the subdivision ring of k((G)) generated by k[G]. The main aim of this paper is to present a proof that if G is nonabelian, then the canonical involution on k[G] extends to a k-involution ⋆ on k(G) and k(G) contains a group k-algebra of a nonabelian free group generated by symmetric elements with respect to ⋆. Moreover, a pair of free symmetric generators is explicitly constructed. In Section 1, we approach the case of a torsion-free nilpotent group of class 2 generated by 2 elements. This case is treated separately for two reasons. First, it is a step in the proof of the general case and, second, it has interest in itself for, in this case, the group algebra k[G] is a noetherian domain and, therefore, has a unique field of fractions, regardless of it being embeddable in a Malcev-Neumann series ring. Section 2 is devoted to a proof of the fact that the canonical involution on k[G] can be extended to an involution on k(G) for an arbitrary ordered group. To be more precise, we consider the larger class of locally indicable groups which have a crossed product with a Hughes-free division ring of fractions. The existence of free symmetric generators of a free group algebra inside k(G) for G orderable is tackled in Section 3. The proof itself splits in two cases, depending on properties of the ordering of G, according to the trichotomy described in [20]. Here, again, we consider the more general case of crossed products of locally indicable groups with a Hughes-free division ring of fractions. The main result in the paper is Theorem 3.5 which states that a class of groups including poly-orderable groups satisfies Makar-Limanov's conjecture with respect to free group algebras generated by symmetric elements. Finally, in Section 4 we propose some questions and indicate possible develop- ments springing from the ideas in the paper. In what follows, if D is a division ring, which might be a field, we shall let D× denote the set of nonzero elements of D. 1. The nilpotent case Let G be the free nilpotent group of class 2 generated by two elements, that is, G = hx, y : [[x, y], x] = [[x, y], y] = 1i, where [x, y] = x−1y−1xy. Let k be a field and let ⋆ denote the canonical k-involution on the group algebra k[G] of G over k. It is well known that k[G] is a noetherian domain and, thus, ⋆ extends to a k-involution (still denoted by ⋆) on the Ore field of fractions D of k[G]. Our aim in this section is to provide a proof of the following fact. Theorem 1.1. Let k be a field and consider the group G = hx, y : [[x, y], x] = [[x, y], y] = 1i. Let D denote the Ore field of fractions of the group algebra k[G]. Then 1 + y(1 − y)−2 and 1 + y(1 − y)−2x(1 − x)−2y(1 − y)−2 FREE GROUP ALGEBRAS IN DIVISION RINGS 3 are symmetric elements with respect to the canonical involution on D and freely generate a free group k-algebra in D. Let k(λ) denote the field of rational functions in the indeterminate λ over k and let K = k(λ)(t) be the field of rational functions in the indeterminate t over k(λ). It is known that D is k-isomorphic to the Ore field of fractions Q of the skew polynomial ring K[X; σ] =(cid:8)Pi≥0 X iai : ai ∈ K all but a finite number nonzero(cid:9), where σ is the k(λ)-automorphism of K satisfying tσ = λt and aX = Xaσ, for all a ∈ K. An explicit isomorphism Q → D is given by λ 7→ [x, y], t 7→ x, X 7→ y. The main ingredient in the proof of Theorem 1.1 is the following proposition, whose proof is given below, after the proof of Theorem 1.1. Proposition 1.2. Let k be a field and let Q denote the Ore field of fractions of the skew polynomial ring K[X; σ], where K = k(λ)(t) and σ is the k(λ)-automorphism of K such that tσ = λt. Then the elements (1 − X)−1 and t(1 − t)−2(1 − X)−1 freely generate a free k-subalgebra of Q. Proof of Theorem 1.1. We work in the field of fractions Q of the skew polyno- mial ring K[X; σ], with the notation introduced in the paragraph preceding the statement of Proposition 1.2. In particular, the k-involution ⋆ of D extending the canonical involution of k[G] defines a k-involution ⋆ on Q satisfying λ⋆ = λ−1, t⋆ = t−1 and X ⋆ = X −1. Set α = (1 − X)−1 and β = t(1 − t)−2. By Proposition 1.2, α and βα freely generate a free k-subalgebra A of Q. Clearly, α2 − α, β(α2 − α) ∈ A and they do not commute. It then follows from [2, Corollary 6.7.4] that α2 − α and β(α2 − α) freely generate a free k-subalgebra of A and, thus, of Q. Therefore, γ = α2 − α and δ = (α2 − α)β(α2 − α) freely generate a free k-subalgebra of Q too. But γ = X(1 − X)−2 and so γ⋆ = γ. Also β⋆ = β and this implies that γ, δ form a pair of symmetric free generators of a free k-subalgebra of Q. Now consider the X-adic valuation on Q, that is, the valuation ν in Q which is zero in K and ν(X) = 1. We have ν(γ) = 1 and ν(δ) = 2. By [10, Corollary 1 on p. 524], 1 + γ and 1 + δ is a pair of symmetric elements in Q which freely generates a free group k-algebra in Q. Finally, one uses the isomorphism between Q and D to obtain the desired result. (cid:3) We now proceed to a proof of Proposition 1.2 using a criterion of Bell and Rogalski [1] for a pair of elements to generate a free subalgebra in the field of fractions of a skew polynomial ring. Let k be a field and let F be an extension field of k. Given σ ∈ Autk F , let Fσ = {f ∈ F : f σ = f } denote the fixed subfield of F with respect to σ. We shall make use of the following simplified form of the result of Bell and Rogalski. Theorem 1.3 ([1, Theorem 2.2]). Let F/k be a field extension and let σ ∈ Autk F . Denote by Q the Ore field of fractions of the skew polynomial ring F [X; σ]. Let b ∈ F \ Fσ be such that for all f ∈ F , f σ − f ∈ Fσ + Fσb implies f ∈ Fσ. Then the elements (1 − X)−1 and b(1 − X)−1 freely generate a free k-subalgebra of Q. (cid:3) We shall apply the above criterion to the case where F is a field of rational functions over k and in order to verify the hypothesis of Theorem 1.3, we shall use the notion of a pole of a rational function. So let k(t) denote the field of rational functions over k on the indeterminate t and let ¯k denote the algebraic closure of k. We add an element ∞ to ¯k and regard the elements of k(t) as functions on ¯k ∪ {∞} 4 V. O. FERREIRA, J. Z. GONC¸ ALVES, AND J. S ´ANCHEZ with the following conventions: given h ∈ k(t), say h = p gcd(p, q) = 1, we set h(r) = ∞, for all r ∈ ¯k such that q(r) = 0, and q , with p, q ∈ k[t] and ∞ 0 ab−1 if deg(p) > deg(q), if deg(p) < deg(q), if deg(p) = deg(q), h(∞) =  where a stands for the leading term of p and b for the leading term of q. We say that an element r ∈ ¯k ∪ {∞} is a pole of h ∈ k(t) if h(r) = ∞. Given h ∈ k(t) we shall let Sh denote the set of poles of h. Then, clearly, Sh is a finite set which is nonempty whenever h 6∈ k. Also, h ∈ k[t] \ k if and only if Sh = {∞}. Note, finally, that if σ ∈ Autk k(t) and tσ = h, then for all f ∈ k(t), we have Sf = h(Sf σ ). The following lemma is based on an argument of Lorenz (see [14, Lemma 1]). Lemma 1.4. Let k be a field, let λ ∈ k× be an element which is not a root of unity and let σ ∈ Autk k(t) be the automorphism such that tσ = λt. Let g ∈ k(t) \ k[t] be a rational function which has a unique pole and this pole is nonzero. If f ∈ k(t) satisfies f σ − f ∈ k + kg, then f ∈ k. Proof. We start by remarking that, taking Laurent series representing the rational functions, for all u ∈ k(t), uσ − u ∈ k implies that u ∈ k. So it is enough to show that f σ − f ∈ k. Let A = Sf σ and B = Sf . If tσ = h, then, as we have seen, B = h(A). Let D = (A ∪ B) \ (A ∩ B) and let C = Sf σ−f . Then D is a subset of C with an even number of elements, for A = B. Suppose that f σ − f = a + bg for some a, b ∈ k. Then C = Sa+bg ⊆ Sg. But, by hypothesis, Sg contains just one element; it follows that D = ∅ and B = A. Thus, B = h(B) = h2(B) = . . . . If there existed r ∈ B, with r 6= 0 and r 6= ∞, we would have {r, h(r), h2(r), . . . } ⊆ B. Since B is finite, this would imply that λmr = hm(r) = hn(r) = λnr, for some 0 ≤ m < n, which is impossible, because λ is not a root of 1. If either B = {0} or B = {0, ∞}, we would have that f = p/tn, for some n ≥ 1 and some p ∈ k[t] which is not a multiple of t. In this case, f σ − f =(cid:0)λ−npσ − p(cid:1)/tn. Now the numerator λ−npσ − p does not vanish, because t does not divide p and λn 6= 1. So, 0 would be a pole of f σ − f , that is, we would have 0 ∈ C ⊆ Sg and this would contradict the hypothesis on g. We are left, then, with two possibilities: either B = ∅ or B = {∞}. In either case, a + bg = f σ − f ∈ k[t], and this can only be possible if b = 0, for, by hypothesis, g 6∈ k[t]. So f σ − f ∈ k, which is what we needed to show. (cid:3) Theorem 1.5. Let k be a field, let λ ∈ k× be an element which is not a root of unity and let σ ∈ Autk k(t) be the automorphism such that tσ = λt. Let Q denote the Ore field of fractions of the skew polynomial ring k(t)[X; σ]. Then the elements (1 − X)−1 and t(1 − t)−2(1 − X)−1 freely generate a free k-subalgebra in Q. Proof. We shall apply the criterion in Theorem 1.3. So let F = k(t). We have seen in the first paragraph of the proof of Lemma 1.4 that, for all u ∈ F , uσ − u ∈ k implies u ∈ k. In particular, it follows that Fσ = k. The element b = t(1 − t)−2 ∈ F is a rational function which is not a polynomial and has a unique pole which is nonzero. So, by Lemma 1.4, the hypothesis of Theorem 1.3 are satisfied and (1 − X)−1 and t(1 − t)−2(1 − X)−1 freely generate a free k-subalgebra in Q. (cid:3) FREE GROUP ALGEBRAS IN DIVISION RINGS 5 To get Proposition 1.2 from the theorem above, let k = k(λ) and note that since (1 − X)−1 and t(1 − t)−2(1 − X)−1 freely generate a free k-subalgebra in Q, they freely generate a free k-subalgebra in Q. We end this section with a direct consequence of Theorem 1.1. Corollary 1.6. Let k be a field and let G be a nonabelian torsion-free nilpotent group. Then the Ore field of fractions of the group algebra k[G] contains a free group k-algebra of rank 2 freely generated by symmetric elements with respect to the canonical involution. Proof. Since k[G] is a noetherian domain, the Ore field of fractions Q of k[G] exists and the canonical involution of k[G] extends to a k-involution on Q. Since G is nilpotent and nonabelian, there exist a, b ∈ G such that [a, b] = c 6= 1 and [a, c] = [b, c] = 1. Then the subgroup H of G generated by a and b is a free nilpotent group of class 2 generated by 2 elements. By Theorem 1.1, the subdivision ring D of Q generated by k[N ] contains a pair of symmetric elements with respect to the canonical involution on k[N ], and, thus, with respect to the canonical involution on k[G], which freely generate a free group k-algebra in D ⊆ Q. (cid:3) 2. The general case: extending the canonical involution In this section we shall consider more general groups. More precisely, the main result in the section, Theorem 2.9, will show that an involution on a crossed product of a locally indicable group with a Hughes-free division ring of fractions can be uniquely extended to this division ring. We must start by recalling the necessary definitions, notation and basic results. 2.1. General crossed products. Given a group G and a ring R, a crossed product of G over R is an associative ring R ∗ G containing R as a subring and such that R ∗ G is a free right R-module with basis G = {¯x : x ∈ G}, a set in bijection with G, with multiplication satisfying (1) ¯x¯y = xyτ (x, y) and a¯x = ¯xaσ(x), for all x, y ∈ G and a ∈ R, where τ : G × G → U (R) and σ : G → Aut(R) are maps and U (R) denotes the group of units of R. We can always assume that the unity element of K ∗ G is ¯1. The set {¯xa : a ∈ U (R), x ∈ G} is a subgroup of the group of units of R ∗ G, and its elements are called trivial units of R ∗ G. (We refer to [16] for general facts about crossed products.) Since we shall be looking at involutions on crossed products, the following fact will come in handy. Lemma 2.1. Let R be a ring, let G be a group and let R ∗ G be a crossed product of G over R. Then (R ∗ G)op is a crossed product of the opposite group Gop over the opposite ring Rop. Proof. By definition, R ∗ G is a free right R-module with basis G. It follows that R ∗ G is also a free left R-module with basis G and, hence, (R ∗ G)op is a free right Rop-module with basis Gop. Further, if σ : G → Aut(R) and τ : G × G → U (R) are the maps defining the multiplication in R ∗ G, then it is easy to see that (R ∗ G)op = Rop ∗ Gop, with maps Gop −→ Aut(Rop) x 7−→ σ(x)−1 and Gop × Gop −→ U (Rop) (x, y) 7−→ τ (y, x)σ(yx)−1 . 6 V. O. FERREIRA, J. Z. GONC¸ ALVES, AND J. S ´ANCHEZ (cid:3) The following is a remark on crossed products embeddable in division rings that will be used in this section and in the next one. Remark 2.2. Let R ∗ G be a crossed product of the group G over the ring R. Suppose that there exists a division ring D and a ring embedding R ∗ G ֒→ D. For each subgroup N of G, there is a natural ring homomorphism R ∗ N ֒→ R ∗ G. Let D(R, N ) denote the subdivision ring of D generated by R ∗ N . Then, clearly, D(R, N ) = ∪i≥0Di, where D0 = R ∗ N and for each i ≥ 1, Di is the subring of D generated by Di−1 and all the inverses of the nonzero elements of Di−1. Further, if N is a normal subgroup of a subgroup H of G and t ∈ H, then the inner automorphism σ : D → D defined by σ(f ) = ¯t−1f ¯t, for all f ∈ D, restricts to an automorphism of D(R, N ). This is so, because σ leaves D0 invariant and, hence, σ(Di) ⊆ Di, for all i ≥ 0. 2.2. Hughes-free embeddings. A group G is locally indicable if for every non- trivial finitely generated subgroup H of G there exists a normal subgroup N of H such that H/N is infinite cyclic. Clearly, locally indicable groups are torsion-free. Subgroups of locally indicable groups are locally indicable. Moreover, extensions of locally indicable groups are locally indicable (cf. [5]) and orderable groups are locally indicable (by [8]); it follows that poly-orderable groups are locally indicable. Let G be a group and let δ be an ordinal. A subnormal series of G (indexed on δ) is a chain of subgroups {Gγ}γ≤δ of G such that (i) G0 = {1}, Gδ = G, (ii) Gγ is a normal subgroup of Gγ+1, for all γ < δ, and (iii) for each limit ordinal γ′ ≤ δ, Gγ ′ =Sγ<γ ′ Gγ. Let X be a class of groups closed under isomorphisms. We say that G is a δ- poly-X group if there exists a subnormal series {Gγ}γ≤δ of G such that for all γ < δ the quotient Gγ+1/Gγ ∈ X . Given an ordinal δ, we shall consider δ-poly-orderable groups. Note that such a group is locally indicable, because orderable groups are locally indicable and extensions of locally indicable groups are locally indicable. Since local indicability is a local property of a group, it follows by transfinite induction that a δ-poly- orderable group is locally indicable. Let G be a locally indicable group, let K be a division ring and let K ∗ G be a crossed product of G over K. Suppose that there exists a ring embedding K∗G ֒→ D of K ∗ G into a division ring D. We say that the embedding K ∗ G ֒→ D is Hughes- free if for each nontrivial finitely generated subgroup H of G, for each normal subgroup N of H for which H/N is infinite cyclic, and for each t ∈ H such that H/N = htN i, the set {¯tn : n ≥ 0} is right D(K, N )-linearly independent inside D. If, moreover, D = D(K, G), we say that D is a Hughes-free division ring of fractions of K ∗ G. If a crossed product K ∗ G of a locally indicable group G over a division ring K is an Ore domain, then clearly any embedding of K ∗ G into a division ring will be Hughes-free. FREE GROUP ALGEBRAS IN DIVISION RINGS 7 If G is an ordered group, K is a division ring and K ∗ G is a crossed product of G over K, then K ∗ G can be embedded in the division ring K ∗ ((G)) of Malcev- Neumann series ([11, 15]), whose elements are of the form ¯xax, f = Xx∈G with ax ∈ K and supp(f ) = {x ∈ G : ax 6= 0} a well-ordered subset of G, where It is known that the embedding K ∗ G ֒→ multiplication satisfies the laws (1). K ∗ ((G)) is Hughes-free [6]. Hughes-free division rings of fractions are unique: Theorem 2.3 ([6, Theorem]). Let K be a division ring, let G be a locally indicable group and let K ∗ G be a crossed product. Suppose that D1 and D2 are Hughes-free division rings of fractions of K ∗ G. Then there exists a unique ring isomorphism ϕ : D1 → D2 such that ϕ(α) = α, for all α ∈ K ∗ G. (cid:3) A proof of Theorem 2.3 can be found in [3]. We recall that a locally indicable group Γ is said to be Hughes-free embeddable if every crossed product F ∗ Γ of Γ over a division ring F has a Hughes-free division ring of fractions. For example, torsion-free nilpotent groups, or more generally orderable groups, are Hughes-free embeddable. It was proved in [7] that extensions of Hughes-free embeddable groups are Hughes- free embeddable. Thus poly-orderable groups are Hughes-free embeddable. Lemma 2.4. Let X be the class of Hughes-free embeddable groups. Let δ be an ordinal and let G be a δ-poly-X group. Then G is a Hughes-free embeddable group. Hence a δ-poly-orderable group is Hughes-free embeddable. Proof. Let {Gγ}γ≤δ be a subnormal series of G such that Gγ+1/Gγ is Hughes-free embeddable for each γ < δ. We shall show that Gγ is Hughes-free embeddable for all γ ≤ δ. If γ = 1, then clearly Gγ is Hughes-free embeddable. Let γ > 1 and suppose that the result holds for all ordinals ρ < γ. If γ = ρ + 1 for some ordinal ρ, then Gγ is the extension of the Hughes-free embeddable groups Gρ and Gγ/Gρ. By [7], Gγ is Hughes-free embeddable. Finally, suppose that γ is a limit ordinal. Let K be a division ring and let K ∗ Gγ be a crossed product. If ε < ρ < γ, then it follows from Theorem 2.3 that the following diagram is commutative K ∗ Gε K ∗ Gρ Dε / Dρ where Dε and Dρ are the Hughes-free division ring of fractions of K ∗ Gε and K ∗ Gρ, respectively. Let Dγ = lim−→ρ<γ Dρ. Then Dγ is a division ring that contains K ∗ Gγ = lim−→ρ<γ K ∗ Gρ and Dρ for all ρ < γ. Let H be a finitely generated subgroup of Gγ. Then H is contained in some Gρ for some ρ < γ. Thus K ∗ H is contained in K ∗ Gρ and Dρ, which proves the linear independence condition that characterizes Hughes-freeness, because K ∗ Gρ ֒→ Dρ is Hughes-free. Since H, K and K ∗ Gγ were arbitrary, we get that Gγ is Hughes-free embeddable. / / _ _ / 8 V. O. FERREIRA, J. Z. GONC¸ ALVES, AND J. S ´ANCHEZ The second statement follows because orderable groups are Hughes-free embed- (cid:3) dable. 2.3. Extending involutions. We shall show that involutions of a crossed product of a locally indicable group by a division ring extend to involutions on a Hughes-free division ring of fractions. In order to prove this we shall also present a version of Corollary 7.3 from [3]. We start with two lemmas which will also be used in the proof of Theorem 2.9, the main result in this section. Lemma 2.5. Let K and K ′ be division rings, let G and G′ be groups and let K ∗ G and K ′ ∗ G′ be crossed products. Suppose that K ∗ G and K ′ ∗ G′ have only trivial units. Then, if φ : K ∗ G → K ′ ∗ G′ is a ring isomorphism, we have (i) φ(K) = K ′, (ii) φ(K ×G) = (K ′)×G′, and (iii) φ induces a group isomorphism G → G′. Proof. Let x ∈ G be a fixed element. Since K ′ ∗ G′ has only trivial units, φ(¯x) = ¯yb, for some y ∈ G′ and some b ∈ (K ′)×. We claim that φ(¯xK) = ¯yK ′. Indeed, given a ∈ K \ {0, 1}, we have φ(¯xa) = φ(cid:0)¯x(a − 1)(cid:1) + ¯yb. Since ¯x(a − 1) is a unit, φ(cid:0)¯x(a − 1)(cid:1) = ¯zc for some z ∈ G′ and some c ∈ (K ′)×. But φ(¯xa) is also a (trivial) unit. Hence z = y and we have φ(¯xK) ⊆ ¯yK ′. Using that φ−1(¯yb) = ¯x, one proves similarly that φ−1(¯yK ′) ⊆ ¯xK. The above proves (ii) and (i), by setting x = 1. To obtain (iii), note that the surjective group homomorphism K ×G → (K ′)×G′/(K ′)×, obtained by composing the isomorphism K ×G → (K ′)×G′, given by the restriction of φ, with the canonical surjection (K ′)×G′ → (K ′)×G′/(K ′)×, has kernel K ×. Since G ∼= K ×G/K × and G′ ∼= (K ′)×G′/(K ′)×, (iii) follows. (Explicitly, the isomorphism G → G′ is given (cid:3) by x 7→ y, where supp(cid:0)φ(¯x)(cid:1) = {y}.) Lemma 2.6. Let K and K ′ be division rings, let G and G′ be locally indicable groups and let K ∗ G and K ′ ∗ G′ be crossed products. Suppose that K ∗ G and K ′ ∗ G′ are isomorphic as rings and that K ′ ∗ G′ has a Hughes-free division ring of fractions D. Then D is also a Hughes-free division ring of fractions for K ∗ G. Proof. Denote by φ : K ∗ G → K ′ ∗ G′ an isomorphism. Then K ∗ G embeds into D via φ and D is generated by its image. We are left to prove that this embedding is Hughes-free. That K ∗G and K ′ ∗G′ have only trivial units was proved in [5], since the grading arguments of his Section 4 apply to crossed products. By Lemma 2.5, φ(K) = K ′ and there is a group isomorphism η : G → G′ induced by φ. For a subgroup N of G, we then have φ(K ∗ N ) = K ′ ∗ η(N ). This implies that the subdivision ring of D generated by (the image of) K ∗ N is D(K ′, η(N )). Let H be a nontrivial finitely generated subgroup of G, let N be a normal subgroup of H such that H/N is infinite cyclic and pick t ∈ H such that H/N = htN i. We must show that the powers of φ(¯t) are right linearly independent over D(K ′, η(N )). Indeed, let d0, . . . , dn ∈ D(K ′, η(N )) be such that d0 + φ(¯t)d1 + · · · + φ(¯t)ndn = 0. By definition of η, for each i = 1, . . . , n, there exists bi ∈ (K ′)× i such that φ(¯t)i = η(t) bndn = 0. But η(t)η(N ) generates η(H)/η(N ), which is infinite cyclic; since η(H) is a nontrivial finitely generated subgroup of G′ and K ′ ∗ G′ ֒→ D is Hughes-free, it follows that bi. So, we have d0 + η(t)b1d1 + · · · + η(t) n FREE GROUP ALGEBRAS IN DIVISION RINGS 9 d0 = 0 and bidi = 0 for all i = 1, . . . , n. But, then, di = 0 for all i = 1, . . . , n. So the embedding of K ∗ G into D through φ is Hughes-free. (cid:3) Let G be a locally indicable group, let K be a division ring and let D be a Hughes-free division ring of fractions of a crossed product K ∗ G. Given a ring automorphism φ of K ∗ G, the embedding of K ∗ G into D through φ is a Hughes- free division ring of fractions of K ∗ G, by Lemma 2.6. If follows from Theorem 2.3 that φ extends to a ring isomorphism of D. This proves the following result. Proposition 2.7. Let K be a division ring, let G be a locally indicable group, and let D be a Hughes-free division ring of fractions of a crossed product K ∗ G. Then there is a natural injective group homomorphism Aut(K ∗ G) → Aut(D). More precisely, if φ is a ring automorphism of K ∗ G, and ι : K ∗ G → D denotes the Hughes-free embedding of K ∗ G into G, then there exists a unique ring automor- phism ψ of D such that ψι = ιφ. (cid:3) At first sight, it could seem that Proposition 2.7 is a more general version of [3, Corollary 7.3]. We remark that Lemma 2.5 together with the fact that K ∗ G has only trivial units imply that any automorphism of K ∗ G sends K to K, and thus Proposition 2.7 and [3, Corollary 7.3] are the same result. The next lemma allows us to apply Theorem 2.3 to involutions. Lemma 2.8. Let K be a division ring, let G be a locally indicable group, and let K ∗ G be a crossed product of G over K. Let D be a division ring. Then an embedding K ∗G ֒→ D is Hughes-free if and only if (K ∗G)op ֒→ Dop is Hughes-free. Proof. By Lemma 2.1, (K ∗ G)op = K op ∗ Gop. It follows that if N is a subgroup of G, then D(K, N )op = Dop(K op, N op). Given a finitely generated subgroup H op of Gop and a normal subgroup N op of H op such that H op/N op is infinite cyclic, pick t ∈ H such that H op/N op = hN opti. We must prove that {¯tn : n ≥ 0} is a right D(K, N )op-linearly independent subset of Dop, or equivalently, that it is a left D(K, N )-linearly independent subset of D. So suppose that d0, d1, . . . , dn ∈ D(K, N ) are such that 0 = d0 + d1¯t + · · · + dn¯tn = d0 + ¯t(¯t−1d1¯t) + · · · + ¯tn(¯t−ndn¯tn). By Remark 2.2, ¯t−idi¯t ∈ D(K, N ) for all i = 1, . . . , n. Then, Hughes-freeness of K ∗ G ֒→ D implies that ¯t−1di¯t = 0, and therefore di = 0, for all i = 0, . . . , n. (cid:3) Theorem 2.9. Let K be a division ring, let G be a locally indicable group and let K ∗ G be a crossed product. Suppose that K ∗ G has a Hughes-free division ring of fractions D. Then any involution on K ∗ G extends to a unique involution on D. Proof. Let ⋆ : K ∗ G → K ∗ G be an involution on K ∗ G. Then the map φ : K ∗ G −→ (K ∗ G)op α 7−→ α⋆ is a ring isomorphism. By Lemma 2.1, (K ∗ G)op = K op ∗ Gop and, by Lemma 2.8, K op ∗ Gop ֒→ Dop is a Hughes-free embedding. But then Lemma 2.6 implies that the embedding of K ∗ G into Dop through φ is Hughes-free. So it follows from Theorem 2.3, that there exists a ring isomorphism Φ : D → Dop extending φ. Hence, the map D → D, defined by ζ 7→ Φ(ζ), for all ζ ∈ D, is an involution on D extending ⋆. (cid:3) 10 V. O. FERREIRA, J. Z. GONC¸ ALVES, AND J. S ´ANCHEZ 3. The general case: free symmetric generators Here we go back to our original problem, that of searching for free symmetric elements in a division ring generated by a group over a field and endowed with an involution. Let k be a field and let G be a locally indicable group. Suppose that the group ring k[G] has a Hughes-free division ring of fractions D. By Theorem 2.9, D has an involution ⋆ extending the canonical k-involution on k[G], which will also be called the canonical involution on D. Our main result, Theorem 3.5, is stated for δ-poly-orderable groups. To prove it, we need two propositions which are valid for more general locally indicable groups. Proposition 3.1. Let k be a field and let G be a locally indicable group. Suppose that the group algebra k[G] has a Hughes-free division ring of fractions D. If G contains a free monoid of rank 2, then D contains a free group k-algebra of rank 2 freely generated by symmetric elements with respect to the canonical involution on D. Proof. Here we simplify the notation introduced in Section 2 and given a subgroup N of G write D(N ) for the division subring of D generated by k[N ]. Let x, y ∈ G be elements that freely generate a free submonoid of G and let H be the subgroup of G generated by x and y. We shall prove that D(H) contains a free group k-algebra freely generated by symmetric elements. Since G is locally indicable, there exists a subgroup N of H such that H/N is infinite cyclic. Let t ∈ H be such that tN generates H/N . By Remark 2.2, there exists an automorphism σ of D(N ) extending the inner automorphism of k[H] given by conjugation by t. Since k[G] ֒→ D is Hughes-free, the powers of t are right linearly independent over D(N ). Hence, D(N )[t; σ] embeds into D(H). Thus, D(H) = D(N )(t; σ). Now let ν denote the discrete valuation on D(H) extending the t-adic valuation on D(N )[t; σ]. In other words, the division ring D(H) has a discrete valuation ν such that tnfn! = min{n : fn 6= 0}, for all finite sums of the form Pn∈Z tnfn, with fn ∈ D(N ). ν Xn∈Z Since H/N is infinite cyclic, there exist unique m, n ∈ Z and u, v ∈ N such that x = tmu and y = tnv. So ν(x) = m and ν(y) = n, and m 6= 0 or n 6= 0 because H 6= N . Suppose m 6= 0. Now, if m < 0, substitute the pair {x, y} for the pair {x−1, y−1} and, then, substituting y for yxk, for an appropriate choice of k, if necessary, we can assume that (i) x and y freely generate a free monoid in G, and (ii) ν(x) > 0 and ν(y) > 0. Consider the following elements of k[H], f = x + x−1 and g = y + y−1. They are both symmetric and, thus, their inverses f −1, g−1 are symmetric elements in D(H). We shall regard the elements of D(H) as series via the embedding of D(H) = D(N )(t; σ) into the skew Laurent series ring D(N )((t; σ)). We have (2) f −1 = (x + x−1)−1 =(cid:0)(x2 + 1)x−1(cid:1)−1 = x(1 + x2)−1. FREE GROUP ALGEBRAS IN DIVISION RINGS 11 And, similarly, g−1 = y(1 + y2)−1. We claim that v(f −1) > 0. Indeed, since ν(x) > 0, then expressing 1 + x2 as a polynomial in t, we see that ν(1 + x2) = 0, which implies ν(f −1) = ν(x) − ν(1 + x2) = ν(x) > 0. Analogously, ν(g−1) > 0. First we show that f −1 and g−1 freely generate a free monoid in D(H). Suppose that there exist non-negative integers a1, b1, . . . , ar, br and c1, d1, . . . , cs, ds such that (f −1)a1 (g−1)b1 · · · (f −1)ar (g−1)br = (g−1)c1(f −1)d1 · · · (g−1)cs(g−1)ds. (3) If follows from (2) that, as elements of D(N )((t; σ)), f −1 and g−1 are written in a unique way as (4) f −1 = tmu + Xi≥m+1 tiαi and g−1 = tnv + Xj≥n+1 with αi, βj ∈ D(N ). But, then, (3) reads (5) (tmu)a1(tnv)b1 · · · (tmu)ar (tnv)br + i>m(a1+···+ar )+n(b1+···+br ) X tjβj, tiγi = (tnv)c1 (tmu)d1 · · · (tnv)cr (tmu)dr + tjδj, j>m(d1+···+ds)+n(c1+···+cs) X with γi, δj ∈ D(N ). Now (5) implies a relation of the form xa1 yb1 · · · xar ybr = yc1xd1 · · · ycsxds, which must be trivial. Now we must show that monomials in f −1 and g−1 are linearly independent over k. We again regard D(H) = D(N )(t; σ) ⊆ D(N )((t; σ)), endowed with the t-adic valuation. By what we have just seen, given a monomial w(f −1, g−1) in f −1 and g−1, there exists h ∈ D(N )((t; σ)) such that w(f −1, g−1) = w(x, y) + h, with ν(h) > ν(cid:0)w(x, y)(cid:1) = ν(cid:0)w(f −1, g−1)(cid:1). Suppose that the monomials in f −1 and g−1 are not linearly independent over k and pick a linear dependence relation r wi(f −1, g−1)λi = 0, Xi=1 where wi(f −1, g−1) are monomials in f −1 and g−1, and λi ∈ k×, with r minimal. Write I = {1, . . . , r} and let q = min(cid:8)ν(cid:0)wi(f −1, g−1)(cid:1) : i ∈ I(cid:9). Let J = (cid:8)i ∈ I : ν(cid:0)wi(f −1, g−1)(cid:1) = q(cid:9). For each j ∈ J, let hj ∈ D(N )((t; σ)) be such that wj(f −1, g−1) = wj(x, y) + hj and ν(hj) > q. We have 0 =Xi∈I =Xj∈J wi(f −1, g−1)λi =Xj∈J wj(x, y)λj +Xj∈J hjλj + Xi∈I\J wj (f −1, g−1)λj + Xi∈I\J wi(f −1, g−1)λi. wi(f −1, g−1)λi Now, ν(cid:16)Pj∈J hjλj +Pi∈I\J wi(f −1, g−1)λi(cid:17) > q. So we must havePj∈J wj(x, y)λj = 0. But since wj (x, y) ∈ H and the elements of H are linearly independent over k, it follows that λj = 0 for all j ∈ J. If J 6= I, we would have a nontrivial k-linear de- pendence relation among monomials in f −1 and g−1 with less than r terms, which is impossible. Thus, λi = 0 for all i ∈ I. 12 V. O. FERREIRA, J. Z. GONC¸ ALVES, AND J. S ´ANCHEZ Finally, it follows from [10, Corollary 1 on p. 524] that 1 + f −1 and 1 + g−1, which are symmetric elements of D, freely generate a free group k-subalgebra of D(H) and, hence, of D. (cid:3) It is worth mentioning at this point that a locally indicable group with at least one right order which is not of Conrad type contains a free monoid of rank 2 (cf. [17, Corollary 2.6]). We shall need the following result from I. Hughes which, although not explicitly stated there, can be extracted from [7] or [19]. Lemma 3.2. Let G be a locally indicable group with a normal subgroup N such that G/N is locally indicable. Let K be a division ring and let K ∗G be a crossed product. If K ∗ N has a Hughes-free division ring of fractions D and G/N is Hughes-free embeddable, then K ∗ G has a Hughes-free division ring of fractions. More precisely, regarding K ∗ G = (K ∗ N ) ∗ (G/N ) ֒→ D ∗ (G/N ), if E is a Hughes-free division ring of fractions of D ∗ (G/N ), then E is a Hughes-free division ring of fractions of K ∗ G. (cid:3) In what follows, given a field k and a locally indicable group G if the group algebra k[G] has a Hughes-free division ring of fractions, this division ring will be denoted by k(G). It follows from Theorem 2.9 that the canonical involution of k[G] extends to an involution on k(G). The next result is a more general version of [20, Proposition 3.4]. Proposition 3.3. Let A be a locally indicable group generated by two elements and let B be normal subgroup of A such that A/B is a torsion-free nilpotent group of class 2. Let k be a field and suppose that the group algebra k[B] has a Hughes-free division ring of fractions. Then k[A] has a Hughes-free division ring of fractions k(A) which contains a free group k-algebra of rank 2 freely generated by symmetric elements with respect to the canonical involution. Proof. Denote Γ = A/B. Since Γ is a torsion-free nilpotent group, it is orderable. Let < denote an order in Γ. We shall fix a transversal of B in A in the following way, for each γ ∈ Γ, with 1 < γ, choose xγ ∈ A such that Bxγ = γ and take x−1 to be the representative of γ−1; set x1 = 1. Now regard k[A] = k[B] ∗ Γ. With this choice of a transversal, in k[B] ∗ Γ we have ¯γ⋆ = γ−1, for all γ ∈ Γ, where ⋆ stands for the canonical involution on k[A]. γ By Lemma 3.2, k[A] has a Hughes-free division ring of fractions k(A), which coincides with the division ring of fractions of k(B) ∗ Γ inside k(B) ∗ ((Γ)). Thus, we can consider the following commutative diagram k[B] ∗ Γ = k[A] k(A) k[B] ∗ ((Γ)) / k(B) ∗ ((Γ)) By Theorem 1.1, there exist U, V ∈ k(Γ) such that U ⋆ = U , V ⋆ = V , and U and V freely generate a free group k-subalgebra of rank 2. Then, clearly, X = U V and X ⋆ = V U freely generate a free group k-algebra in k(Γ) of rank 2. Let ε : k[B] → k / / _ _ / FREE GROUP ALGEBRAS IN DIVISION RINGS 13 denote the augmentation map and consider the homomorphism ϕ : k[B] ∗ ((Γ)) −→ k((Γ)) Pγ∈Γ ¯γhγ 7−→ Pγ∈Γ γε(hγ) . We shall regard k(Γ) embedded into k((Γ)) and shall show that with an appropriate choice of a preimage of X, we can pullback the free group k-algebra from k(Γ) to k(A) through ϕ. Choose E, F ∈ k[Γ] such that X = EF −1. If E =Pγ∈Γ γeγ and F =Pγ∈Γ γfγ, with eγ, fγ ∈ k, let E =Pγ∈Γ ¯γeγ ∈ k[B]∗Γ = k[A] and F =Pγ∈Γ ¯γfγ ∈ k[B]∗Γ = k[A]. Then ϕ( E) = E, ϕ( F ) = F , and ϕ( E⋆) = ϕ Xγ∈Γ Similarly, ϕ( F ⋆) = F ⋆. ¯γ⋆eγ  = Xγ∈Γ ϕ(cid:0)γ−1(cid:1)eγ = Xγ∈Γ γ−1eγ = E⋆. Now, since the coefficients of F lie k, both F −1 and ( F ⋆)−1 lie in k[B]∗((Γ)), and we have ϕ( E F −1) = EF −1 = X and ϕ(cid:0)( F ⋆)−1 E⋆(cid:1) = (F ⋆)−1E⋆ = (EF −1)⋆ = X ⋆. Letting W = E F −1, we get W ∈ k(A) with the property that W and W ⋆ freely generate a free group k-subalgebra of k(A). Thus, with respect to the canonical involution, W W ⋆ and W ⋆W are symmetric elements of k(A), which freely generate a free group k-subalgebra. (cid:3) We are ready to tackle our main objective. But first we recall a result of Longo- bardi, Maj and Rhemtulla, which will be used in the proof. Lemma 3.4 ([12, Corollary 3]). Let G be a finitely generated group with no free submonoid. Then for every positive integer n, the nth derived subgroup G(n) of G is finitely generated. In particular if G is solvable, then it is polycyclic. (cid:3) Theorem 3.5. Let δ be an ordinal, let G be a δ-poly-orderable group, let k be a field and let k(G) be the Hughes-free division ring of fractions of the group algebra k[G]. Then the following are equivalent: (1) k(G) contains a free group k-algebra of rank 2 freely generated by symmetric elements with respect to the canonical involution. (2) k(G) is not a locally P.I. k-algebra. (3) G is not locally abelian-by-finite. Proof. We claim that for each x ∈ k(G), there exists a finitely generated subgroup Hx of G such that x ∈ k(Hx). Indeed, first observe that k(G) = [n≥0 Qn, where Q0 = k[G] and, for each n ≥ 0, Qn+1 is the subring of k(G) generated by Qn and the inverses of its nonzero elements. If x ∈ Q0, the result is clear. Suppose εini , ini where xij ∈ Qn and εij ∈ {1, −1}. For each i and each j, let Hxij be the finitely generated subgroup given by the induction hypothesis. If we set Hx as the subgroup generated by all Hxij , then x ∈ k(Hx). by induction that the claim holds for n ≥ 0. If x ∈ Qn+1, then x =P xεi1 i1 · · · x From this claim it follows that for each finitely generated k-subalgebra R of k(G) there exists a finitely generated subgroup H of G such that R ⊆ k(H). It is well 14 V. O. FERREIRA, J. Z. GONC¸ ALVES, AND J. S ´ANCHEZ known that if H is an abelian-by-finite group, then k(H) is of finite dimension over its center (see, e.g., [9]), and thus a P.I. algebra. So condition (2) implies condition (3). Condition (1) implies condition (2) because a P.I. algebra does not contain a noncommutative free subalgebra. Now, suppose that condition (3) holds. If G contains a free monoid, the result follows from Proposition 3.1. An ordered group belongs to one and only one of the three classes defined in [20]. Ordered groups belonging to two of these classes contain a free monoid of rank 2. Thus, if, for some γ < δ, Gγ+1/Gγ belongs to one of these, then Gγ+1/Gγ and, a fortiori, G contain a free monoid. On the other hand, if a nonabelian factor Gγ+1/Gγ, for some γ < δ, lies in the remaining class of ordered groups, then Gγ+1/Gγ contains a subgroup A generated by two elements α, β with a normal subgroup B such that A/B is a nonabelian torsion-free nilpotent group of class 2. Pick x, y ∈ Gγ+1 such that xGγ = α and yGγ = β, and let A be the subgroup of Gγ generated by x and y. Then A is a subgroup of Gγ containing a normal subgroup B such that A/B is a nonabelian torsion-free nilpotent group of class 2. By Proposition 3.3, k(A) ⊆ k(G) contains a free group k-algebra of rank 2 freely generated by symmetric elements with respect to the canonical involution. Thus we can suppose that G does not contain a free monoid and that Gγ+1/Gγ is abelian for all γ < δ. Let H be a finitely generated subgroup of G which is not abelian-by-finite. We claim that H is solvable. Indeed, by Lemma 3.4, we know that for each positive integer n, the derived subgroup H (n) is finitely generated. Let γn be the first ordinal such that H (n) ⊆ Gγn . Note that γn is not a limit ordinal. But, for each nonlimit ordinal γ = ε + 1, since Gε+1/Gε is abelian, we have that [Gε+1, Gε+1] ⊆ Gε. Thus, if H is not a solvable group, using the fact that H (n) is finitely generated and locally indicable, we get that {γn} is a strictly decreasing sequence of ordinals, which is impossible. Hence H is solvable. Moreover, since H does not contain a free monoid, [18, Theorems 4.7 and 4.12] or [12, Theorem 1] imply that H is nilpotent-by-finite, but not abelian-by-finite by hypothesis. Hence, being H torsion-free, it contains a torsion-free nilpotent subgroup L of class 2. Therefore, if follows from Theorem 1.1, that k(L) ⊆ k(G) contains a free group k-algebra of rank 2 freely generated by symmetric elements with respect to the canonical involution. (cid:3) As a first remark on Theorem 3.5, note that it generalizes [20, Theorem 3.1], but the proof of the latter is more elementary since no use of [6], [7] and [18] is made Secondly, if G is an orderable group, then condition (3) in Theorem 3.5 is equiv- alent to G being not abelian, because any abelian-by-periodic orderable group is abelian. Indeed, suppose that H is an abelian normal subgroup of G and that G/H is periodic. Let z ∈ H and let g ∈ G. There exists n ≥ 1 such that gn ∈ H. Thus zgn = gnz and (zgz−1)n = gn. Since roots are unique in an orderable group, we get that zg = gz. Because z ∈ H and g ∈ G are arbitrary, we get that H is contained in the center Z of G. It is well known that in an orderable group the quotient of the group by its center is orderable. Thus, G/Z ∼= (G/H)/(Z/H) is both orderable and periodic. Therefore G = Z, as desired. Corollary 3.6. Let k be a field and let G be a torsion-free polycylic group. Then the Ore division ring of fractions of the group algebra k[G] contains a free group FREE GROUP ALGEBRAS IN DIVISION RINGS 15 k-algebra of rank 2 freely generated by symmetric elements with respect to the canon- ical involution if and only if G is not abelian-by-finite. Proof. Let D denote the Ore division ring of fractions of k[G]. It is well known that G is poly-{infinite cyclic}-by-finite. Let H be a normal poly-{infinite cyclic} (and, thus, poly-orderable) subgroup of G of finite index. By Theorem 3.5, k(H) ⊆ D contains a free group k-algebra of rank 2 freely generated by symmetric elements with respect to the canonical involution if and only if H is not abelian-by-finite. If G is not abelian-by-finite, then H is not abelian-by-finite, and the result follows. The proof of the reverse implication is the same as the proof that condition (1) implies condition (3) in Theorem 3.5 (in which the hypothesis on G is not needed). (cid:3) Remark 3.7. In view of Propositions 3.1 and 3.3, we think it might be possible to prove that a Hughes-free division ring of fractions of the group algebra of a locally indicable group which is not locally abelian-by-finite will always contain a free group algebra freely generated by symmetric elements with respect to the canonical involution. The evidence we have for this is as follows. Let G be a locally indicable group, let k be a field and suppose that the group ring k[G] has a Hughes-free division ring of fractions D. Thus the canonical involution can be uniquely extended to D. We have proved that if there exists a finitely generated subgroup H of G such that either (i) H contains a free monoid of rank 2, or (ii) there exists a normal subgroup N of H such that H/N is torsion-free nilpo- tent of class 2, then D contains a free group k-algebra of rank 2 generated by symmetric elements. Incidentally, if G is a δ-poly-orderable group which is not locally abelian-by-finite, then it contains such a finitely generated subgroup H. Of course, if G is abelian- by-finite, then D cannot contain a free algebra of rank 2. Our methods do not apply in the remaining case: G is not locally abelian-by- finite and it does not contain a finitely generated subgroup H satisfying (i) or (ii). In this case, G does not contain a free monoid and each finitely generated subgroup H is either abelian-by-finite or not solvable. (Indeed, if H were solvable, then [18, Theorems 4.7 and 4.12] or [12, Theorem 1] would imply that H was nilpotent- by-finite. Being H torsion free, H would either be abelian-by-finite or contain a torsion-free nilpotent subgroup of class 2 and Theorem 1.1 would apply.) If H is neither abelian-by-finite nor solvable, then the n-th derived subgroup H (n) of H is finitely generated and H/H (n) is polycyclic for every positive integer n by Lemma 3.4. Moreover, since H (n) 6= {1} and H is locally indicable, H (n+1) ⊆ H (n) and H (n)/H (n+1) is a finitely generated abelian group which contains torsion-free elements for every positive integer n. Also H/H (n) is polycyclic with no noncom- mutative free monoid, and thus nilpotent-by-finite. Let N be a normal subgroup of H such that H (n) ⊆ N , N/H (n) is nilpotent and H/N is finite. Since the elements of finite order in a nilpotent group form a characteristic subgroup, there exists a subgroup H (n) ⊆ N + ⊆ N such that N +/H (n) is finite, N + is normal in H and N/N + torsion-free nilpotent. Thus N/N + has to be abelian because N is finitely generated and (ii) is not satisfied. We do not have examples of such a group G. 16 V. O. FERREIRA, J. Z. GONC¸ ALVES, AND J. S ´ANCHEZ 4. Further developments In this last section, we shall point to possible generalizations of the results ob- tained in this paper. We present two possible directions towards which new research can be done. 4.1. General involutions. We have seen in Theorem 3.5 that if k is a field and G is a nonabelian ordered group, then the division ring k(G) generated by k[G] inside the Malcev-Neumann series ring k((G)) has an involution induced by the canonical involution in G. Moreover, with respect to this involution, k(G) contains a free group k-algebra of rank 2 freely generated by symmetric elements. A natural question that arises is whether the same can be proved for other kinds of involutions. We shall address this question for involutions in k(G) which, although not being the canonical one, still are induced by involutions on G. We restrict to the nilpotent case and, in fact, raise more questions than answers. By and involution on a group G we understand a map ⋆ : G → G satisfying (i) (xy)⋆ = y⋆x⋆, for all x, y ∈ G, and (ii) (x⋆)⋆ = x, for all x ∈ G. If k is a field and ⋆ is an involution on a group G, then the group algebra k[G] = For instance, if G = hx, y : can be endowed with a k-involution, still denoted by ⋆, defined by (cid:0)Px∈G xax(cid:1)⋆ Px∈G x⋆ax, with x ∈ G and ax ∈ k, all but a finite number of which nonzero. [[x, y], x] = [[x, y], y] = 1i is the free nilpotent group of class 2 generated by two elements, considered in Section 1, then G has an involution ⋆ satisfying x⋆ = x−1 and y⋆ = y. The map x† = x, y† = y also induces an involution on G (satisfying [x, y]† = [x, y]−1). A third example of a non canonical involution ‡ on G is given by x‡ = y and y‡ = x. These involutions extend to k-involutions on k[G] and, therefore, to k-involutions on the Ore field of fractions D of k[G]. Clearly, the elements 1 + y(1 − y)−2 and 1 + y(1 − y)−2x(1 − x)−2y(1 − y)−2, found in Theorem 1.1, which freely generate a free group k-algebra in D, are sym- metric with respect to both ⋆ and †. The involution ‡ on G seems more mysterious and our methods do not provide an answer. We, thus, suggest the following prob- lem. Problem 4.1. Let k be a field and consider the group G = hx, y : [[y, x], x] = [[y, x], y] = 1i. Let D be the field of fractions of the group algebra k[G] and let ‡ denote the k-involution on D induced by the involution on G such that x‡ = y and y‡ = x. Find a pair of symmetric elements in D that freely generate a free group k-algebra. Rather, more generally, one could ask whether for every k-involution on D which is induced by an involution on G, there will always exist a pair of symmetric ele- ments freely generating a free group k-algebra inside D. 4.2. Twisted involutions. Given an involution ⋆ on a group G and a group ho- momorphism c : G → k× into the multiplicative group of a field k, the map k[G] −→ k[G] Px∈G xax 7−→ Px∈G x⋆c(x)ax, FREE GROUP ALGEBRAS IN DIVISION RINGS 17 where, for all x ∈ G, ax are elements of k all but a finite number of which nonzero, is a k-involution on k[G]. These kind of involutions in k[G] will be called twisted involutions. One might ask whether the results in this paper hold for general twisted invo- lutions or, particularly, for twisted involutions induced by the canonical involution on G. We have not explored twisted involutions throughly, but some of the results on the paper do have a version for them. For instance, the following version of Proposition 3.1 can be proved in the same way. Proposition 4.2. Let k be a field, let G be a locally indicable group and let c : G → k× be a group homomorphism. Suppose that the group algebra k[G] has a Hughes- free division ring of fractions D. If G contains a free monoid of rank 2, then D contains a free group k-algebra of rank 2 freely generated by symmetric elements with respect to the c-twisted canonical involution on D. Here we have called c-twisted canonical involution the extension to D of the twisted involution induced by the canonical involution on G and c. For the proof we need slight modifications of the free generators: we take f = c(x−1)x + x−1 and g = c(y−1)y + y−1 for the same x and y. References [1] J. P. Bell and D. Rogalski, Free subalgebras of quotient rings of Ore extensions, to appear in Algebra Number Theory (available at arXiv:1101.5829v2). [2] P. M. Cohn, Free Ideal Rings and Localization in General Rings, Cambridge University Press, Cambridge, 2006. [3] W. Dicks, D. Herbera and J. S´anchez, On a theorem of Ian Hughes about division rings of fractions, Comm. Algebra 32 (2004), 1127 -- 1149. [4] J. Z. Gon¸calves and M. Shirvani, A survey on free objects in division rings and in division rings with an involution, Comm. Algebra 40 (2012), 1704 -- 1723. [5] G. Higman, The units of group-rings, Proc. London Math. Soc. (2) 46 (1940), 231 -- 248. [6] I. Hughes, Division rings of fractions for group rings, Comm. Pure Appl. Math. 23 (1970), 181 -- 188. [7] I. Hughes, Division rings of fractions for group rings II, Comm. Pure Appl. Math. 25 (1972), 127 -- 131. [8] F. W. Levi, Contributions to the theory of ordered groups, Proc. Indian Acad. Sci., Sect. A. 17 (1943), 199 -- 201. [9] A. I. Lichtman, Free subgroups of normal subgroups of the multiplicative group of skew fields, Proc. Amer. Math. Soc. 71 (1978), 174 -- 178. [10] A. I. Lichtman, On matrix rings and linear groups over fields of fractions of group rings and enveloping algebras II, J. Algebra 90 (1984), 516 -- 527. [11] A. I. Malcev, On the embedding of group algebras in division algebras, Doklady Akad. Nauk SSSR (N.S.) 60 (1948), 1499 -- 1501. [12] P. Longobardi, M. Maj and A. H. Rhemtulla, Groups with no free subsemigroups, Trans. Amer. Math. Soc. 347 (1995), 1419 -- 1427. [13] L. Makar-Limanov, On free subobjects of skew fields, Methods in Ring Theory (Antwerp, 1983), 281 -- 285, NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., 129, Reidel, Dordrecht, 1984. [14] M. Lorenz, On free subalgebras of certain division algebras, Proc. Amer. Math. Soc. 98 (1986), 401 -- 405. [15] B. H. Neumann, On ordered division rings, Trans. Amer. Math. Soc. 66 (1949), 202 -- 252. [16] D. S. Passman, Infinite Crossed Products, Academic Press, Boston, 1989. [17] A. Rhemtulla and D. Rolfsen, Local indicability in ordered groups: braids and elementary amenable groups, Proc. Amer. Math. Soc. 130 (2002), 2569 -- 2577. [18] J. M. Rosenblatt, Invariant measures and growth conditions, Trans. Amer. Math. Soc. 193 (1974), 33 -- 53. 18 V. O. FERREIRA, J. Z. GONC¸ ALVES, AND J. S ´ANCHEZ [19] J. S´anchez, Localization: On Division Rings and Tilting Modules, PhD thesis, Universitat Aut`onoma de Barcelona, Barcelona, Spain, 2008 [20] J. S´anchez, Free group algebras in Malcev-Neumann skew fields of fractions, to appear in Forum Math. (doi:10.1515/form.2011.170). Department of Mathematics - IME, University of Sao Paulo, Caixa Postal 66281, Sao Paulo, SP, 05314-970, Brazil E-mail address: [email protected] Department of Mathematics - IME, University of Sao Paulo, Caixa Postal 66281, Sao Paulo, SP, 05314-970, Brazil E-mail address: [email protected] Department of Mathematics - IME, University of Sao Paulo, Caixa Postal 66281, Sao Paulo, SP, 05314-970, Brazil E-mail address: [email protected]
1601.04313	2	1601	2016-08-10T07:01:48	On nilpotent Lie algebras of derivations of fraction fields	[ "math.RA" ]	Let $K$ be an arbitrary field of characteristic zero and $A$ a commutative associative $ K$-algebra which is an integral domain. Denote by $R$ the fraction field of $A$ and by $W(A)=RDer_{\mathbb K}A,$ the Lie algebra of $\mathbb K$-derivations of $R$ obtained from $Der_{\mathbb K}A$ via multiplication by elements of $R.$ If $L\subseteq W(A)$ is a subalgebra of $W(A)$ denote by $rk_{R}L$ the dimension of the vector space $RL$ over the field $R$ and by $F=R^{L}$ the field of constants of $L$ in $R.$ Let $L$ be a nilpotent subalgebra $L\subseteq W(A)$ with $rk_{R}L\leq 3$. It is proven that the Lie algebra $FL$ (as a Lie algebra over the field $F$) is isomorphic to a finite dimensional subalgebra of the triangular Lie subalgebra $u_{3}(F)$ of the Lie algebra $Der F[x_{1}, x_{2}, x_{3}], $ where $u_{3}(F)=\{f(x_{2}, x_{3})\frac{\partial}{\partial x_{1}}+g(x_{3})\frac{\partial}{\partial x_{2}}+c\frac{\partial}{\partial x_{3}}\}$ with $f\in F[x_{2}, x_{3}], g\in F[x_3]$, $c\in F.$ In particular, a characterization of nilpotent Lie algebras of vector fields with polynomial coefficients in three variables is obtained.	math.RA	math	ON NILPOTENT LIE ALGEBRAS OF DERIVATIONS OF FRACTION FIELDS A.P. PETRAVCHUK Abstract. Let K be an arbitrary field of characteristic zero and A an integral K-domain. Denote by R the fraction field of A and by W (A) = RDerKA, the Lie algebra of K-derivations on R obtained from DerKA via multiplication by elements of R. If L ⊆ W (A) is a subalgebra of W (A) denote by rkRL the dimension of the vector space RL over the field R and by F = RL the field of constants of L in R. Let L be a nilpotent subalgebra L ⊆ W (A) with rkRL ≤ 3. It is proven that the Lie algebra F L (as a Lie algebra over the field F ) is isomorphic to a finite dimensional subalgebra of the triangular Lie subalgebra u3(F ) of the Lie algebra DerF [x1, x2, x3], where u3(F ) = {f (x2, x3) ∂ } with ∂x1 f ∈ F [x2, x3], g ∈ F [x3], c ∈ F. In particular, a characterization of nilpotent Lie algebras of vector fields with polynomial coefficients in three variables is obtained. + g(x3) ∂ ∂x2 + c ∂ ∂x3 1. Introduction Let K be an arbitrary field of characteristic zero and A an associative commutative K- algebra that is a domain. The set DerKA of all K-derivations of A is a Lie algebra over K and an A-module in a natural way: given a ∈ A, D ∈ DerKA, the derivation aD sends any element x ∈ A to a · D(x). The structure of the Lie algebra DerKA is of great interest because in case K = R and A = R[[x1, . . . , xn]], the ring of formal power series, the Lie algebra of all K-derivations of the form D = f1 ∂ ∂x1 + · · · + fn ∂ ∂xn , fi ∈ R[[x1, . . . , xn]] can be considered as the Lie algebra of vector fields on Rn with formal power series coefficients. Such Lie algebras with polynomial, formal power series, or analytical coefficients were studied by many authors. Main results for fields K = C and K = R in case n = 1 and n = 2 were obtained in [7] [4], [5] (see also [1], [3], [9], [10]). One of the important problems in Lie theory is to describe finite dimensional subalgebras of the Lie algebra W3(C) consisting of all derivations on the ring C[[x1, x2, x3]] of the form a1 ∂ ∂x1 + a2 ∂ ∂x2 + a3 ∂ ∂x3 , ai ∈ C[[x1, . . . , xn]]. In order to characterize nilpotent subalgebras of the Lie algebra W3(C) we consider more general situation. Let R = Frac(A) be the field of fractions of an integral domain A and W (A) = RDerK(A) the Lie algebra of derivations of the field R obtained from derivations on A by multiplying by elements of the field R (obviously DerKA ⊆ W (A)). For a subalgebra L of the Lie algebra W (A) let us define rkR(L) = dimR RL and denote by F = RL = {r ∈ Date: October 15, 2018. 2000 Mathematics Subject Classification. Primary 17B66; Secondary 17B05, 13N15. Key words and phrases. Lie algebra, vector field, nilpotent algebra, derivation. 1 2 A.P. PETRAVCHUK R D(r) = 0, ∀D ∈ L} the field of constants of the Lie algebra L. The K-space F L is a vector space over the field F and a Lie algebra over F. If L is a nilpotent subalgebra of W (A), then F L is finite dimensional over F (by Lemma 5). The main result of the paper: If L is a nilpotent subalgebra of rank k ≤ 3 over R from the Lie algebra W (A), then F L is isomorphic to a finite dimensional subalgebra of the triangular Lie algebra uk(F ) (Theorem 2). Triangular Lie algebras were studied in [1] and [2], they are locally nilpotent but not nilpotent, the structure of their ideals was described in these papers. We use standard notation, the ground field is arbitrary of characteristic zero. The quotient field of the integral domain A under consideration is denoted by R. Any derivation D of A can be uniquely extended to a derivation of R by the rule: D(a/b) = (D(a)b − aD(b))/b2. If F is a subfield of the field R and r1, . . . , rk ∈ R, then the set of all linear combinations of these elements with coefficients in F is denoted by F hr1, . . . , rki, it is a subspace of the F -space R. The triangular subalgebra un(K) of the Lie algebra Wn(K) = Der(K[x1, . . . , xn]) consists of all the derivations on the ring K[x1, . . . , xn] of the form D = f1(x2, . . . xn) ∂ + ∂x1 · · · + fn−1(xn) , where fi ∈ K[xi+1, . . . xn], fn ∈ K. + fn ∂ ∂xn−1 ∂ ∂xn 2. Some properties of nilpotent subalgebras of W (A) We will use some statements about derivations and nilpotent Lie algebras of derivations from the paper [8]. The next statement can be immediately checked. Lemma 1. Let D1, D2 ∈ W (A) and a, b ∈ R. Then it holds: 1. [aD1, bD2] = ab[D1, D2] + aD1(b)D2 − bD2(a)D1. 2. If a, b ∈ RD1 ∩ RD2, then [aD1, bD2] = ab[D1, D2]. Let L be a subalgebra of rank k over R of the Lie algebra W (A) and F = RL its field of constants. Denote by RL the set of all linear combinations over K of elements aD, where a ∈ R and D ∈ L. The set F L is defined analogously. Lemma 2. ([8], Lemma 2). Let L be a nonzero subalgebra of W (A) and let F L, RL be K-spaces defined as above. Then: 1. F L and RL are K-subalgebras of the Lie algebra W (A). Moreover, F L is a Lie algebra over the field F. 2. If the algebra L is abelian, nilpotent, or solvable then the Lie algebra F L has the same property, respectively. Lemma 3. ([8], Lemma 3). Let L be a subalgebra of finite rank over R of the Lie algebra W (A), Z = Z(L) the center of L, and F = RL the field of constants of L. Then rkRZ = dimF F Z and F Z is a subalgebra of the center Z(F L). In particular, if L is abelian, then F L is an abelian subalgebra of W (A) and rkRL = dimF F L. Lemma 4. ([8], Lemma 4). Let L be a subalgebra of the Lie algebra W (A) and I be an ideal of L. Then the vector space RI ∩ L (over K) is also an ideal of L. Lemma 5. ([8], Proposition 1, Theorem 1). Let L be a nilpotent subalgebra of W (A) and F = RL be its field of constants. Then: 1. If rkRL < ∞, then dimF F L < ∞. 2. If rkRL = 1, then L is abelian and dimF F L = 1. ON NILPOTENT LIE ALGEBRAS OF DERIVATIONS OF FRACTION FIELDS 3 3. If rkRL = 2, then there exist elements D1, D2 ∈ F L and a ∈ R such that F L = F hD1, aD1, . . . , ak k! D1, D2i, k ≥ 0 (if k = 0, then put F L = F hD1, D2i), where [D1, D2] = 0, D1(a) = 0, D2(a) = 1. Lemma 6. Let D1, D2, D3 ∈ W (A) and a ∈ R be such elements that D1(a) = D2(a) = 0, D3(a) = 1 and let F = ∩3 i=1RDi. If there exists an element b ∈ R such that D1(b) = D2(b) = 0, D3(b) ∈ F h1, a, . . . , as/s!i for some s ≥ 0, then b ∈ F h1, a, . . . , as+1/(s + 1)!i. Proof. Write down D3(b) = Ps Ps tions of Lemma) D1(b − c) = 0 and D2(b − c) = 0. Then we have b − c ∈ ∩3 and therefore b = γ + Ps b ∈ F h1, a, . . . , as+1/(s + 1)!i. i=0 βiai/i! with βi ∈ F and take the element c = i=0 βiai+1/(i + 1)! of the field R. It holds obviously D3(b − c) = 0 and (by the condi- i=1RDi = F, i=0 βiai+1/(i + 1)! for some element γ ∈ F. The latter means that (cid:3) Lemma 7. Let D1, D2, D3 ∈ W (A) and a, b ∈ R be such elements that D1(a) = D1(b) = 0, i=1RDi. If there exists an element D2(a) = 1 D2(b) = 0, D3(a) = 0, D3(b) = 1 and let F = ∩3 c ∈ R such that D1(c) = 0, [D2, D3](c) = 0, D2(c) ∈ F h{ aibj i!j! }, 0 ≤ i ≤ m − 1, 0 ≤ j ≤ ki, D3(c) ∈ F h{ aibj i!j! }, 0 ≤ i ≤ m, 0 ≤ j ≤ k − 1i, then c ∈ F h{ aibj i!j! }, 0 ≤ i ≤ m, 0 ≤ j ≤ ki ∂ag(a, b), D3(f ) = ∂ Proof. The elements D2(c) and D3(c) can be written (by conditions of the lemma) in the form D2(c) = f (a, b), D3(c) = g(a, b) where f, g ∈ F [u, v] are some polynomials it holds D2(g) = D3(f ). It follows from the relations of u, v. Since [D2, D3](c) = 0, D2(g) = ∂ ∂bf (a, b). Hence there exists a poly- nomial h(a, b) ∈ F [a, b] (the potential of the vector field f (a, b) ∂ ∂b) such that D3(h(a, b)) = g, D2(h(a, b)) = f. The polynomial h(a, b) is obtained from the polynomials f, g by formal integration on a and on b, so we have h(a, b) ∈ F h{ aibj i!j! }, 0 ≤ i ≤ m, 0 ≤ j ≤ ki. Further, using properties of the element h(a, b) we get D2(h − c) = D3(h − c) = 0. Besides, it holds D1(h − c) = 0. The latter means that h − c ∈ F = ∩3 i=1RDi. But then c = γ + h for some γ ∈ F and therefore c ∈ F h{ aibj i!j! }, 0 ≤ i ≤ m, 0 ≤ j ≤ ki. (cid:3) ∂bf (a, b) that ∂ ∂a g(a, b) = ∂ ∂a + g(a, b) ∂ 3. On nilpotent subalgebras of small rank of W (A) Lemma 8. Let L be a nilpotent subalgebra of rank 3 over R from the Lie algebra W (A) and F = RL be its field of constants. If the center Z(L) of the algebra L is of rank 2 over R and dimF F L ≥ 4, then there exist D1, D2, D3 ∈ L, a ∈ R such that the Lie algebra F L is contained in a nilpotent Lie algebra eL of the Lie algebra W (A) of the form eL = F hD3, D1, aD1, . . . , (an/n!)D1, D2, aD2, . . . , (an/n!)D2i for some n ≥ 1, with [Di, Dj] = 0, i, j = 1, 2, 3, D1(a) = D2(a) = 0, D3(a) = 1. Proof. Take any elements D1, D2 ∈ Z(L) that are linearly independent over R and denote I = (RD1 + RD2) ∩ L. Then I is an ideal of the Lie algebra L (by Lemma 4). Take an arbitrary element D ∈ I and write down D = a1D1 + a2D2 for some elements ai ∈ R. Since [Di, D] = 0 = Di(a1)D1 + Di(a2)D2, i = 1, 2 we get Di(aj) = 0, i, j = 1, 2. It follows easily 4 A.P. PETRAVCHUK that for any element D′ ∈ I it holds the equality [D, D′] = 0, so the ideal I is abelian. The Lie algebra F L is finite dimensional over F and dimF F L/F I = 1 by Lemma 5. Take any element D3 ∈ L \ I. Then F L = F I + F D3 and D1, D2, D3 are linearly independent over R. Since rkRZ(L) = 2, (by conditions of the lemma) we have dimF F Z(L) = 2 by Lemma 3. The ideal I of the Lie algebra L is abelian by the above proven, so the ideal F I of the Lie algebra F L over the field F is also abelian. Since F L = F I + F D3, there exists a basis of the F -space F I in which the nilpotent linear operator adD3 has a matrix consisting of two Jordan blocks. Let J1 and J2 be the correspondent Jordan bases; without loss of generality one can assume that D1 ∈ J1, D2 ∈ J2 and the elements D1, D2 are the first members of the bases J1 and J2 respectively. If dimF F hJ1i = dimF F hJ2i = 1, then F L = F hD3, D1, D2i is of dimension 3 over F which contradicts the conditions of the lemma. So, we may assume that dimF F hJ1i ≥ dimF F hJ2i and dimF F hJ1i = n+1, n ≥ 1. Denote the elements of the basis J1 by D1, a1D1 + b1D2, . . . , anD1 + bnD2, where the elements ai, bi belong to R and put for convenience a = a1. Let us prove by induction on i that ai, bi ∈ F h1, a, . . . , ai/i!i. If i = 1, then a1 = a ∈ F h1, ai by definition. It follows from the relation [D3, a1D1 + b1D2] = D1 = D3(a1)D1 + D3(b1)D2 that D3(b1) = 0. Since F I is abelian (by the above proven), we have D1(b1) = D2(b1) = 0. The latter means that b1 ∈ F ⊂ F h1, ai. Further, the relation [D3, aiD1 + biD2] = ai−1D1 + bi−1D2 = D3(ai)D1 + D3(bi)D2 gives the equalities D3(ai) = ai−1 and D3(bi) = bi−1. By the inductive assumption, ai−1, bi−1 ∈ F h1, a, . . . , ai−1/(i − 1)!i and taking into account the relations Dj(ai) = Dj(bi) = 0, j = 1, 2 (they hold because F I is abelian) we get by Lemma 6 that ai, bi ∈ F h1, . . . , ai/i!i. The latter relation means that the F -subspace F hJ1i of F I lies in the subalgebra eL from the conditions of the lemma. Now let J2 = {D2, c1D1 + d1D2, . . . , ckD1 + dkD2} be a basis corresponding to the second Jordan block. The relation [D3, c1D1 + d1D2] = D2 implies the equality D3(d1) = 1 and therefore D3(a−d1) = 0. Since D1(a−d1) = D2(a−d1) = 0, we get a − d1 ∈ F, i.e. d1 = a + γ for some γ ∈ F. Applying the above considerations to the Jordan basis J2 we obtain that F hJ2i ⊂ eL. But then the Lie algebra L is entirely contained in eL. Lemma 9. Let L be a nilpotent subalgebra of rank 3 over R from the Lie algebra W (A) and F = RL be its field of constants. If the center Z(L) of the algebra L is of rank 1 over R and dimF F L ≥ 4, then the Lie algebra F L is contained either in the nilpotent Lie algebra eL from the conditions of Lemma 8 or in a subalgebra L of W (A) of the form (cid:3) L = F hD3, D2, aD2, . . . , (an/n!)D2, { aibj i!j! D1}, 0 ≤ i, j ≤ mi where n ≥ 0, m ≥ 1, Di ∈ L, [Di, Dj] = 0, for i, j = 1, 2, 3, and a, b ∈ R such that D1(a) = D2(a) = 0, D3(a) = 1, D1(b) = D3(b) = 0, D2(b) = 1. Proof. Take any nonzero element D1 ∈ Z(L) and denote I1 = RD1 ∩ L. Then I1 is an abelian ideal of the algebra Lie L and rkRI1 = 1 by Lemma 3. Choose any nonzero element ON NILPOTENT LIE ALGEBRAS OF DERIVATIONS OF FRACTION FIELDS 5 D2 + I1 in the center of the quotient Lie algebra L/I1 and denote I2 = (RD1 + RD2) ∩ L. By the same Lemma 3, I2 is an ideal of the Lie algebra L and rkRI2 = 2. Further, take any element D3 ∈ L \ I2. Since dimF F L/F I2 = 1 by Lemma 5 from the paper [8], we have F L = F I2 + F D3. Case 1. The ideal I2 is abelian. Let us show that F L is contained in the Lie algebra eL from the conditions of Lemma 8. It is obvious that F I2 is an abelian ideal of codimension 1 of the Lie algebra F L over the field F. By Lemma 3, rkRZ(L) = dimF F Z(L) and by the conditions of the lemma, we see that dimF F Z(L) = 1. The linear operator adD3 acts on the F -space F I2 and dimF Ker(adD3) = dimF F Z(L). Therefore dimF Ker(adD3) = 1 and there exists a basis of F I2 in which adD3 has a matrix in the form of a single Jordan block. The same is true for the action of adD3 on the vector space F I1 (since [D3, I1] ⊆ I1, the ideal F I1 is invariant under adD3). The subalgebra F I1 + F D3 is of rank 2 over R. If dimF F I1 > 1, then the center of the Lie algebra F1 + F D3 is of dimension 1 over F. By Lemma 5, there exists a Jordan basis in F I1 of the form {D1, aD1, . . . , (as/s!)D1}, where s ≥ 0, D3(a) = 1, [D3, D1] = 0. If dimF F I1 = 1, then s = 0 and the desired basis of F1 is of the form {D1}. Let first s > 0. Since (F D2 +F I1)/F I1 is a central ideal of the quotient algebra F L/F I1, we have [D3, D2] ∈ F I1 and hence one can write [D3, D2] = γ0D1 +. . .+(γsas/s!)D1 for some γi ∈ F. Taking D2 −Ps−1 i=0 (γiai+1/(i+1)!)D1 instead D2 we may assume that [D3, D2] = γsas/s!D1. Note that γs 6= 0. Really, in the opposite case [D3, D2] = 0 and therefore D2 ∈ Z(L). Then rkRZ(L) = 2 which is impossible because of the conditions of the lemma. After changing D3 by γ −1 s D3 we may assume that [D3, D2] = (as/s!)D1. Since the linear operator adD3 has in a basis of the F -space F I2 a matrix, consisting of a single Jordan block, the same is true for the linear operator adD3 on the vector space F I2/F I1. Let dimF F I2/F I1 = k and {S 1, . . . , Sk} be a Jordan basis for adD3 on F I2/F I1, where Si = (ciD1 +diD2)+F I1, i = 1, . . . , k, ci, di ∈ R. The representatives ciD1 +diD2 of the cosets Si can be chosen in such a way that [D3, ciD1 + diD2] = ci−1D1 + di−1D2, i = 2, . . . , k and [D3, c1D1 + d1D2] = βi(ai/i!)D1 sX (1) (2) for some βi ∈ F. Let us show by induction on i that the relations hold: di ∈ F h1, . . . , ai−1/(i − 1)!i, ci ∈ F h1, . . . , as+i/(s + i)!i. i=0 Really, for i = 1 it follows from the relation (1) that [D3, c1D1 + d1D2] = sX i=0 βiai/i!D1 = D3(c1)D1 + (d1as/s!)D1 + D3(d1)D2. (3) It follows from (3) that D3(d1) = 0, and since the ideal F I2 is abelian, it holds D1(d1) = D2(d1) = 0. The latter means that d1 ∈ F = F h1i. We also get from (3) that D3(c1) ∈ F h1, . . . , as/s!i and obviously it holds D1(c1) = D2(c1) = 0. Then, by Lemma 6, c1 ∈ F h1, a, . . . , as+1/(s+1)!i and the relations (2) hold for i = 1. Assume they hold for i−1. Let us prove that the relations (2) hold for i. Using the equalities [D3, ciD1 +diD2] = ci−1D1 +di−1D2 and [D3, D2] = as/s!D1 we get D3(di) = di−1, D3(ci) = dias/s! + ci−1. By the inductive as- sumption, we have di−1 ∈ F h1, a, . . . , ai−2/(i − 2)!i, hence di ∈ F h1, a, . . . , ai−1/(i − 1)!i by 6 A.P. PETRAVCHUK Lemma 6. Analogously, by the inductive assumption it holds ci−1 ∈ F h1, a, . . . , as+i−1/(s + i − 1)!i and therefore D3(ci) ∈ F h1, a, . . . , as+i−1/(s + i − 1)!i. Since D1(ci) = D2(ci) = 0 we get by Lemma 6 that ci ∈ F h1, a, . . . , as+i/(s + i)!i. But then we have inclusion F I2 ⊆ F hD1, aD1, . . . , (as+k/(s + k)!)D1, D2, aD2, . . . , (ak/k!)D2i. The last subalgebra of the Lie algebra W (R) is contained in the subalgebra of the form F hD1, aD1, . . . , (as+k/(s + k)!)D1, D2, aD2, . . . , (as+k/(s + k)!)D2i. But then the Lie algebra L is contained in the subalgebra eL from the conditions Lemma 8. Let now s = 0. Then F I1 = F D1 and without loss of generality we may assume that [D3, D2] = D1. Repeating the above considerations we can build a Jordan basis {(ciD1 + diD2) + F I1, i = 1, . . . , k} of the quotient algebra F I2/F I1 with [D3, ciD1 + diD2] = ci−1D1 + di−1D2, i = 2, . . . , k and [D3, c1D1 + d1D2] = αD1 for some α ∈ F. It follows from the last equality that D3(d1) = 0 and taking into account the equalities D1(d1) = 0 and D2(d1) = 0 we see that d1 ∈ F. Since a1D1 + d1D2 6∈ F I1, we have d1 6= 0. By conditions of the lemma, dimF F L > 3, so we have k ≥ 2 and the relation [D3, c2D1 + d2D2] = c1D1 + d1D2 implies the equality D3(d2) = d1. But then D3(d2d−1 1 ) = 1 and multiplying all the elements of the Jordan basis considered above by d−1 1 we may assume that D3(d2) = 1. Denoting a = d2 and repeating the considerations from the subcase s > 0 we see that the Lie algebra L is contained in the subalgebra eL from the conditions of Lemma 8. Case 2. The ideal I2 is nonabelian. We may assume without loss of generality that I1 coincides with its centralizer in L, i.e. CL(I1) = I1. Really, let CL(I1) ⊃ I1 with strong containment. Choose a one-dimensional (central) ideal (D4 + I1)/I1 in the ideal CL(I1)/I1 of the quotient algebra L/I1. Then I4 := (RD1 + RD4) ∩ L is an abelian ideal of rank 2 of the algebra L and dimF F L/F I4 = 1 by Lemma 5 from [8]. Thus the problem is reduced to the case 1 (one should take F I4 instead of F I2). So, we assume that CL(I1) = I1. It follows from this equality that CF L(F I1) = F I1. As in the case 1 we write F L = F I2 + F D3 and [D3, D2] = rD1 for some r ∈ R. Since the ideal F I1 is abelian, the linear operator ad[D3, D2] = ad(rD1) acts trivially on the vector space F I1, and therefore the linear operators adD2 and adD3 commute on F I1. Denote by M2 the kernel Ker(adD2) on the F -space F I1. It is obvious that M2 is an abelian subalgebra of F I1 and M2 is invariant under the action of adD3. Since [D1, M2] = [D2, M2] = 0 the linear operator adD3 has on the F -space F I1 the kernel of dimension 1 (in other case the center of the Lie algebra F L would have dimension ≥ 2 over F which contradicts our assumption). Using Lemma 5 one can easily show that M2 = F hD1, aD1, . . . , (ak/k!)D1i for some a ∈ R with D1(a) = 0, D2(a) = 0, D3(a) = 1 (if k = 0, then put M2 = F D1). Further denote M3 = Ker(adD3) on the vector space F I1. As above one can prove that M3 is invariant under action of adD2, this linear operator has one-dimensional kernel on M3, and for some b ∈ R with D1(b) = D3(b) = 0 and D2(b) = 1 (if m = 0 put M3 = F D1). M3 = F hD1, bD1, . . . , (bm/m!)D1i ON NILPOTENT LIE ALGEBRAS OF DERIVATIONS OF FRACTION FIELDS 7 Take now any element cD1 of the ideal F I1, c ∈ R. Since the linear operators adD2 and adD3 act nilpotently on F I1, there exist the least positive integers k0 and m0 (depending on the element cD1) such that (adD2)k0(cD1) = 0, (adD3)m0(cD1) = 0. Let us show by induction on s = m0 + k0 that the element cD1 is a linear combination (with coefficients from F ) of elements of the form aibj i!j! D1 ∈ W (A) for some 0 ≤ i ≤ k0 − 1, 0 ≤ j ≤ m0 − 1 (note that the elements aibj i!j! D1 can be outside of F I1). If s = 2 (obviously s ≥ 2), then we must only consider the case m0 = 1, k0 = 1. In this case, we have [D3, cD1] = 0, [D2, cD1] = 0. These equalities imply that cD1 ∈ Z(F L) = F D1 and all is done. Let s ≥ 3. The element [D2, cD1] can be written by the inductive assumption in the form Analogously we get [D2, cD1] = k0−2X m0−1X i=0 j=0 γij aibj i!j! D1 for some γij ∈ F. [D3, cD1] = k0−1X m0−2X i=0 j=0 δij aibj i!j! D1 for some δij ∈ F. It follows from the previous two equalities that D2(c) = k0−2X m0−1X i=0 j=0 γij aibj i!j! , D3(c) = k0−1X m0−2X i=0 j=0 δij aibj i!j! . Note that [D3, D2](c) = rD1(c) = 0 and therefore by Lemma 7 c ∈ F h aibj i!j! , 0 ≤ i ≤ k0 − 1, 0 ≤ j ≤ m0 − 1i. Since cD1 is arbitrarily chosen we have F I1 ⊆ F h aibj i!j! D1, 0 ≤ i ≤ k0 − 1, 0 ≤ j ≤ m0 − 1i. One can straightforwardly check that k0 ≤ k, where k = dim M2 − 1 and analogously m0 ≤ m = dim M3 − 1. Let, for example, m ≥ n. Then F I1 ⊆ F h aibj i!j! D1, 0 ≤ i, j ≤ mi. Further, by the above proven, the linear operator adD3 on the vector space F I2/F I1 has a matrix in a basis in the form of a single Jordan block. This basis can be chosen in the form (u1D1 + v1D2) + F I1, . . . , (utD1 + vtD2) + F I1 such that [D3, uiD1 + viD2] = ui−1D1 + vi−1D2, i ≥ 2, [D3, u1D1 + v1D2] = f D1 (4) for some element f, f ∈ F h aibj i!j! , 0 ≤ i, j ≤ mi. Let us show by induction on s that us ∈ F h aibj i!j! If s = 1, then the equalities , 0 ≤ i, j ≤ m + si, vs ∈ F h1, . . . , as−1/(s − 1)!i. [D3, u1D1 + v1D2] = f D1 = D3(u1)D1 + D3(v1)D2 + v1rD1 (5) imply D3(v1) = 0 (let us recall here that [D3, D2] = rD1). Taking into account the relations [D1, u1D1 + v1D2] = 0 and [D2, u1D1 + v1D2] ∈ F I1 we obtain that v1 ∈ ∩3 i=1RDi = F, that is v1 ∈ F h1i. It follows from the relations (4) that D3(u1) + v1r ∈ F h{ aibj i!j! }, 0 ≤ i, j ≤ m.i. 8 A.P. PETRAVCHUK Analogously the inclusion [D2, u1D1 + v1D2] ∈ F I1 implies the relation D2(u1) ∈ F h{ aibj i!j! }, 0 ≤ i, j ≤ m.i. Since [D3, D2] = rD1 and rD1(u1) = 0, we see (using Lemma 7) that u1 ∈ F h{ aibj i!j! }, 0 ≤ i, j ≤ m + 1.i. By inductive assumption, we have us−1 ∈ F h{ aibj i!j! }, 0 ≤ i, j ≤ m + s − 1i, vs−1 ∈ F h1, . . . , as−2 (s − 2)! i. Note that the relations (4) imply the equalities D3(us) = us−1 − rvs, D3(vs) = vs−1 (here [D3, D2] = rD1). Analogously it follows from the relation [D2, usD1 + vsD2] ∈ F I1 that D2(vs) = 0, D2(us) ∈ F h{ aibj i!j! D1}, 0 ≤ i, j ≤ mi. i!j! }, 0 ≤ i, j ≤ m + s − 1i. But then by Lemma 7 us ∈ F h{ aibj Since D1 ∈ Z(L), we have the equalities D1(us) = D1(vs) = 0. Therefore we get by Lemma 6 that vs ∈ F h1, . . . , as−1/(s − 1)!i. By Lemma 7, us ∈ F h{ aibj i!j! }, 0 ≤ i, j ≤ m + si (since D3(us) ∈ F h{ aibj i!j! }, 0 ≤ i, j ≤ m + s − 1i by the relations (4)). Since rD1 ∈ F I1, we have rvs ∈ F h{ aibj i!j! }, 0 ≤ i, j ≤ m + si. So, we have proved that the Lie algebra L is contained in the subalgebra L from the conditions of the lemma. To finish with the proof we must prove that the element D2 can be chosen in W (A) in such a way that [D3, D2] = 0. Take the element D2 − r0D1 instead D2, where the element r0 is obtained from r by formal integration on variable a (recall that r ∈ F h{ aibj (cid:3) i!j! }, 0 ≤ i, j ≤ mi). Then [D3, D2] = 0. The proof is complete. Theorem 1. Let K be a field of characteristic zero, A an integral K-domain with fraction field R. Denote by W (A) the subalgebra RDerKA of the Lie algebra DerKR. Let L be a nilpotent subalgebra of rank 3 over R from W (A) and F = RL its field of constants. If dimF F L ≥ 4, then there exist integers n ≥ 0, m ≥ 0, elements D1, D2, D3 ∈ F L such that [Di, Dj] = 0, i, j = 1, 2, 3 and the Lie algebra F L is contained in one of the following subalgebras of the Lie algebra W (A) : 1) L1 = F hD3, D1, aD1, . . . , (an/n!)D1, D2, aD2, . . . , (an/n!)D2i, where a ∈ R is such that D1(a) = D2(a) = 0, D3(a) = 1. 2) L2 = F hD3, D2, aD2, . . . , (an/n!)D2, { aibj i!j! D1}, 0 ≤ i, j ≤ mi where a, b ∈ R are such that D1(a) = D2(a) = 0, D3(a) = 1, D1(b) = D3(b) = 0, D2(b) = 1. As a corollary we get the next characterization of nilpotent Lie algebras of rank ≤ 3 from the Lie algebra W (A). Theorem 2. Under conditions of Theorem 1, every nilpotent subalgebra L of rank k ≤ 3 over R from the Lie algebra W (A) is isomorphic to a finite dimensional subalgebra of the triangular Lie algebra uk(F ). ON NILPOTENT LIE ALGEBRAS OF DERIVATIONS OF FRACTION FIELDS 9 Proof. If k = 1 then the Lie algebra F L is one-dimensional over F and therefore is isomorphic to u1(F ) = F ∂ ∂x1 . In the case k = 2, the Lie algebra F L is (by Lemma 4) of the form F L = F hD1, aD1, . . . , ak k! D1, D2i, k ≥ 0 (if k = 0, then put F L = F hD1, D2i), 7→ ∂ ∂xi where [D1, D2] = 0, D1(a) = 0, D2(a) = 1. The Lie algebra F L is isomorphic to a suitable subalgebra of the triangular Lie algebra u2(F ) = {f (x2) ∂ } : the correspondence ∂x1 , i = 1, 2 and a 7→ x2 can be extended to an isomorphism between F L and a Di subalgebra of u2(F ). Let now k = 3. If dimF F L = 3, then F L is either abelian or has a basis D1, D2, D3 with multiplication rule [D3, D2] = D1, [D2, D1] = [D3, D1] = 0. In the first case, F L is isomorphic to the subalgebra F h ∂ i, in the second case it is isomorphic to the ∂x1 subalgebra F h ∂ ∂x1 i of the triangular Lie algebra u3(F ). + F ∂ ∂x2 + ∂ ∂x2 , ∂ ∂x3 , ∂ ∂x2 , ∂ ∂x3 , x3 ∂ ∂x1 Let now dimF F L ≥ 4. The Lie algebra F L is contained (by Theorem 1) in one of the Lie algebras L1 or L2 from the statement of that theorem. Note that the Lie algebra L1 is isomorphic to the subalgebra L1 of the Lie algebra u3(F ) of the form L1 = F h ∂ ∂x3 , ∂ ∂x1 , . . . , (xn 3 /n!) ∂ ∂x1 , ∂ ∂x2 , . . . , (xn 3 /n!) ∂ ∂x2 i, Analogously the Lie algebra L2 is isomorphic the the subalgebra L2 of u3(F ) of the form L2 = F h ∂ ∂x3 , ∂ ∂x2 , . . . , (xn 3 /n!) ∂ ∂x2 , { 2xj xi 3 i!j! ∂ ∂x1 }, 0 ≤ i, j ≤ mi. (cid:3) Corollary 1. Let L be a nilpotent subalgebra of the Lie algebra W3(K) = Der(K[x1, x2, x3]) and F the field of constants for the Lie algebra L in the field K(x,x2, x3). Then the Lie algebra F L (over the field F ) is isomorphic to a finite dimensional subalgebra of the triangular Lie algebra u3(F ). Remark 1. If L is a nilpotent subalgebra of rank 3 over R from the Lie algebra W3(K), then L being isomorphic to a subalgebra of the triangular Lie algebra u3(K) can be not conjugated (by an automorphism of W3(K)) with any subalgebra of u3(K). Indeed, the subalgebra L = i is nilpotent but not conjugated with any subalgebra of u3(K) (L is Khx1 selfnormalized in W3(K), but any finite dimensional subalgebra of u3(K) is not, because of locally nilpotency of the Lie algebra u3(K)). , x2 , x3 ∂x3 ∂x1 ∂x2 ∂ ∂ ∂ The author is grateful to V. Bavula and V.M.Bondarenko for useful discussions and advice. References [1] V.V. Bavula, Lie algebras of triangular polynomial derivations and an isomorphism criterion for their Lie factor algebras Izv. RAN. Ser. Mat., 77 (2013), Issue 6, 3-44. [2] V. V. Bavula, "The groups of automorphisms of the Lie algebras of triangular polynomial derivations", J. Pure Appl. Algebra, 218:5 (2014), 829 -- 851. [3] J. Draisma, Transitive Lie algebras of vector fields: an overview. Qual. Theory Dyn. Syst. 11 (2012), no. 1, 39-60. [4] A. Gonz´alez-L´opez, N. Kamran and P.J. Olver, Lie algebras of differential operators in two complex variavles, Amer. J. Math., 1992, v.114, 1163-1185. 10 A.P. PETRAVCHUK [5] A. Gonz´alez-L´opez, N. Kamran and P.J. Olver, Lie algebras of vector fields in the real plane. Proc. London Math. Soc. (3) 64 (1992), no. 2, 339-368. [6] D.A. Jordan, On the ideals of a Lie algebra of derivations. J. London Math. Soc. (2) 33 (1986), no. 1, 33 -- 39. [7] S. Lie, Theorie der Transformationsgruppen, Vol. 3, Leipzig, 1893. [8] Ie. O. Makedonskyi and A.P. Petravchuk, On nilpotent and solvable Lie algebras of derivations, J. Algebra, 401 (2014), 245-257. [9] R.O. Popovych, V.M. Boyko, M.O. Nesterenko and M.W. Lutfullin, Realizations of real low-dimensional Lie algebras, J. Phys. A 36 (2003), no. 26, 7337-7360. [10] G. Post, On the structure of graded transitive Lie algebras, J. Lie Theory, 2002, V.12, N 1, 265 -- 288. Anatoliy P. Petravchuk: Department of Algebra and Mathematical Logic, Faculty of Mechanics and Mathematics, Kyiv Taras Shevchenko University, 64, Volodymyrska street, 01033 Kyiv, Ukraine E-mail address: [email protected] , [email protected]
1601.03438	1	1601	2016-01-13T22:59:21	Modules with ascending chain condition on annihilators and Goldie modules	[ "math.RA" ]	Using the concepts of prime module, semiprime module and the concept of ascending chain condition (ACC) on annihilators for an $R$-module $M$ . We prove that if \ $M$ is semiprime \ and projective in $\sigma \left[ M\right] $, such that $M$ satisfies ACC on annihilators, then $M$ has finitely many minimal prime submodules. Moreover if each submodule $N\subseteq M$ contains a uniform submodule, we prove that there is a bijective correspondence between a complete set of representatives of isomorphism classes of indecomposable non $M$-singular injective modules in $\sigma \left[ M\right] $ and the set of minimal primes in $M$. If $M$ is Goldie module then $% \hat{M}\cong E_{1}^{k_{1}}\oplus E_{2}^{k_{2}}\oplus ...\oplus E_{n}^{k_{n}}$ where each $E_{i}$ is a uniform $M$-injective module. As an application, new characterizations of left Goldie rings are obtained.	math.RA	math	Modules with ascending chain condition on annihilators and Goldie modules Jaime Castro P´erez Escuela de Ingenier´ıa y Ciencias Instituto Tecnol´ogico y de Estudios Superiores de Monterrey Calle del Puente 222, Tlalpan 14380 M´exico, D.F. M´exico Mauricio Medina B´arcenas and Jos´e R´ıos Montes Instituto de Matem´aticas Universidad Nacional Aut´onoma de M´exico Area de la Investigaci´on Cient´ıfica Circuito Exterior, C.U. 04510 M´exico, D.F. M´exico Abstract Using the concepts of prime module, semiprime module and the concept of ascending chain condition (ACC) on annihilators for an R-module M . We prove that if M is semiprime and projective in σ [M], such that M satisfies ACC on annihilators, then M has finitely many minimal prime submodules. Moreover if each submod- ule N ⊆ M contains a uniform submodule, we prove that there is a bijective correspondence between a complete set of representatives of isomorphism classes of indecomposable non M-singular injective modules in σ [M] and the set of minimal primes in M. If M is Goldie n where each Ei is a uniform M-injective module. As an application, new characterizations of left 2 ⊕ ... ⊕ Ekn module then cM ∼= Ek1 1 ⊕ Ek2 Goldie rings are obtained. Key Words: Prime modules; Semiprime modules; Goldie Mod- ules; Indecomposable Modules; Torsion Theory 1 2000 Mathematics Subject Classification: 16S90; 16D50; 16P50; 16P70 Introduction The Rings with ascending chain condition (ACC) on annihilators has been stud- ied by several authors. For non commutative semiprime rings A.W. Chatters and C. R. Hjarnavis in [6] showed that if R is semiprime ring such that R satis- fies ACC on annihilator ideals, then R has finitely many minimal prime ideals. K.R. Goodearl and R. B. Warfield in [8] showed that if R is a semiprime right Goldie ring, then R has finitely many minimal prime ideals. In this paper we work in a more general situation, using the concept of prime and semiprime submodules defined in [11], [13] and the concept of as- cending chain condition (ACC) on annihilators for a R-module M defined in [4], we show that if M is projective in σ [M] and semiprime module, such that M satisfies ACC on annihilators, then M has finitely many minimal prime submod- ules in M. Moreover if each submodule N ⊆ M contains a uniform submodule, we prove that there is a bijective correspondence between a complete set of iso- morphism classes of indecomposable non M-singular injective modules in σ [M] Eχ(M )(M) and the set of minimal prime in M submodules SpecM in (M). Also we prove that there is a bijective correspondence between sets Eχ(M )(M) and P M in M , where P M in M = {χ (M/P ) P ⊆ M is a minimal prime in M}. So we ob- tain new information about semiprime rings with ACC on annihilators. As an application, of these results we obtain a new characterization of Goldie Rings in terms of the of minimal prime idelas of R and a new characterization of these Rings in terms of the continuous modules with ACC on annihilators. In order to do this, we organized the article in three sections. Section 1, we give some results about prime and semiprime modules and we define the concept right annihilator for a submodule N of M. In particular we prove in Proposition 1.16, that if M is projective in σ [M] and M is semiprime mod- ule, then AnnM (N) = Annr M (N) for all N ⊆ M. In Section 2, we define the concept of ascending chain condition (ACC) on annihilators for an R-module 2 M and we give the main result of this paper Theorem 2.2, we prove that if M is a semiprie module projective in σ [M] and M satisfies ACC on annihilators, then M has finitely many minimal prime submodules in M. Moreover if P1, P2, ..., Pn are the minimal prime in M submodules, then P1 ∩ P2∩... ∩Pn = 0. In Theorem 2.7, we show that there is a bijective correspondence between sets Eχ(M )(M) and SpecM in (M). Moreover in Corollary 2.13 we obtain an applica- tion for semiprime rings. We prove that if R is a semiprime ring such R satisfies ACC on left annihilator and each 0 6= I ⊆ R left ideal of R, contains a uni- form left ideal of R, then there is a bijective correspondence between the set of representatives of isomorphism classes of indecomposable non singular injective R-modules and the set of minimal prime ideals of R. In particular in Corollary 2.15 we show, that for R a semiprime left Goldie ring the bijective correspon- dence above mentioned is true. Moreover in this Corollary we show that R has local Gabriel correspondence with respect to χ (R) = τg. Other important re- sults are the Theorem 2.18 and 2.20. In the first Theorem we show that, if M satisfies ACC on annihilators and for each 0 6= N ⊆ M, N contains a uniform M (Pi) for 1 ≤ i ≤ n. that if R is a semiprime ring, such that R satisfies ACC on left annihilators and for each non zero left ideal I ⊆ R, I contains a uniform left ideal, then the following conditions are equivalents: i) R is left Goldie ring, ii) Pi has finite uniform dimension for each 1 ≤ i ≤ n. In the Section 3 we use the concept of continuous module and we show in the Theorem 3.6 that if M is continuous, retractable, non M-singular module and M satisfies ACC on annihilators, then M is a semiprime Goldie module. Finally as one more application we obtain the Corollary 3.7, where we show that, if R is a continuous, non singular ring and R satisfies ACC on left annihilators, then R is semiprime left Goldie ring. In what follows, R will denote an associative ring with unity and R-Mod will denote the category of unitary left R-modules. Let M and X be R-modules. X is said to be M-generated if there exists an R-epimorphism from a direct sum 3 is semiprime Goldie module, then there are E1, E2, ..., En uniform injective And in the second Theorem we prove that if M submodule, then cM =cN1 ⊕cN2 ⊕ ... ⊕cNn where Ni = Annr modules such that cM ∼= Ek1 1 ⊕ Ek2 n and AssM (Ei) = {Pi}. As other application of these results we give the Theorem 2.27, where we prove 2 ⊕ ... ⊕ Ekn of copies of M onto X. The category σ [M] is defined as the full subcategory of R-Mod containing all R-modules X which are isomorphic to a submodule of an M-generated module. Let M-tors be the frame of all hereditary torsion theories on σ [M]. For a family {Mα} of left R-modules in σ [M], let χ ({Mα}) be the greatest element of M-tors for which all the Mα are torsion free, and let ξ ({Mα}) denote the least element of M-tors for which all the Mα are torsion. χ ({Mα}) is called the torsion theory cogenerated by the family {Mα}, and ξ ({Mα}) is the torsion theory generated by the family {Mα}. In particular, the greatest element of M- tors is denoted by χ and the least element of M-tors is denoted by ξ. If τ is an element of M-tors, gen (τ ) denotes the interval [τ, χ]. Let τ ∈ M-tors. By Tτ , Fτ , tτ we denote the torsion class, the torsion free class and the torsion functor associated to τ , respectively. For N ∈ σ [M], N is called τ -cocritical if N ∈ Fτ and for all 0 6= N ′ ⊆ N, N/N ′ ∈ Tτ . We say that N is cocritical if N is τ -cocritical for some τ ∈ M-tors. If N is an essential submodule of M, we write N ⊆ess M. If N is a fully invariant submodule of M we write N ⊆F I M. For τ ∈ M-tors and M ′ ∈ σ [M], a submodule N of M ′ is τ -pure in M ′ if M ′/N ∈ Fτ . A module N ∈ σ [M] is called singular in σ [M], or M-singular, if there is an exact sequence in σ [M] 0 → K → L → N → 0, with K essential in L. The class S of all M-singular modules in σ [M] is closed by taking submodules, factor modules and direct sums. Therefore, any L ∈ σ [M] has a largest M-singular submodule Z (L) =P {f (N) N ∈ S and f ∈ HomR (N, L)}. L is called non If N is a fully invarint submodule of M, we write M-singular if Z (L) = 0. N ⊆F I M. Injective modules and injective hulls exist in σ [M], since injective hull of X in σ [M] is bX = trM (E (X)) = Pf ∈HomR(M,E(X)) f (M), where E (X) denotes the injective hull of X in R-Mod. Let M ∈ R-Mod. In [1, Definition 1.1] the annihilator in M of a class C of R-modules is defined as AnnM (C) = ∩ K∈Ω K, where Ω = {K ⊆ M there exists W ∈ C and f ∈ HomR(M, W ) with K = ker f }. Also in [1, Definition 1.5] a product is defined in the following way. Let N be 4 a submodule of M. For each module X ∈ R-Mod, N · X = AnnX (C), where C is the class of modules W , such that f (N) = 0 for all f ∈ HomR(M, W ). For M ∈ R-Mod and K, L submodules of M, in [2] is defined the product that if M ∈ R-Mod and C be a class of left R-modules, then AnnM (C) = KM L as KM L = P {f (K) f ∈ HomR (M, L)}. We in [3,Proposition 1.9] P {N ⊂ M NM X = 0 for all X ∈ C}. Moreover Beachy showed in [1, Proposition 5.5] that, if M is projective in σ [M] and N is a any submodule of M, then N · X = NM X, for any R-module X ∈ σ [M]. Let M ∈ R-Mod and N 6= M a fully invariant submodule of M, N is prime in M if for any K, L fully invariant submodules of M we have that KM L ⊆ N implies that K ⊆ N or L ⊆ N. We say that M is a prime module if 0 is prime in M see [11, Definition 13 and Definition 16]. For N ∈ σ[M], a proper fully invariant submodule K of M is said to be associ- ated to N, if there exists a non-zero submodule L of N such that AnnM (L′) = K for all non-zero submodules L′ of L. By [3, Proposition 1.16] we have that K is prime in M. We denote by AssM (N) the set of all submodules prime in M associated to N. Also note that, if N is a uniform module, then AssM (N) has at most one element. A module M is retractable if HomR (M, N) 6= 0 for all 0 6= N ⊆ M. For details about concepts and terminology concerning torsion theories in σ [M], see [16 ] and [17]. 1 Preliminaries The following definition was given in [4, Definition 3.2] we include it for the convenience of the reader. Definition 1.1. Let M ∈ R-Mod and N 6= M a fully invariant submodule of M. We say that N is semiprime in M if for any K ⊆F I M such that KM K ⊆ N, then K ⊆ N. We say M is semiprime if 0 is semiprime in M. 5 Remark 1.2. Notice that if M is projective in σ [M] and N ⊆F I M, then by [4, Remark 3.3], N is semiprime in M if and only if for any submodule K of M such that KM K ⊆ N we have that K ⊆ N. Analogously by [4 Remark 3.22] we have that if P ⊆F I M, then P is prime in M if and only if for any K, L submodules of M such that KM L ⊆ P we have that K ⊆ P or L ⊆ P . Proposition 1.3. Let M ∈ R-Mod be projective in σ [M]. If N ⊆F I M, then the following conditions are equivalent i) N is semiprime in M ii) For any submodule K of M such that N ⊆ K and KM K ⊆ N, then K = N. Proof. i) ⇒ ii) It is clear by Remark 1.2. ii) ⇒ i) Let K ⊆ M a submodule such that KM K ⊆ N. We claim that (K + N)M K ⊆ N and KM (K + N) ⊆ N. In fact by [3, Proposition 1.3] we have that (K + N)M K = KM K + NM K. As N is a fully invariant submodule of M, then NM K ⊆ N. Thus (K + N)M K ⊆ N. Now as KM K ⊆ N, then KM K ⊆ N ∩ K. So by [1, Proposition 5.5] we have that KM(cid:18) K Hence KM(cid:18) K + N N ∩ K(cid:19) = 0. N (cid:19) = 0. Newly by [1, Proposition 5.5] KM (K + N) ⊆ N. Hence by [3, Proposition 1.3] we have that (K + N)M (K + N) = KM (K + N)+ NM (K + N) ⊆ N. By hypothesis we have that (K + N) = N. Thus K = N. Remark 1.4. Let M ∈ R-Mod be projective in σ [M]. Similarly to the proof of Proposition 1.3, we can prove that a fully invariant submodule P of M is prime in M if and only if for any submodules K and L of M containing P , such that KM L ⊆ P , then K = P or L = P . The following definition was given in [4, Definition 3.2], and we include it here for the convenience of the reader. Definition 1.5. Let M ∈ R- Mod. If N is a submodule of M, then successive powers of N are defined as follows: First, N 2 = NM N. Then by induction, for any integer k > 2, we define N k = NM(cid:0)N k−1(cid:1). 6 Notice that in general is false that NM(cid:0)N k−1(cid:1) =(cid:0)N k−1(cid:1)M N. In the Exam- ple 2.8 (1) we can see this. Lemma 1.6. Let M ∈ R-Mod be projective in σ [M] and N ⊆ M semiprime in M submodule. If J is a submodule of M such that J n ⊆ N for some positive positive integer n, then J ⊆ N. Proof. In case n = 1, there is nothing to prove. Now let n > 1 and assume the Lemma holds for lower powers. Since n ≥ 2, then 2n − 2 ≥ n. Hence J 2n−2 ⊆ J n. Now by [1, Proposition 5.6] we have that (J n−1)2 = J 2n−2 ⊆ N. Thus by Remark 1.2 we have that J n−1 ⊆ N. So by the induction hypothesis we have that J ⊆ N. Lemma 1.7. Let M ∈ R-Mod be projective in σ [M] and semiprime module . If N and L are submodules of M such that NM L = 0, then i) LM N = 0 ii) N ∩ L = 0 Proof. i) Since M is projective in σ [M], then by [1 Proposition 5.6] we have that (LM N)M (LM N) = LM (NM L)M N = 0. As M is semiprime and by Remark 1.2, then LM N = 0. ii) By [3, Proposition 1.3] we have that (N ∩ L)M (N ∩ L) ⊆ NM L = 0. Newly as M is semiprime, then N ∩ L = 0. Remark 1.8. Let M ∈ R-Mod be projective in σ [M] and a semiprime module, then we claim that L ∩ AnnM (L) = 0 for all L submodule of M. In fact, if K = L ∩ AnnM (L) , then by [3, Proposition 1.3] we have that KM K ⊆ AnnM (L)M L = 0. Since M is semiprime, then K = 0. Moreover we claim that L ∩ L′ 6= 0 for all L′ ⊆F I M such that AnnM (L) L′. In fact, let L′ ⊆F I M be such that AnnM (L) L′. Suppose that L ∩ L′ = 0, then L′ M L ⊆ L′ ∩ L = 0. Hence L′ ⊆ AnnM (L). So AnnM (L) = L′ a contradiction. Proposition 1.9. Let M ∈ R-Mod be projective in σ [M] and a semiprime module, then HomR (M, N) 6= 0 for all 0 6= N ⊆ M. 7 Proof. Suppose that HomR (M, N) = 0. Since AnnM (N) = ∩ {ker f f ∈ HomR (M, N)}, then AnnM (N) = M. Hence MM N = 0. Now by Lemma 1.7 we have that NM M = 0. As NM M =Pf ∈EndR(M ) f (M), then N ⊆ NM M. So N = 0 a contradiction. Notice that if M is as in Proposition 1.9, then AnnM (N) = M if and only if HomR (M, N) = 0 if and only if N = 0. Therefore AnnM (N) 6= M for all 0 6= N ⊆ M. Proposition 1.10. Let M ∈ R-Mod and τ ∈ M-tors. If P is a prime in M and P is τ -pure submodule of M, then there exists P ′ ⊆ P such that P ′ is τ -pure and minimal prime in M. Proof.Let X = {Q ∈ M Q ⊆ P and Q is τ -pure and prime in M }. Since P ∈ X , then X 6= ∅. We claim that any chain Y ⊆ X , has a lower bound in X . In fact, let Y = {Qi} a descending chain chain in X . Let Q = ∩Y is clear that Q is fully invariant submodule of M and Q ⊆ P . Since each Qi is τ -pure in M, then Q is τ -pure in M. Now K and L be fully invariant submodule of M such that KM L ⊆ Q. If L * Q, then there exists Qt ∈ Y such that L * Qt. Hence L * Ql for all Ql ⊆ Qt. Thus K ⊆ Ql for all Ql ⊆ Qt. As Y is a descending chain, then K ⊆ Q. Therefore by Zorn's Lemma X has minimal elements. Proposition 1.11. Let M ∈ R-Mod be projective in σ [M], semiprime module and τ ∈ M-tors. If U ⊆ M is a uniform submodule, such that U ∈ Fτ , then AnnM (U) is τ -pure and prime in M. Proof. By Proposition 1.9 we have that AnnM (U) M. Now we claim that AnnM (U) = AnnM (U ′) for all 0 6= U ′ ⊆ U. In fact let 0 6= U ′ ⊆ U, then AnnM (U) ⊆ AnnM (U ′). Suppose that AnnM (U) AnnM (U ′). By [3, Proposition 1.9] we have that AnnM (U ′) is a fully invariant submodule of M. Now by Remark 1.8 we have that U ∩ AnnM (U ′) 6= 0. As U is uniform module, then [U ∩ AnnM (U ′)] ∩ U ′ 6= 0. But by Remark 1.8 we have that AnnM (U ′) ∩ U ′ = 0. So [U ∩ AnnM (U ′)] ∩ U ′ = 0 a contradiction. Therefore 8 AnnM (U) = AnnM (U ′). Now by [3, Lemma 1.16] AnnM (U) is prime in M. On the other hand we know that AnnM (U) = ∩ {ker f f ∈ HomR (M, U)}. Since U ∈ Fτ , ker f is τ -pure in M. Hence AnnM (U) is τ -pure in M. Remark 1.12. Let M be as in the Proposition 1.11 and U ⊆ M uniform submodule, then AnnM (U) = AnnM (U ′) = P for all 0 6= U ′ ⊆ U, where P is prime in M. So by [3, Definition 4.3] we have that AssM (U) = {P }. Moreover by [3, Proposition 4.4] we have that AssM (U ′) = AssM (U) = {P } for all 0 6= U ′ ⊆ U. Lemma 1.13. Let M ∈ R-Mod be projective in σ [M] and semiprime mod- ule. Suppose that N is a submodule of M such that P = AnnM (N) is prime in M, then P is minimal prime in M. Proof. By Proposition 1.10, there exists P ′ a minimal prime in M such that P ′ ⊆ P . As [AnnM (N)]M N = 0 ⊆ P ′, then N ⊆ P ′ or AnnM (N) ⊆ P ′. If N ⊆ P ′, then N ⊆ P = AnnM (N). Thus NM N = 0. Since M is semiprime module, then N = 0. Hence P = AnnM (N) = AnnM (0) = M, it is not possible. So the only possibility is P = AnnM (N) ⊆ P ′. Therefore P = P ′ and we have the result. Note that if M is as in the Lemma 1.13 and U is a uniform submodule of M, then by Proposition 1.11 we have that P = AnnM (U) is a prime in M submodule. Now by Lemma 1.13 we obtain that P = AnnM (U) is minimal prime in M. Moreover by proof of the Proposition 1.11 we have that P = AnnM (U) = AnnM(cid:16)bU(cid:17). Definition 1.14. Let M ∈ R-Mod be projective in σ [M]. If N is a submod- ule of M, the right annihilator of N in M is Annr M (N) =P {L ⊆ M NM L = 0}. Note that Annr M (N) is the largest submodule of M such that NM [Annr M (N)] = 0. In fact, if L and K are submodules of M such that NM L = 0 and NM K = 0, then by [3, Proposition 1.3] we have that 9 NM (L ⊕ K) = 0. Now since M is projective in σ [M] then by [1, Proposition 5.5] NM(cid:18) L ⊕ K T (cid:19) = 0 for all T ⊆ L ⊕ K. So NM (L + K) = 0. Remark 1.15. If M is an R-module projective in σ [M] and N and L are submodules of M such that NM L = 0, then we claim that NM (LM M) = 0. In fact by [1, Proposition 5.6] we have that NM (LM M) = (NM L)M M = 0. Since L ⊆ LM M and LM M is a fully invariant submodule of M, then Annr M (N) = M (N) is a fully invariant submodule of P {L ⊆F I M NM L = 0}. Hence Annr M. Proposition 1.16. Let M ∈ R-Mod be projective in σ [M]. If M is a semiprime module, then AnnM (N) = Annr M (N) for all N ⊆ M. Proof. Since AnnM (N)M N = 0, then by Lemma 1.7, NM AnnM (N) = 0. So AnnM (N) ⊆ Annr M (N). Newly as NM Annr M (N) = 0, then Annr M (N)M N = 0. Thus Annr M (N) ⊆ AnnM (N) Hence AnnM (N) = Annr M (N). Definition 1.17. Let M ∈ R-Mod. A submodule N of M is named annihi- lator submodule, if N = AnnM (K) for some K ⊆ M. Notice that if N is an annihilator submodule, then N is a fully invariant submodule of M. Proposition 1.18. Let M ∈ R-Mod be projective in σ [M]. If M is a semiprime module, then AnnM (AnnM (N)) = Annr M (Annr M (N)) = N for all N annihilator submodule of M. Proof. As NM Annr Annr M (N)M N = 0. Thus N ⊆ Annr M (Annr M (N) = 0, then by Lemma 1.7 we have that for some submodule of M, then NM K = 0. Hence K ⊆ Annr Annr M (K). So by Proposition 1.16 we have that M (N)) ⊆ Annr M (N)). Since N = Ann (K) M (N). Therefore M (Annr M (N)) ⊆ AnnM (K) = N. Hence Annr M (Annr M (N)) = N. M (Annr Annr Moreover by Proposition 1.16 we have that AnnM (AnnM (N)) = Annr M (Annr M (N)). 10 2 Modules with ACC on annihilators and Goldie Modules The following definition was given in [4, Definition 3.1]. We include here this definition for the convenience of the reader. Let M ∈ R-Mod. For a subset X ⊆ EndR (M), let AX = ∩ {ker f f ∈ X}. Now we consider the set AM = {AX X ⊆ EndR (M) }. Definition 2.1. Let M ∈ R-Mod, we say M satisfies ascending chain con- dition (ACC) on annihilators , if AM satisfies ACC. Notice that if M = R, then R satisfies ACC on annihilators in the sense of Definition 2.1, if and only if R satisfies ACC on left annihilators in the usual sense for a ring R. Also note that if K is a submodule of M, we have that AnnM (K) = ∩ f ∈X {ker f X = HomR (M, K)}. Since HomR (M, K) ⊆ EndR (M), then AnnM (K) = AX, where X = HomR (M, K). Hence AnnM (K) ∈ AM Theorem 2.2. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. If M satisfies ACC on annihilators, then: i) M has finitely many minimal of minimal prime in M submodules. ii) If P1, P2, ..., Pn are the minimal prime in M submodules, then P1∩ P2∩ , ..., ∩Pn = 0 iii) If P ⊆ M is prime in M, then P is minimal prime in M if and only if P is an annihilator submodule. Proof. i) By a "prime annihilator " we mean a prime in M submodule which is an annihilator submodule. We shall first show that every annihilator submodule of M contains a finite product of annihilators primes in M. Suppose not, then by hypothesis there is an annihilator submodule N which is maxi- mal with respect to not containing a finite product of annihilators primes in 11 M. Hence N is not prime in M. Thus by Remark 1.4 there are fully invari- ant submodules L and K of M such that N L, N K and LM K ⊆ N. Since AnnM (N)M N = 0, then by [3, Proposition 1.3] AnnM (N)M (LM K) = 0. Since M is projective in σ [M], then by [ 1, Proposition 5.6] we have that (AnnM (N)M L)M K = 0. Therefore K ⊆ Annr claim that LM [Annr M (AnnM (N)M L)] ⊆ N. We have that M (AnnM (N)M L). Now we [AnnM (N)M L]M [Annr M (AnnM (N)M L)] = 0. Newly as M is projective in σ [M], then AnnM (N)M (LM [Annr M (AnnM (N)M L)]) = 0. Hence LM [Annr M (AnnM (N)M L)] ⊆ Annr M (AnnM (N)). Since N is annihilator submodule, then by Lemma 1.7 and Proposition 1.18 we have that LM [Annr M (AnnM (N)M L)] ⊆ N. Since K ⊆ Annr M (AnnM (N)M L) = AnnM (AnnM (N)M L), then we can take K = AnnM (AnnM (N)M L). Thus K is an annihilator submodule. Similarly we can prove that L ⊆ AnnM [KM Annr M (N)] and (AnnM [KM Annr M (N)])M K ⊆ N. So also we can take L = AnnM [KM Annr M (N)]. Hence L is an annihilator submodule. Therefore K and L are annihilator submodules such that N K, N L, and LM K ⊆ N. So L and K contain a finite product of prime annihilators. Hence N contains a finite product of prime annihilator, this is a contradiction. Therefore every annihilator submodule of M contains a finite product of prime annihilators. Since 0 = AnnM (M), then the zero is an annihilator submodule. Thus there are prime annihilators P1, P2, ..., Pn of M such that (P1)M (P2)M , ..., Pn = 0. Now if Q is a minimal prime in M, then (P1)M (P2)M ... M (Pn) = 0 ⊆ Q. Thus there exists Pj ⊆ Q for some 0 ≤ j ≤ n. Whence Pj = Q. Therefore the minimal primes in M are contained in the finite set {P1, P2, ... , Pn}. Moreover each Pi is an annihilator submodule for 0 ≤ i ≤ n. submodules. Now by [3 Proposition 1.3] we have that ii) By i) we can suppose that P1, P2, ..., Pn are the minimal prime in M [P1 ∩ P2∩, ..., ∩Pn]n ⊆ (P1)M (P2)M ...M (Pn) = 0. Since M is semiprime module, then by Lemma 1.6 we have that P1 ∩ P2∩, ..., ∩Pn = 0 iii) Let P ⊆ M be a prime in M submodule. If P is minimal prime in M, 12 then by ii) we have that P = Pj for some 1 ≤ j ≤ n. By proof of i) we know that each Pt is an annihilator submodule for 1 ≤ t ≤ n. Hence P so does. Now we obtain the converse by By Lemma 1.13. . Notice that in Theorem 2.2 i) we only use the condition M satisfies ACC on annihilator submodules. Also note that in iii) each minimal prime Pi in M is an annihilator submodule. Thus Pi = AnnM (K) for some K ⊆ M . As K ∈ Fχ(M ) and AnnM (K) = ∩ {ker f f ∈ HomR (M, K)}, then Pi is χ (M)-pure in M for all 1 ≤ i ≤ n. Note that in Theorem 2.2 M satisfies ACC on annihilators is a necessary condition. In order to see this, consider de following example Example 2.3. Let R a ring such that {Si}i∈I is a family of non isomorphic It is clear that M is projective in simple R-modules and let M = Li∈I Si . σ [M]. Now let N ⊆ M such that NM N = 0. Since M is semisimple module, then N = 0. Thus M is a semiprime module. We claim that M does not satisfy ACC on annihilators. In fact, let L be a submodule of M, then there exists K such that M = K ⊕ L. So we can consider the canonical projection π : M → K. Hence ker π = L. Thus each submodule of M is the kernel of some morphism. Hence M does not satisfy ACC on annihilators. Now since Si ≇ Sj, then Si is minimal prime in M. So M has infinitely many minimal prime in M submodules. Lemma 2.4. Let M ∈ R-Mod. If C ∈ σ [M] is χ (M)-cocritical, then there are submodules C ′ ⊆ C and M ′ ⊆ M such that C ′ is isomorphic to M ′. Proof. Since C is χ (M)-cocritical, then C ∈ Fχ(M ). Thus HomR(cid:16)C, cM(cid:17) 6= 0. So there exists f : C →cM a non zero morphism. Since C is χ (M)-cocritical and M ⊆ess cM , then there exists C ′ submodule of C such that C ′ ֒→ M. Hence there exists M ′ ⊆ M such that C ′ ∼= M ′. Also note that M ′ is χ (M)-cocritical. Remark 2.5. If M is as in the Theorem 2.2 and C ∈ σ [M] is χ (M)- cocritical, then by Lemma 2.4 there are C ′ ⊆ C and M ′ ⊆ M such that C ′ ∼= 13 M ′. Hence M ′ is a uniform module, then by Proposition 1.11 we have that AnnM (M ′) = P where P is prime in M. By Lemma 2.13 P is minimal prime in M. Moreover by Remark 1.12 we have that AssM (M ′) = {P }. Hence AssM (C ′) = {P }. Since C ′ ⊆ess C, then by [3, Proposition 4.4] we obtain AssM (C) = {P }. Let SpecM in (M) denote the set of minimal primes in M and Eτ (M) a com- plete set of representatives of isomorphism classes of indecomposable τ -torsion free injective modules in σ [M]. Also we denote by P M in M = {χ (M/P ) P ⊆ M is minimal prime in M}. Remark 2.6. Let τg be the hereditary torsion theory generated by the family of M-singular modules in σ [M]. If M is non M-singular, then τg = χ (M). In fact, if M is non M-singular, then τg ≤ χ (M). Now if τg < χ (M), then there exists 0 6= N ∈ Tχ(M ) such that N ∈ Fτg. Thus HomR(cid:16)N,cM(cid:17) = 0. By [17, Proposition 10.2] N is M-singular. Thus N ∈ Tτg a contradiction. So τg = χ (M). Theorem 2.7 . Let M ∈ R-Mod be projective in σ [M] and semiprime module. If M satisfies ACC on annihilators and for each 0 6= N ⊆ M, N contains a uniform submodule, then there is a bijective correspondences between Eχ(M ) (M) and SpecM in (M). Proof. Since M is semiprime and M satisfies ACC on annihilators then by [4, Proposition 3.4] we have that M is non M-singular. Thus by Remark 2.6 τg = χ (M). Let E ∈ Eχ(M ) (M). As E is a uniform module and E ∈ Fχ(M ), then E is χ (M)-cocritical. Now by Remark 2.5, we have that AssM (E) = {P } with P ∈ SpecM in (M). Hence we define the function Ψ : Eχ(M ) (M) → SpecM in (M) as Ψ (E) = P . We claim that Ψ is bijective. Suppose that Ψ (E1) = Ψ (E2) = P . Since E1 and E2 are uniform modules, then E1 and E2 are χ (M)-cocritical. 14 Since τg = χ (M), then by Lemma 2.4 there are C ′ submodules of M such that C ′ 1 form modules, then by Remark 1.12 we have that. AnnM (C ′ P = AnnM (M2) = AnnM (C ′ 2 ⊆ E2 and M1, M2 ∼= M2. Hence M1 and M2 are uni- 1) = AnnM (M1) = 2). So AnnM (M1) = AnnM (M2) = P . ∼= M1 and C ′ 1 ⊆ E1, C ′ 2 On the other hand we have that (M1 + P ) /P and (M2 + P ) /P are submod- ules of M/P . By [3, Proposition 2.2 and 2.7 ] χ ((M1 + P ) /P ) = χ (M/P ) = χ ((M2 + P ) /P ). Since AnnM (M1) = AnnM (M2) = P , then by Remark 1.8, M1 ∩ P = 0 and M2 ∩ P = 0. Therefore (M1 + P ) /P ∼= M1/ (P ∩ M1) = M1 and (M2 + P ) /P ∼= M2/ (P ∩ M1) = M2. Thus χ (M1) = χ (M/P ) = χ (M2). So HomR(cid:16)M1,cM2(cid:17) 6= 0. Since M1 and M2 are τg-cocritical, then there exists N1 ⊆ M1 such that N1 ֒→ M2. As M1 and M2 are uniform modules, then cM1 ∼= cM2. Thus E1 =cC ′ cN1 Now let P ∈ SpecM in (M) since M satisfies ACC on annihilators, then by 2 = E2. So Ψ is injective. ∼= cM2 ∼= cM1 ∼=cC ′ ∼= 1 Theorem 2.2 P = AnnM (N) for some 0 6= N ⊆ M. By hypothesis there exists a uniform module U such that U ⊆ N. Thus P = AnnM (N) ⊆ AnnM (U). By Proposition 1.11, we have that AnnM (U) is prime in M. As AnnM (U) is clearly an annihilator submodule, then by Theorem 2.2 iii) we have that AnnM (U) is minimal prime in M. Therefore P = AnnM (U). Furthermore by Remark 1.12 we have that AssM (U) = {P }. Hence AssM(cid:16)bU(cid:17) = {P } Thus Ψ(cid:16)bU(cid:17) = P . As U ⊆ M, then U ∈ Fχ(M ). Therefore bU is an indecomposable χ (M)-torsion free injective module in σ [M]. Hence Ψ is surjective. Notice that if P is prime in M such that P is χ (M)-pure, then P is minimal prime in M. In fact by Proposition 1.10 there exists P ′ ⊆ P such that P ′ is χ (M)-pure and minimal prime in M. Now let N be a submodule of M such that P ∩ N = 0, then PM N ⊆ P ∩ N = 0. So PM N ⊆ P ′. Thus P ⊆ P ′or N ⊆ P ′. Suppose that P * P ′, then N ⊆ P ′. Thus N ⊆ P . Hence 0 = P ∩ N = N. So P ⊆ess M. Thus M/P ∈ Tχ(M ) a contradiction. So P ⊆ P ′. Thus P = P ′. On the other hand , if Specχ(M ) (M) denotes the set of χ (M)-pure submodules prime in M, then SpecM in (M) = Specχ(M ) (M). So if M is as in the Theorem 15 2.7, then M has local Gabriel correspondence with respect to χ (M). Example 2.8. Let R = Z2 ⋊ (Z2 ⊕ Z2), the trivial extension of Z2 by Z2 ⊕ Z2. This ring can be described as 0 0 0 R = (cid:26)(cid:18) a (x, y) ideal I = (cid:26)(cid:18) 0 (x, y) J1, J2, J3 which are isomorphic, where J1 = (cid:26)(cid:18) 0 (0, 0) 0 (cid:19)(cid:27), J3 = (cid:26)(cid:18) 0 (0, 0) J2 = (cid:26)(cid:18) 0 (0, 0) a (cid:19) a ∈ Z2, (x, y) ∈ Z2 ⊕ Z2(cid:27). R has only one maximal 0 (cid:19) (x, y) ∈ Z2 ⊕ Z2(cid:27) and it has three simple ideals: 0 (cid:19)(cid:27), 0 (cid:19)(cid:27). 0 (cid:19) ,(cid:18) 0 (1, 0) 0 (cid:19) ,(cid:18) 0 (1, 1) 0 (cid:19) ,(cid:18) 0 (0, 1) 0 0 0 0 0 Then the lattice of ideals of R has the following form R • I • J2• 0 • J1• J3• R is artinian and R-Mod has only one simple module up to isomorphism. Let S be the simple module. By [12, Theorem 2.13], we know that there is a lattice anti-isomorphism between the lattice of ideals of R and the lattice of fully invariant submodules of E (S). Thus the lattice of fully invariant submodules of E (S) has tree maximal elements K, L and N. Moreover, E (S) contains only one simple module S. Therefore, the lattice of fully invariant submodules of E (S) has the following form. K • N • E(S) • L • S • 0 • 16 Let M = E (S). As K ∩ L = S and KM L ⊆ K ∩ L, then KM L ⊆ S. On the other hand, we consider the morphism. f : M π→ (M/N) ∼= S is the canonical projection and i is the inclusion. Thus f (K) = S. Therefore i ֒→ L, where π S ⊆ KM L. Thus we have that KM L = S. Thus KM L ⊆ N but K * N and L * N. Therefore N is not prime in M. Analogously, we prove that neither K nor L are prime in M. Also we note that KM K = S and SM K = 0. Thus AnnM (K) = S. We claim that KM S = S. In fact, if π is the natural projection from M onto M/N, then π (K) = S since (M/N) ∼= S. Thus KM S = S. Analogously we prove that LM S = S and NM S = S. Moreover, it is not difficult to prove that AnnM (S) = K∩L∩N = S and that S is not a prime in M submodule. Soc (R) = J1+ J2 + J3 = J1⊕ J2 and Soc (R) ⊂ess R. As S is the only simple module, then Soc (R) ∼= S ⊕ S. So E (S ⊕ S) = E (R). Thus E (R) = E (S) ⊕ E (S) = M ⊕ M. Therefore σ [M] = σ [R] = R-Mod. Moreover HomR (M, K) 6= 0 for all K ∈ σ [M] but M is not a generator of σ [M]. Now let 0 6= M ′ a submodule of M. As S ⊆ess M, then S ⊆ M ′. Therefore every submodule of M contains a uniform submodule of M. From the previous arguments we can conclude that: 1) KM (KM K) = KM S = S and (KM K)M K = SM K = 0. Thus KM (KM K) 6= (KM K)M K 2) M does not have prime in M submodules. Thus SpecM in (M) = ∅ 3) M is the only one indecomposable injective module in σ[M]. Thus Eχ(M ) (M) = {M} 4) M is noetherian. So M satisfies ACC on annihilators 5) M is not projective in σ [M] Remark 2.9. If M is projective in σ[M] and we define the function Φ : SpecM in (M) → P M in M as Φ (P ) = χ (M/P ), then Φ is bijective. In fact. Let P and P ′ are minimal primes in M submodules, such that χ (M/P ) = χ (M/P ′), then P ′ is not χ (M/P )-dense in M. As P ′ is fully invariant submodule of M, then by [3, Lemma 2.6] we obtain P ′ ⊆ P . But P is minimal prime in M, then P ′ = P . So Φ is injective. Moreover it is clear that Φ is surjective. 17 Corollary 2.10. Let M ∈ R-Mod be projective in σ [M] and semiprime module. If M satisfies ACC on annihilators and for each 0 6= N ⊆ M, N contains a uniform submodule, then there is a bijective correspondence between Eχ(M ) (M) and P M in M . Proof. It follows from Theorem 2.7 and Remark 2.9. Definition 2.11. An R-module M is Goldie Module if it satisfies ACC on annihilators and it has finite uniform dimension. Notice that if M = R, then R is Goldie module in the sense of the definition 2.11 if and only if R is left Goldie ring. Corollary 2.12. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. If M is a Goldie module, then there is a bijective correspondences between Eτg (M) and SpecM in (M). Proof. It follows from Definition 2.11 and Theorem 2.7. Corollary 2.13. Let R be a semiprime ring such that satisfies ACC on left annihilators. Suppose that for each 0 6= I ⊆ R left ideal of R, I contains a uniform left ideal of R, then there is a bijective correspondence between the set of representatives of isomorphism classes of indecomposable non singular injective R-modules and the set of minimal prime ideals of R. Proof. From [3, Definition 1.10] we have that P is prime in R if and only if P is prime ideal of R. Since R is a semiprime ring and satisfies ACC on left annihilators, R is non singular On the other hand we know that σ [R] = R-Mod. So by Theorem 2.7 we have the result. Notice that the condition " for each non zero left ideal I of R, I contains a uniform left ideal of R " in the Corollary 2.13 is necessary. In order to see this, consider de following example Example 2.14. Let R which is an Ore domain to the right but not to the left. See [15 p 53] for examples of rings of this sort. It is proved in [7, p 486] that 18 there are no χ (R)-cocritical left R-modules. As R is a domain, then R is non singular. Thus χ (R) = τg. Moreover R is a prime ring and R clearly satisfies ACC on left annihilators. On the other hand we know that if U is a uniform τg- torsion free module, then U is τg-cocritical. But this is not possible. Thus there are no τg-torsion free uniform modules in R-mod. Hence Eτg (R) = ∅. As R is a domain, then SpecM in (R) = {0}. Thus there are no bijective correspondence between Eτg (R) and SpecM in (R). Corollary 2.15. Let R be a semiprime left Goldie ring, then there is a bijective correspondence between a complete set of representatives of isomor- phism classes of indecomposable non singular injective R-modules and the set of minimal prime ideals of R. Proof. As R is a left Goldie ring, then R is a Goldie module. So by the Corollary 2.13 we have the result. Notice that if R is as the Corollary 2.15, then by Theorem 2.7 and Corollary 2.13 Specτg (R) = {I I is a τg-pure prime ideal of R} = SpecM in (R). Thus R has local Gabriel correspondence with respec to χ (R) = τg Lemma 2.16. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. Suppose that M satisfies ACC on annihilators. If P is a minimal prime in M and Annr M (P ) = L, then P = AnnM (L′) for all 0 6= L′ ⊆ L. Proof. By Theorem 2.2, P is annihilator submodule, then by Proposition M (P )) = AnnM (L). Now let 0 6= L′ ⊆ L 1.18 we have that P = Annr and AnnM (L′) = K. Thus P = AnnM (L) ⊆ AnnM (L′) = K. M (Annr Suppose P K. As P is prime in M and AnnM (K)M K = 0 ⊆ P , then AnnM (K) ⊆ P . So AnnM (K) ⊆ K. Thus AnnM (K)K AnnM (K) ⊆ AnnM (K)M K = 0. Since M is semiprime, then AnnM (K) = 0. By Propo- sition 1.18 we have that AnnM (AnnM (K)) = K. But AnnM (AnnM (K)) = AnnM (0) = M. Thus M = K = AnnM (L′). So by Proposition 1.9 L′ = 0 a contradiction. Therefore P = K = AnnM (L′). 19 Note that if M is as in Lemma 2.16 and Annr M (P ) = L, then AssM (L′) = {P } for all 0 6= L′ ⊆ L. Hence {P } = AssM (N) = AssM(cid:16)bL(cid:17) for all 0 6= N ⊆bL. Proposition 2.17. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. Suppose that M satisfies ACC on annihilators. If P1, P2, ..., Pn are the minimal primes in M, then {N1, N2, ...Nn} is an independent family, where Ni = Annr M (Pi) for 1 ≤ i ≤ n. Proof. By induction. If n = 1, then we have the result. Suppose that {N1, N2, ...Nn−1} is an independent family. If (N1 ⊕ N2 ⊕ ... ⊕ Nn−1) ∩ Nn 6= 0, then there are x ∈ Nn and 1 ≤ i ≤ n − 1, such that Rx ∼= N ′ i ⊆ Ni. By Lemma 2.16 we have that Pn = Annr i ) = Annr M (Ni) = Pi a {N1, N2, ...Nn} contradiction. Therefore (N1 ⊕ N2 ⊕ ... ⊕ Nn−1)∩Nn = 0. So is a independent family. M (Rx) = Annr M (N ′ Theorem 2.18. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. Suppose that M satisfies ACC on annihilators and for each 0 6= N ⊆ M, N contains a uniform submodule. If P1, P2, ..., Pn are the minimal primes in M, then cN1 ⊕cN2 ⊕ ... ⊕cNn = cM where Ni = Annr Proof. By Proposition 2.17 we have that {N1, N2, ...Nn} is an independent In fact let 0 6= L ⊆ family. We claim that N1 ⊕ N2 ⊕ ... ⊕ Nn ⊆ess M. M. By hypothesis there exists U a uniform module shut that U ⊆ L. By M (Pi) for 1 ≤ i ≤ n. Proposition 1.11 and Theorem 2.2 we have that AnnM (U) is minimal prime in M. So AnnM (U) = Pj for some 1 ≤ j ≤ n. Hence (Pj)M U = 0. Thus U ⊆ Annr M (Pj) = Nj. Whence (N1 ⊕ N2 ⊕ ... ⊕ Nn) ∩ L 6= 0. Therefore cN1 ⊕cN2 ⊕ ... ⊕cNn = cM . Corollary 2.19. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. Suppose that M is Goldie Module and P1, P2, ..., Pn are the minimal primes in M, thencN1 ⊕cN2 ⊕ ... ⊕cNn = cM where Ni = Annr Proof. As M is Goldie module, then M has finite uniform dimension and M (Pi) for 1 ≤ i ≤ n. M satisfies ACC on annihilators. So by the Theorem 2.18 we have the result. 20 Theorem 2.20. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. Suppose that M is Goldie Module and P1, P2, ..., Pn are the minimal primes in M submodules. If Ni = Annr M (Pi) for 1 ≤ i ≤ n, then there exists E1, E2, ... , En uniform injective modules such that cM ∼= Ek1 and AssM (Ei) = {Pi}. 1 ⊕ Ek2 2 ⊕ ... ⊕ Ekn n Proof. Since Ni has finite uniform dimension, then there are Ui1, Ui2, ... , ⊆ess Ni. Thus uniform submodules of Ni, such that Ui1 ⊕ Ui2 ⊕ ... ⊕ Uiki Uiki =cNi. Now by Lemma 2.16, we have that AnnM(cid:0)Uij(cid:1) = Pi cUi1 ⊕cUi2 ⊕ ... ⊕dUiki for all 1 ≤ j ≤ ki. Thus AssM(cid:0)Uij(cid:1) = AssM(cid:16)cUij(cid:17) = {Pi} for all 1 ≤ j ≤ ki. Hence by Theorem 2.7, we have that cUi1 Ei = cUi1. Hence cNi that cM ∼= Ek1 for 1 ≤ i ≤ n. Therefore by Theorem 2.18 we have ∼= .... ∼= dUiki Notice that by Theorem 2.2 Ei ≇ Ej for i 6= j. ∼= cUi2 . So we can denote 2 ⊕ ... ⊕ Ekn n . 1 ⊕ Ek2 ∼= Eki i Proposition 2.21. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. Suppose that M is a Goldie Module and P1, P2, ..., Pn are the minimal M (Pi) for 1 ≤ i ≤ n. primes in M, thencNi is a fully invariant submodule of cM where Ni = Annr Proof. We claim that if i 6= j, then HomR(cid:16)cNi, cNj(cid:17) = 0. In fact by The- =cNi where Ui1, Ui2, ... , Uiki orem 2.20 we have that cUi1 ⊕ cUi2 ⊕ ... ⊕dUiki =cNj where Uj1, uniform submodules of Ni analogously cUj1 ⊕ cUj2 ⊕ ... ⊕dUjkj are uniform submodules of Nj. Let 0 6= f ∈ HomR(cid:16)cNi, cNj(cid:17), : cUir → cUjt is non zero. By [4, Proposition 3.4] M is non M-singular. Hence cM is non M-singular. Since cUir is uniform submodule of cM , then fdUir phism. So AssM(cid:16)cUir(cid:17) = AssM(cid:16)cUjt(cid:17). But AssM(cid:16)cUir(cid:17) = AssM(cid:16)cNi(cid:17) = Pi and then there exist ir and jt such that the restriction morphism. fdUir is a monomor- Uij2, ... , Ujkj are 21 AssM(cid:16)cUjt(cid:17) = AssM(cid:16)cNj(cid:17) = Pj a contradiction. Therefore HomR(cid:16)cNi, cNj(cid:17) = 0. Now let g : cM → cM be a morphism. By Theorem 2.18 we have g(cid:16)cNi(cid:17) ⊆ cN1 ⊕cN2 ⊕ ... ⊕ cNn. Since HomR(cid:16)cNi, cNj(cid:17) = 0 for i 6= j, then g(cid:16)cNi(cid:17) ⊆ cNi. Thus cNi is a fully invariant submodule of cM. Proposition 2.22. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. Suppose that M satisfies ACC on annihilators and for each 0 6= N ⊆ M, N contains a uniform submodule. If P1, P2, ..., Pn are the minimal primes in M and Ni = Annr M (Pi) has finite uniform dimension for each 1 ≤ i ≤ n, then M has finite uniform dimension. Proof. By Theorem 2.18 we obtain that N1 ⊕ N2 ⊕ ... ⊕ Nn ⊆ess M. Now since Ni has finite uniform dimension for each 1 ≤ i ≤ n, then M has finite uniform dimension. Theorem 2.23. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. Suppose that M satisfies ACC on annihilators and for each 0 6= N ⊆ M, N contains a uniform submodule. If P1, P2, ..., Pn are the minimal primes in M then the following conditions are equivalent. i) M is a Goldie Module. ii) Ni = Annr M (Pi) has finite uniform dimension for each 1 ≤ i ≤ n. Proof i) ⇒ ii) As M is left Goldie module, then M has finite uniform dimension. Hence we have the result. ii) ⇒ i) By Proposition 2.22 we have that M has finite uniform dimension. Thus M is left Goldie module. Corollary 2.24. Let R be a semiprime ring such that R satisfies ACC on left annihilators. Suppose that for each non zero left ideal I ⊆ R, I contains a uniform left ideal. If P1, P2, ..., Pn are the minimal prime ideals of R, then the following conditions are equivalent. i) R is a left Goldie ring. 22 ii) AnnR (Pi) = Ni has finite uniform dimension for each 1 ≤ i ≤ n. Proof. It follows from Theorem 2.23 Lemma 2.25. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. Suppose that M satisfies ACC on annihilators. If P1, P2, ..., Pn are the minimal primes in M, then Annr Pj for every 1 ≤ i ≤ n. M (Pi) = ∩ i6=j Proof. As each Pi is a fully invariant submodule of M, then (Pi)M(cid:18) ∩ Pi ∩(cid:18) ∩ (Pi)M(cid:18) ∩ Pj(cid:19) = ∩n Pj(cid:19) = 0. So ∩ j=1Pj. By Theorem 2.2 we have that ∩n i6=j i6=j Pj(cid:19) ⊆ i6=j j=1Pj = 0. Thus Pj ⊆ Annr M (Pi). Now let K ⊆ M such that i6=j (Pi)M K = 0, then (Pi)M K ⊆ Pj Pi ⊆ Pj or K ⊆ Pj. But Pi ⊆ Pj is not possible. Thus K ⊆ Pj for all j 6= i. for all i 6= j. As Pj is prime in M, then Hence K ⊆ ∩ i6=j Pj. Thus Annr M (Pi) = ∩ i6=j Pj Notice that by Proposition 1.16 we have that AnnM (Pi) = Annr M (Pi). Thus AnnM (Pi) = ∩ i6=j Pj for every 1 ≤ i ≤ n. Corollary 2.26. Let M ∈ R-Mod be projective in σ [M] and a semiprime module. Suppose that M satisfies ACC on annihilators and for each 0 6= N ⊆ M, N contains a uniform submodule. If P1, P2, ..., Pn are the minimal primes in M and Pi has finite uniform dimension for each 1 ≤ i ≤ n, then M has finite uniform dimension. Proof. By Lemma 2.25 Annr M (Pi) = ∩ i6=j Pj. Hence Annr M (Pi) has finite uniform dimension. So by Proposition 2.23 we have that M has finite uniform dimension. Theorem 2.27. Let R be a semiprime ring such that R satisfies ACC on left annihilators. Suppose that for each non zero left ideal I ⊆ R, I contains a uniform left ideal. If P1, P2, ..., Pn are the minimal prime ideals of R, then the following conditions are equivalent. 23 i) R is left Goldie ring. ii) Pi has finite uniform dimension for each 1 ≤ i ≤ n. Proof. i) ⇒ ii) It is clear. ii) ⇒ i) By Corollary 2.26 we have that R has finite uniform dimension. So R is left Goldie ring. Finally we obtain the following result which extends the result given in [ Bo Stemtrom Lemma 2.5] Definition 2.28. Let M ∈ R-Mod. We say that a submodule N of M is a nilpotent submodule if N k = 0 for some positive integer k. Proposition 2.29. Let M be projective in σ [M] . If M satisfies ACC on annihilators, then Z (M) is a nilpotent submodule. Proof. We consider the descending chain Z (M) ⊇ Z (M)2 ⊇ ....We suppose that Z (M)n 6= 0 for all n ≥ 1.So we have the ascending chain AnnM (Z (M)) ⊆ As Z (M)n+2 6= 0, then by [1, Proposition 5.6] we have that [Z (M)M Z (M)]M Z (M)n 6= 0. AnnM(cid:0)Z (M)2(cid:1) ⊆ ...Since M satisfies ACC on annihilators, there exists n > 0 such that AnnM (Z (M)n) = AnnM(cid:0)Z (M)n+1(cid:1) = .... Since Z (M)M Z (M) =P {f (Z (M)) f : M → Z (M) }, then [P {f (Z (M)) f : M → Z (M) }]M Z (M)n 6= 0. So by [3, Proposition1.3 ] there exits f : M → Z (M) such that f (Z (M))M Z (M)n 6= 0. Thus f (M)M Z (M)n 6= 0 Now consider the set Γ = {ker f f : M → Z (M) and f (M)M Z (M)n 6= 0} By hypothesis Γ has maximal elements. Let f : M → Z (M) such that ker f is a maximal element in Γ. 24 We claim that h (f (M))M Z (M)n = 0 for all h : M → Z (M). In fact let h : M → Z (M) be a morphism. By [5, Lemma 2.7] ker h ⊆ess M. So ker h ∩ f (M) 6= 0. Thus there exists 0 6= f (m) such that h (f (m)) = 0. So ker f ker h ◦ f . If h (f (M))M Z (M)n 6= 0, then h ◦ f ∈ Γ. But ker f is a maximal element in Γ and ker f ker h ◦ f , then h ◦ f /∈ Γ a contradiction. Therefore h (f (M))M Z (M)n = 0. On the other hand by [1, Proposition 5.6] we have that . f (M)M Z (M)n+1 = [f (M)M Z (M)]M Z (M)n = [P {h (f (M)) h : M → Z (M)}]M Z (M)n. By [3, Proposition 1.3] we have that [P {h (f (M)) h : M → Z (M)}]M Z (M)n = P {h (f (M))M Z (M)n h : M → Z (M)} = 0. Hence f (M) ⊆ AnnM(cid:0)Z (M)n+1(cid:1) = AnnM (Z (M)n). So f (M)M Z (M)n = Thus f (M)M Z (M)n+1 = 0. 0 a contradiction. Corollary 2.30. Let M ∈ R- Mod be projective in σ [M] and S = EndR (M). If M is retractable and satisfies ACC on annihilators, then Zr (S) is nilpotent. Where Zr (S) is the right singular ideal of S. Proof. Notice that if, N and L are submodules of M then, HomR (M, L) HomR (M, N) ⊆ HomR (M, NM L). Now we consider the ideal ∆ = {f ∈ S ker f ⊆ess M}. We claim that Zr (S) ⊆ ∆, in fact let α ∈ Zr (S), then there exists I essential right deal of S such that αI = 0. Thus α ◦ g = 0 g (M)! = 0. Now let 0 6= N ⊆ M . As M is for all g ∈ I. Hence α Pg∈I N ∩Pg∈I g (M). Thus Pg∈I retractable, then there exits ρ : M → N a non zero morphism. Thus ρS ∩ I 6= 0. So there exists h ∈ S such that 0 6= ρ ◦ h ∈ I. Hence 0 6= (ρ ◦ h) (M) ⊆ g (M) ⊆ess M. So ker α ⊆ess M. Hence Zr (S) ⊆ ∆. So Zr (S) M ⊆ ∆M ⊆ Z (M). Moreover HomR (M, ∆M) ⊆ (M, Z (M)). On the other hand we have that ∆ ⊆ HomR (M, ∆M). By Proposition 2.29 there exists n > 0 such that Z (M)n = 0. Thus ∆n ⊆ HomR (M, ∆M)n ⊆ (M, Z (M))n ⊆ (M, Z (M)n) = 0. So ∆ is nilpotent. Since Zr (S) ⊆ ∆, then Zr (S) is nilpotent. 25 In the Corollary 2.30 we have that ∆ is nilpotent. Notice that to obtain this resul it is not necessary the condition M is retractable. Corollary 2.31. Let M ∈ R- Mod be projective in σ [M] and S = EndR (M). If I is an ideal of S such that ∩f ∈I ker f ⊆ess M, then I is nilpotent. Proof. As ∩f ∈I ker f ⊆ess M, then ker f ⊆ess M for all f ∈ I. Thus I ⊆ ∆. By Corollary 2.30 we have that ∆ is nilpotent. So I is nilpotent Definition 2.32. Let M ∈ R. A submodule N of M is TM -nilpotent in case every sequence {f1, f2, ..., fn, ...} in HomR (M, N) and any a ∈ N, there exists n ≥ 1 such that fnfn−1...f1 (a) = 0. Notice that if I is a left ideal of a ring R, then I is left T -nilpotent in the usual sense if and only if I is TR-nilpotente. Proposition 2.33. Let M ∈ R- Mod be projective in σ [M] and retractable. Suppose that M satisfies ACC on annihilators. If N ⊆ M is TM -nilpotent, then N is nilpotent. Proof. Let N ⊆ M be TM -nilpotent. Consider the chain N ⊇ N 2 ⊇ N 3 ⊇ .... Then we have the chain AnnM (N) ⊆ AnnM (N 2) ⊆ AnnM (N 3) ⊆ .... Since M satisfies ACC on annihilators, then there exists k ≥ 1 such that AnnM(cid:0)N K(cid:1) = AnnM(cid:0)N k+1(cid:1) = AnnM(cid:0)N k+2(cid:1) ..... Let L = N k. If L2 = LM L 6= 0, then there exists f : M → L and a ∈ L ⊆ N such that f (a) 6= 0. Hence (Ra)M L 6= 0. If (Ra)M (LM L) = 0, then Ra ⊆ AnnM (L2) = AnnM (L). So RaM L = 0 a contradiction. Thus (Ra)M (LM L) 6= 0. By [1, M-injective 5.6] 0 6= (Ra)M (LM L) = ((Ra)M L)M L. As (Ra)M L =P {f (Ra) f : M → L}, then by [3, Proposition 1.3] there exists f1 : M → L such that f1 (Ra)M L 6= 0. So (Rf1 (a))M L 6= 0. Newly (Rf1 (a))M (LM L) 6= 0. Hence (Rf1 (a)M L)M L 6= 0. So there exists f2 : M → L such that f2 (Rf1 (a))M L 6= 0. Hence Rf2 (f1 (a))M L 6= 0. Continuing in this way, we have that Rfn.fn−1...f1 (a)M L 6= 0 for all n ≥ 1. Hence fn.fn−1...f1 (a) 6= 0 for all n ≥ 1 a contradiction because N is TM - nilpotent. Hence L2 = 0. So N is nilpotent. 26 3 Continuous Modules with ACC on annihila- tors The following definitions were given in [14]. We include here these definitions for convenience of the reader. Definition 3.1. Let M ∈ R-Mod. M is called continuous module if it satisfies the followings conditions: C1) Every submodule of M is essential in a direct summand of M C2) Every submodule of M that is isomorphic to a direct summand of M is itself a direct summand. Definition 3.2. An R-module M is K-nonsingular if for every f ∈ EndR (M) such that ker f ⊆ess M implies f = 0. Notice that by [14, Proposition 2.3] we have that if M is non M-singular module, then M is K-nonsingular Definition 3.3. Let M ∈ R-Mod. The K-singular submodule of M is defined as Z K (M) =P {f (M) f ∈ EndR (M) and ker f ⊆ess M } Remark 3.4. If M is a continuous module and S = EndR (M), then by [10, Proposition 1.25 ] we have that J (S) = {f ∈ S ker f ⊆ess M}. Hence Z K (M) = J (S) M. On the other hand it is clear that Z K (M) ⊆ Z (M). Therefore J (S) M ⊆ Z (M). Proposition 3.5. Let M ∈ R- Mod be projective in σ [M] and S = EndR (M). Suppose that M is a continuous module . If M satisfies ACC on annihilators, then J(S) is nilpotent Proof. As M is a continuous module, then by Remark 3.4 we have that J (S) = {f ∈ S ker f ⊆ess M}. By Corollary 2.30 J (S) = ∆ is nilpotent. Theorem 3.6. Let M ∈ R- Mod be projective in σ [M] and S = EndR (M). Suppose that M is a continuous, retractable, non M-singular module and satis- fies ACC on annihilators. Then M is a semiprime Goldie module. 27 Proof. Since M is non M-singular and continuous, then by Remark 3.4. J (S) M = 0. Hence J (S) = 0. Thus S is a semiprime ring. We claim that M is semiprime module. In fact let L be a fully invariant submodule of M such that L2 = LM L = 0. As HomR (M, L) HomR (M, L) ⊆ HomR (M, LmL), then HomR (M, L) HomR (M, L) = 0. Since S is semiprime ring the HomR (M, L) = 0. Now as M is retractable, then L = 0. Since M is non M-singular, then M is K-nonsingular, then by [14, Proposition 3.1, Corollary 3.3 and Theorem 1.5] we have that ker f is a direct summand of M for all f ∈ S. As satisfies ACC on annihilators, then M satisfies ACC on direct summands. Since M satisfies the condition C1, then every closed submodule of M is a direct summand . Thus M satisfies ACC on closed submodules. By [9, Proposition 6.30] M has finite uniform dimension. Hence M is a semiprime Goldie module. Corollary 3.7. Let R be a continuous and non singular ring. Suppose that R satisfies ACC on left annihilators, then R is a semiprime left Goldie ring. Notice that in the Corollary 3.7 the inverse is not true in general. Consider the following example Example 3.8. Let R = Z be the ring of integers. We know Z is a prime Goldie ring, but Z is not a continuous ring. References [1] Beachy, J. (2002). M-injective modules and prime M-ideals, Comm. Algebra 30(10):4639-4676. [2] Bican, L., Jambor, P., Kepka, T., Nemec, P. (1980). Prime and coprime modules, Fundamenta Matematicae 107:33-44 [3] Castro, J., R´ıos, J. (2012). Prime submodules and local Gabriel correspon- dence in σ [M], Comm. Algebra. 40(1):213-232. [4] Castro, J., R´ıos, J. (2014). Krull dimension and classical Krull dimension of modules. Comm. Algebra. 42(7):3183-3204 28 [5] Castro, J., R´ıos, J. (2015). M tame modules and Gabriel correpondence, to appear in Comm. Algebra. [6] Chatters, A. , Hajarnavis, C. (1980) Rings with chain conditions, Boston, London, Melbourne: Pitman Advanced Publishing Program [7] Golan, J. (1986). Torsion Theories, Longman Scientific & Technical, Har- low. [8] Goodearl, K, Warfield, R ( 2004). An introduction to Noncommutative Noetherian Rings, Cambridge University Press. [9] Lam, T. (1998), Lectures on Modules and Rings, Graduate Texts in Math- ematics., vol 139, New York: Springer-Verlag. [10] Nicholson, W. Yousif, M. (2003). Quasi-Frobenius Rings. Cambridge Uni- versity Press. [11] Raggi, F., Rios, J., Rinc´on, H., Fern´andez-Alonso, R., Signoret, C. (2005). Prime and irreducible preradicals, J. Algebra Appl. 4(4):451-466. [12] Raggi, F., R´ıos, J., Rinc´on, H., Fern´andez-Alonso, R. (2009). Basic Prerad- icals and Main Injective Modules, J. Algebra Appl. 8(1):1-16. [13] Raggi, F., R´ıos, J., Rinc´on, H., Fern´andez-Alonso, R. (2009). Semiprime Preradicals, Comm. Algebra. 37(7):2811-2822. [14] Rizvi, S., Roman, C. (2007). On K-nonsigular Modules and Applications. Comm. Algebra. 35:2960-2982.. [15] Stenstrom, B. (1975). Rings of Quotients, Graduate Texts in Mathematics, New York: Springer-Verlag. [16] Wisbauer, R. (1991). Foundations of Module and Ring Theory, Gordon and Breach: Reading. [17] Wisbauer, R.(1996). Modules and Algebras: Bimodule Structure and Group Actions on Algebras, England: Addison Wesley Longman Limited. 29
1907.11288	1	1907	2019-07-25T19:35:20	Algebras with Laurent polynomial identity	[ "math.RA" ]	In this article we shows some results about algebra with the group of units having special polynomial identity.	math.RA	math	ALGEBRAS WITH LAURENT POLYNOMIAL IDENTITY By Claudenir Freire Rodrigues, Ramon C´odamo B. da Costa 9 1 0 2 l u J 5 2 ] [ . A R h t a m 1 v 8 8 2 1 1 . 7 0 9 1 : v i X r a Universidade Federal do Amazonas - Departamento de Matem´atica Av. General Rodrigo Octavio Jordao Ramos, 1200 - Coroado I, Manaus - AM - Brasil, 69067-005 [email protected] (corresponding author) [email protected] ALGEBRAS WITH LAURENT POLYNOMIAL IDENTITY CLAUDENIR FREIRE, RAMON CODAMO Abstract. In this article we shows some results about algebra with the group of units having special polynomial identity. Keywords: Laurent polynomial identity, unit group, group identity, nil ideal, Brian Hartley Conjecture. 1. Introduction A Laurent polynomial f = f (x1, ..., xl) in the noncommutative variables xi, i = 1, . . . , l is an element non-zero in the group algebra RFl over a ring R with free group Fl =< x1, ..., xl >. One says f in RFl is a Laurent Polynomial Identity (LP I) for an R algebra A (respc. for U(A) the group of units in A) if f (a1, ..., al) = 0 for all sequence a1, ..., al in A (respc. in U(A)). A word w in RFl is group identity for all U(A) if w(a1, ..., al) = 1, for a1, ..., al in U(A) . Brian Hartleys Conjecture: Let G be a torsion group and R a field. If the unit group U(RG) of RG satisfies a group identity w = 1, then RG satisfies a polynomial identity. In [2] A.Giambruno, E.Jespers and A.Valenti shows that the Brian Hartleys Con- jecture is true for a group algebra RG over an infinite commutative domnain R and a torsion group G and G has no divisible order elements by p with characteristic of R is p. For this purpose they proved the following crucial result. Proposition 1.1. Let A be an algebra over an infinite commutative domain R and suppose that U(A) satisfies a group identity. There exists a positive m such that if a, b, c, u in A and a2 = bc = 0, then bacA is nil right ideal of bounded exponent less or equal than m. The case identity group w = 1 in the conjecture says the unity group satisfy equivalently the Laurent polynomial F = 1 − w. We use this idea and study the case F = a1 + a2w2 + ... + anwn. Clearly the case F = 1 − w is a special case of our generalization. In this paper we prove the following results. 1 2 CLAUDENIR FREIRE, RAMON CODAMO Proposition 1.2. Let A be an algebra with the group of units U(A) admits a LP I over a ring R whith unit whose non-constants words has the sum of exponents non- zero at least one of the variables. Then there exists a polynomial f ∈ R[X] with the limited degree d ≤ 4(−l+r)+3, where l = min {P expwi} and r = max{P exp wi}, determinated by the LP I such that for all a, b, c, u in A with a2 = bc = 0, f (bacu) = 0. In particular, bacA is an algebraic ideal. As a consequence of Proposition 1.2 we prove the following: Corollary 1.3. Let A be an algebra over a commutative domain R with R > d whose unit group U(A) has a LP I. If a, b, c ∈ A with a2 = bc = 0, then bacA is a nil ideal with limited exponent. Notice that the restriction on LP I is necessary because we will show that without this assumption Proposition 1.2 i not valid. In this article we always adopt LP I with this restriction. Also, by changin xi by x−iyxi we may assume that LP I in two variables. ALGEBRAS WITH LAURENT POLYNOMIAL IDENTITY 3 Proof of Proposition 1.2 : Let P = a1 + a2w2 + ... + arwt the LP I with wi in the form wi = xr1 1 xs2 2 · · · xrk 1 xsk 2 , k ≥ 1, ri, si integers. If P exp wi = P expx1 wi + P expx2 wi = 0 then we substitute x1 = xk 2 with k > 1 big enough, We get a new LP I in the form P = a1 + a2w′ x2 = xk ... + arw′ r, where 1 or 2 + X exp w′ i = X expx1 w′ i + X expx2 w′ i = k X expx1 wi + X expx2 wi or X exp w′ i = X expx1 wi + k X expx2 wi. In any case that is not zero. In general, P (1, 1) = a1 + a2 + ... + ar = 0, so a2 + ... + ar 6= 0 , because a1 6= 0. Hence any collection of sums with distinct parcels involving all coefficients a2, ..., ar has a non-zero term. From this P (α, α) = a1 + a2αP expw2 + ... + akαP expwk = 0, 1 < k ≤ r, is a non-zero polynomial with integers exponents over R, for all α ∈ U(A). In the following we see that all exponents are not necessarily positive. Let α−l be where l = min {P expwi}, so α−lP (α, α) = f0(α) = 0, where f0 6= 0 is a polynomial with degree ≤ −l + r, over R. Therefore U(A) is algebraic over R. Now, we are going to use this polynomial with a particular unit in U(A). First let b = c be, so a2 = b2 = 0. Then we have (1 + aua), (1 + bauab) ∈ U(A), for all u ∈ A. Let Qn be the set of all products in aua and bauab with at most n − 1 factors, and let W be the set of all products in coeficients ai's, aua and bauab. Thus α = (1 + aua).(1 + bauab) ∈ U(A), and for this unit, αn = 1 + aua + bauab + X t + (auabauab)n t∈Qn for all n ≥ 1. It follows that there exists a polynomial f0 6= 0 over R such that f0(α) = a1 + a2 + ... + ar + X t + t∈W = X t + t∈W r X i=2 ai(auabauab)si = 0. r X i=2 ai(auabauab)si As a2 + ... + ar 6= 0, after to organize the powers of auabauab which appears in the last sum, has a non-zero coefficients. This term has the form (aj1 + ... + ajk)(auabauab)s where s ≤ (−l + r). So, abf0(α)au = 0 is a non-zero polynomial g over R such that g(abau) = 0 with degree ≤ 2(−l + r) + 1. This finishs the case b = c. 4 CLAUDENIR FREIRE, RAMON CODAMO In general, if a, α, β ∈ A and a2 = αβ = 0, then we have (βuα)2 = 0 , for all u ∈ A. So βuαaβuαA is algebraic and there exists g over R which is not zero and g((βuαaβuα)a) = 0, from this there exists a non-zero polynomial f = (αa)f0(βu) with f (αaβu) = 0 such that d = degree f ≤ 4(−l + r) + 3. Therefore αaβA is algebraic over R. ALGEBRAS WITH LAURENT POLYNOMIAL IDENTITY 5 Proof of Corollary 1.3 : By Proposition 1.2, there exists f ∈ R[x] \ {0} with degree d ≤ m = 4(−l + r) + 3, such that f (bacλu) = 0 for all λ ∈ R, u ∈ A. That is, λp1 + λ2p2 + ... + λdpd = 0 where pi is polynomial on bacu with pd = bd(bacu)d . Now, by the Vandermond argument, p1 = ... = pd = 0, so (bacu)d = 0 . The next results are LPI versions of Lema 3.2, Lema 3.3 of [6] and Corolary 3 of [2]. Their proofs are analogous to the group identity case changing GI by LPI. Corollary 1.4. Let A be a semiprime algebra over an commutative domain R with R > d. If U(A) satisfies a LP I, then every idempotent element of R−1A is central. Corollary 1.5. Let A be an algebra over a field K with U(A) satisfying a LP I, and a, b ∈ A such that a2 = b2 = 0. (1) If K > 2d, then (ab)2d = 0. (2) If ab is nilpotent, then (ab)2d = 0. Corollary 1.6. Let K be any field and n ≥ 2. Then the following are equivalent (1) U(Mn(K)) satisfies the LP I. (2) K is a finite field. (3) U(Mn(K)) satisfies a group identity. In addition, if one of these conditions is hold, then K ≤ 2d and n ≤ 2logK2d + 2 ≤ 2log22d + 2. In general, Proposition 1.2 is not hold without the restrictions about exponents. If it does, then bacu is algebraic for all u ∈ A, in particular, for c = b and u = a, ba is algebraic. It follows that by Vandermonde augument with R a infinite commutative domain ba is nilpotent. In general, ba is not a nilpotent element. For example, let A = Mn(R), n > 1. By Amitsur − Levitzki theorem [3] A satisfies a LP I f1(x1, ..., x2n) = S2n · (x1 · · · x2n)−1 where s2n is the polynomial standard. However A with a = e21 and b = e12, ba = e11 is not nilpotent. Note that this algebra also satisfies f2(x1, ..., x2n) = f1(x1, ..., x2n) + S2n (an LP I with some words having sum of exponents zero at every variable). By the same arguments we can see that if A = Mn(R) be n > 1 and R a infinite commutative domain, there is not exist a non-zero polynomial g ∈ R[X] not zero such that g(ab) = 0 for all a, b ∈ A with a2 = b2 = 0. Note that the other side of Proposition 1.2 is true. In fact, suppose that R is finite, so A is finite. From this for all matrices x ∈ A there exist integers r > t > 0 such that xr = xt. So x satisfies the polynomial px(X) = X r − X t. Then defining g = Y px one has g(ab) = 0 for all a, b ∈ A with a2 = b2 = 0, which is a contradiction. x∈A 6 CLAUDENIR FREIRE, RAMON CODAMO Lemma 1.7. Let A = Mn(R) be n > 1 and R a commutative domain. Then R is infinite if and only if there is not any non-zero polynomial g ∈ R[X] such that g(ab) = 0 for all a, b ∈ A with a2 = b2 = 0. Proposition 1.8. Let F = R[x, y] (x2, y2) where R is a infinite commutative domain. Then F satisfies any polynomial identity f of Mn(R) where n ≥ 2. Also, U(F ) does not satisfy the standard polynomial S2 and S3. In particular, F satisfies S2n if only if n ≥ 2. Proof. Suppose by contradiction that F does not satisfy an identity f of Mn(R) where n ≥ 2. Then there exist u1 = g1(x, y), ..., u2n = g2n(x, y) ∈ F with So (multiplying by x or y on left or right if necessary) we obtain f (u1, ..., u2n) 6= 0. g(xy) = f (u1, ..., u2n) 6= X fiu2gi, i u = x, y. Then, for all a, b ∈ A = Mn(R) such that a2 = b2 = 0, we have g(ab) = 0 it is a contradiction by Lemma 1.7. By evaluation of S2 and S3 on units 1 + x, 1 + y and (1 + x)(1 + y), we see that U(F ) does not satisfy S2 and S3. (cid:3) Corollary 1.9. Let B be an algebra over a commutative domain R whose U(A) satisfies the standard polynomial S3. Then there exists a polynomial g ∈ R[X] such that g(ab) = 0 for all a, b ∈ B with a2 = b2 = 0. Proof. By applying S3 on X, Y and XY we obtain S3(X, Y, XY ) = (Y X)2 − X 2Y 2 − Y X 2Y + XY 2X. So S3(1 + x, 1 + y, (1 + x)(1 + y)) 6= 0 in F , then multiplying by x on the left and y on the right we obtain a polynomial g(xy) = xS3y such that g(ab) = 0. (cid:3) Corollary 1.10. Let F be the algebra in Proposition 1.8. The F does not satisfy the group identity w = 1, with R a infinite field. Proof. If F satisfies a group identity w = 1. By Lemma 3.1 [6] there exists a polynomial g(t) ∈ R[t] such that g(x y) = 0. Thus g(xy) = 0, that is, g(xy) = P fiu2gi, u = x, y. Then, g(ab) = 0 for all a, b ∈ Mn(R) such that a2 = b2 = 0 and this contradicts Lemma 1.1. (cid:3) With some adaptations Lemma 3.1 [6] holds whenever R is an infinite commu- tative domain. It follows that Corollary 1.5 holds if R is an infinite commutative domain. Note that F is an algebra with LP I ( having more than one word which not constant) with P expwi = 0 without group identity (LP I with only one non- constant word). ALGEBRAS WITH LAURENT POLYNOMIAL IDENTITY 7 References [1] A. Giambruno, S.Sehgal, and A. Valenti, Group algebras whose units satisfy a group identity, Proc. Amer. Math. Soc. 125 (1997), 629-634. MR 97f:16056 [2] A. Giambruno, E. Jespers, and A. Valenti, Group identities on units of rings, Arch. Mathj. (Basel) 63 (1994), no. 4, 291-296. MR 95:16044 [3] D.S. Passman, The algebraic structure of group rings, Dover, New York 2011 [4] J. Z. Gonalves, M.A. Dokuchaev Identities on units of alge braic algebras, Journal of Algebra 250, 638-646 (2002) [5] J.Z. Gonalves, Free subgroups of units in group rings, Canad. Math. Bull.27 (1984), no. 3, 309-312. MR 85k:20021 [6] Liu, Chia-Hsin, Group Algebras With Units Satisfying a Group Identity, Proc. Amer. Math. Soc. 127, (1999f), 327-336. Departmento de Matemtica, Universidade Federal do Amazonas, Amazonas, Man- aus, Brasil E-mail address: [email protected] E-mail address: [email protected]
1009.3904	2	1009	2010-09-22T18:19:43	Unitary SK_1 of semiramified graded and valued division algebras	[ "math.RA", "math.KT" ]	We prove formulas for the unitary SK_1 of a semiramified graded division algebra (or valued division algebra over a Henselian field) with a unitary involution. These formulas generalize earlier formulas of Yanchevskii, (and Platonov and Ershov for the nonunitary SK_1).	math.RA	math	UNITARY SK1 OF SEMIRAMIFIED GRADED AND VALUED DIVISION ALGEBRAS A. R. WADSWORTH 1. Introduction Let D be a division algebra finite-dimensional over its center K. Then, SK1(D) = {d ∈ D∗ NrdD(d) = 1}(cid:14) [D∗, D∗], where NrdD denotes the reduced norm and [D∗, D∗] is the commutator group of the group of units D∗ of D. If D has a unitary involution τ (i.e., an involution τ on D with τK 6= id), then the unitary SK1 for τ on D is (1.1) SK1(D, τ ) = Σ′ τ (D)(cid:14) Στ (D), 0 1 0 2 p e S 2 2 ] . A R h t a m [ where Σ′ τ (D) = {d ∈ D∗ NrdD(d) = τ (NrdD(d))} and Στ (D) = (cid:10){d ∈ D∗ d = τ (d)}(cid:11). The groups SK1(D) and SK1(D, τ ) are of considerable interest as subtle invariants of D, and as reduced Whitehead groups for certain algebraic groups (cf. [Ti], [P6], [G]). 2 v 4 0 9 3 . 9 0 0 1 : v i X r a In this paper we will prove formulas for SK1(E) and SK1(E, τ ) for E a semiramified graded division algebra E of finite rank over its center. In view of the isomorphisms in [HW1, Th. 4.8] and [HW2, Th. 3.5], the formulas for E imply analogous formulas for SK1 and unitary SK1 for a tame semiramified division algebra D over a Henselian valued field K. The formulas thus obtained in the Henselian case generalize ones given by Platonov for SK1(D) and Yanchevskiı for SK1(D, τ ) for bicyclic decomposably semiramified division algebras over iterated Laurent fields. Most of our work will be in the unitary setting, which is not as well developed as the nonunitary setting. Ever since Platonov gave examples of division algebras with SK1(D) nontrivial there has been ongoing interest in SK1. Platonov showed in [P5] that nontriviality of SK1(D) implies that the algebraic group group SL1(D) (with K-points {d ∈ D NrdD(d) = 1}) is not a rational variety. Also, Voskresenskiı showed in [V1] and [V2, Th., p. 186] that SK1(D) ∼= SL1(D)/R, the group of R-equivalence classes of the variety SL1(D). The corresponding unitary result, SK1(D, τ ) ∼= SU1(D, τ )/R was given in [Y5, Remark, p. 537] and [CM, Th. 5.4]. More recently, Suslin in [Su1] and [Su2] has related SK1(D) to certain 4-th cohomology groups associated to D, and has conjectured that whenever the Schur index ind(D) is not square-free then SK1(D ⊗K L) is nontrivial for some field L ⊇ K. (This has been proved by Merkurjev in [M1] and [M4] if 4 ind(D), but remains open otherwise.) Nonetheless, explicit computable formulas for SK1(D) and SK1(D, τ ) have remained elusive, and are principally available, when ind(D) > 4, only for algebras over Henselian fields (cf. [E2] and [HW1, Th. 3.4]) and quotients of iterated twisted polynomial algebras (cf. [HW1, Th. 5.7]). Platonov's original examples with nontrivial SK1 in [P1] and [P2] were division algebras D over a twice iterated Laurent power series field K = k(((x))((y)), where k is a local or global field or an infinite algebraic extension of such a field. His K has a naturally associated rank 2 Henselian valuation which extends uniquely to a valuation on D. With respect to this valuation, his D is tame and "decomposably The author would like to thank R. Hazrat and the Queen's University, Belfast and J.-P. Tignol and the Universit´e Catholique de Louvain for their hospitality while some of the research for this paper was carried out. 1 2 A. R. WADSWORTH semiramified" and, in addition, its residue division algebra D is a field with D = L1 ⊗k L2, where each Li is cyclic Galois over k. His basic formula for such D is: SK1(D) ∼= Br(D/k)(cid:14)(cid:2) Br(L1/k) · Br(L2/k)(cid:3), (1.2) where k is any field, Br(k) is the Brauer group of k, and for a field M ⊇ k, Br(M/k) denotes the relative Brauer group ker(Br(k) → Br(M )), a subgroup of Br(k). That D is tame and semiramified means that value group the valuation on D. We say that D is decomposably semiramified (abbreviated DSR) if D is a tensor product of cyclic tame and semiramified division algebras. Using (1.2) with k a global field or an algebraic extension of a global field, Platonov showed in [P4] that every finite abelian group and some infinite abelian groups of bounded torsion appear as SK1(D) for suitable D. [D : K] = ΓD : ΓK =p[D : K] and D is a field separable (hence abelian Galois) over K, where ΓD is the Shortly after Platonov's work, Yanchevskiı obtained in [Y2], [Y3], [Y4] similar results for the unitary SK1 for similar types of division algebras, namely D decomposably semiramified over K = k((x))((y)), with k any field, given that D has a unitary involution τ with fixed field K τ = ℓ((x))((y)) for some field ℓ ⊆ k with [k : ℓ] = 2. Yanchevskiı's key formula (when D = L1 ⊗k L2 as above) is: SK1(D, τ ) ∼= Br(D/k; ℓ)(cid:14)(cid:2) Br(L1/k; ℓ) · Br(L2/k; ℓ)(cid:3), where for a field M ⊇ k, Br(M/k; ℓ) = ker(cid:0)cork→ℓ : Br(M/k) → Br(ℓ)(cid:1); this is the subgroup of Br(k) consisting of the classes of central simple k-algebras split by M and having a unitary involution τ with fixed field kτ = ℓ. He used this in [Y4] with k and ℓ global fields to show that any finite abelian group is realizable as SK1(D, τ ). He obtained remarkably similar analogues for the unitary SK1 to other results of Platonov for the nonunitary SK1, but generally with substantially more difficult and intricate proofs. (1.3) Ershov showed in [E1] and [E2] that the natural setting for viewing Platonov's examples of nontrivial SK1(D) is that of tame division algebras D over a Henselian valued field K. (Platonov considered his K in a somewhat cumbersome way as a field with complete discrete valuation with residue field which also has a complete discrete valuation.) The Henselian valuation on K has a unique extension to a valuation on D, and Ershov gave exact sequences that describe SK1(D) in terms of various data related to the residue division ring D. In particular he showed (combining [E2, p. 69, (6) and Cor. (b)]) that if D is DSR (with K Henselian), then ∗ ). (1.4) More recently, there has been work on associated graded rings of valued division algebras, see especially [HwW2], [Mou], [TW]. The tenor of this work has been that for a tame division algebra D over a Henselian valued field, most of the structure of D is inherited by its associated graded ring gr(D), while gr(D) is often much easier to work with than D itself. This theme was applied quite recently by R. Hazrat and the author in [HW1] and [HW2] to calculations of SK1 and unitary SK1. It was shown in [HW1, Th. 4.8] that if D is tame over K with respect to a Henselian valuation, then SK1(D) ∼= SK1(gr(D)); the corresponding result for unitary SK1 was proved in [HW2, Th. 3.5]. Calculations of SK1 in the graded setting are significantly easier and more transparent than in the original ungraded setting, allowing almost effortless recovery of Ershov's exact sequences, with some worthwhile improvements. Notably, it was shown in [HW1, Cor. 3.6(iii)] that if K is Henselian and D is tame and semiramified (but not necessarily DSR), then there is an exact sequence SK1(D) ∼= bH −1(Gal(D/K), D H ∧ H −→ bH −1(H, D ∗ ) −→ SK1(D) −→ 1, ∗ When D is DSR, the image of H ∧ H in bH −1(H, D ) is trivial, yielding (1.4). Then, Platonov's formula (1.2) is obtained from (1.4) via the following isomorphism: For a field M = L1 ⊗k L2 where each Li is cyclic Galois over k, where H = Gal(D/K) ∼= ΓD/ΓK . (1.5) bH −1(Gal(M/k), M ∗) ∼= Br(M/k)/(cid:2) Br(L1/k) · Br(L2/k)(cid:3). (1.6) UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 3 See (3.6) -- (3.9) below for a short proof of (1.6) using facts about abelian crossed products. When D is semiramified but not DSR, the contribution of the first term in (1.5) can be better understood in terms of the I ⊗ N decomposition of D: Our semiramified D is equivalent in Br(K) to I ⊗K N , where I is inertial (= unramified) over K and N is DSR, so N ∼= D and ΓN = ΓD. Thus, the bH −1 term in (1.5) coincides with SK1(N ). We will show in Cor. 3.8(i) below that the image of H ∧ H in bH −1(H, D ) is expressible in terms of parameters describing the residue algebra I of I, which is central simple over K and split by the field D. This I does not show up within D or D, but nonetheless has significant influence on the structure of D. (For example, it determines whether D can be a crossed product or nontrivially decomposable -- see [JW, pp. 162 -- 166, Remarks 5.16]. In [JW] DSR algebras were called "nicely semiramified," and abbreviated NSR. We prefer the more descriptive term decomposably semiramified.) Also, I is not uniquely determined by D, but determined only modulo the group Dec(D/K) of simple K- algebras which "decompose according to D " -- see §3 below for the definition of Dec(D/K). In the bicyclic case where D is semiramified and K Henselian and D ∼= L1 ⊗K L2 with each Li cyclic Galois over K, we will show in Cor. 3.8(ii) that ∗ SK1(D) ∼= Br(D/K)(cid:14)(cid:2) Br(L1/K) · Br(L2/K) · h[I]i(cid:3), (1.7) which is a natural generalization of Platonov's formula (1.2). The principal aim of this paper is to prove unitary versions of the results described above for nonuni- tary SK1, especially (1.4), (1.6), and (1.7). The unitary versions of these are, respectively, Th. 7.1(i), Prop. 6.2, and Th. 7.3(ii). Along the way, it will be necessary to develop a unitary version of the I ⊗ N decomposition for semiramified division algebras. This is given in Prop. 4.5. In the final section we will apply some of these formulas to give an example where the natural map SK1(D, τ ) → SK1(D) is not injective. This paper is a sequel to [HW2], which describes the equivalence of the graded setting and the Henselian valued setting for computing unitary SK1, and has calculations of SK1(D, τ ) for several cases other than the semiramified one considered here. However, the present paper can be read independently of [HW2]. We will work here primarily with graded division algebras, where the calculations are more transparent than for valued algebras. Some basic background on the graded objects is given in §2. But we reiterate that by [HW2, Th. 3.5] every result in the graded setting yields a corresponding result for tame division algebras over Henselian valued fields. While what is proved here is for a rather specialized type of algebra, we note that detailed knowledge of SK1 in special cases sometimes has wider consequences. See, e.g., the paper [RTY] where Suslin's conjecture is reduced to the case of cyclic algebras. See also [W, Th. 4.11], where the proof of nontriviality of a cohomological invariant of Kahn uses a careful analysis of SK1(D) for the D in Platonov's original example. From the perspective of algebraic groups, it is perhaps unsurprising that there should be results for the unitary SK1 similar to those in the nonunitary case. For, SL1(D) is a group of inner type An−1 where n = deg(D), and SU1(D, τ ) is a group of outer type An−1 (cf. [KMRT, Th. (26.9)]). Nonetheless, the similarities in formulas for SK1(D, τ ) given in Yanchevskiı's work and in [HW2] and here to those for SK1(D) seem quite striking. Likewise, the results by Rost on SK1(D) for biquaternion algebras (see [KMRT, §17A]) and by Merkurjev in [M2] for arbitrary algebras of degree 4, have a unitary analogue proved by Merkurjev in [M3]. This suggests that a further analysis of the unitary SK1 would be worthwhile, notably to investigate whether there are unitary versions of the deep results by Suslin [Su2] and Kahn [K] relating SK1(D) to higher ´etale cohomology groups. 4 A. R. WADSWORTH 2. Graded division algebras and simple algebras We will be working throughout with graded algebras graded by a torsion-free abelian group. We now set up the terminology for such algebras and recall some of the basic facts we will use frequently. Let Γ be a torsion-free abelian group, and let R be a ring graded by Γ, i.e., R =Lγ∈Γ Rγ, where each Rγ is an additive subgroup of R and Rγ · Rδ ⊆ Rγ+δ for all γ, δ ∈ Γ. The homogeneous elements of R are those lying inSγ∈Γ Rγ. If r ∈ Rγ, r 6= 0, then we write deg(r) = γ. The grade set of R is ΓR = {γ ∈ Γ Rγ 6= {0}}. (We work only with gradings by torsion-free abelian groups because we are interested in the associated graded rings determined by valuations on division algebras; for such rings the grading is indexed by the value group of the valuation, which is torsion-free abelian.) If R′ = Lγ∈Γ R′ γ is another graded ring, a graded ring homomorphism ϕ : R → R′ is a ring homomorphism such that ϕ(Rγ) ⊆ R′ γ for all γ ∈ Γ. If ϕ is an isomorphism, we say that R and R′ are graded ring isomorphic, and write R ∼=g R′. For example, if a ∈ R is homogeneous and a ∈ R∗, the group of units of R, then the map int(a) : R → R given by r 7→ ara−1 is a graded ring automorphism of R. A graded ring E =Lγ∈Γ Eγ is said to be a graded division ring if every nonzero homogeneous element of E lies in the multiplicative group E∗ of units of E. See [HwW2] for background on graded division ring and proofs of the properties mentioned here. Notably (as Γ is torsion-free abelian), E has no zero divisors, E∗ consists entirely of homogeneous elements, ΓE is a subgroup of Γ, E0 is a division ring, and each nonzero homogeneous component Eγ of E is a 1-dimensional left and right E0-vector space. Furthermore, if M is then M is a free E-module with a homogeneous base, and any two such bases have the same cardinality; this cardinality is called the dimension of M and denoted dimE(M). Any such M is therefore called a left graded E-vector space. any left graded E- module (i.e., an E-module such that M =Lγ∈Γ Mγ with Eγ·Mδ ⊆ Mγ+δ for all γ, δ ∈ Γ), A commutative graded division ring T =Lγ∈Γ Tγ is called a graded field. Such a T is an integral domain; let q(T) denote the quotient field of T. A graded ring A which is a T-algebra is called a graded T-algebra if the module action of T on A makes A into a graded T-module. When this occurs, T is graded isomorphic to a graded subring of the center of A, which is denoted Z(A). All graded T-algebras considered in this paper are assumed to be finite-dimensional graded T-vector spaces. Note that if A is a graded T-algebra, then A⊗T q(T) is a q(T)-algebra of the same dimension. That is, [A : T] = [A⊗T q(T) : q(T)], where [A : T] denotes dimT(A) and [A ⊗T q(T) : q(T)] = dimq(T)(A ⊗T q(T)). with (A ⊗T B)γ = Pδ∈Γ Aδ ⊗T0 Bγ−δ for all γ ∈ Γ. Clearly, ΓA⊗TB = ΓA + ΓB. Also, if C is a finite- Note that if A and B are graded algebras over a graded field T then A ⊗T B is also a graded T-algebra dimensional T0-algebra, then C ⊗T0 A is a graded T-algebra with (C ⊗T0 A)γ = C ⊗T0 Aγ for all γ ∈ Γ, and ΓC⊗T0 A = ΓA. A graded T-algebra A is said to be simple if it has no homogeneous two-sided ideals except A and {0}. A is called a central simple T-algebra if in addition its center Z(A) is T. The theory of simple graded algebras is analogous to the usual theory of finite-dimensional simple algebras. This is described in [HwW2, §1], where proofs of the following facts can be found. There is a graded Wedderburn Theorem for simple graded algebras: Any such A is graded isomorphic to EndE(M) for some finite-dimensional graded vector space M over a graded division algebra E, and E is unique up to graded isomorphism. Also, while A0 need not be simple, it is always semisimple, and A0 ∼=Qs j=1 Mℓj (E0) for some ℓj × ℓj matrix rings over E0 (see the proof of Lemma 2.2 below). We write [A] for the equivalence class of A under the equivalence relation ∼g given by: A ∼g A′ iff A ∼=g EndE(M) and A′ ∼=g EndE(M′) for the same graded division algebra E. The Brauer group (of graded algebras) for T is Br(T) = {[A] A is a graded central simple T-algebra}, UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 5 with the well-defined group operation [A] · [A′] = [A ⊗T A′]. When A ∼=g EndE(M) as above, then [A] = [E], and up to graded isomorphism E is the only graded division algebra with A ∼g E. There is a graded version of the Double Centralizer Theorem, see [HwW2, Prop. 1.5] and also the Skolem-Noether Theorem, see [HwW2, Prop. 1.6]. We recall the latter, since it has an added condition not appearing in the ungraded version. Proposition 2.1 ([HwW2, Prop. 1.6(b),(c)]). Let A be a central simple graded algebra over the graded field T, and let B and B′ be simple graded T-subalgebras of A. Let C = CA(B), the centralizer of B in A, and let Z = Z(C) = Z(B) and C′ = CA(B′). Let α : B → B′ be a graded T-algebra isomorphism. Then there is a homogeneous a ∈ A∗ such that α(b) = aba−1 for all b ∈ B if and only if there is a graded T-algebra isomorphism γ : C → C′ such that γZ = αZ. Such a γ exists whenever C0 is a division ring. If E is a graded division algebra over a graded field T, we write [E : T] for dimT(E). A basic fact is the Fundamental Equality (2.1) where ΓE : ΓT denotes the index in ΓE of its subgroup ΓT. Also, it is known that Z(E0) is abelian Galois over T0, and there is a well-defined group epimorphism [E : T] = [E0 : T0]ΓE : ΓT, ΘE : ΓE → Gal(Z(E0)/T0) given by ΘE(γ)(z) = aza−1 for any z ∈ Z(E0) and a ∈ Eγ \ {0}. (2.2) Clearly, ΓT ⊆ ker(ΘE), so ΘE induces an epimorphism of finite groups ΘE : ΓE/ΓT → Gal(Z(E0)/E0). The terminology for different cases in (2.1) is carried over from valuation theory: We say that a graded field S ⊇ T is inertial over T if [S0 : T0] = [S : T] < ∞ and the field S0 is separable over T0. When this occurs, ΓS = ΓT, and the graded monomorphism S0 ⊗T0 T → S given by multiplication in S is surjective by dimension count; so S ∼=g S0 ⊗T0 T. At the other extreme, we say that a graded field J ⊇ T is totally ramified over T if ΓJ : ΓT = [J : T] < ∞. When this occurs, J0 = T0 and, more generally, for any γ ∈ ΓT, we have Jγ = Tγ since dimT0(Jγ) = dimJ0(Jγ) = 1 = dimT0(Tγ). There is an extensive theory of finite-degree graded field extensions; [HwW1] is a good reference for what we need here. Notably, there is a version of Galois theory: For graded fields T ⊆ F, with [F : T] < ∞, the (graded) Galois group of F over T is defined to be: Gal(F/T) = {ψ : F → F ψ is a graded field automorphism of F and ψT = id}. Galois theory for graded fields follows easily from the classical ungraded theory since for the quotient fields of F and T we have q(F) ∼= F ⊗T q(T), so [q(F) : q(T)] = [F : T], and there is a canonical isomorphism Gal(F/T) → Gal(q(F)/q(T)) (the usual Galois group) given by ψ 7→ ψ ⊗ idq(T) (see [HwW1, Cor. 2.5(d), Th. 3.11 ]). Thus, F is Galois over T iff q(F) is Galois over q(T), iff Gal(F/T) = [F : T], iff T is the fixed ring of Gal(F/T). This will arise here primarily in the inertial case: Suppose S is a graded field which contains and is inertial over T, with [S : T] < ∞. For any ψ ∈ Gal(S/T) clearly the restriction ψS0 lies in Gal(S0/T0). Moreover, as S ∼=g S0 ⊗T0 T, for any ρ ∈ Gal(S0/T0) we have ρ ⊗ idT ∈ Gal(S/T). Thus, the restriction map ψ 7→ ψS0 yields a canonical isomorphism Gal(S/T) → Gal(S0/T0). Hence, as [S : T] = [S0 : T0], S is Galois over T iff S0 is Galois over T0. Just as in the ungraded case, we can use Galois graded field extensions to build central simple graded al- gebras. If F is a Galois graded field extension of T, set G = Gal(F/T) and take any 2-cocycle f ∈ Z 2(G, F∗). Then we can build a crossed product graded algebra B = (F/T, G, f ) =Lσ∈G Fxσ with multiplication given by (axσ)(bxρ) = aσ(b)f (σ, ρ)xσρ for all a, b ∈ F, σ, ρ ∈ G. The grading is given by viewing B as a left graded GPρ∈G deg(f (σ, ρ)). A short calcu- F-vector space with (xσ)σ∈G as a homogeneous base with deg(xσ) = 1 lation shows that deg(f (σ, τ ) xστ ) = deg(xσ) + deg(xτ ) for all σ, τ ∈ G; it follows easily that B is a graded T-algebra. Indeed, B is a simple graded algebra with Z(B) ∼=g T. Conversely, if A is any central simple graded T-algebra containing F as a strictly maximal graded subfield (i.e., [F : T] = deg(A) (=pdimT(A) ), SK1(B) = {b ∈ B NrdB(b) = 1}(cid:14) [B∗, B∗]. (2.3) 6 A. R. WADSWORTH then by the graded Double Centralizer Theorem CA(F) = F = Z(F); so the graded Skolem-Noether The- orem, Prop. 2.1 above, applies to the graded isomorphisms in G, which yields that A ∼=g (F/T, G, f ) for some f ∈ Z 2(G, F∗). From this one deduces, as in the ungraded case, that Br(F/T) ∼= H 2(G, F∗), where Br(F/T) denotes the kernel of the canonical map Br(T) → Br(F) given by [A] 7→ [A ⊗T F]. In particu- lar, if Gal(F/T), is cyclic, say with generator σ, then for any b ∈ T∗ we have the graded cyclic algebra C = (F/T, σ, b) = Lr−1 i=0 Fyi, in which ya = σ(a)y for all a ∈ F and yr = b, where r = [F : T]. For the grading, we view C as a left graded F-vector space with homogeneous base (1, y, y2, . . . , yr−1) with deg(yi) = i r deg(b). Then C is a central simple graded T-algebra. There are also norm maps in the graded setting: If T ⊆ F are graded fields with [F : T] < ∞, then because F is a free module the norm NF/T : F → T can be defined by c 7→ det(λc), where for c ∈ F, λc ∈ HomT(F, F) is the map a 7→ ca. Clearly, NF/T(c) = Nq(F)/q(T)(c), where Nq(F)/q(T) is the usual norm for the quotient fields. Also, if c ∈ F is homogeneous, say c ∈ Fγ, then NF/T(c) ∈ T[F:T]γ. Likewise, if B is a central simple graded T-algebra, then it is known that B is an Azumaya algebra of constant rank [B : T] over T; hence there is a reduced norm map NrdB : B → T. It is easy to see that for the central ring of quotients q(B) = B⊗T q(T) of B, we have q(B) is a central simple algebra over the field q(T), and it is known (see [HW1, proof of Prop. 3.2(i)]) that for any b ∈ B, NrdB(b) = Nrdq(B)(b), where Nrdq(B) : q(B) → q(T) is the reduced norm for q(B). As usual, b ∈ B∗ iff NrdB(b) ∈ T∗. Also, if b ∈ Bγ, then NrdB(b) ∈ Tdeg(B)γ. Now assume further that B is a graded division algebra, so that all its units are homogeneous. Then for the commutator group [B∗, B∗] of B, we have [B∗, B∗] ⊆ {b ∈ B NrdB(b) = 1} ⊆ B∗ 0. We define The fact that both terms in the right quotient lie in B∗ tractable in this graded setting than for ungraded division algebras. 0 often makes that calculation of SK1(B) much more We need terminology for some types of simple graded algebras and graded division algebras over a graded field T. A central simple graded T-algebra I is said to be inertial (or unramified) if [I0 : T0] = [I : T]. When this occurs, the injective graded T-algebra homomorphism I0 ⊗T0 T → I is surjective by dimension count. So, ΓI = ΓT and I ∼=g I0 ⊗T0 T. Hence, I0 must be a central simple T0-algebra. Moreover, if we let D be the T0-central division algebra with I0 ∼= Mℓ(D), then D ⊗T0 T is clearly a graded division algebra over T which is also inertial over T, and D ⊗T0 T ∼g I (see Lemma 2.2 below). The principal focus of this paper is on calculating SK1 and unitary SK1 for semiramified graded di- vision algebras. Let E be a central graded division algebra over a graded field T. This E is said to be semiramified if [E0 : T0] = ΓE : ΓT = deg(E) and E0 is a field. Since E0 = Z(E0), E0 is abelian Ga- lois over T0 and the epimorphism ΘE : ΓE/ΓT → Gal(E0/T0) (see (2.2)) must be an isomorphism as ΓE/ΓT = [E0 : T0] = Gal(E0/T0). Furthermore, E has the graded subfield E0T ∼=g E0 ⊗T0 T, which is inertial and Galois over T with Gal(E0T/T) ∼= Gal(E0/T0). Because [E0T : T] = deg(E), the graded Double Centralizer Theorem [HwW2, Prop. 1.5] shows that CE(E0T) = E0T, and hence E0T is a maximal graded subfield of E; thus, E is a graded abelian crossed product, as will be discussed in §3. There is a significant special class of semiramified graded division algebras which are building blocks for all semiramified algebras. We say that a T-central graded division algebra N is decomposably semiramified (abbreviated DSR) if N has a maximal graded subfield S which is inertial over T and another maximal graded subfield J which is totally ramified over T. The graded Double Centralizer Theorem yields that [S : T] = [J : T] = deg(N). We thus have deg(N) = [J : T] = ΓJ : ΓT ≤ ΓN : ΓT and deg(N) = [S : T] = [S0 : T0] ≤ [N0 : T0]. (2.4) Since ΓN : ΓT [N0 : T0] = [N : T] = deg(N)2, the inequalities in (2.4) must be equalities, showing that N0 = S0 and ΓN = ΓJ, hence N is semiramified. We call such an N decomposably semiramified because it UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 7 is always decomposable into a tensor product of cyclic semiramified graded division algebras (see Prop. 4.4 below for the unitary analogue to this). The older term for such algebras is nicely semiramified (NSR). While our focus in this paper is on central graded division algebras we will often take tensor products of such algebras, obtaining simple graded algebras which may have zero divisors. The next lemma allows us to recover information about the graded division algebra Brauer equivalent to such a tensor product. Lemma 2.2. Let B be a central simple graded algebra over the graded field T. Let D be the graded division algebra Brauer equivalent to B. Suppose B0 is a simple ring. Then, (i) B ∼=g Mℓ(D) for some ℓ, where the matrix ring Mℓ(D) is given the standard grading in which B = ΓD, and ΘB = ΘD, where (cid:0)Mℓ(D)(cid:1)γ = Mℓ(Dγ) for all γ ∈ ΓD. Hence, B0 ∼= Mℓ(D0), ΓB = Γ′ B = {deg(b) b ∈ B∗ and b is homogeneous}, and Γ′ ΘB : Γ′ B → Gal(Z(B0)/T0) is given by deg(b) 7→ int(b)Z(B0), for any homogeneous b ∈ B∗ (2.5) where int(b) denotes conjugation by b. (ii) B is a graded division algebra if and only if B0 is a division ring. Proof. (i) By the graded Wedderburn Theorem [HwW2, Prop. 1.3], B ∼=g EndD(V) for some right graded vector space V of D. The grading on EndD(V) is given by (cid:0)EndD(V)(cid:1)ε = {f ∈ EndD(V) f (Vδ) ⊆ Vε+δ for all δ ∈ ΓV}. Take a homogeneous D-base (v1, . . . , vℓ) of V, and let γi = deg(vi), for 1 ≤ i ≤ ℓ; then, ΓV =Sℓ i=1 γi + ΓD. Let δ1 + ΓD, . . . , δs + ΓD be the distinct cosets of ΓD appearing in ΓV, and let tj be the number of i with γi ∈ δj + ΓD. So, t1 + . . . + ts = ℓ. By replacing each vi by a D∗-multiple of it, we may assume that deg(vi) = δj whenever γi ∈ δj + ΓD. Then, we can reindex (v1, . . . , vℓ) = (v11, . . . , v1t1 , . . . , vs1, . . . , vsts) with deg(vjk) = δj for all j, k. Then, Vδj = D0-span(vj1, . . . , vjtj ) for j = 1, 2, . . . , s, and This is a direct product of s simple algebras. Since we have assumed that B0 is simple, we must have s = 1, i.e., all the vi have degree δ1. It is then clear that when we use the base (v1, . . . , vℓ) for the isomorphism EndD(V) ∼= Mℓ(D), the grading on Mℓ(D) induced by the isomorphism is the standard grading. Thus, B ∼=g Mℓ(D) and hence B0 ∼= Mℓ(D0) and ΓB = ΓD. Then, Γ′ Mℓ(D) = ΓD and, when we identify Z(B0) with Z(Mℓ(D0)) and with Z(D0), clearly ΘB = ΘMℓ(D) = ΘD. B = Γ′ (ii) If B is a graded division algebra, then every nonzero homogeneous element of B lies in B∗. In particular, B0 \ {0} ⊆ B∗, so B0 is a division ring. Conversely, suppose B0 is a division ring. Since B0 is then simple, part (i) applies, showing that for some graded division algebra D, we have B ∼=g Mℓ(D) where B0 ∼= Mℓ(D0). Necessarily ℓ = 1, as B0 is a division ring. Corollary 2.3. Let I and E be central graded division algebras over a graded field T, with I inertial, and (cid:3) let D be the graded division algebra with D ∼g I ⊗T E. Then, D0 ∼ I0 ⊗T0 E0, Z(D0) ∼= Z(E0), ΓD = ΓE, and ΘD = ΘE. Since I ∼=g I0 ⊗T0 E. Hence, B0 ∼= I0 ⊗T0 E0, I0 ⊗T0 T, we have B ∼=g Proof. Let B = I ⊗T E. Z(B0) ∼= Z(I0) ⊗T0 Z(E0) ∼= Z(E0), and ΓB = ΓE. Moreover, B0 is simple as I0 is central simple over T0, so Lemma 2.2 applies to B. In particular, Γ′ B = ΓB and ΘB = ΘE. Since D is the graded division algebra with D ∼g B, the Lemma yields D0 ∼ B0 ∼= I0 ⊗T0 E0, so Z(D0) ∼= Z(B0) ∼= Z(E0), and ΓD = ΓB = ΓE, and ΘD = ΘB = ΘE. (cid:3) (cid:0)EndD(V)(cid:1)0 = (cid:8)f ∈ EndD(V) f (Vε) ⊆ Vε for all ε ∈ ΓV(cid:9) sQj=1 EndD0(D0-span(vj 1, . . . , vj tj )) ∼= sQj=1 ∼= Mtj (D0). For i, j ∈ I, if we set i∗j to be the element of I congruent to i + j mod r1Z × . . . × rkZ in Zk, and set A = Li ∈ I M zi. f (σi, σj) = zizj(zi∗j)−1 ∈ M ∗, 8 A. R. WADSWORTH 3. Abelian crossed products and nonunitary SK1 for semiramified algebras Let M be a finite degree abelian Galois extension of a field K, and let H = Gal(M/K). Let X(M/K) = Hom(H, Q/Z), the character group of H. Take any cyclic decomposition H = hσ1i× . . .×hσki, and let ri be the order of σi in H. Let (χ1, . . . , χk) be the base of X(M/K) dual to (σ1, . . . , σk); so χi(σj) = δij/ri + Z, where δij = 1 if j = i and = 0 if j 6= i. Let Li be the fixed field of ker(χi). So, M = L1⊗K . . .⊗K Lk, and for each i, Li is cyclic Galois over K with [Li : K] = ri and Gal(Li/K) = hσiKi. Let A be any central simple K-algebra containing M as a strictly maximal subfield (i.e., M is a maximal subfield of A with [M : K] = deg(A)). By the Double Centralizer Theorem, the centralizer CA(M ) is M . Recall that every algebra class in Br(M/K) is represented by a unique such A. By Skolem-Noether, for each i there is zi ∈ A∗ with int(zi)M = σi, where int(zi) denotes conjugation by zi. Set uij = zizjz−1 i z−1 j and bi = zri i . Since int(uij)M = σiσjσ−1 Take the index set I = Qk zi = zi1 product decomposition i σ−1 j = idM and int(bi)M = σri = idM , all the uij and bi lie in CA(M )∗ = M ∗. i=1{0, 1, 2, . . . , ri − 1} ⊆ Zk. For i = (i1, . . . , ik) ∈ I, set σi = σi1 k and k . So, int(zi)M = σi and, as the map i 7→ σi is a bijection I → H, we have the crossed 1 . . . σik 1 . . . zik then f ∈ Z 2(H, M ∗) and the multiplication in A is given by azi · czj = aσi(c)f (σi, σj) zi∗j, for all a, c ∈ M and i, j ∈ I. Since each f (σi, σj) is expressible as a computable product of the uij and the bi and their images un- der H, the multiplication for A is completely determined by M , H, and the uij and bi. Thus, we write A = A(M/K, σ, u, b), where σ = (σ1, . . . , σk), u = (uij)k i=1, j=1, and b = (b1, . . . , bk). k It is easy to check (cf. [AS, Lemma 1.2] or [T2, p. 423]) that the uij and the bi satisfy the following relations, for all i, j, ℓ, and uii = 1, uji = u−1 ij , σi(ujℓ)σj(uℓi)σℓ(uij) = ujℓuℓiuij (3.1) (3.2) where M hσii is the fixed field of M under hσii. It is known (cf. [AS, Th. 1.3]) that for any family of uij and bi in M ∗ satisfying (3.1) and (3.2) there is a central simple K-algebra A(M/K, σ, u, b). NM/M hσii(uij) = bi/σj(bi), Lemma 3.1. Let A = A(M/K, σ, u, b) as above, and let B = A(M/K, σ, v, c). Then, there is a well- defined abelian crossed product A(M/K, σ, w, d) where wij = uijvij and di = bici for all i, j. Moreover, A ⊗K B ∼ A(M/K, σ, w, d) (Brauer equivalent). Proof. Because the uij and bi satisfy (3.1) and (3.2) as do the vij and ci, and the σi and the norm maps are multiplicative, the wij and di also satisfy (3.1) and (3.2). Therefore A(M/K, σ, w, d) is a well-defined abelian crossed product. We have the 2-cycle f ∈ Z 2(H, M ∗) representing A defined as above by, f (σi, σj) = zizj(zi∗j)−1. The relations zri i = bi and [zi, zj] = uij are encoded in f by f (σℓ i , σi) =(1, bi, if 0 ≤ ℓ ≤ ri − 2 if ℓ = ri − 1 and f (σi, σj) =(1 uij if i < j if i > j. (3.3) UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 9 We likewise build a cocycle g ∈ Z 2(H, M ∗) for B = A(M/K, σ, v, c). Then, the cocycle f·g satisfies condi- tions corresponding to those for f in (3.3), so f·g is a cocycle for C = A(M/K, σ, w, d) where wij = uijvij and di = bici. From the group isomorphism H 2(H, M ∗) ∼= Br(M/K) it follows that A ⊗K B ∼ C. (cid:3) In Tignol's terminology in [T2], a central simple K-algebra containing M as a strictly maximal subfield decomposes according to M if A ∼= (L1/K, σ1, b1)⊗K . . .⊗K (Lk/K, σk, bk) for some b1, . . . , bk ∈ K ∗. Clearly then, A ∼= A(M/K, σ, 1, b), i.e., each uij = 1. Conversely, for any algebra A(M/K, σ, 1, b) (i.e., the zi commute with each other), each zj centralizes bi = zri i , so bi ∈ M H = K and the algebra decomposes according to M . The collection of such algebras yields an important distinguished subgroup Dec(M/K) of Br(M/K), i.e. Dec(M/K) = { [A] ∈ Br(M/K) A decomposes according to M} = { [A(M/K, σ, u, b)] every uij = 1 and every bi ∈ K ∗ }. (3.4) Since Br(Li/K) = { [(Li/K, σi, b)] b ∈ K ∗ }, we have also Dec(M/K) = Qk i=1 Br(Li/K) ⊆ Br(M/K). Tignol also also points out in [T2, p. 426] a homological characterization: From the short exact sequence of trivial H-modules 0 → Z → Q → Q/Z → 0 the long exact cohomology sequence yields the connecting homomorphism δ : H 1(H, Q/Z) → H 2(H, Z), which is an isomorphism since H i(H, Q) = 1 for i ≥ 1 as Q is uniquely divisible. For any χ ∈ X(M/K) = H 1(H, Q/Z) and any c ∈ K ∗ = H 0(H, M ∗) it is known (cf. [Se, p. 204, Prop. 2]) that under the cup product pairing ∪ : H 2(H, Z) × H 0(H, M ∗) → H 2(H, M ∗) = Br(M/K), we have δ(χ) ∪ c = [(N/K, ρN , c)], where N is the fixed field of ker(χ) and ρ ∈ H is determined by χ(ρ) = (1/χ) + Z ∈ Q/Z. Thus, the algebra class [(L1/K, σ1, b1) ⊗K . . . ⊗K (Lk/K, σk, bk)] in Br(M/K) corresponds to (δ(χ1) ∪ b1)+ . . . + (δ(χk) ∪ bk) in H 2(H, M ∗). Since the cup product is bimultiplicative and X(M/K) = hχ1, . . . , χki, we have Dec(M/K) = (cid:10) im(cid:0) ∪ : H 2(H, Z) × H 0(H, M ∗) → H 2(H, M ∗)(cid:1)(cid:11) = QK⊆L⊆M Gal(L/K) cyclic Br(L/K), (3.5) showing that Dec(M/K) is independent of the choice of the σi and the Li. (Actually, Tignol uses (3.5) as his definition of Dec(M/F ), and proves in [T2, Cor. 1.4] that this is equivalent to the definition given here in (3.4).) The case when H is bicyclic is of particular interest, i.e., H = hσ1i × hσ2i and M = L1 ⊗K L2. Then, 12 and for any algebra A = A(M/K, σ, u, b), if we set u = u12, then u determines all the uij as u21 = u−1 u11 = u22 = 1. We write, for short, A = A(u, b1, b2). The conditions in (3.2) can then be restated: b1 ∈ M hσ1i = L2, b2 ∈ M hσ2i = L1, NM/L2(u) = b1/σ2(b1), NM/L1(u) = σ1(b2)/b2. Note that NM/K (u) = NL2/K (b1/σ2(b1)) = 1. An easy calculation (cf. [AS, Th. 1.4]) shows that (3.6) A(u, b1, b2) ∼= A(u′, b′ 1, b′ 2) if and only if there exist c1, c2 ∈ M ∗ such that b′ 1 = NM/L2(c1)b1, u′ = (cid:2)c1/σ2(c1)(cid:3)(cid:2)σ1(c2)/c2(cid:3)u, These observations can be formulated homologically: Recall that bH −1(H, M ∗) = ker(NM/K )/IH (M ∗), where ker(NM/K ) = {m ∈ M ∗ NM/K (m) = 1} and, as H = hσ1i × hσ2i, IH(M ∗) = (cid:8) [a/σ1(a)] [b/σ2(b)] a, b ∈ M ∗(cid:9). 2 = NM/L1(c2)b2. We define a map and b′ (3.7) η : Br(M/K) −→ bH −1(H, M ∗) given by (cid:2)A(u, b1, b2)(cid:3) 7→ uIH(M ∗). By (3.7) above η is well-defined, and Lemma 3.1 shows that η is a group homomorphism. Given any u ∈ M ∗ with NM/K (u) = 1, Hilbert 90 gives b1 ∈ L∗ 1 so that the conditions in (3.6) are 2 and b2 ∈ L∗ (3.8) 10 A. R. WADSWORTH satisfied and the algebra A(u, b1, b2) exists. Therefore η is surjective. By (3.7), so η yields an isomorphism ker(η) = (cid:8)[A(u, b1, b2)] u = 1(cid:9) = Dec(M/K), Br(M/K)(cid:14) Dec(M/K) ∼= bH −1(Gal(M/K), M ∗) This isomorphism is known (see, e.g., [T2, Remarque, pp. 427 -- 428]); indeed, it follows by comparing Draxl's formula [D, Kor. 8, p. 133] for SK1 of the division algebras considered by Platonov in [P2] with Platonov's formula in [P2, Th. 4.11, Th. 4.17]. Its relevance for SK1 calculations is shown in the next proposition, which is the graded version of (1.4) and (1.2) above. I learned of this description of the isomorphism from Tignol. whenever M is bicyclic over K. (3.9) Proposition 3.2. Suppose N is a DSR central graded division algebra over the graded field T. Then, (i) SK1(N) ∼= bH −1(H, N∗ (ii) If N0 ∼= L1 ⊗T0 L2 with each Li cyclic Galois over T0, then 0) where H = Gal(N0/T0). SK1(N) ∼= Br(N0/T0)(cid:14) Dec(N0/T0). Proof. (i) was given in [HW1, Cor. 3.6(iv)], and (ii) follows from (i) and (3.9) above. (cid:3) We will generalize Prop. 3.2 in Th. 3.7 below by giving formulas for SK1(E) when E is semiramified but not necessarily DSR. For this we need, first, a graded version of the abelian crossed products described at the beginning of this section. Second, we need a graded version of the I⊗N decomposition for semiramified division algebras over a Henselian valued field. Here I is inertial and N is DSR. (See [JW, Lemma 5.14, Th. 5.15] for the valued I ⊗ N decomposition.) is available as CB(S) = S = Z(S); Here is the graded version of abelian crossed products. Let B be a central simple graded algebra over a graded field T. Assume that B contains a maximal graded subfield S with [S : T] = deg(B) (=p[B : T] ) such that S is Galois over T and H = Gal(S/T) is abelian. We have CB(S) = S by the graded Dou- ble Centralizer Theorem. For any cyclic decomposition H = hσ1i × . . . × hσki, the graded Skolem- Noether Theorem, Prop. 2.1, it shows that for each i there is yi ∈ B∗ with yi homogeneous and int(yi)S = σi. Set ci = yri i where ri is the order of σi in H, and set vij = yiyjy−1 0. For each i = (i1, . . . , ik) ∈ I =Qk 1 . . . yik j=1{0, 1, 2, . . . , rj − 1}, set yi = yi1 B = Li∈I S yi. For, the sum in the equation is direct since B ⊗T q(T) =Li∈I(S ⊗T q(T)) yi by the ungraded case. Then base(cid:0)yi(cid:1)i∈I, and equality holds in (3.10) by dimension count. Note that B is a left graded S-vector space with homogeneous . Then, each ci ∈ CB(S)∗ = S∗ with deg(yi) = 1 k . Then, int(yi)S = σi, and we have deg(ci), and each vij ∈ S∗ i y−1 j (3.10) ri deg(yi) = (3.11) So, r1 deg(c1), . . . , 1 rk ΓB = (cid:10) 1 deg(ck)(cid:11) + ΓS each Bδ = Li∈I Since B is determined as a graded T-algebra by S, the σi, the vij, and the ci, we write B = A(S/T, σ, v, c), where σ = (σ1, . . . , σk), v = (vij)k i=1,j=1, and c = (c1, . . . , ck). Note that the vij and the ci satisfy the identities corresponding to (3.1) and (3.2). Conversely, given any vij ∈ S∗ 0 and ci ∈ S∗ satisfying those identities there is a central simple graded T algebra A(S/T, σ, v, c). This is obtainable as B =Li∈I S yi within the ungraded abelian crossed product A = A(q(S)/q(T), σ, v, c), with the grading on B determined by that on S and deg(yi) = 1 deg(ci), as described above. To see that B is a graded ring, one uses that ri S(δ−deg(yi))yi. (3.12) and k deg(cj). ij rj kPj=1 UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 11 each σ ∈ H is a (degree-preserving) graded automorphism of S and that deg(yi · yj) = deg(yi) + deg(yj) for all i, j ∈ I, since all the vij have degree 0. This B is graded simple, since any nontrivial proper homogeneous ideal would localize to a nontrivial proper ideal of the simple q(T)-algebra A. Remark 3.3. The graded analogue to Lemma 3.1 holds, with the same proof, since for S Galois over T, we have Br(S/T) ∼= H 2(Gal(S/T), S∗). The graded abelian crossed products we work with here will have S inertial over T and will be semi- ramified, as described in the next lemma. Lemma 3.4. Let S be an inertial graded field extension of T with S abelian Galois over T. Let H = Gal(S/T) = hσ1i × . . . × hσki as above with ri the order of σi, and let B = A(S/T, σ, v, c) be a graded abelian crossed product. Let δi = 1 deg(ci) ∈ ΓB and δi = δi + ΓT ∈ ΓB/ΓT. Then, B is a semiramified ri graded division algebra if and only if each δi has order ri and δ1, . . . , δk are independent in ΓB/ΓT. When this occurs, B0 = S0 and ΓB/ΓT = hδ1i × . . . × hδki ∼= H. Proof. Since S is inertial and Galois over T, S0 is Galois over T0 with Gal(S0/T0) ∼= Gal(S/T) = H. We identify H with Gal(S0/T0). We have S0 ⊆ B0 and [S0 : T0] = [S : T] = deg(B). Suppose B is a semiramified graded division algebra. Then, [B0 : T0] = deg(B) = [S0 : T0], so B0 = S0. Since B is semiramified, the epimorphism ΘB : ΓB/T → H is an isomorphism, as noted in §2. When we deg(ci) = δi, we have ΘB(δi) = σi. represent B =Li∈I Syi as above, since int(yi) = σi and deg(yi) = 1 Hence, δi has the same order ri as σi, and ri ΓB/ΓT = Θ −1 B (H) = Θ −1 B (hσ1i) × . . . × Θ −1 B (hσki) = hδii × . . . × hδki, so the δi are independent in ΓB/ΓT. Conversely, suppose each δi has order ri and the δi are independent in ΓB/ΓT. Then, Hence, ΓB : ΓT ≥ kQi=1hδii = r1 . . . rk = H = deg(B). [B0 : T0] = [B : T](cid:14)ΓB : ΓT ≤ deg(B)2/ deg(B) = [S0 : T0]. (3.13) Since S0 ⊆ B0, (3.13) shows that B0 = S0, so equality holds in (3.13). Since B0 is a field, B is a graded division algebra by Lemma 2.2(ii), and it is semiramified by the equality in (3.13). (cid:3) Observe that if E is any semiramified graded T-central division algebra, then E is a graded abelian crossed product as described in Lemma 3.4. For, E0T is a maximal graded subfield of E which is inertial and Galois over T with Gal(E0T/T) ∼= Gal(E0/T0), which is abelian. Proposition 3.5. Let E be a semiramified central graded division algebra over the graded field T. Then, (i) There exist graded T-central division algebras I and N such that I is inertial, N is DSR, and E ∼g I⊗T N in Br(T). When this occurs, N0 ∼= E0, ΓN = ΓE, ΘN = ΘE, and E0 splits I0. (ii) For any other decomposition E ∼g I′⊗TN′ with I′ inertial and N′ DSR, we have I′ We do not give a proof of Prop. 3.5 because it is a simpler version of the proof of the analogous unitary result, which is Prop. 4.5 below. Also, Prop. 3.5 is the graded analogue of a known result for semiramified division algebras over Henselian valued fields, [JW, Lemma 5.14, Th. 5.15], and the graded result given here is deducible from the Henselian one. 0 ≡ I0 (mod Dec(E0/T0) ). 12 A. R. WADSWORTH Lemma 3.6. For the semiramified graded division algebra E = A(E0T/T, σ, v, c) as above, write E ∼g I ⊗T N with I inertial and N DSR; so [I0] ∈ Br(E0/T0). If I0 ∼ A(E0/T0, σ, u, b), then by chang- ing the chioce of the yi ∈ E∗ inducing σi on E0T we have E = A(E0T/T, σ, u, e) with the same u as for I0. i x−1 Proof. Let J be a maximal graded subfield of N which is totally ramified over T, so ΓN = ΓJ. Because N is semiramified, the map ΘN : ΓN/ΓT → Gal(N0/T0) is an isomorphism. But also N0 = E0. Thus, for each i, we can choose xi ∈ J∗ with ΘN(deg(xi)) = σiE0. Let di = xri i ∈ (N0T)∗ = (E0T)∗. Then, N ∼=g A(E0T/T, σ, w, d), where each wij = xixjx−1 j = 1, as all the xi lie in the graded field J. Let I′ 0 = A(E0/T0, σ, u, b), which is Brauer equivalent to I0. Then set I′ = I′ 0 ⊗T0 T, which is an inertial T-algebra with I′ ∼g I. Since I′ ⊗T N ∼g I ⊗T N ∼g E, we may without any loss replace I by I′. Then, as I0 ∼= A(E0/T0, σ, u, b), clearly I ∼=g I0 ⊗T0 T ∼=g A(E0T/T, σ, u, b). Let E′ = A(E0T/T, σ, u, e), where each k be the associated generators of E′ over E0T. Then, E ∼g I ⊗T N ∼g E′, by ei = bidi, and let y′ Remark 3.3, as uijwij = uij. Note that for each i, deg(ei) = deg(di), as deg(bi) = 0. Hence, ΓE′ = ΓN by (3.12). Furthermore, E′ is a semiramified graded division algebra since N is, because Lemma 3.4 shows that this is determined by the deg(ei), resp. deg(di). Because E′ is a graded division algebra (not just a graded simple algebra), as is E, from E ∼g E′ the uniqueness in the graded Wedderburn Theorem [HwW2, Prop. 1.3] yields a graded T-isomorphism η : E → E′. By the graded Skolem-Noether Theorem, Prop. 2.1, η can be chosen so that ηE0T = id. Then replacing the yi by η−1(y′ i) in the presentation of E changes each vij to uij. (cid:3) 1, . . . , y′ Theorem 3.7. Suppose E is a semiramified T-central graded division algebra, and take any decomposition E ∼g I ⊗T N where I is an inertial graded T-algebra and N is DSR. Then, (i) Since I0 ∈ Br(E0/T0) with E0 abelian Galois over T0, we can write I0 ∼ A(E0/T0, σ, u, b) in Br(T0). Then, (ii) If E0 ∼= L1 ⊗T0 L2 with each Li cyclic Galois over T0, then SK1(E) ∼= bH −1(H, E∗ 0)(cid:14)(cid:10) im{uij 1 ≤ i, j ≤ k}(cid:11), where H = Gal(E0/T0). where Dec(E0/T0) = Br(L1/T0) · Br(L2/T0). SK1(E) ∼= Br(E0/T0)(cid:14)(cid:2) Dec(E0/T0) · h[I0]i(cid:3), Proof. The definition of SK1 for graded division algebras is given in (2.3) above. H = Gal(E0/T0) ∼= Gal(E0T/T). Since E is semiramified, H ∼= ΓE/ΓT via Θ Cor. 3.6(ii)] there is an exact sequence (i) We have (see §2). By [HW1, −1 E 0 −→ H ∧ H Φ−→ bH −1(G, E∗ 0) Ψ−→ SK1(E) −→ 0. The maps in (3.14) are given as follows: Let ker(NrdE) = {a ∈ E∗ NrdE(a) = 1} ⊆ E∗ 0, and let ker(NE0/T0 ) = {a ∈ E∗ 0 NE0/T0(a) = 1}. Because E is semiramified, by [HW2, Remark 2.1(iii), Lemma 2.2], ker(NrdE) = ker(NE0/T0). For every ρ ∈ H, choose any yρ ∈ E∗ with int(yρ)E0 = ρ. The map Φ is given by: for ρ, π ∈ H, (3.14) The map Ψ is given by: for a ∈ ker(NE0/T0), Φ(ρ ∧ π) = yρyπy−1 ρ y−1 π IH(E∗ 0) ∈ ker(NE0/T0)(cid:14)IH(E∗ 0) = bH −1(H, E∗ Ψ(cid:0)a IH(E0)∗(cid:1) = a [E∗, E∗] ∈ ker(NrdE)(cid:14)[E∗, E∗] = SK1(E). 0). By Lemma 3.6, we can assume E = A(E0T/T, σ, u, c) (with the same uij as for I0). Since H ∼= Gal(E0T/T) = hσ1i × . . . × hσki, we have H ∧ H = hσi ∧ σj 1 ≤ i, j ≤ ki. There are y1, . . . , yk ∈ E∗, with int(yi)E0 = σi and yiyjy−1 j = uij. So we can take yσi = yi, 1 ≤ i ≤ k, yielding for the Φ in (3.14), i y−1 Φ(σi ∧ σj) = uijIH(E∗ the exact sequence (3.14). 0) ∈ bH −1(H, E∗ UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 13 0). Thus, im(Φ) = him(uij) 1 ≤ i, j ≤ ki, and part (i) follows from (ii) When E0 = L1 ⊗T0 L2, H = Gal(E0/T0) has rank 2, say H = hσ1i × hσ2i. So, H ∧ H = hσ1 ∧ σ2i and im(Φ) = hu12IH(E∗ 0)i. As we saw in discussion of (3.9) above, the isomorphism Br(E0/T0)(cid:14) Dec(E0/T0) −→ bH −1(H, E∗ 0). Thus using part (i), 0), maps [ I0] + Dec(E0/T0) to u12IH(E∗ SK1(E) ∼= bH −1(H, E∗ 0)(cid:14)(cid:10) im(u12)(cid:11) ∼= Br(E0/T0)(cid:14)(cid:2) Dec(E0/T0) · h[ I0]i(cid:3). For any division algebra D over a Henselian valued field F , the valuation on F extends uniquely to a valuation on D, and we write D for its residue division algebra and ΓD for its value group. Recall the isomorphism SK1(D) ∼= SK1(gr(D)) for a tame such D, proved in [HW1, Th. 4.8]. By using this isomorphism, Th. 3.7 yields the following: (cid:3) Corollary 3.8. Let F be field with Henselian valuation v, and let D be an F -central division algebra which (with respect to the unique extension of v to D) is tame and semiramified. Take any decomposition D ∼ I ⊗F N , where I and N are F -central division algebras with I inertial and N DSR. (i) Since I ∈ Br(D/F ) with D abelian Galois over F , we can write I ∼ A(D/F , σ, u, b) in Br(F ). Then, (ii) If D ∼= L1 ⊗F L2 with each Li cyclic Galois over F , then SK1(D) ∼= bH −1(H, D ∗ )(cid:14)(cid:10) im{uij 1 ≤ i, j ≤ k}(cid:11), where H = Gal(D/F ). SK1(D) ∼= Br(D/F )(cid:14)(cid:2) Dec(D/F ) · h[I]i(cid:3), where Dec(D/F ) = Br(L1/F ) · Br(L2/F ). Proof. (That D is tame and semiramified means [D : F ] = ΓD : ΓF = p[D : F ] and D is a field sep- arable over F .) Let T = gr(F ), the associated graded ring of F with respect to the filtration on it induced by the valuation (cf. [HwW2] or [HW1]). Since F is a field, T is a graded field with T0 = F and ΓT = ΓF . Since v is Henselian, it has unique extensions to valuations on D, I, and N ; with respect to these valuations, let E = gr(D), I = gr(I), and N = gr(N ). These are graded division rings, with E0 = D, I0 = I ∼ A(E0/T0, σ, u, b), and N0 = N ∼= D = E0. Moreover, as D, I, and N are each tame over F , it follows by [HwW2, Prop. 4.3] that T is the center of E, I, and N, and [E : T] = [D : F ], [ I : T] = [I : F ], and [N : T] = [N : F ]. Since I is inertial over F , we have I is inertial over T. That N is DSR means (cf. [JW, p. 149], where the term NSR is used) that N has maximal subfields S and J with S inertial over F and J totally ramified of radical type over F . Then, gr(S) and gr(J) are maximal graded subfields of N with gr(S) inertial over T and gr(J) totally ramified over T. So, N is DSR. Similarly, E is semiramified since D is tame and semiramified. Let Brt(F ) be the tame part of the Brauer group Br(F ). From the isomorphism Brt(F ) ∼= Br(T) given by [HwW2, Th. 5.3], we obtain E ∼g I ⊗T N from D ∼ I ⊗F N . Thus, Th. 3.7 applies to E with the decomposition E ∼g I ⊗T N, and the assertions of Cor. 3.8 follow immediately as SK1(D) ∼= SK1(E) by [HW1, Th. 4.8]. Example 3.9. Take any integer n ≥ 2 and let K be any field containing a primitive n2-root of unity ω. Let T = K[x, x−1, y, y−1], the Laurent polynomial ring, graded as usual by Z × Z with T(k,ℓ) = Kxkyℓ; in particular, T0 = K. (So T ∼=g gr(cid:0)K((x))((y))(cid:1) where the iterated Laurent power series ring K((x))(y)) is given its usual rank 2 Henselian valuation.) Take any a, b ∈ K ∗ such that(cid:2)K( n√a, n√b ) : K(cid:3) = n2, and let E be the graded symbol algebra E = (axn, byn, T)ω, of degree n2. That is, E is the graded central simple T-algebra with homogenous generators i and j such that in2 = byn, and ij = ωji, and nZ), and E0 = K(inx−1, jny−1) ∼= K( n√a, n√b ). deg(i) = ( 1 Since E0 is a field, by Lemma 2.2(ii) E is a graded division ring, which is clearly semiramified. We can n , 0), deg(j) = (0, 1 n ). Then, ΓE = ( 1 nZ) × ( 1 = axn, jn2 (cid:3) 14 A. R. WADSWORTH write E0 = L1 ⊗K L2 where L1 = K( n√a ) and L2 = K( n√b ), and H = Gal(E0/K) = hσ1i × hσ2i where σ1( n√a) = ωn n√a, σ1( n√b) = n√b and σ2( n√b) = ωn n√b, σ2( n√a) = n√a. Since int(j−1)E0 = σ1 and int(i)E0 = σ2, we can express E as a graded abelian crossed product with y1 = j−1 and y2 = i, obtaining E = A(T( n√a, n√b)/T, σ, u, d), where u11 = u22 = 1, u12 = ω, u21 = ω−1, and d1 = 1/(y n√b), d2 = x n√a. Graded symbol algebras satisfy the same multiplicative rules in the graded Brauer group as do the usual ungraded symbol algebras in the Brauer group. (This follows, e.g., by the injectivity of the scalar extension map Br(T) → Br(q(T)), cf. [HwW2, p. 90].) Thus, in Br(T), we have E ∼g (a, b, T)ω ⊗T (xn, b, T)ω ⊗T (a, yn, T)ω ⊗T (xn, yn, T)ω ∼g (a, b, T)ω ⊗T (x, b, T)ωn ⊗ (a, y, T)ωn . (The last two terms are symbol algebras of degree n.) Thus, E ∼g I ⊗T N where I = (a, b, T)ω and N = (x, b, T)ωn ⊗T (a, y, T)ωn . Then, I ∼=g I0 ⊗T0 T, where I0 = (a, b, T0)ω = A(K( n√a, n√b )/K, σ, u, b), with the same u as for E and b1 = 1/ n√b, b2 = n√a. So, I is an inertial central simple graded T-algebra. We have N0 is the field K( n√a, n√b ), so N is a graded division algebra by Lemma 2.2(ii). N is DSR since it has the inertial maximal graded subfield T( n√a, n√b ) = N0T and the totally ramified maximal graded subfield T( n√x, n√y ). As a graded abelian crossed product, N ∼=g A(T( n√a, n√b)/T, σ, 1, c), where c1 = 1/y, c2 = x. Let M = K( n√a, n√b ). By Prop. 3.2, SK1(N) ∼= bH −1(H, M ∗) ∼= Br(M/K)(cid:14) Dec(M/K), where H = Gal(M/K) and Dec(M/K) = Br(K( n√a )/K) · Br(K( n√b )/K); but, by Th. 3.7, SK1(E) ∼= bH −1(H, M ∗)(cid:14)him(ω)i ∼= Br(M/K)(cid:14)(cid:2) Dec(M/K) · h[(a, b, K)ω ]i(cid:3). This example is the graded version of Platonov's example in [P3] and [P4] of a cyclic algebra with nontrivial SK1, where K is a suitably chosen global field. (Platonov worked with the Henselian valued ground field K ′ = K((x))((y)) in place of the graded field T = gr(K ′) considered here.) In [P4, Th. 2] the added term distingushing SK1(E) from SK1(N) is omitted. This error is corrected in [Y5, p. 536, footnote 1] and in [E2, p. 70], giving the first isomorphism of (3.15) but not the second. (3.15) 4. Unitary graded I ⊗ N decomposition The goal for §§4 -- 7 is to give a unitary version of the formulas for SK1 in Prop. 3.2 and Th. 3.7 for semiramified graded division algebras with graded unitary involution. In this section we consider abelian crossed products with unitary involution and prove a unitary analogue to the I ⊗ N decomposition of Prop. 3.5. A unitary involution on a central simple algebra A over a field K is a ring antiautomorphism τ of A such that τ 2 = idA and τK 6= id. (Such a τ is also called an involution on A of the second kind.) Let F = K τ = {c ∈ K τ (c) = c}, which is a subfield of K with [K : F ] = 2 and K Galois over F with Gal(K/F ) = {τK , idK}. Our τ is also called a unitary K/F -involution. The unitary SK1(A, τ ) is defined just as for SK1(D, τ ) in (1.1). Recall (see [KMRT, Prop. (17.24)(2)]) that if τ ′ is another unitary K/F - involution on A, then SK1(A, τ ′) = SK1(A, τ ). Thus, we will freely pass from one unitary K/F -involution on A to another when convenient. In the unitary setting generalized dihedral Galois groups often arise where abelian Galois groups appear in the nonunitary setting. A group G is said to be generalized dihedral with respect to a subgroup H if G : H = 2 and for some θ ∈ G \ H, θ2 = 1 and θhθ−1 = h−1 for every h ∈ H. Equivalently, every element of G\ H has order 2. See [HW2, §2.4] for some remarks on such groups. Note that H is necessarily abelian. If H is cyclic, we say that G is dihedral. (This includes the trivial cases where H = 1 or 2.) For fields F ⊆ K ⊆ M , we say that M is K/F -generalized dihedral if [M : F ] < ∞, M is Galois over F , and G = Gal(M/F ) is generalized dihedral with respect to its subgroup H = Gal(M/K). UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 15 Lemma 4.1. Let F ⊆ K ⊆ M be fields, and suppose M is K/F -generalized dihedral. Let A be a central simple K-algebra containing M as a strictly maximal subfield. Let G = Gal(M/F ) and H = Gal(M/K), and fix any θ ∈ G \ H (so θ2 = idM ). Then, the following conditions are equivalent: (i) A has a unitary K/F -involution. (ii) A has a unitary K/F -involution τ such that τM = θ. (iii) A ∼= A(M/K, σ, u, b) where (in addition to conditions (3.1) and (3.2)) for all i, j. and (4.1) The A in (iii), has a unitary K/F -involution τ with τM = θ and τ (zi) = zi for each of the standard generators zi of A. uij · σiσjθ(uij) = 1 bi = θ(bi) Proof. Note that as θ /∈ H and K is Galois over F , we have θ(K) = K and Gal(K/F ) = {idK , θK}. (i) ⇒ (ii) This is a special case of a substantial result [KMRT, Th. 4.14] on simple subalgebras with compatible involutions. For the convenience of the reader we give a short direct proof. Let ρ be a unitary K/F -involution on A, so ρK = θK. Since ρθ is a K-linear homomorphism M → A, by the Skolem-Noether Theorem, there is y ∈ A∗ with int(y)M = ρθ. For any a ∈ M , as ρ2 = θ2 = idM , we have ρ(y)aρ(y)−1 = ρ(y−1ρ(a)y) = ρ(ρθ)−1ρ(a) = ρθ(a) = yay−1. Therefore, letting c = y−1ρ(y), we have c ∈ CA(M )∗ = M ∗ and ρ(y) = yc. Hence, y = ρ2(y) = ρ(yc) = ρ(c)yc = ρ(c)ρθ(c)y = ρ(cθ(c))y; so, cθ(c) = 1. Since θ2 = idM , by Hilbert 90 applied to the quadratic extension M/M θ there is d ∈ M ∗ with c = dθ(d)−1. Let z = yd. Then, as θ(c) = θ(d)d−1, ρ(z) = ρ(d)yc = ρ(d)ρθ(c) y = ρθ(d) y = yd = z. Let τ = ρ◦int(z), which is an involution on A, as ρ(z) = z. Then, τM = ρ int(z)M = ρ int(y)M = ρ2θ = θ, as desired. (ii) ⇒ (iii) Let τ be a unitary K/F -involution on A such that τM = θ. For any σ ∈ H, we claim that there is z ∈ A∗ with int(z)M = σ and τ (z) = z. For this, first apply Skolem-Noether to obtain y ∈ A∗ with int(y)M = σ. For any a ∈ M we have, as τ σ−1τ = σ on M since τ σ−1M ∈ G \ H, τ (y)aτ (y)−1 = τ (y−1τ (a)y) = τ σ−1τ (a) = σ(a) = yay−1. Hence, τ (y) = cy, where c ∈ CA(M )∗ = M ∗. Now, y = τ 2(y) = τ (cy) = τ (y)τ (c) = cyθ(c) = cσθ(c) y, so c σθ(c) = 1. Since σθ has order 2, Hilbert 90 applied to the quadratic extension M/M σθ shows that there is d ∈ M ∗ with c = d σθ(d)−1. Let z = dy. Then, int(z)M = int(y)M = σ and τ (z) = cyθ(d) = [d σθ(d)−1] σθ(d) y = z, proving the claim. Thus, with our cyclic decomposition H = hσ1i×. . .×hσki, we can choose z1, . . . , zk ∈ A∗ with int(zi)M = σi and τ (zi) = zi. Then, for bi = zri i ) = bi. Also, for uij = zizjz−1 i ∈ M ∗, we have θ(bi) = τ (bi) = τ (zri , we have i z−1 j σiσjθ(uij) = zizj τ (zizjz−1 i z−1 j )z−1 j z−1 i = zizj(z−1 j z−1 i zjzi)z−1 j z−1 i = zjziz−1 j z−1 i = u−1 ij , so uij σiσjθ(uij) = 1. Thus, A ∼= A(M/K, σ, u, b) with the uij and bi satisfying the equations in (4.1). (iii) ⇒ (i) Assume A = A(M/K, σ, u, b) where the uij and bi satisfy the conditions in (4.1). Take z1, . . . , zk ∈ A∗ with int(zi)M = σ, zri j = uij. We show that there is a unitary K/F -involution τ on A satisfying (and determined by) τM = θ and τ (zi) = zi for each i. Basically, this is a matter of checking that the τ just described is compatible with the defining relations of A. Here i = bi and zizjz−1 i z−1 16 A. R. WADSWORTH 1], . . . , Bℓ = Bℓ−1[yℓ; σ∗ ℓ of Bℓ−1 is defined by σ∗ ℓM = σℓ and σ∗ ℓ satisfies int(uℓi)σ∗ i σ∗ ℓ = σ∗ ℓ σ∗ ℓ ], . . . , Bk = Bk−1[yk; σ∗ k], where σ∗ ℓ extends from Bi−1 to Bi; thus, σ∗ ℓ τℓ−1σ∗ ℓ = τℓ−1. For this, note first that σ∗ ℓ τℓ−1σ∗ is a more complete argument, based on the description of A(M/K, σi, uij, bi) given in the proof of [AS, Th. 1.3]. First, take any ring B with an automorphism σ, and let B[y; σ] be the twisted polynomial ring {P ciyi ci ∈ B} with the multiplication determined by yc = σ(c)y for all c ∈ B. It is easy to check that an involution ρ on B extends to an involution ρ′ on B[y; σ] with ρ′(y) = y iff σρσ = ρ. Also, for d ∈ B∗, an automorphism η of B extends to an automorphism η′ of B[y; σ] with η′(y) = dy iff int(d)ση = ησ. Here, let B0 = M , B1 = B0[y1; σ∗ 1 = σ1 and for ℓ > 1, the automorphism σ∗ ℓ (yi) = uℓiyi for 1 ≤ i < ℓ. (One checks inductively using the identities in (3.1) that for 1 ≤ i ≤ ℓ − 1, σ∗ i on Bi−1, hence σ∗ ℓ is an automorphism of Bℓ−1.) Define inductively involutions τi on Bi by τ0 = θ and for ℓ > 0, τℓBℓ−1 = τℓ−1 and τℓ(yℓ) = yℓ. Given τℓ−1, the condition for the existence of τℓ is that σ∗ ℓM = σℓθσℓ = θ = τℓ−1M as G is generalized dihedral. Furthermore, for 1 ≤ i < ℓ, ℓ τℓ−1(uℓiyi) = σ∗ ℓ τℓ−1σ∗ σ∗ Thus, σ∗ ℓ agrees with τℓ−1 throughout Bℓ−1, as needed. By induction, we have the involution τk on Bk. As pointed out in [AS, p. 79], A ∼= Bk/I, where I is the two-sided ideal of Bk generated by {yri i − bi 1 ≤ i ≤ k}. Since τk(bi) = θ(bi) = bi, τk maps each generator of I to itself. Therefore, τk induces an involution τ on A ∼= Bk/I which clearly restricts to θ on M ; so τ is a unitary K/F -involution on A. (cid:3) We write Br(M/K; F ) for the subgroup of Br(M/K) of algebra classes [A] such that A has a unitary K/F -involution. By Albert's theorem [KMRT, Th. 3.1(2)], Br(M/K; F ) is the kernel of the corestriction map corK→F : Br(M/K) → Br(M/F ). For M a K/F -generalized dihedral extension of F , as above, there is in addition a corresponding subgroup of Dec(M/K). For this, note first that for any field L with K ⊆ L ⊆ M and L cyclic Galois over K, say Gal(L/K) = hσi, L is K/F -dihedral, so Lemma 4.1 (with k = 1) implies that Br(L/K; F ) = {[(L/K, σ, b)] b ∈ F ∗}. For H = Gal(M/K) = hσ1i × . . . × hσki and (χ1, . . . , χk) the base of X(M/K) dual to (σ1, . . . , σk), and Li the fixed field of ker(χi), as at the beginning of §3, define ℓ(cid:0)yi θ(uℓi)(cid:1) = σ∗ ℓ [σiθ(uℓi) yi] = [σℓσiθ(uℓi)]uℓi yi = yi = τℓ−1(yi). ℓ (yi) = σ∗ ℓ τℓ−1σ∗ Dec(M/K; F ) = {[(L1/K, σ1, b1) ⊗K . . . ⊗K (Lk/K, σk, bk)] each bi ∈ F ∗} ⊆ Br(M/K; F )}. (4.2) Note that Dec(M/K; F ) is generated as a group by the image under the cup product of H 2(H, Z) × F ∗. Thus Dec(M/K; F ) is independent of the choice of cyclic decomposition of H, and we have analogously to (3.5), Dec(M/K; F ) = Br(Li/K; F ) = kQi=1 QK⊆L⊆M Gal(L/K) cyclic Br(L/K; F ). (4.3) For the rest of this section we fix a graded field T and a graded subfield R ⊆ T such that [T : R] = 2 and T is inertial and Galois over R. Let ψ be the nonidentity graded R-automorphism of T, and let ψ0 be the restriction ψT0 . Thus, ΓT = ΓR, [T0 : R0] = 2, T ∼=g T0 ⊗R0 R, T0 is Galois over R0, and ψ on T corresponds to ψ0 ⊗ idR on T0 ⊗R0 R. We are interested in central simple graded T-algebras A with graded unitary T/R-involutions τ . This means that τ is a degree-preserving ring antiautomorphism of A with τ 2 = idA and the ring of invariants Tτ = R; the last condition is equivalent to τT = ψ. Suppose now that A is a graded division algebra. Set τ0 = τA0, which is a unitary involution on A0, as τ0T0 = ψ0 6= id and T0 ⊆ Z(A0). Just as for any graded division algebra, Z(A0) is abelian Galois over T0. But the presence of the involution τ implies further that Z(A0) is actually T0/R0-generalized dihedral, by [HW2, Lemma 4.6(ii)]. A central graded division algebra N over T is said to be decomposably semiramified for T/R (abbreviated DSR for T/R) if N has a unitary graded T/R-involution τ and a maximal graded subfield M inertial over T and another maximal graded subfield J with J totally ramified over T and τ (J) = J. When this occurs, N is UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 17 semiramified with N0 = M0, a field, which as just noted is T0/R0-generalized dihedral. Also, ΓN = ΓJ and ΘN induces an isomorphism ΓN/ΓT ∼= Gal(N0/T0). Furthermore, as M = M0T = N0T, we have τ (M) = M. Example 4.2. Let L be any cyclic Galois field extension of T0 with L dihedral over R0. (That is, L is Galois over R0 and there is θ ∈ Gal(L/R0) \ Gal(L/T0) with θ2 = idL and θhθ−1 = h−1 for every h ∈ Gal(L/T0). Thus, the group Gal(L/R0) is either dihedral or isomorphic to Z/2Z or Z/2Z× Z/2Z.) Let r = [L : T0], and take any b ∈ R∗ with the image of deg(b) having order r in ΓT/rΓT. Take any generator σ of Gal(L/T0), and let σ denote also its canonical extension σ ⊗ idT in Gal((L ⊗T0 T)/T). Let N = ((L ⊗T0 T)/T, σ, b), a cyclic graded algebra over T. Gal(LT/T) = hσi. Our N isLr−1 We show that N is a central graded division algebra over T of degree r, and N is DSR for T/R. For, letting LT denote L⊗T0T, note that LT is a graded field which is inertial over T and is Galois over T with i=0 LTzi, where zcz−1 = σ(c) for all c ∈ LT, and zr = b, with the grading on N extending that on LT by setting deg(z) = 1 r deg(b). A graded cyclic T-algebra is always graded simple with center T. Note that for j ∈ Z, if j deg(b)/r ∈ ΓLT = ΓT, then j deg(b) ∈ rΓT, so by hypothesis r j. Hence, N0 = (LT)−i deg(b)/r zi = (LT)0 = L. r−1Pi=0 r−1Pi=0 r−1Pi=0 ziθ(ci) = σiθ(ci)zi. τ ( cizi) = Since N0 is a division ring, the simple graded algebra N is a graded division ring, by Lemma 2.2(ii). Also, as [LT : T] = [L : T0] = r = deg(N), LT is a maximal graded subfield of N which is inertial over T. Take any θ ∈ Gal(L/R0) with θT0 = ψ0, and let θ denote also its canonical extension θ ⊗ idR to Gal(LT/R). Define a map τ : N → N by r−1Pi=0 Since θT = ψ, θ2 = id, and θσθ−1 = σ−1 (as L is T0/R0-dihedral), it is easy to check that τ is a graded T/R-involution of N. Moreover, if we let J =Lr−1 i=0 Tzi = T[z], then J is a maximal graded subfield of N, and the hypothesis on deg(b) assures that J is totally ramified over T; also τ (J) = J. This verifies that N is DSR for T/R. Note that N0 = L and ΓN = h 1 Lemma 4.3. Let N and N′ be graded division algebras which are each DSR for T/R. Suppose N0 and N′ 0 are linearly disjoint over T0 and ΓN ∩ ΓN′ = ΓT. Then, N ⊗T N′ is a graded division algebra which is DSR for T/R. Also, (N ⊗T N′)0 ∼= N0 ⊗T0 N′ Proof. Let B = N ⊗T N′, which is a central simple graded T-algebra, since this is true for N and N′ by [HwW2, Prop. 1.1]. For each γ ∈ ΓT choose a nonzero tγ ∈ Tγ. Then, 0 t−1 0 and ΓN⊗TN′ = ΓN + ΓN′. r deg(b)i + ΓT. N0 tγ ⊗T0 N′ γ = N0 ⊗T0 N′ 0. B0 = Pγ∈ΓN∩ΓN′ Nγ ⊗T0 N′ −γ = Pγ∈ΓT The linear disjointness hypothesis assures that B0 is a field, and hence B is a graded division ring, by Lemma 2.2(ii). Moreover, by dimension count B0T is a graded maximal subfield of B which is inertial over T. Let τ be a graded T/R-involution of N, and let J be a graded maximal subfield of N with τ (J) = J. Take τ ′ and J′ correspondingly for N′. Then, JJ′ = J ⊗T J′ and τ ⊗ τ ′ is a graded T/R-involution on B with (τ ⊗ τ ′)(JJ′) = JJ′. Moreover, JJ′ is a maximal graded subfield of B by dimension count, and, as ΓJ ∩ ΓJ′ = ΓN ∩ ΓN′ = ΓT, we have ΓJJ′ : ΓT ≥ ΓJ + ΓJ′ : ΓT = ΓJ : ΓT · ΓJ′ : ΓT = [J : T] · [J′ : T] = [JJ′ : T]. Hence, JJ′ is totally ramified over T. Thus, B is DSR for T/R. (cid:3) The next proposition shows that all graded division algebras N which are DSR for T/R are obtain- able from those in Ex. 4.2 by iterated application of Prop. 4.3. This justifies the term "decomposably semiramified" for such N. 18 A. R. WADSWORTH Proposition 4.4. Let N be a graded division algebra which is DSR for T/R. Take any decomposition N0 = L1 ⊗T0 . . . ⊗T0 Lk with correspondingly σ1 . . . , σk ∈ Gal(N0T/T) ∼= Gal(N0/T0) such that σiLj = id whenever j 6= i and Gal(Li/T0) = hσiLii for each i. (So Gal(N0T/T) = hσ1i × . . . × hσki.) Let ri be the order of σi. For each i choose γi ∈ ΓN with ΘN(γi) = σi. Then, there exist b1, . . . , bk ∈ R∗ such that deg(bi) = riγi and cyclic Galois over T0, each Li choose and N ∼=g (L1T/T, σ1, b1) ⊗T . . . ⊗T (LkT/T, σk, bk) ∼=g A(N0T/T, σ, 1, b). Proof. Since N is DSR for T/R, there is a graded T/R-involution τ of N and a maximal graded subfield J of N with J totally ramified over T and τ (J) = J. As noted earlier, we have ΓJ = ΓN. Since τ is a graded automorphism of J of order 2, the fixed set S = Jτ = {a ∈ J τ (a) = a} is a graded subfield of J with 2 = [J : S] = [J0 : S0]ΓJ : ΓS. Since S0 ∩ T0 = R0 $ T0 = J0 ∩ T0 we have S0 $ J0, so [J0 : S0] = 2, and hence ΓS = ΓJ (= ΓN). Thus, for each i there is a nonzero xi ∈ Sγi, and for any such xi, int(xi)N0T = σi as ΘN(γi) = σi. Let bi = xri i = id, so deg(bi) ∈ ker(ΘN) = ΓT; hence, bi ∈ Jdeg(bi) = Tdeg(bi) as J is totally ramified over T. Therefore, b ∈ S∗ ∩ T = R∗. Let Ci be the graded T-subalgebra of N generated by Li and xi. Since int(xi)LiT = σiLiT, there is a graded T-algebra epimorphism (LiT/T, σi, bi) → Ci, which is a graded isomorphism as the domain is graded sim- ple. Since the xi all lie in the graded field S and σiLj T = id for j 6= i, the distinct Ci centralize each other. Hence, there is a graded T-algebra homomorphism (L1T/T, σ1, b1) ⊗T . . . ⊗T (LkT/T, σk, bk) → N which is injective as the domain is graded simple, and surjective by dimension count. Clearly also, (L1T/T, σ1, b1) ⊗T . . . ⊗T (LkT/T, σk, bk) ∼=g A(N0T/T, σ, 1, b). (cid:3) i ∈ S∗. Then, ΘN(deg(bi)) = σri Proposition 4.5. Let E be a semiramified central graded division algebra over T, and suppose E has a graded T/R-involution, where T is inertial over R. Then, E0 is T0/R0-generalized dihedral and N DSR for T/R. (i) E ∼g I ⊗T N in Br(T) for some T-central graded division algebras I and N with I inertial and (ii) Take any decomposition T ∼g I′⊗T N′ in Br(T) with graded T-central division algebras I′ and N′ with 0] ∈ Br(E0/T0; R0). 0 ∼= E0, ΓN′ = ΓE, ΘN′ = ΘE, and [ I′ I′ inertial and N′ DSR for T/R. Then, N′ Furthermore, I′ 0 is uniquely determined modulo Dec(E0/T0; R0). Proof. (i) Since E is semiramified, E0T is an inertial maximal graded subfield of E. Moreover, as E has a graded T/R-involution, E0 is T0/R0-generalized dihedral, by [HW2, Lemma 4.6(ii)]. Because E has an inertial graded maximal subfield, it is a graded abelian crossed product: Say E0 = L1 ⊗T0 . . . ⊗T0 Lk, where each field Li is cyclic Galois over T0 (so dihedral over R0). Then G = Gal(E0T/T) ∼= Gal(E0/T0) has a corresponding cyclic decomposition G = hσ1i × . . . × hσki, where each σiLj T = id for j 6= i, and σiLiT generates Gal(LiT/T). Let ri = hσii = [Li : T0]. By Lemma 3.4, E = A(E0T/T, σ, u, b) where each uij ∈ E∗ 0, bi ∈ E0T∗, 1 ri (4.4) So, deg(bi) ∈ ΓE0T = ΓT = ΓR and the image of deg(bi) has order ri in ΓT/riΓT. For each i, choose ci ∈ R∗ with deg(ci) = deg(bi). Let deg(bk) + ΓTi. rk deg(bi) + ΓT has order ri in ΓE/ΓT, and deg(b1) + ΓTi × . . . × h 1 ΓE/ΓT = h 1 r1 N = C1 ⊗T . . . ⊗T Ck, where each Ci = (LiT/T, σi, ci). By Ex. 4.2 each Ci is DSR for T/R with (Ci)0 ∼= Li and ΓCi = h 1 deg(bi)i + ΓT. It follows by induction on k using Lemma 4.3 and (4.4) that N is a graded division algebra which is DSR for T/R. Choose zi ∈ C∗ i = ci. Then, when we view zi ∈ N∗, we have int(zi) = σi on all of N0T. Since further zizj = zjzi for all i, j, our N is the graded abelian crossed product N = A(E0T/T, σ, 1, c). For its opposite algebra Nop we then have Nop ∼=g A(E0T/T, σ, 1, d) where each di = c−1 0. The uij and bi satisfy i with int(zi)LiT = σi and zri deg(ci)i + ΓT = h 1 ri ri i . LetbI = A(E0T/T, σ, u, e) where each ei = bidi = bic−1 i ∈ E∗ UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 19 conditions (3.1) and (3.2), as do the ci with the corresponding uij = 1; hence the uij here and ei satisfy τ0 = τbI 0 0⊗T0 N′ [I0 : T0] = 1 in Br(T), xri i = ei, and xixjx−1 i x−1 deg(ei) = 0; hence, deg(xi) = 0 for each j = uij for all i, j. Then, deg(xi) = 1 ri (3.1) and (3.2); also, deg(uij) = 0 for all i, j. So,bI is a well-defined graded abelian crossed product. By Remark 3.3, we havebI ∼g E ⊗T Nop. There are homogeneous x1, . . . , xk ∈bI ∗ such that int(xi)E0T = σi, i ∈ I = Qk i=1{0, 1, 2, . . . , ri − 1}. Thus, inbI = Li∈I E0Txi we havebI 0 = Li∈I E0xi ∼= A(E0/T0, σ, u, e), which is a central simple T0-algebra with dimT0(bI 0) = [E0 : T0]2 = dimT(cid:0)bI(cid:1). Hence,bI is inertial over T. SincebI is simple, by Lemma 2.2bI ∼=g Mℓ(I) for a graded division algebra I withbI 0 ∼= Mℓ(I0). Then, ℓ2 dimT(cid:0)bI(cid:1) = [I : T], showing that I is inertial over T. Since I ∼gbI, we have ℓ2 dimT0(bI 0) = 1 for Nop, and E has a graded T/R-involution τE. So, τ = τE ⊗ τN is a graded T/R-involution onbI , and (ii) Take any decomposition E ∼g I′⊗ N′ as in (ii). Since I′ is inertial and E is the graded division algebra with E ∼g I′⊗TN′, Cor. 2.3 yields E0 ∼ I′ 0; furthermore, 0 ∼= E0 ΓE = ΓN′ and ΘE = ΘN′. We now use the bi, ci, N, and I of part (i). Because N′ is DSR with N′ and ΘN′( 1 i) = deg(ci) ri such that N′ ∼=g A(E0T/T, σ, 1, c′). Let B = A(E0T/T, σ, 1, f ) where each fi = cic′−1 0. So, in Br(T), [E] = [E] [N]−1[N] = [E ⊗T Nop] [N] = [bI ] [N] = [ I ] [N] = [ I ⊗T N ], is a T0/R0-involution onbI 0. So, in Br(T0) we have [ I0] = [bI 0] ∈ Br(E0/T0; R0). i.e., E ∼g I ⊗T N, proving (i). Also, N has a graded T/R-involution τN, which is also a graded involution deg(bi)) = σi, by Prop. 4.4 there exist c′ 0 and E0 = Z(E0) ∼= Z(N′ k ∈ R∗ with deg(c′ inertial over T with B ∼g N ⊗T N′ op ∼g I′ ⊗T I op. Because deg(fi) = 0 for each i, the argument forbI in (i) shows that B is Thus, [B0] ∈ Dec(E0/T0; R0), as each fi ∈ R∗ 0 (see Ex. 4.2). Let C be the graded division algebra with C ∼g B ∼g I′ ⊗T Iop. Since B0 is simple and I′ is inertial, Lemma 2.2 and Cor. 2.3 yield C0 ∼ B0 and C0 ∼ (I′ ⊗T Iop)0 ∼= I′ B0 ∼= A(E0/T0, σ, 1, f ) ∼= (L1/T0, σ1, f1) ⊗T0 . . . ⊗T0 (Lk/T0, σk, fk). 0] = [C0] [ I0] = [B0] [ I0] = [B0] [bI 0] ∈ Br(E0/T0; R0). Since [B0] ∈ Dec(E0/T0; R0), we have I′ modulo Dec(E0/T0; R0) independent of the choice of decomposition of E as I′ ⊗T N′. Remark 4.6. The I ⊗ N decomposition described in Prop. 4.5 for E semiramified actually holds more generally for E inertially split (with graded T/R-involution), i.e., when E has a maximal graded subfield inertial over T. One then has N0 ∼= Z(E0) and I0 ⊗T0 Z(E0) ∼ E0. See [JW, Lemma 5.14, Th. 5.15] for the nonunitary nongraded Henselian valued analogue of this. 0 ≡ I0 (mod Dec(E0/T0; R0)). This yields the uniqueness of I′ 0 ⊗T0 Iop 0 ; so, in Br(T0), [ I′ 0 (cid:3) 0) = N′ 0, so E0 splits I′ deg(ci)) = ΘE( 1 ri 1, . . . , c′ i ∈ R∗ 5. Galois cohomology with twisted coefficients twisted action also allows us to give a new interpretation of Albert's corestriction condition for an algebra to have a unitary involution, see Prop. 5.1 below. Where bH −1(H, M ∗) occurs in formulas for SK1 as in §3, analogous formulas for the unitary SK1 involve bH −1(G,gM ∗) for a twisted action of G on the multiplicative group M ∗. In this section, we recall the relevant twisted action, and give some calculations concerning bH −1 which will be used later. The cohomology with Let G be a profinite group with a closed subgroup H with G : H = 2. From the mappping G/H ∼−→ Z/2Z ∼−→ Aut(Z) we obtain a nontrivial discrete G-module structure on Z for which for g ∈ G, j ∈ Z, g ∗ j = ( j, −j, if g ∈ H, if g /∈ H. 20 A. R. WADSWORTH LeteZ denote Z with this new G-action. Then, for any discrete G-module A we have an associated discrete G-module eA = A ⊗ZeZ. That is, eA = A as an abelian group, but the G-action on eA (denoted by ∗, while · denotes the G-action on A) is given by g ∗ a = ( g · a, −g · a, if g ∈ H, if g /∈ H, for all g ∈ G, a ∈ A. (5.1) sequence of Tate cohomology groups: (This is stated in [KMRT, (30.10)] and [AE] for nonnegative indices, but it is valid for i < 0 as well.) 0 −→ eA −→ IndH→G(A) −→ A −→ 0 is discussed in [AE, Appendix], [KMRT, §30.B], [HKRT, §5]. Notably, there is a canonical short exact sequence of G-modules So, the actions of H on eA and on A coincide, and eeA = A as G-modules. The cohomology of such modules Since Shapiro's Lemma says that bH i(G, IndH→G(A)) ∼= bH i(H, A) for all i ∈ Z, this yields a long exact . . . −→ bH i−1(G, A) −→ bH i(G, eA) −→ bH i(H, A) −→ bH i(G, A) −→ bH i+1(G, eA) −→ . . . For the trivial G-module Z we have H 1(G,eZ) = 2, as (5.2) shows, and each connecting homomorphism δ : bH i−1(G, A) → bH i(G, eA) is given by the cup product with the nontrivial element of H 1(G,eZ). of G of index 2. Then, M ∗ is a discrete G-module, andgM ∗ denotes M ∗ with the twisted G-action relative Proposition 5.1. H 2(G,gM ∗) ∼= Br(M/K; F ). We will invoke the twisted cohomology typically in the following setting: Let F ⊆ K ⊆ M be fields with [K : F ] = 2, and M Galois over F . Let G = Gal(M/F ) and H = Gal(M/K), which is a closed subgroup to H described above. Recall that Br(M/K; F ) denotes the subgroup of Br(M/K) consisting of classes of central simple K-algebras split by M and having a unitary K/F -involution. Proof. Part of the long exact sequence (5.2) is (5.2) cor −→ H 2(G, M ∗) sequence (5.3) above yields the desired isomorphism. (5.3) By Albert's theorem [KMRT, Th. 3.1(2)], for [A] ∈ Br(M/K), the algebra A has a K/F -involution iff corK→F (A) is split. Thus, in the isomorphism Br(M/K) ∼= H 2(H, M ∗), Br(M/K; F ) maps isomorphically H 1(G, M ∗) −→ H 2(G,gM ∗) −→ H 2(H, M ∗) to ker(cid:0)H 2(H, M ∗) cor−→ H 2(G, M ∗)(cid:1). Because H 1(G, M ∗) = 0 by the homological Hilbert 90, the exact Remark 5.2. Here are formulas for bH i(G,gM ∗) for small i, which are easily derived from standard group cohomology formulas and (5.2) above. We assume [M : K] < ∞, and let θ be any element of G \ H. So, Gal(K/F ) = {idK, θK}. We write b1−θ for b/θ(b). (i) H 1(G,gM ∗) ∼= F ∗(cid:14)NK/F (K ∗) ∼= bH 0(Gal(K/F ), K ∗). (ii) H 0(G,gM ∗) ∼= {c ∈ K ∗ NK/F (c) = 1}. bH 0(G,gM ∗) ∼= {c ∈ K ∗ NK/F (c) = 1}(cid:14) {NM/K (m)1−θ m ∈ M ∗} = {b1−θ b ∈ K ∗}(cid:14) {NM/K (m)1−θ m ∈ M ∗}. (iii) (cid:3) eN (m) = Qg∈G We will be working particularly with bH −1(G,gM ∗). For this, let eN : gM ∗ → K ∗ be given by h(m) · (θh)(m)−1 = NM/K (m)(cid:14)θ(NM/K(m)). So, eN is the norm map forgM ∗ as a G-module. Note that g ∗ m = Qh∈H ker(eN ) = {m ∈ M ∗ NM/K(m) ∈ F ∗}. (5.4) UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS Also, let IG(gM ∗) = (cid:10) (g ∗ m)m−1 m ∈ M ∗, g ∈ G(cid:11) = (cid:10) h(m)/m, hθ(m) m m ∈ M ∗, h ∈ H(cid:11). Then, by definition, 21 (5.5) (5.6) bH −1(G,gM ∗) ∼= ker(eN )(cid:14)IG(gM ∗). In the following useful lemma, part (ii) is an abstraction of an argument of Yanchevskiı [Y3, proof of Cor. 4.13]. Lemma 5.3. Let D be a finite dihedral group, i.e., D = hh, θi where θ2 = 1, θ 6= 1, and θhθ−1 = h−1, and h has finite order. Let H = hhi. Let A be a D-module such that H 1(H, A) = 0 and H 1(hθi, AH ) = 0. Let Aθ = {a ∈ A θ · a = a} and NH(a) =Ph∈H h·a. Then, (i) AH + Aθ = {a ∈ A a − θ·a ∈ AH}. (ii) Aθ + Ahθ = {a ∈ A NH(a) ∈ Aθ} = Aθ + Aθh. (iii) The map corhθi→D × corhhθi→D : bH −1(hθi, eA) × bH −1(hhθi, eA) → bH −1(D, eA) is surjective. Proof. (i) We have the short exact sequence of hθi-modules 0 → AH → A → A/AH → 0. Since H 1(hθi, AH ) = 1, the long exact cohomology sequence shows that Aθ maps onto (A/AH )θ, which yields (i). (ii) Note that for a ∈ A, NH(θ·a) = Pk∈H(kθ)· a = Pk∈H(θk−1)· a = θ· NH (a). The left inclu- sion ⊆ in (ii) follows immediately. For the inverse inclusion, take a ∈ A with NH(a) ∈ Aθ. Then, Since H 1(H, A) = 0, with H = hhi, there is c ∈ A with NH(a − θ·a) = NH(a) − θ·NH(a) = 0. a − θ·a = c − h·c. So, 0 = a − θ·a + θ·(a − θ·a) = c − h·c + θ·c − (θh)·c = c − h·c + (hθh)·c − (θh)·c = [c − (θh)·c] − h·[c − (θh)·c], i.e., c− (θh)·c ∈ AH. Since the group action of hθhi on AH coincides with the action of hθi on AH , we have H 1(hθhi, AH ) ∼= H 1(hθi, AH ) = 0. Therefore, part (i) applies, with θh replacing θ. Thus, we can write c = d + e with d ∈ AH and e ∈ Aθh, hence θ·e = h·e = (hθ)·(θ·e). Now, as d = h·d, a − θ·a = c − h·c = e − h·e = e − θ·e, showing that a + θ·e ∈ Aθ. Thus, a = [a + θ·e] − θ·e ∈ Aθ + Ahθ, completing the proof of the first equality in (ii). Since θh = h−1θ, the second equality in (ii) follows from the first by replacing h by h−1. (iii) We have bH −1(hθi, eA) ∼= Aθ(cid:14){a + θ·a a ∈ A}, bH −1(hhθi, eA) ∼= Ahθ(cid:14){a + (hθ)·a a ∈ A}, and The map corhθi→D : bH −1(hθi, eA) → bH −1(D, eA) arises from the inclusion Aθ ֒→ {a ∈ A NH(a) ∈ Aθ}; like- wise for corhhθi→D : bH −1(hhθi, eA) → bH −1(D, eA). Thus, the surjectivity asserted in part (iii) is immediate bH −1(D, eA) ∼= {a ∈ A NH(a) ∈ Aθ}(cid:14) ha − k·a, a + (kθ)·a a ∈ A, k ∈ Hi. from part (ii). (cid:3) Proposition 5.4. Let F ⊆ K ⊆ M be fields with [M : F ] < ∞ and M a K/F -generalized dihedral exten-sion. Let G = Gal(M/F ) and H = Gal(M/K). Take any θ ∈ G \ H. Then there is an exact sequence: Qh∈H bH −1(hhθi,gM ∗) −→ bH −1(G,gM ∗) −→ ker(eN )(cid:14) Π −→ 1 where ker(eN ) = {m ∈ M ∗ NM/K(m) ∈ F ∗} and Π =Qh∈H M ∗hθ. In particular, if M/K is cyclic Galois, then ker(eN )/Π = 1. (5.7) 22 A. R. WADSWORTH Proof. Here, M ∗hθ = {m ∈ M ∗ hθ(m) = m}. We have bH −1(G,gM ∗) ∼= ker(eN )(cid:14)IG(gM ∗) as in (5.4) -- (5.6). For any h ∈ H and m ∈ M ∗, m/h(m) = [m · θ(m)](cid:14)[θ(m) · h(m)] = [m · θ(m)](cid:14)[θ(m) · hθ(θ(m))] ∈ M ∗θM ∗hθ and m·hθ(m) ∈ M ∗hθ. Hence, by (5.5), M ∗hθ = Π. (5.8) Thus, there is a well-defined epimorphism ζ : bH −1(G,gM ∗) → ker(eN )(cid:14) Π, with ker(ζ) = Π(cid:14)IG(gM ∗). Now, for h ∈ H, we have bH −1(hhθi,gM ∗) ∼= M ∗hθ(cid:14)NM/M hhθi(M ∗). So, Qh∈H bH −1(hhθi,gM ∗) clearly maps If H is cyclic, then G is dihedral, and ker(eN ) = Π by onto ker(ζ), proving the exactness of (5.7). Lemma 5.3(ii). Remark 5.5. In the context of Prop. 5.4, suppose H = hh1, . . . , hmi. Then, the following lemma shows that (cid:3) IG(gM ∗) ⊆ Qh∈H M ∗hθ = Qh∈H m θ, (ε1,...,εm)∈{0,1}m 1 ...hεm M ∗ hε1 Q (ε1,...,εm)∈{0,1}m bH −1(hhε1 Q (5.9) 1 . . . hεm m θi,gM ∗). One can see by looking at examples that the product on the right in (5.9) is minimal in that if we delete any of the terms in that product, then the equality no longer holds in general. so the left term in (5.7) could be replaced by Lemma 5.6. Let G = hH, θi be a generalized dihedral group, where H is an abelian subgroup of G with G : H = 2, θ has order 2, and θhθ = h−1 for all h ∈ H. Let A be any G-module. Suppose H = hh1, . . . , hmi. Then, Ahθ = Ph∈H P Ah ε1 1 ...hεm m θ. (ε1,...εm)∈{0,1}m Proof. This follows from [HW2, Lemma 4.9] (with A for U , H for the abelian group A and Wh = Ahθ for all h ∈ H), once we establish that Ahθ ⊆ Akθ + Ak2h−1θ for all h, k ∈ H. For this, take any a ∈ Ahθ. Then θ(a) = h−1(a). Hence, k2h−1θ(kθ(a)) = k2h−1k−1(a) = kθ(a), showing that kθ(a) ∈ Ak2h−1θ. Thus a = [a + kθ(a)] − kθ(a) ∈ Akθ + Ak2h−1θ, proving the required inclusion. (cid:3) 6. Unitary relative Brauer Groups, bicylic case In this section we prove a unitary version of the formula Br(M/K)(cid:14) Dec(M/K) ∼= bH −1(Gal(M/K), M ∗), for M a bicyclic Galois extension of K, see (3.9) above. The unitary version was inspired by the result of Yanchevskiı [Y3, Prop. 5.5], which was a key part of his proof in [Y4, Th. A] that any finite abelian group can be realized as the unitary SK1 of some division algebra with involution of the second kind. Let F ⊆ K ⊆ M be fields with [K : F ] = 2 and K Galois over F , and M = L1 ⊗K L2 with each Li cyclic Galois over F . Assume M is K/F -generalized dihedral, as described at the beginning of §4. Let G = Gal(M/F ) and H = Gal(M/K), and choose and fix an element θ ∈ G\H. So, Gal(K/F ) = {θK, idK}. To simplify notation, let σ (not σ1) be a fixed generator of Gal(M/L2), and ρ (not σ2) a fixed generator of Gal(M/L1); so, H = hσi × hρi. Let n = [L1 : K], which is the order of σ in H, and let ℓ = [L2 : K], which is the order of ρ. As in Prop. 5.4, let and (See (5.9) for the second equality.) ker(eN ) = {a ∈ M ∗ NM/K (a) ∈ F ∗} Π = Qh∈HM ∗hθ = M ∗θM ∗ρθM ∗σθM ∗ρσθ. (6.1) UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 23 Proposition 6.1. We have Br(M/K; F )(cid:14) Dec(M/K; F ) ∼= ker(eN )/Π. Proof. This follows by combining the formulas for unitary SK1 given in [Y3, Prop. 5.5] with the Henselian version of the formula in [HW2, Cor. 4.11]. However, we give a direct proof avoiding the use of Yanchevskiı's special unitary conorms, since we will later need an explicit description of the isomorphism. Define a map as follows: By Lemma 4.1, a Brauer class in Br(M/K; F ) is represented by an algebra A = A(u, b1, b2), where u, b1, b2 satisfy the conditions in (3.6) and b1 ∈ L∗θ 1 , and u ρσθ(u) = 1. By Hilbert 90 (for the group hρσθi), there is q ∈ M ∗ with u = q/ρσθ(q). Define Ψ : Br(M/K; F ) −→ ker(eN )/Π 2 , b2 ∈ L∗θ Ψ(cid:0)A(u, b1, b2)(cid:1) = q Π ∈ ker(eN )/Π. i=0 Lℓ−1 b1 ∈ M hσi = L2, We will show that Ψ is a well-defined, surjective homomorphism with kernel Br(L1/K; F ) Br(L2/K; F ), which equals Dec(M/K; F ) (see (4.3)). For the well-definition of Ψ, first note that 1 = NM/K (u) = NM/K (q/ρσθ(q)) = NM/K (q)(cid:14)NM/K (θ(q)) = NM/K(q)(cid:14)θ(NM/K (q)), so, q ∈ ker(eN ). Also, given u, the choice of q with q/ρσθ(q) = u is unique up to a multiple in M ∗ρσθ. Since M ∗ρσθ ⊆ Π, Ψ(cid:0)A(u, b1, b2)(cid:1) is independent of the choice of q from u. Now, suppose A(u, b1, b2) ∼= A(u′, b′ 1, b′ 1, b′ 2 each satisfying the conditions of Lemma 4.1(iii). We have the presentation A(u, b1, b2) =Ln−1 j=0 M xiyj, where int(x)M = σ, xn = b1, int(y)M = ρ, yℓ = b2, and xyx−1y−1 = u, with u, b1, b2 and u′, b′ so, (see (3.6)) 2), b2 ∈ M hρi = L1, NM/L2(u) = b1/ρ(b1), NM/L1(u) = σ(b2)/b2. (6.2) The conditions of Lemma 4.1(iii) we are also assuming are that b1 ∈ Lθ 2, b2 ∈ Lθ 1, and (6.3) 2 and u′. By Lemma 4.1, there is a The corresponding conditions in (6.2) and (6.3) hold for b′ K/F -involution τ of A = A(u, b1, b2) with τM = θ, τ (x) = x, τ (y) = y. We have an isomorphism A(u, b1, b2) ∼= A(u′, b′ 2), and by Skolem-Noether there is such an isomorphism which restricts to the 1, int(y′)M = ρ, identity on M . Therefore, there exist x′ and y′ in A∗ such that int(x′)M = σ, x′n = b′ 2, and x′y′x′−1y′−1 = u′. Since int(x′)M = int(x)M there is c1 ∈ CA(M )∗ = M ∗ with x′ = c1x, y′ℓ = b′ 2 = (c2y)ℓ, and 1 = (c1x)n, b′ and likewise c2 ∈ M ∗ with y′ = c2y. By simplifying the expressions b′ u′ = (c1x)(c2y)(c1x)−1(c2y)−1, we find that u ρσθ(u) = 1. 1, b′ 1, b′ b′ 2 = NM/L1(c2) b2, b′ 1 = NM/L2(c1) b1, (6.4) By Lemma 4.1, there is a K/F -involution τ ′ on A with τ ′(x′) = x′, τ ′(y′) = y′, and τ ′M = θ. Since τ ′τ −1 is a K-automorphism of A, there exists e ∈ A∗ with τ ′ = int(e)τ . Because τ ′M = τM , e ∈ CA(M ) = M . The condition that τ ′2 = idA implies that e/θ(e) ∈ K ∗. Since e/θ(e)(cid:0)θ(e/θ(e))(cid:1) = 1, Hilbert 90 for K/F shows that there is d ∈ K ∗ with d/θ(d) = e/θ(e). By replacing e by e/d, we may assume that θ(e) = e. The conditions that c1x = τ ′(c1x) = int(e)τ (c1x) and c2y = τ ′(c2y) = int(e)τ (c2y) yield u′ = (cid:0)c1/ρ(c1)(cid:1)(cid:0)σ(c2)/c2(cid:1) u. c1 = σθ(c1) e/σ(e) and c2 = ρθ(c2) e/ρ(e), hence, ρ(c1) = ρσθ(c1) ρ(e)/ρσ(e) and σ(c2) = ρσθ(c2) σ(e)/ρσ(e). The equations (6.5) yield c1/ρ(c1) = (cid:0)c1/ρσθ(c1)(cid:1)(cid:0)ρσ(e)/ρ(e)(cid:1) and σ(c2)/c2 = (cid:0)ρσθ(c2)/c2(cid:1)(cid:0)σ(e)/ρσ(e)(cid:1). (6.5) (6.6) 24 A. R. WADSWORTH Let eq = (c1/c2)σ(e). Then, using (6.6), (6.4) and θ(e) = e, = (cid:0)c1/ρ(c1)(cid:1)(cid:0)ρ(e)/ρσ(e)(cid:1)(cid:0)σ(c2)/c2(cid:1)(cid:0)ρσ(e)/σ(e)(cid:1)(cid:0)σ(e)/ρθ(e)(cid:1) = (cid:0)c1/ρ(c1)(cid:1)(cid:0)σ(c2)/c2(cid:1) = u′/u. eq/ρσθ(eq) = (cid:0)c1/ρσθ(c1)(cid:1)(cid:0)ρσθ(c2)/c2(cid:1)(cid:0)σ(e)/ρσθσ(e)(cid:1) 2/b2 ∈ L∗θ 1/b1 ∈ L∗θ When q ∈ M ∗ is chosen so that q/ρσθ(q) = u, set q′ = eqq; then (6.7) shows that q′/ρσθ(q′) = u′. We check that eq ∈ Π: We have (see (6.4) and (6.3)) NM/L2(c1) = b′ 2 . Therefore, by Lemma 5.3(ii) 2), c1 ∈ M ∗θM ∗σθ ⊆ Π. Likewise, c2 ∈ M ∗θM ∗ρθ ⊆ Π as applied to the dihedral group hσ, θi = Gal(M/Lθ 1 . Finally, since θ(e) = e, we have σ(e) = σθ(e) = σθσ−1(σ(e)) = σ2θ(σ(e)). So, NM/L1(c2) = b′ σ(e) ∈ M ∗σ2θ ⊆ Π. Thus, q′ ≡ q (mod Π), which shows that Ψ is well-defined independent of the choice of presentation of A as A(u, b1, b2) with u, b1, b2 as in Lemma 4.1(iii). For the surjectivity of Ψ, take any q ∈ ker(eN ) and set u = q/ρσθ(q). So, u ρσθ(u) = 1. Furthermore, as NM/K (q) ∈ F ∗, Since NL2/K (NM/L2(u)) = NM/K(u) = 1, by Hilbert 90 for L2/K there is b1 ∈ L∗ Then, NM/K (u) = NM/K (q)/NM/K (ρσθ(q)) = NM/K(q)/θ(NM/K (q)) = 1. 2 with b1/ρ(b1) = NM/L2(u). (6.7) Hence, b1/ρ(b1) = NM/L2(q)(cid:14)NM/L2(ρσθ(q)) = NM/L2(q)/ρθ(NM/L2(q)). 1 = (b1/ρ(b1)) ρθ(b1/ρ(b1)) = (b1/θ(b1))/ρ(b1/θ(b1)), which shows that b1/θ(b1) ∈ Lρ it follows that b1 = kbb1 with k ∈ K ∗ and bb1 ∈ L∗θ 2 . Likewise, there is b2 ∈ L∗θ 1 with NM/L1(u−1) = b2/σ(b2). Then, as u, b1, b2 satisfy the conditions of (3.6) (where σ1 = σ and σ2 = ρ) the algebra A(u, b1, b2) exists, and by Lemma 4.1 [A(u, b1, b2)] ∈ Br(M/K; F ). Clearly, Ψ[A(u, b1, b2)] = q Π. 2 . By replacing b1 with bb1, we may assume that b1 ∈ L∗θ 2 = K. By Lemma 5.3(i) applied to the dihedral group Gal(L2/F ) = hρL2 , θL2i, Finally, we determine ker(Ψ): If [B] ∈ Br(L1/K; F ) then we can assume that B has L1 as a maximal sub- field. Then, by Lemma 4.1, B ∼= (L1/K, σ, b1), where b1 ∈ K ∗θ = F ∗. Likewise, for any [C] ∈ Br(L2/K; F ), we have C ∼ (L2/K, ρ, b2) for some b2 ∈ F ∗. Then, (cid:2)B ⊗K C(cid:3) =(cid:2)(L1/K, σ, b1) ⊗K (L2/K, ρ, b2)(cid:3) =(cid:2)A(1, b1, b2)(cid:3) ∈ ker(Ψ), 2 , b2 ∈ L∗θ since when u = 1 we can take q = 1. So Br(L1/K; F ) Br(L2/K; F ) ⊆ ker(Ψ). For the reverse inclusion, take any A = A(u, b1, b2) with [A] ∈ ker(Ψ). Since [A] ∈ Br(M/K; F ), by Lemma 4.1 we may assume that b1 ∈ L∗θ 1 and u ρσθ(u) = 1. Since, [A] ∈ ker(Ψ), we have u = q/ρσθ(q) with q ∈ Π, so q = qθqρθqσθqρσθ, where qθ ∈ M ∗θ, qρθ ∈ M ∗ρθ, qσθ ∈ M ∗σθ, and qρσθ ∈ M ∗ρσθ. Thus, u = q/ρσθ(q) = (cid:0)qθ/ρσ(qθ)(cid:1)(cid:0)qρθ/σ(qρθ)(cid:1)(cid:0)qσθ/ρ(qσθ)(cid:1) = (cid:0)qθqρθ/σ(qθqρθ)(cid:1)(cid:0)qσθσ(qθ)/ρ(qσθσ(qθ)(cid:1) = (cid:0)c2/σ(c2)(cid:1)(cid:0)ρ(c1)/c1(cid:1), where c2 = qθqρθ and c1 = (qσθσ(qθ))−1. Then by (3.7), A = A(u, b1, b2) ∼= A(u′, b′ 2) where u′ = (c1/ρ(c1))(σ(c2)/c2)u = 1, and b′ 2 = NM/L1(c2)b2. Since c2 ∈ M ∗θM ∗ρθ, an easy calculation or an application of Lemma 5.3(ii) for the dihedral group Gal(M/Lθ 1) = hρ, θi shows θ, as b2 ∈ L∗θ that NM/L1(c2) ∈ L∗θ 1 . But also, as in (6.2), σ(b′ 1 = K ∗θ = F ∗. Likewise, as qσθ ∈ M ∗θ and σ(qθ) ∈ M ∗σ2θ ⊆ M ∗θM ∗σθ (see (5.9)), we have c1 ∈ M ∗θM ∗σθ. Therefore, an easy calculation or Lemma 5.3(ii) for the dihedral group Gal(M/Lθ 2 . So, arguing just as for b′ 2 ∈ L∗θ 2) = hσ, θi shows that NM/L2(c1) ∈ L∗θ 2 = NM/L1(c2)b2 ∈ L∗ 1 ∩ L∗σ 2 = NM/L1(u′) = NM/L1(1) = 1. Hence, b′ 1 = NM/L2(c1)b1 and b′ 1 . Therefore, b′ 2)/b′ 1, b′ 2, we find that b′ 1 1 ∈ F ∗. Thus, A ∼= A(1, b′ 1, b′ 2) ∼= (L1/K, σ, b′ 1) ⊗K (L2/K, ρ, b′ 2), UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 25 and since the b′ Thus, ker(Ψ) = Br(L1/K; F ) Br(L2/K; F ) = Dec(M/K; F ). 1)(cid:3) ∈ Br(L1/K; F ) and(cid:2)(L2/K, ρ, b′ i ∈ F ∗,(cid:2)(L1/K, σ, b′ 2)(cid:3) ∈ Br(L2/K; F ), by Lemma 4.1. (cid:3) This yields our unitary analogue to (3.9) above. Proposition 6.2. For M bicyclic Galois over K with M K/F -generalized dihedral, setting G = Gal(M/F ), H = Gal(M/K), and θ any element of G \ H as above, there is an exact sequence Qh∈H bH −1(hhθi,gM ∗) −→ bH −1(G,gM ∗) −→ Br(M/K; F )/ Dec(M/K; F ) −→ 0 Proof. This follows from Prop. 6.1 and Prop. 5.4. (6.8) (cid:3) 7. Semiramfied algebras We now apply the results of the preceding sections to the calculation of unitary SK1 for semiramified graded division algebras with graded T/R-involution Throughout this section, fix a graded field T and a graded subfield R of T with [T : R] = 2 and T Galois over R, say with Gal(T/R) = {id, ψ}. Assume further that T is inertial over R. Thus, ΓT = ΓR, [T0 : R0] = 2, T0 is Galois over with Gal(T0/R0) = {id, ψ0}, where ψ0 = ψT0 , and ψ = ψ0 ⊗ idR when we identify T with T0 ⊗R0 R. By definition, for a central simple graded division algebra B over T with a graded unitary T/R-involution τ , the unitary SK1 is given by SK1(B, τ ) = Σ′ τ (B)(cid:14) Στ (B), where Σ′ τ (B) = {b ∈ B∗ NrdB(b) ∈ R} and Στ (B) = (cid:10){b ∈ B∗ τ (b) = b}(cid:11) We are assuming that T/R is inertial because otherwise T/R is totally ramified and SK1(B, τ ) = 1, by [HW2, Th. 4.5]. It is known by [HW2, Lemma 2.3(iii)] that [B∗, B∗] ⊆ Στ (B), so SK1(B, τ ) is an abelian group. Also, if τ ′ is another graded T/R-involution on B, then Σ′ τ (B) and Στ ′(B) = Στ (B), so SK1(B, τ ′) = SK1(B, τ ). The easy proof is analogous to the ungraded proof given in [Y1, Lemma 1]. τ ′(B) = Σ′ Let E be a semiramified T-central graded division algebra. So, as we have seen, E0 is a field abelian Galois over T0, and ΘE : ΓE/ΓT → Gal(E0/T0) is a canonical isomorphism. Suppose E has a graded T/R- involution τ ; so τT0 = ψ0. We have seen in Prop. 4.5 that E0 is then a T0/R0-generalized dihedral Galois extension. Let H = Gal(E0/T0) and G = Gal(E0/R0), and let τ = τE0 ∈ G \ H. For each γ ∈ ΓE choose and fix xγ ∈ Eγ with xγ 6= 0 and τ (xγ) = xγ. (Such xγ exist, by [HW2, Lemma 4.6(i)].) Our starting point is the formula proved in [HW2, Th. 4.7] where SK1(E, τ ) ∼= (cid:0)Στ (E)′ ∩ E∗ 0(cid:1)(cid:14)(cid:0)Στ (E) ∩ E∗ 0(cid:1) = ker(eN )(cid:14)(cid:0) Π · X(cid:1), (7.1) 0 NE0/T0(a) ∈ R0}; , where E∗hτ 0 = {a ∈ E∗ 0 hτ (a) = a}; E∗hτ ker(eN ) = {a ∈ E∗ Π = Qh∈H X = (cid:10)xγxδx−1 0 γ+δ γ, δ ∈ ΓE(cid:11) ⊆ E∗ 0. ∗ Note that H maps ker(eN ) (resp. Π) to itself, so H acts on ker(eN )/Π. But this action is trivial since IH(ker(eN )) ⊆ IG(fE0 Theorem 7.1. Suppose E is DSR for T/R, i.e., in addition to the hypotheses above, E has a maximal graded subfield J with τ (J) = J. Then, ) ⊆ Π (see (5.8) above). 26 A. R. WADSWORTH (i) SK1(E, τ ) ∼= ker(eN )/Π, and there is an exact sequence ) −→ bH −1(G,fE0 Qh∈H bH −1(hhτi,fE0 (ii) If E0 = L1 ⊗T0 L2 with each Li cyclic Galois over T0, then ∗ ∗ ) −→ SK1(E, τ ) −→ 1. SK1(E, τ ) ∼= Br(E0/T0; R0)(cid:14) Dec(E0/T0; R0). Proof. (i) The first formula for SK1(E, τ ) was given in [HW2, Cor. 4.11]. The point is that the xγ can all ∗ ⊆ Π, so the X term in (7.1) drops out. The exact sequence in (i) then be chosen in J; then X ⊆ J∗τ follows by Prop. 5.4. Part (ii) is immediate from (i) and Prop. 6.1. 0 = R0 (cid:3) Note that Th. 7.1 is the unitary analogue to Prop. 3.2 for nonunitary SK1 in the DSR case. To improve the formula (7.1) in the manner of Th. 7.1 for E semiramified but not DSR we need more information on the contribution of the X term. This contribution is measured by (Π · X)/Π. For γ ∈ ΓE we write γ for γ + ΓT ∈ ΓE/ΓT. Proposition 7.2. There is a well-defined 2-cocycle g ∈ Z 2(ΓE/ΓT, ker(eN )/Π) given by g(γ, δ) = xγxδx−1 γ+δ Π. (7.2) This g is independent of the choice of nonzero symmetric elements xγ, xδ, xγ+δ in Eγ, Eδ, Eγ+δ. Further- more, for all γ, δ ∈ ΓE/ΓT and i, j, k, ℓ ∈ Z, we have k ℓ(cid:1) . g(iγ + jδ, kγ + ℓδ) = g(γ, δ)∆ where ∆ = det(cid:0) i j (In particular, g(γ, γ) = 1 Π and g(δ, γ) = g(γ, δ)−1.) Moreover, him(g)i = (cid:0)Π · X(cid:1)/Π, which is a finite group. (7.3) Proof. For γ, δ ∈ ΓE, set cγ,δ = xγxδx−1 γ+δ ∈ E∗ 0. we work with the function Note that cγ,δ ∈ ker(eN ), since it is a product of τ -symmetric elements of E∗. For notational convenience f : ΓE × ΓE −→ ker(eN )/Π given by f (γ, δ) = cγ,δ Π. Thus, g(γ, δ) = f (γ, δ) We first show that the definition of f is independent of the choices made of xγ, xδ, xγ+δ. Fix γ and δ in ΓE for the moment. Take any a ∈ E∗ 0 with τ (axγ) = axγ. Then, axγ = τ (axγ) = xγτ (a) = ΘE(γ)(τ (a))xγ ; so a = ΘE(γ)(τ (a)), i.e. a ∈ E∗ΘE(γ)τ ⊆ Π. Hence, if we let x′ γ = axγ, then x′ 0 with τ (bxδ) = bxδ, then ΘE(γ)(b) ∈ E∗ΘE (2γ+δ)τ γ+δ ≡ xγxδx−1 γ+δ (mod Π). Again, for d ∈ E∗ ⊆ Π, so for x′ γ+δ = dxγ+δ, we have xγxδx′−1 γ+δ (mod Π). Thus, each such change does not affect the value of f (γ, δ), and we are free to make such changes when convenient. 0 with τ (dxγ+δ) = dxγ+δ, we have d ∈ E∗ΘE(γ+δ)τ γ+δ ≡ xγxδx−1 γ+δ (mod Π). Likewise, if we take any b ∈ E∗ γ+δ = ΘE(γ)(b)xγ xδx−1 γ+δ ≡ xγxδx−1 ⊆ Π so xγx′ γxδx−1 δx−1 0 0 0 We prove further identities for the function f which hold for all γ, δ, ε ∈ ΓE and i, j, k, ℓ ∈ Z: (i) f (γ + β, δ) = f (γ, δ) = f (γ, δ + β) for any β ∈ ΓT. For, as ΓR = ΓT, there is a nonzero a ∈ Rβ. Since a ∈ Z(E) and τ (a) = a, we could have chosen xγ+β = axγ, xδ+β = axδ, and xγ+δ+β = axγ+δ. Then, f (γ + β, δ) = (axγ)xδ(axγ+δ)−1 Π = xγxδx−1 γ+δ Π = f (γ, δ), and likewise f (γ, δ + β) = f (γ, δ). This proves (i), which shows that the g of the Prop. is well-defined. (ii) f (iγ, jγ) = 1 Π. UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 27 For, we can choose xiγ = xi γ, xjγ = xj γ, and xiγ+jγ = xi+j γ . Then, ciγ,jγ = 1. (iii) f (δ, γ) = f (γ, δ)−1. For, by applying τ to the equation xγxδ = cγ,δxγ+δ, we obtain xδxγ = xγ+δτ (cγ,δ) = ΘE(γ + δ)(τ (cγ,δ))xδ+γ , yielding cδ,γ = ΘE(γ + δ)(τ (cγ,δ)), so cδ,γcγ,δ ∈ E∗ΘE(γ+δ)τ 0 ⊆ Π. Formula (iii) then follows. (iv) f (γ, δ)f (γ + δ, ε) = f (γ, δ + ε)f (δ, ε), follows from (xγxδ)xε = xγ(xδxε), which yields cγ,δ cγ+δ,ε = ΘE(γ)(cδ,ε) cγ,δ+ε. Then (iv) follows, given the i.e., f ∈ Z 2(ΓE, ker(eN )/Π), since ΓE (acting via ΘE(ΓE) = H) acts trivially on ker(eN )/Π. This identity trivial action of H on ker(eN )/Π. and f (γ, γ + δ) = f (γ, δ). f (γ + δ, δ) = f (γ, δ) (v) For, as τ (xδxγxδ) = xδxγxδ, we can take xγ+2δ = xδxγxδ. Then, xδxγxδ = cδ,γ cδ+γ,δ xγ+2δ = cδ,γ cδ+γ,δ xδxγxδ. Hence, 1 Π = f (δ, γ)f (δ + γ, δ), so f (δ + γ, δ) = f (δ, γ)−1 = f (γ, δ), using (iii). This proves the first formula in (v), and the second formula follows analogously, or from the first by using (iii). (vi) f (γ + jδ, δ) = f (γ, δ) = f (γ, jγ + δ) for all j ∈ Z. This follows from (v) by induction on j. (vii) f (iγ, jδ) = f (γ, δ)ij . For, by (iv) with jδ for δ and δ for ε, f (γ, jδ)f (γ + jδ, δ) = f (γ, (j + 1)δ)f (jδ, δ), which by (vi) and (ii) reduces to f (γ, jδ)f (γ, δ) = f (γ, (j + 1)δ). Then (vii) for i = 1 follows by induction on j with the initial case j = 0 given by (ii). From the i = 1 case the result for arbitrary i follows by using (iii). (viii) k ℓ(cid:1) . f (iγ+jδ, kγ+ℓδ) = f (γ, δ)∆ where ∆ = det(cid:0) i j For this note first that this is true if i = 0, as f (jδ, kγ + ℓδ) = f (δ, kγ + ℓδ)j = f (δ, kγ)j = f (γ, δ)−jk, by (vii), (vi), (vii), and (iii). Analogously, (viii) is true if k = 0. To verify (viii) in general, we argue by induction on i + k. By invoking (iii) and interchanging iγ + jδ with kγ + ℓδ if necessary, we can assume i ≤ k. We can assume i ≥ 1, since the case i = 0 is already done. Let η = ±1, with the sign chosen so that k − ηi = k − i. Since i + k − ηi = k < i + k, we have by (vi) and induction, f (iγ + jδ, kγ + ℓδ) = f(cid:0)iγ + jδ, (kγ + ℓδ(cid:1) − η(iγ + jδ)) = f(cid:0)iγ + jδ, (k − ηi)γ + (ℓ − ηj)δ(cid:1) = f (γ, δ)∆′ j where ∆′ = det(cid:16) i k−ηi ℓ−ηj(cid:17) = det(cid:0) i j k ℓ(cid:1) . Thus, (viii) is proved, and when (viii) is restated in terms of g, it is formula (7.3). It is clear from the definition and well-definition of g that him(g)i = (Π · X)/Π. This abelian group is finite since the domain of g is finite, and each g(γ, δ) has finite order by formula (7.3). Identity (iv) above shows that f is a 2-cocycle, so g is also a 2-cocycle. (cid:3) 28 A. R. WADSWORTH Remark. If the finite abelian group ΓE/ΓT has exponent e, then formula (7.3) shows that him(g)i has exponent dividing e. So, we have the crude upper bound him(g)i ≤ eΓE/ΓT2 We can now prove a formula for unitary SK1 of semiramified graded algebras. This is a unitary analogue . to Th. 3.7 above. Theorem 7.3. Let E be a semiramified T-central graded division algebra with a unitary graded T/R- involution τ , where T is unramified over R. Take any decomposition E ∼g I ⊗T N where I is inertial with [I0] ∈ Br(E0/T0; R0) and N is DSR for T/R, as in Prop. 4.5 above. Then, (i) SK1(E, τ ) ∼=(cid:0) ker(eN )/Π(cid:1)(cid:14)him(g)i, where g is the function of Prop. 7.2. If I0 ∼ A(E0/T0, σ, u, b) as in Lemma 4.1(iii) with θ = τE0 , then im(g) is computable from the uij . (ii) If E0 ∼= L1 ⊗T0 L2 with each Li cyclic Galois over T0, then SK1(E, T) ∼= Br(E0/T0; R0)(cid:14)[Dec(E0/T0; R0) · h[I0]i]. Proof. (i) From (7.1) and Prop. 7.2, we have It remains to relate im(g) to the uij describing I0. SK(E, τ ) ∼= (cid:0) ker(eN )/Π(cid:1)(cid:14)(cid:2)(Π · X)/Π(cid:3) ∼= (cid:0) ker(eN )/Π(cid:1)(cid:14)him(g)i. We have I0 ∼ A(E0/T0, σ, u, b), as in Lemma 4.1(iii), with θ = τ = τT0 . Since N is DSR for T/R with N0 ∼= E0 and ΘN = ΘE by Prop. 4.5, Prop. 4.4 yields N ∼=g A(E0T/T, σ, 1, c), with each ci ∈ R∗ with deg(ci) = riγi for some γi ∈ ΓN = ΓE with ΘE(γi) = σi. Therefore, by Remark 3.3, 0R∗. So, E ∼g E′, where E′ = A(E0T/T, σ, u, d), with the same u as for I0 and each di = bici ∈ E∗ τ (di) = τ (ci)τ (bi) = cibi = bici = di. Since N is a semiramified graded division algebra and deg(di) = deg(ci) for each i, Lemma 3.4 applied to N and to E′ shows that ΓE′ = ΓN and E′ is a semiramified graded division algebra. Therefore, as E and E′ are each graded division algebras with E ∼g E′, we have E ∼=g E′ by the graded Wedderburn Theorem. So, we may assume E = E′ = A(E0T/T, σ, u, d). Take y1, . . . , yk ∈ E∗ with int(yi)E0T = σi, yri j = uij. Now, the graded field E0T is T/R generalized dihedral, and θ = τE0T lies in Gal(E0T/R) \ Gal(E0T/T). Therefore, the proof of Lemma 4.1 (iii) ⇒ (i) shows that there is a graded T/R-involution τ ′ of E with each yi = τ ′(yi) and τ ′E0T = θ. Since SK1(E, τ ) = SK1(E, τ ′) we may replace τ by τ ′, so each yi = τ (yi), while τ is unchanged. i = di, and yiyjy−1 i y−1 Fix any η ∈ ΓE/ΓT, and let ση = ΘE(η) ∈ H. Take the unique i ∈ I with σi = ση (notation as in §3), let γ = deg(yi) ∈ ΓE, and set yγ = yi. Since ΘE(γ) = int(yγ)E0 = ΘE(η) and ΘE : ΓE/ΓT → H is an isomorphism for E semiramified (see §2), η = γ in ΓE/ΓT. Since τ (yi) = yi for each i, τ (yγ) is the product of the yi appearing in yγ but with the order reversed. Hence, the commutator identities show that τ (yγ) = aγyγ where aγ in E0 is a computable product of the uij and their conjugates under the yi. Since each yℓuijy−1 (For example, τ (y1y2y3) = y3y2y1 = [u32σ2(u31)u21]y1y2y3.) By applying τ to the equation τ (yγ) = aγyγ, we find ℓ = σℓ(uij), aγ is a computable product of terms σℓ(uij). ση τ Therefore, from Hilbert 90 for the quadratic extension E0/E 0 , there is tγ ∈ E∗ 0 with tγ [σητ (tγ)]−1 = aγ. aγ σητ (aγ) = 1. Then, τ (tγyγ) = tγyγ, so for the xγ in X we can set xγ = tγyγ. Now take any ζ ∈ ΓE/ΓT and carry out the same process for ζ as we have just done for η, obtaining δ ∈ Γ with δ = ζ, and yδ with deg(yδ) = δ and int(yδ)E0 = σζ, then determining aδ, tδ, xδ. Then set yγ+δ = yγyδ, so int(yγ+δ)E0 = σησζ. Let aγ+δ = τ (yγ+δ)y−1 0 with tγ+δ[σησζτ (bγ+δ)]−1 = aγ+δ. Then set xγ+δ = tγ+δyγ+δ, so that τ (xγ+δ) = xγ+δ. By the definition of Since aγ+δσησζτ (aγ+δ) = 1, by Hilbert 90 there is tγ+δ ∈ E∗ γ+δ ∈ E∗ 0. UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 29 the function g of Prop. 7.2, we have in ker(eN )/Π, g(η, ζ) = xγxδx−1 γ+δ Π = (tγyγ)(tδyδ)(tγ+δyγyδ)−1 Π = tγση(tδ)t−1 γ+δ Π. Since the t's are determined by the a's, which are determined by the uij, this shows that im(g) is determined by the uij. (ii) Suppose now that E0 = L1 ⊗T0 L2 with each Li cyclic Galois over T0, and let σ = σ1 and ρ = σ2, as in §6. The isomorphism Br(M/K; F )(cid:14) Dec(M/K; F ) ∼= ker(eN )/Π (7.4) 0 with u = q[ρστ (q)]−1. Take standard gener- of Prop. 6.1 maps [I0] = [A(u, b1, b2)] to q Π, where q ∈ E∗ ators y1, y2 of A(u, b1, b2). As noted for (i), we can assume after modifying τ (without changing τ ) that τ (y1) = y1 and τ (y2) = y2. Let γ = deg(y1) and δ = deg(y2) in ΓE, so ΘE(γ) = int(y1)E0 = σ and ΘE(δ) = int(y2)E0 = ρ. Since ΓE/ΓT ∼= H = hσ, ρi, we have ΓE/ΓT = hγ, δi. As τ (y1) = y1, we can take xγ = y1, and likewise xδ = y2. Because τ (y2y1) = uy2y1 = q[ρστ (q)]−1y2y1, we have τ (qy2y1) = qy2y1; thus, we can take xδ+γ = qy2y1. Then, g(δ, γ) = xδxγx−1 δ+γ Π = y2y1(qy2y1)−1 Π = q−1 Π. Since δ and γ generate ΓE/ΓT formula (7.3) shows that im(g) = hg(δ, γ)i = hq−1Πi = hq Πi. Therefore, the isomorphism of (7.4) maps h[I0]i to hq Πi = him(g)i. Thus, the isomorphism asserted for (ii) follows from (i). (cid:3) Example 7.4. Here is a unitary version of Ex. 3.9. Take any integer n ≥ 2, and let F ⊆ K be fields with [K : F ] = 2, K Galois over F , and K = F (ω) where ω is a primitive n2-root of unity. Suppose further that for the nonidentity element ψ0 of Gal(K/F ) we have ψ0(ω) = ω−1. (For example, we could take K = Q(ω), the n2-cyclotomic extension of Q, and F = K ∩R.) Let T = K[x, x−1, y, y−1], the Laurent polynomial ring, with its usual grading by Z× Z; so, T is a graded field. Let R = F [x, x−1, y, y−1], which is a graded subfield of T with [T : R] = 2, T Galois over R, and T inertial over R. Also, Gal(T/R) = {ψ, idT}, where ψ = ψ0⊗ idR on T = T0 ⊗R0 R. Take any a, b ∈ F ∗ such that [K( n√a, n√b ) : K] = n2, and let M = K( n√a, n√b ). Then, it is easy to check that M is K/F -generalized dihedral. (One can think of such field extensions M/F as the generalized dihedral analogue to Kummer extensions.) Indeed, ψ0 on K extends to θ ∈ Gal(M/F ) given by θ( n√a) = n√a, θ( n√b) = n√b, and θK = ψ0; so, θ2 = idM , and for h ∈ Gal(M/K), we have θhθ = h−1. As in Ex. 3.9, take the graded symbol algebra E = (axn, byn, T)ω of degree n2, with its generators i, j satisfying in2 = byn, ij = ωji. For σ1, σ2 as in Ex. 3.9, it was noted there that E = A(M T/T, σ, u, d) where u12 = ω, and d1 = 1/(y n√b) and d2 = x n√a. We extend θ to an element of Gal(M T/R) by setting θR = id. Since θ(d1) = d1, θ(d2) = d2, and u12 σ1σ2θ(u12) = ωω−1 = 1, the graded version of Lemma 4.1 shows that there is a graded T/R-involution τ on E given by τ (j−1) = j−1, τ (i) = i, and τM E = θ. That is, τ is the R-linear map E → E such that τ (c iℓjm) = ψ(c)jmiℓ for all c ∈ T, ℓ, m ∈ Z. We have the decomposition of E noted in Ex. 3.9, = axn, jn2 E ∼g I ⊗T N where I = (a, b, T)ω and N = (x, b, T)ωn ⊗T (a, y, T)ωn . These I and N are T-central graded division algebras with I inertial and N DSR. Furthermore, as a, b, x, y ∈ R∗, there are unitary graded T/R-involutions τI on I and τN on N defined analogously to τ on E. So, by Th. 7.1(ii) SK1(N, τN) ∼= Br(cid:0)M/K; F(cid:1)(cid:14) Dec(cid:0)M/K; F(cid:1), where M = K( n√a, n√b ), with Dec(cid:0)M/K; F(cid:1) = Br(cid:0)K( n√a )/K; F(cid:1) · Br(cid:0)K( n√b )/K; F(cid:1) by (4.3). Since I0 ∼= (a, b, K)ω, Th. 7.3(ii) yields SK1(E, τ ) ∼= Br(M/K; F )(cid:14)(cid:2) Dec(M/K; F ) · h(a, b, K)ωi(cid:3). 30 A. R. WADSWORTH Note that E is semiramified, but it may or may not be DSR. Indeed, by Prop. 4.5(ii) E is DSR if and only if I0 ∈ Dec(cid:0)M/K; F(cid:1); the formulas above show that this holds if and only if the obvious surjection SK1(N, τN ) → SK1(E, τ ) is an isomorphism. Note also that Dec(M/K; F ) may be strictly smaller than Dec(M/K) ∩ Br(M/K; F ), i.e., there may be an algebra in Br(M/K) which decomposes according to M and has a K/F -involution, but in any decomposition the factors do not have K/F -involutions. Examples of this are given in Remark 8.2 below. For an ungraded version of this example, let K, F , a, and b be as above; then let K ′ = K((x))((y)) and F ′ = F ((x))((y)), and D = (axn, byn, K ′)ω. Then, with respect to the usual rank 2 Henselian valuations vK ′ on K ′ and vF ′ on F ′, K ′ is inertial of degree 2 over F ′. Furthermore, with respect to the valuation vD on D extending vK ′ on K ′, D is a semiramified K ′-central division algebra with a unitary K ′/F ′- involution τD defined just as for τ on E. For the associated graded ring gr(D) of D determined by vD, we have gr(D) ∼=g E, so by [HW2, Th. 3.5] SK1(D, τD) ∼= SK1(E, τ ). 8. Noninjectivity For any T-central graded division algebra B with unitary T/R-involution τ , there are well-defined canonical homomorphisms and α : SK1(B, τ ) → SK1(B) β : SK1(B) → SK1(B, τ ) given by a Στ (B) 7→ τ (a)a−1 [B∗, B∗] given by b [B∗, B∗] 7→ b Στ (B) for b ∈ B∗ with NrdB(b) = 1. for a ∈ Σ′ τ (B), (8.1) It is easy to check that β ◦ α and α ◦ β are each the squaring map. As pointed out in [Y3, Lemma, p. 185], since the exponent of the abelian group SK1(B, τ ) divides deg(B), if deg(B) is odd, then α must be injective. It seems to have been an open question up to now whether α is always injective, even when deg(B) is even. We now settle this question by using some of the results above to give examples of B of degree 4 with α not injective. We thank J.-P. Tignol for pointing out the relevance of indecomposable division algebras of degree 8 and exponent 2, and for calling his paper [T1] to our attention. Let F be a field with char(F ) 6= 2. Let M = F (√a,√b,√c ) with a, b, c ∈ F ∗ and [M : F ] = 8. Let K = F (√a ). We write Br2(F ) for the 2-torsion subgroup of Br(F ), and set Br2(M/F ) = Br(M/F ) ∩ Br2(F ), Br2(M/K; F ) = Br(M/K; F ) ∩ Br2(K), etc. Note that as Gal(M/F ) is an elementary abelian 2-group, M is a K/F -generalized dihedral extension. Also, resF →K maps Br2(M/F ) to Br(M/K; F ), since for [A] ∈ Br2(M/F ), corK→F [A⊗F K] = [A][K:F ] = 1 in Br(F ), so by Albert's Theorem A⊗F K has a unitary K/F -involution. Proposition 8.1. There is an exact sequence: 0 −→ Br2(M/F )(cid:14) Dec(M/F ) −→ Br(M/K; F )(cid:14) Dec(M/K; F ) −→ Br(M/K)(cid:14) Dec(M/K) Proof. The kernel of the right map in (8.2) is(cid:2) Br(M/K; F )∩Dec(M/K)(cid:3)(cid:14) Dec(M/K; F ). So, the exactness of (8.2) is equivalent to two assertions: (8.2) and (a) Br(M/K; F ) ∩ Dec(M/K) = Br2(M/K; F ). (b) Br2(M/F )(cid:14) Dec(M/F ) ∼= Br2(M/K; F )(cid:14) Dec(M/K; F ) The equality (a) is immediate from the fact that Dec(M/K) = Br2(M/K), as M is a biquadratic exten- sion of K. (This is well-known, and is deducible, e.g., by refining the argument in [KMRT, Prop. 16.2]. It also appears in [T1, Cor. 2.8] as the assertion that property P2(2) holds for K.) The isomorphism (b) appears in [T1, Prop.2.2] as the isomorphism N2(M/F ) ∼= M2(M/K/F ), see the conmments on p. 14 of [T1]. Since the isomorphism (b) is somewhat buried in the general arguments of [T1], we give a short and direct proof of it: If [A] ∈ Dec(M/F ), then A ∼ Q1 ⊗F Q2 ⊗F Q3, where Q1 is the quaternion algebra UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 31 F(cid:1), and Q3 = (cid:0) c,t (cid:0) a,r F(cid:1), for some r, s, t ∈ F ∗. So, A ⊗F K ∼ (Q2 ⊗F K) ⊗K (Q3 ⊗F K). F (cid:1), Q2 = (cid:0) b,s Here, Q2 ⊗F K has the unitary K/F -involution η ⊗ ψ, where η is any involution of the first kind on Q2 and ψ is the nonidentity F -automorphism of K. So [Q2 ⊗F K] ∈ Br(K(√b)/K; F ) ⊆ Dec(M/K; F ); likewise [Q3 ⊗F K] ∈ Br(K(√c)/K; F ) ⊆ Dec(M/K; F ), and hence [A ⊗F K] ∈ Dec(M/K; F ). Thus, resF →K induces a well-defined map f : Br2(M/F )/ Dec(M/F ) → Br2(M/K; F )/ Dec(M/K; F ). From Arason's long exact sequence (see, e.g., [KMRT, Cor. 30.12(1)] or (5.2) above) . . . → H 2(F, µ2) → H 2(K, µ2) → H 2(F, µ2) → . . . , 2 ∈ Br(K(√b )/K; F ) and Q′ 2 ⊗K Q′ 2 with K/F -involution has the form Q′ f is surjective. For injectivity of f , take any [A] ∈ Br2(M/F ) with resF →K[A] ∈ Dec(M/K; F ). We need to show [A] ∈ Dec(M/F ). We have A ⊗F K ∼ Q′ i are quaternion algebras over K 3 ∈ Br(K(√c )/K; F ). By a result of Albert [KMRT, Prop. 2.22], the with Q′ quaternion algebra Q′ 2 is a quaternion algebra over F . Then, [Q′′ 3 ⊗F K, where [Q′′ 2] ∈ Br2(K(√b )/F ) = Dec(K(√b )/F ), as noted for (a) above. Likewise, Q′ 2 ⊗F Q′′ 3] ∈ Dec(K/F ) · Dec(K(√b )/F ) · Dec(K(√c )/F ) ⊆ Dec(M/F ). 3] ∈ Dec(K(√c )/F ). Since [A ⊗F Q′′ 2 ∼= Q′′ 2 ⊗F K, where Q′′ 2 ⊗F Q′′ 3] ∈ Br(K/F ) = Dec(K/F ), we have 3 where the Q′ 3 ∼= Q′′ [A] = [A ⊗F Q′′ 3] [Q′′ 2] [Q′′ Thus, f is an isomorphism, proving (b). (cid:3) Remark 8.2. The term Br2(M/F )/ Dec(M/F ) for M/F triquadratic has arisen in the study of indecompos- able algebras A of degree 8 and exponent 2. Note first that for any A of degree 8 and exponent 2, by Rowen's theorem [R, Th. 6.2] there is a triquadratic field extension M of the center F of A, such that M is a max- imal subfield of A. If A is indecomposable, then [A] yields a nontrivial element of Br2(M/F )/ Dec(M/F ). Examples of indecomposables if degree 8 and exponent 2 were first given in [ART, Th. 5.1]. Subsequently, Karpenko showed in [Kar, Cor. 5.4] that if B is a division algebra with center F of degree 8 and exponent 8, and F ′ is a field generically reducing the exponent of B to 2, then B ⊗F F ′ is an indecomposable division algebra of degree 8 and exponent 2. Also, K. McKinnie in her thesis (unpublished), using lattice methods, gave another example of indecomposables of degree 8 and exponent 2. There is a kind of converse to this as well: Given a division algebra A with [A] ∈ Br2(M/F ) \ Dec(M/F ), Amitsur, Rowen, and Tignol showed in [ART, Th. 3.3] that the associated generic abelian crossed product algebra A′ of A is indecomposable of degree 8 and exponent 2. (It is not stated this way in [ART], but made explicit in [T2, § 2].) This A′ is the ring of quotients of a semiramified graded division algebra E of the type considered in previous sections: E is graded Brauer equivalent to I ⊗T N, where T is a graded field with T0 ∼= F , I is an inertial graded division algebra over T with I0 ∼= A, and N is DSR over T with N0 ∼= M . Using Prop. 8.1 we now construct biquaterion graded algebras where the map α of (8.1) above is not injective. Example 8.3. Let M be a triquadratic extension of a field F (char(F ) 6= 2) with Br2(M/F )/ Dec(M/F ) 6= 0. (Such F and M exist, as noted in Remark 8.2.) Say M = F (√a,√b,√c ) for a, b, c ∈ F ∗. Let K = F (√a ), and let H = Gal(M/K). Let R = F [x, x−1, y, y−1], the Laurent polynomnial ring in indeterminates x and y, with its usual grading in which R(k,ℓ) = F xkyℓ for all (k, ℓ) ∈ Z×Z. So, R is a graded field with R0 = F and ΓR = Z× Z. Let T = K[x, x−1, y, y−1], a graded field with [T : R] = 2, and let E = Q⊗T Q′, where Q and Q′ are the following semiramified graded quaternion division algebras over T: Q =(cid:0) b,x T(cid:1), which is generated over T by homogeneous elements i andj with relations i2 = b, j2 = x, and ij = −ji, with deg(i) = 0 and 2 , 0). So, Q0 ∼= K(√b ) and ΓQ = 1 T(cid:1) with standard generators 2Z × Z. Likewise, set Q′ =(cid:0) c,y deg(j) = ( 1 2Z. Since Q ∼=(cid:0) b,x R(cid:1) ⊗R T, 2 ), so Q0 ∼= K(√c ) and ΓQ′ = Z × 1 i′ and j′, with deg(i′) = 0 and deg(j′) = (0, 1 Q has the graded T/R-involution τQ = η⊗ψ, where η is the canonical symplectic graded involution on(cid:0) b,x R(cid:1), for which η(i) = −i and η(j) = −j, and ψ is the nonidentity graded R-automorphism of T. Likewise Q′ has a graded T/R-involution τQ′ with τQ′(i′) = −i′ and τQ′(j′) = −j′. By Lemma 4.3, E is a graded division 32 A. R. WADSWORTH algebra which is DSR for T/R with E0 ∼= Q0⊗T0 Q′ 2 Z× 1 2 Z; our graded T/R-involution on E is τ = τQ ⊗ τQ′. (Explicitly, S = T[i, i′] ∼=g M [x, x−1, y, y−1] is a maximal graded subfield of E with S inertial over T, and J = T[j, j′] ∼=g T[√x,√x −1,√y,√y −1] is a maximal graded subfield of E which is totally ramified over T with τ (J) = J.) We claim that the following diagram is commutative with all horizontal maps isomorphisms and vertical maps described below: 0 ∼= K(√b )⊗K K(√c ) ∼= M and ΓE = ΓQ+ΓQ′ = 1 (8.3) Br(M/K; F )(cid:14) Dec(M/K; F ) −−−−→ ker(eN )/Π −−−−→ SK1(E, τ ) αy Br(M/K)(cid:14) Dec(M/K) −−−−→ bH −1(H, M ∗) −−−−→ SK1(E) y y The left vertical map is the map in Prop. 8.1, whose kernel is there shown to be isomorphic to Br2(M/F )/ Dec(M/F ). Since we have assumed this kernel is nontrivial, once the claim is established the right vertical map α, which is the map of (8.1) must also have nontrivial kernel, as desired. We now verify the claim. In the top line of (8.3), ker(eN ) = {a ∈ M ∗ NM/K (a) ∈ F} and Π =Qh∈H M ∗hτ , where H = Gal(M/K) and τ = τE0. The middle vertical map sends a Π 7→ a/τ (a) IH (M ∗). It is well defined since if a ∈ ker(eN ), we have NK/F (a/τ (a)) = NK/F (a)/τ (NK/F (a)) = 1, and if b ∈ M ∗hτ , then b/τ (b) = hτ (b)/τ (b) ∈ IH(M ∗). In the right rectangle of (8.3), the top map sends a Π 7→ aΣτ (E), and the bottom map sends b IH(M ∗) 7→ b [E∗, E∗], so the right rectangle is clearly commutative. The horizontal maps in this rectangle are the isomorphisms given in Th. 7.1(i) and Prop. 3.2(i). For the left vertical map take an arbitrary element of Br(M/K; F ), which has the form [A], where A = A(u, b1, b2) in the notation of §6, with u, b1, b2 satisfying the relations in (3.1) and (3.2) and the added relations in Lemma 4.1(iii), notably u σρτ (u) = 1. The horizontal map in the left rectangle is the isomorphism of Th. 6.1 which sends [A] mod Dec(M/K; F ) to q Π for any q ∈ M ∗ with q/σρτ (q) = u. This is mapped downward to u IH(M ∗), since q/τ (q) = u σρτ (q)/τ (q) ≡ u (mod IH(M ∗)). On the other hand, [A] mod Dec(M/K; F ) is mapped downward to [A] mod Dec(M/K), which is mapped to the right to u IH(M ∗) by the isomorphism of (3.9). Thus, the left rectangle of (8.3) is commutative, and its horizontal maps are isomorphisms, completing the proof of the claim. Remark 8.4. For the preceding example with the α of (8.1) noninjective, we have worked with graded division algebras. There are corresponding examples of division algebras over a Henselian valued field with the corresponding α not injective, obtainable as follows: With fields F ⊆ K ⊆ M as in Ex. 8.3, let F ′ = F ((x))((y)), K ′ = K((x))((y)), and M ′ = M ((x))((y)), which are twice iterated Laurent power series fields each with it standard Henselian valuation with value group Z × Z (with right-to-left lexicographic algebra over K ′, and the Henselian valuation vK ′ on K ′ extends uniquely to a valuation vD on D, for which D ∼= M and ΓD = 1 2Z. For the associated graded ring of D determined by vD, we have gr(D) ∼=g E and, as D is tame over K ′, Z(gr(D)) = gr(K ′) ∼=g T, for the E and T of Ex. 8.3. Also, gr(F ′) ∼=g R for the R of Ex. 8.3. This D has a unitary K ′/F ′-involution τD, since each constituent quaternion algebra has such an involution. Because the Henselian valuation vF ′ on F ′ has a unique extension to K ′, namely vK ′, and vD is ordering) and residue fields F ′ ∼= F , K ′ ∼= K, and M ′ ∼= M . Let D =(cid:0) b,x 2Z × 1 K ′(cid:1) ⊗K ′(cid:0) c,y K ′(cid:1), which is a division the unique extension of vK ′ to D, we must have vD ◦ τD = vD. Therefore, τD induces a graded involutioneτ on E, which is a unitary T/R-involution. By [HW2, Th. 3.5] and [HW1, Th. 4.8], SK1(D, τD) ∼= SK1(E,eτ ) and SK1(D) ∼= SK1(E). These isomorphisms are compatible with the map αeτ : SK1(E,eτ ) → SK1(E) and the corresponding map αD : SK1(D, τD) → SK1(D). Also, becauseeτ and the τ of Ex. 8.3 are each graded T/R-involutions on E, we have SK1(E,eτ ) ∼= SK1(E, τ ), and it is easy to check that under this isomorphism αeτ corresponds to the α of Ex. 8.3. Since this α is not injective, αD is also noninjective. UNITARY SK1 OF SEMIRAMIFIED DIVISION ALGEBRAS 33 References [ART] [AS] [AE] [CM] [D] [E1] [E2] [G] S. A. Amitsur, L. H. Rowen, and J.-P. Tignol, Division algebras of degree 4 and 8 with involution, Israel J. Math., 33 (1979), 133 -- 148. 31 S. A. Amitsur and D. J. Saltman, Generic abelian crossed products and p-algebras, J. Algebra, 51 (1978), 76 -- 87. 8, 9, 16 J. Kr. Arason and R. Elman, Nilpotence in the Witt ring, Amer. J. Math., 113 (1991), 861 -- 875. 20 V. Chernousov and A. Merkurjev, R-equivalence and special unitary groups, J. Algebra, 209 (1998), 175 -- 198. 1 P. Draxl, SK1 von Algebren uber vollstandig diskret bewerteten Korpern und Galoiskohomologie abelscher Korpererweiterungen, J. Reine Angew. Math., 293/294 (1977), 116 -- 142. 10 Yu. L. Ershov, Valuations of division algebras, and the group SK1, Dokl. Akad. Nauk SSSR, 239 (1978), 768 -- 771 (in Russian); English transl., Soviet Math. Doklady, 19 (1978), 395 -- 399. 2 Yu. L. Ershov, Henselian valuations of division rings and the group SK1, Mat. Sb. (N.S.), 117 (1982), 60 -- 68 (in Russian); English transl., Math USSR-Sbornik, 45 (1983), 63 -- 71. 1, 2, 14 P. Gille, Le probl`eme de Kneser-Tits, S´eminaire Bourbaki, Exp. No. 983, Vol. 2007/2008, Ast´erisque, 326 (2010), 39 -- 81. 1 [HKRT] D. E. Haile, M.-A. Knus, M. Rost, and J.-P. Tignol, Algebras of odd degree with involution, trace forms and dihedral extensions, Israel J. Math., 96 (1996), part B, 299 -- 340. 20 [HW1] R. Hazrat, A. R. Wadsworth, SK1 of graded division algebras, Israel J. Math., to appear, preprint available at: http://www.math.uni-bielefeld.de/LAG/, No. 318. 1, 2, 6, 10, 12, 13, 32 [HW2] R. Hazrat and A. R. Wadsworth, Unitary SK1 of graded and valued division algebras, preprint, available at arXiv: 0911.3628. 1, 2, 3, 12, 14, 16, 18, 22, 23, 25, 26, 30, 32 [HwW1] Y.-S. Hwang, A. R. Wadsworth, Algebraic extensions of graded and valued fields, Comm. Algebra, 27 (1999), 821 -- 840. 5 [HwW2] Y.-S. Hwang, A. R. Wadsworth, Correspondences between valued division algebras and graded division algebras, J. Algebra, 220 (1999), 73 -- 114. 2, 4, 5, 6, 7, 12, 13, 14, 17 [JW] [K] B. Jacob, A. Wadsworth, Division algebras over Henselian fields, J. Algebra, 128 (1990), 126 -- 179. 3, 10, 11, 13, 19 B. Kahn, Cohomological approaches to SK1 and SK2 of central simple algebras, Documenta Math., Extra Volume, Andrei A. Suslin's Sixtieth Birthday, (2010), 371 -- 392 (electronic). 3 [Kar] N. A. Karpenko, Codimension 2 cycles on Severi-Brauer varieties, K-Theory, 13 (1998), 305 -- 330. 31 [KMRT] M. -A. Knus, A. Merkurjev, M. Rost, J.-P. Tignol, The Book of Involutions, AMS Coll. Pub., 44, Amer. Math. Soc., Providence, RI, 1998. 3, 14, 15, 16, 20, 30, 31 A. S. Merkurjev, Generic element in SK1 for simple algebras, K-Theory, 7 (1993), 1 -- 3. 1 A. S. Merkurjev, Invariants of algebraic groups, J. Reine Angew. Math., 508 (1999), 127 -- 156. 3 A. S. Merkurjev, Cohomological invariants of simply connected groups of rank 3, J. Algebra, 227 (2000), 614 -- 632. 3 A. S. Merkurjev, The group SK1 for simple algebras, K-Theory, 37 (2006), 311 -- 319. 1 [M1] [M2] [M3] [M4] [Mou] K. Mounirh, Nondegenerate semiramified valued and graded division algebras, Comm. Algebra, 36 (2008), 4386 -- 4406. 2 [P1] [P2] [P3] [P4] [P5] [P6] V. P. Platonov, On the Tannaka-Artin problem, Dokl. Akad. Nauk SSSR, 221 (1975), 1038 -- 1041 (in Russian); English transl., Soviet Math. Dokl., 16 (1975), 468 -- 473. 1 V. P. Platonov, The Tannaka-Artin problem and reduced K-theory, Izv. Akad. Nauk SSSR Ser. Mat., 40 (1976), 227 -- 261 (in Russian); English transl., Math. USSR-Izvestiya, 10 (1976), 211 -- 243. 1, 10 V. P. Platonov, The reduced Whitehead group for cyclic algebras, Dokl. Akad. Nauk SSSR, 228 (1976), 38 -- 40 (in Russian); English transl., Soviet. Math. Doklady, 17 (1976), 652 -- 655. 14 V. P. Platonov, The Infinitude of the reduced Whitehead group in the Tannaka-Artin Problem, Mat. Sb., 100 (142) (1976), 191 -- 200, 335 (in Russian); English transl., Math. USSR Sbornik, 29 (1976), 167 -- 176. 2, 14 V. P. Platonov, Birational properties of the reduced Whitehead group, Dokl. Akad. Nauk BSSR, 21 (1977), 197 -- 198, 283 (in Russian); English transl., pp. 7 -- 9 in Selected papers in K-theory, Amer. Math. Soc. Translations, Ser. 2, Vol. 154, Amer. Math. Soc., Providence, RI, 1992. 1 V. P. Platonov, Algebraic groups and reduced K-theory, pp. 311 -- 317 in Proceedings of the International Congress of Mathematicians (Helsinki 1978), ed. O. Lehto, Acad. Sci. Fennica, Helsinki, 1980. 1 34 A. R. WADSWORTH [RTY] U. Rehmann, S. V. Tikhonov, and V. I. Yanchevskiı, Symbols and cyclicity of algebras after a scalar extension, [R] [Se] [Su1] [Su2] [T1] [T2] [TW] [Ti] [V1] [V2] [W] [Y1] [Y2] [Y3] [Y4] [Y5] preprint, available at http://www.math.uni-bielefeld.de/LAG/, No. 315. 3 L. H. Rowen, Central simple algebras, Israel J. Math., 29 (1978), 285 -- 301. 31 J.-P. Serre, Local Fields, Second Ed., Springer-Verlag, New York, 1995; (English trans. of Corps Locaux). 9 A. A. Suslin, SK1 of division algebras and Galois cohomology, pp. 75 -- 99 in Algebraic K-theory, Adv. Soviet Math., Vol. 4, Amer. Math. Soc., Providence, RI, 1991. 1 A. A. Suslin, SK1 of division algebras and Galois cohomology revisited, pp. 125 -- 147 in Proc. St. Petersburg Math. Soc., Vol. XII, English transl., Amer. Math. Soc. Transl. Ser. 2, Vol. 219, Amer. Math. Soc., Providence, RI, 2006. 1, 3 J.-P. Tignol, Corps `a involution neutralis´es par une extension abelienne el´ementaire, pp. 1 -- 34 in, Groupe de Brauer, eds. M.Kervaire and M. Ojanguren, Lecture Notes in Math., No. 844, Springer-Verlag, Berlin, 1981. 30 J.-P. Tignol, Produits crois´es ab´eliens, J. Algebra, 70 (1981), 420 -- 436. 8, 9, 10, 31 J.-P. Tignol and A. R. Wadsworth, Value functions and associated graded rings for semisimple algebras, Trans. Amer. Math. Soc., 362 (2010), 687 -- 726. 2 J. Tits, Groupes de Whitehead de groupes alg´ebriques simples sur un corps (d'apr`es V. P. Platonov et al.), pp. 218 -- 236 in S´eminaire Bourbaki, 29e ann´ee (1976/77), Exp. No. 505, Lecture Notes in Math., Vol. 677, Springer, Berlin, 1978. 1 V. E. Voskresenskiı, The reduced Whitehead group of a simple algebra, Uspehi Mat. Nauk, 32 (1977), 247 -- 248 (in Russian). 1 V. E. Voskresenskiı, Algebraic groups and their birational invariants, Translations of Math. Monographs, Vol. 179, Amer. Math. Soc., Providence, RI, 1998. 1 T. Wouters, Comparing invariants of SK1, preprint, arXiv: 1003.1654v2. 3 V. I. Yanchevskiı, Simple algebras with involutions, and unitary groups, Mat. Sb. (N.S.), 93 (135) (1974), 368 -- 380, 487 (in Russian); English transl., Math. USSR-Sbornik, 22 (1974), 372 -- 385. 25 V. I. Yanchevskiı, Reduced unitary K-theory, Dokl. Akad. Nauk SSSR, 229 (1976), 1332 -- 1334 (in Russian); English transl., Soviet Math. Dokl., 17 (1976), 1220 -- 1223. 2 V. I. Yanchevskiı, Reduced unitary K-Theory and division rings over discretely valued Hensel fields, Izv. Akad. Nauk SSSR Ser. Mat., 42 (1978), 879 -- 918 (in Russian); English transl., Math. USSR Izvestiya, 13 (1979), 175 -- 213. 2, 21, 22, 23, 30 V. I. Yanchevskiı, The inverse problem of reduced K-theory, Mat. Zametki, 26 (1979), 475 -- 482 (in Russian); English transl., A converse problem in reduced unitary K-theory, Math. Notes, 26 (1979), 728 -- 731. 2, 22 V. I. Yanchevskiı, Reduced unitary K-theory. Applications to algebraic groups. Mat. Sb. (N.S.), 110 (152) (1979), 579 -- 596 (in Russian); English transl., Math. USSR Sbornik, 38 (1981), 533 -- 548. 1, 14 Department of Mathematics, University of California at San Diego, La Jolla, California 92093-0112, U.S.A. E-mail address: [email protected]
1610.02418	1	1610	2016-10-07T20:34:12	On the Structure Theorem of Clifford Algebras	[ "math.RA" ]	In this paper, theory and construction of spinor representations of real Clifford algebras $\cl_{p,q}$ in minimal left ideals are reviewed. Connection with a general theory of semisimple rings is shown. The actual computations can be found in, for example, [2].	math.RA	math	ON THE STRUCTURE THEOREM OF CLIFFORD ALGEBRAS Rafa l Ab lamowicz a a Department of Mathematics, Tennessee Technological University [email protected], http://math.tntech.edu/rafal/ Cookeville, TN 38505, U.S.A. Abstract. In this paper, theory and construction of spinor representations of real Clifford algebras Cℓp,q in minimal left ideals are reviewed. Connection with a general theory of semisimple rings is shown. The actual computations can be found in, for example, [2]. Keywords. Artinian ring, Clifford algebra, division ring, group algebra, idempotent, minimal left ideal, semisimple module, Radon-Hurwitz number, semisimple ring, Wedderburn-Artin Theorem Mathematics Subject Classification (2010). Primary: 11E88, 15A66, 16G10; Secondary: 16S35, 20B05, 20C05, 68W30 Contents 1. 2. Introduction Introduction to Semisimple Rings and Modules 3. The Main Structure Theorem on Real Clifford Algebras Cℓp,q 4. Conclusions 5. Acknowledgments References 1 2 7 9 9 9 1. Introduction Theory of spinor representations of real Clifford algebras Cℓp,q over a quadratic space (V, Q) with a nondegenerate quadratic form Q of signature (p, q) is well known [11,18,19,24]. The purpose of this paper is to review the structure theorem of these algebras in the context of a general theory of semisimple rings culminating with Wedderburn-Artin Theorem [26]. Section 2 is devoted to a short review of general background material on the theory of semisimple rings and modules as a generalization of the representation theory of group algebras of finite groups [17, 26]. While it is well-known that Clifford algebras Cℓp,q are associative finite-dimensional unital semisimple R-algebras, hence the representation theory of semisimple rings [26, Chapter 7] applies to them, it is also possible to view these algebras as twisted group algebras Rt[(Z2)n] of a finite group (Z2)n [5–7, 9, 13, 23]. While this last approach is not pursued here, for a connection between Clifford algebras Cℓp,q and finite groups, see [1, 6, 7, 10, 20, 21, 27] and references therein. 2 ON THE STRUCTURE THEOREM OF CLIFFORD ALGEBRAS In Section 3, we state the main Structure Theorem on Clifford algebras Cℓp,q and relate it to the general theory of semisimple rings, especially to the Wedderburn-Artin theorem. For details of computation of spinor representations, we refer to [2] where these computations were done in great detail by hand and by using CLIFFORD, a Maple package specifically designed for computing and storing spinor representations of Clifford algebras Cℓp,q for n = p + q ≤ 9 [3, 4]. Our standard references on the theory of modules, semisimple rings and their represen- tation is [26]; for Clifford algebras we use [11,18,19] and references therein; on representation theory of finite groups we refer to [17, 25] and for the group theory we refer to [12, 14, 22, 26]. 2. Introduction to Semisimple Rings and Modules This brief introduction to the theory of semisimple rings is based on [26, Chapter 7] and it is stated in the language of left R-modules. Here, R denotes an associative ring with unity 1. We omit proofs as they can be found in Rotman [26]. Definition 1. Let R be a ring. A left R-module is an additive abelian group M equipped with scalar multiplication R × M → M, denoted (r, m) 7→ rm, such that the following axioms hold for all m, m′ ∈ M and all r, r′ ∈ R : (i) r(m + m′) = rm + rm′, (ii) (r + r′)m = rm + r′m, (iii) (rr′)m = r(r′m), (iv) 1m = m. Left R-modules are often denoted by RM. In a similar manner one can define a right R-module with the action by the ring elements on M from the right. When R and S are rings and M is an abelian group, then M is a (R, S)-bimodule, denoted by RMS, if M is a left R-module, a right S-module, and the two scalar multiplications are related by an associative law: r(ms) = (rm)s for all r ∈ R, m ∈ M, and s ∈ S. We recall that a spinor left ideal S in a simple Clifford algebra Cℓp,q by definition carries an irreducible and faithful representation of the algebra, and it is defined as Cℓp,qf where f is a primitive idempotent in Cℓp,q. Thus, as it is known from the Structure Theorem (see Section 3), that these ideals are (R, S)-bimodules where R = Cℓp,q and S = f Cℓp,qf . Similarly, the right spinor modules f Cℓp,q are (S, R)-bimodules. Notice that the associative law mentioned above is automatically satisfied because Cℓp,q is associative. We just recall that when k is a field, every finite-dimensional k-algebra A is both left and right noetherian, that is, any ascending chain of left and right ideals stops (the ACC ascending chain condition). This is important for Clifford algebras because, eventually, we will see that every Clifford algebra can be decomposed into a finite direct sum of left spinor Cℓp,q-modules (ideals). For completeness we mention that every finite-dimensional k-algebra A is both left and right artinian, that is, any descending chain of left and right ideals stops (the DCC ascending chain condition). ON THE STRUCTURE THEOREM OF CLIFFORD ALGEBRAS 3 Thus, every Clifford algebra Cℓp,q, as well as every group algebra kG, when G is a finite group, which then makes kG finite dimensional, have both chain conditions by a dimensionality argument. Definition 2. A left ideal L in a ring R is a minimal left ideal if L 6= (0) and there is no left ideal J with (0) ( J ( L. One standard example of minimal left ideals in matrix algebras R = Mat(n, k) are the subspaces COL(j), 1 ≤ j ≤ n, of Mat(n, k) consisting of matrices [ai,j] such that ai,k = 0 when k 6= j (cf. [26, Example 7.9]). The following proposition relates minimal left ideals in a ring R to simple left R- modules. Recall that a left R-module M is simple (or irreducible) if M 6= {0} and M has no proper nonzero submodules. Proposition 1 (Rotman [26]). (i) Every minimal left ideal L in a ring R is a simple left R-module. (ii) If R is left artinian, then every nonzero left ideal I contains a minimal left ideal. Thus, the above proposition applies to Clifford algebras Cℓp,q: every left spinor ideal S in Cℓp,q is a simple left Cℓp,q-module; and, every left ideal in Cℓp,q contains a spinor ideal. Recall that if D is a division ring, then a left (or right) D-module V is called a left (or right) vector space over D. In particular, when the division ring is a field k, then we have a familiar concept of a k-vector space. Since the concept of linear independence of vectors generalizes from k-vector spaces to D-vector spaces, we have the following result. Proposition 2 (Rotman [26]). Let V be a finitely generated1 left vector space over a division ring D. (i) V is a direct sum of copies of D; that is, every finitely generated left vector space over D has a basis. (ii) Any two bases of V have the same number of elements. Since we know from the Structure Theorem, that every spinor left ideal S in simple Clifford algebras Cℓp,q (p − q 6= 1 mod 4) is a right K-module where K is one of the division rings R, C, or H, the above proposition simply tells us that every spinor left ideal S is finite-dimensional over K where K is one of R, C or H. In semisimple Clifford algebras Cℓp,q (p − q = 1 mod 4), we have to be careful as the faithful double spinor representations are realized in the direct sum of two spinor ideals S ⊕ S which are right K ⊕ K-modules, where K = R or H.2 Yet, it is easy to show that K ⊕ K is not a division ring. Thus, Proposition 2 tells us that every finitely generated left (or right) vector space V over a division ring D has a left (a right) dimension, which may be denoted dim V. In [16] Jacobson gives an example of a division ring D and an abelian group V , which is both a right 1The term "finitely generated" means that every vector in V is a linear combination of a finite number of certain vectors {x1, . . . , xn} with coefficients from R. In particular, a k-vector space is finitely generated if and only if it is finite-dimensional [26, Page 405]. 2Here, S = { ψ ∈ S}, and similarly for K, where denotes the grade involution in Cℓp,q. 4 ON THE STRUCTURE THEOREM OF CLIFFORD ALGEBRAS and a left D-vector space, such that the left and the right dimensions are not equal. In our discussion, spinor minimal ideal S will always be a left Cℓp,q-module and a right K-module. Since semisimple rings generalize the concept of a group algebra CG for a finite group G (cf. [17, 26]), we first discuss semisimple modules over a ring R. Definition 3. A left R-module is semisimple if it is a direct sum of (possibly infinitely many) simple modules. The following result is an important characterization of semisimple modules. Proposition 3 (Rotman [26]). A left R-module M over a ring R is semisimple if and only if every submodule of M is a direct summand. Recall that if a ring R is viewed as a left R-module, then its submodules are its left ideals, and, a left ideal is minimal if and only if it is a simple left R-module [26]. Definition 4. A ring R is left semisimple3 if it is a direct sum of minimal left ideals. One of the important consequences of the above for the theory of Clifford algebras, is the following proposition. Proposition 4 (Rotman [26]). Let R be a left semisimple ring. (i) R is a direct sum of finitely many minimal left ideals. (ii) R has both chain conditions on left ideals. From a proof of the above proposition one learns that, while R = Li Li, that is, R is a direct sum of finitely-many left minimal ideals, the unity 1 decomposes into a sum 1 = Pi fi of mutually annihilating primitive idempotents fi, that is, (fi)2 = fi, and fifj = fjfi = 0, i 6= j. Furthermore, we find that Li = Rfi for every i. We can conclude from the following fundamental result [15, 26] that every Clifford algebra Cℓp,q is a semisimple ring, because every Clifford algebra is a twisted group algebra Rt[(Z2)n] for n = p + q and a suitable twist [1, 7, 9]. Theorem 1 (Maschke's Theorem). If G is a finite group and k is a field whose characteristic does not divide G, the kG is a left semisimple ring. For characterizations of left semisimple rings, we refer to [26, Section 7.3]. Before stating Wedderburn-Artin Theorem, which is all-important to the theory of Clifford algebras, we conclude this part with a definition and two propositions. Definition 5. A ring R is simple if it is not the zero ring and it has no proper nonzero two-sided ideals. Proposition 5 (Rotman [26]). If D is a division ring, then R = Mat(n, D) is a simple ring. Proposition 6 (Rotman [26]). If R = Lj Lj is a left semisimple ring, where the Lj are minimal left ideals, then every simple R-module S is isomorphic to Lj for some j. 3One can define a right semisimple ring R if it is a direct sum of minimal right ideals. However, it is known [26, Corollary 7.45] that a ring is left semisimple if and only if it is right semisimple. ON THE STRUCTURE THEOREM OF CLIFFORD ALGEBRAS 5 The main consequence of this last result is that every simple, hence irreducible, left Cℓp,q- module, that is, every (left) spinor module of Cℓp,q, is isomorphic to some minimal left ideal Lj in the direct sum decomposition of R = Cℓp,q. Following Rotman, we divide the Wedderburn-Artin Theorem into the existence part and a uniqueness part. We also remark after Rotman that Wedderburn proved the existence theorem 2 for semisimple k-algebras, where k is a field, while E. Artin generalized this result to what is now known as the Wedderburn-Artin Theorem. Theorem 2 (Wedderburn-Artin I). A ring R is left semisimple if and only if R is isomorphic to a direct product of matrix rings over division rings D1, . . . , Dm, that is (1) R ∼= Mat(n1, D1) × · · · × Mat(nm, Dm). A proof of the above theorem yields that if R = Lj Lj as in Proposition 6, then each division ring Dj = EndR(Lj), j = 1, . . . , m, where EndR(Lj) denotes the ring of all R-endomorphisms of Lj. Another consequence is the following corollary. Corollary 1. A ring R is left semisimple if and only if it is right semisimple. Thus, we may refer to a ring as being semisimple without specifying from which side.4 However, we have the following result which we know applies to Clifford algebras Cℓp,q. More importantly, its corollary explains part of the Structure Theorem which applies to simple Clifford algebras. Recall from the above that every Clifford algebra Cℓp,q is left artinian (because it is finite-dimensional). Proposition 7 (Rotman [26]). A simple left artinian ring R is semisimple. Corollary 2. If A is a simple left artinian ring, then A ∼= Mat(n, D) for some n ≥ 1 and some division ring D. Before we conclude this section with the second part of the Wedderburn-Artin Theo- rem, which gives certain uniqueness of the decomposition (1), we state the following definition and a lemma. Definition 6. Let R be a left semisimple ring, and let (2) R = L1 ⊕ · · · ⊕ Ln, where the Lj are minimal left ideals. Let the ideals L1, . . . , Lm, possibly after re-indexing, be such that no two among them are isomorphic, and so that every Lj in the given decomposition of R is isomorphic to one and only one Li for 1 ≤ i ≤ m. The left ideals (3) Bi = M ∼=Li Lj Lj are called the simple components of R relative to the decomposition R = Lj Lj. Lemma 1 (Rotman [26]). Let R be a semisimple ring, and let (4) R = L1 ⊕ · · · ⊕ Ln = B1 ⊕ · · · ⊕ Bm where the Lj are minimal left ideals and the Bi are the corresponding simple components of R. 4Not every simple ring is semisimple, cf. [26, Page 554] and reference therein. 6 ON THE STRUCTURE THEOREM OF CLIFFORD ALGEBRAS (i) Each Bi is a ring that is also a two-sided ideal in R, and BiBj = (0) if i 6= j. (ii) If L is any minimal left ideal in R, not necessarily occurring in the given decomposi- tion of R, then L ∼= Li for some i and L ⊆ Bi. (iii) Every two-sided ideal in R is a direct sum of simple components. (iv) Each Bi is a simple ring. Thus, we will gather from the Structure Theorem, that for simple Clifford algebras Cℓp,q we have only one simple component, hence m = 1, and thus all 2k left minimal ideals generated by a complete set of 2k primitive mutually annihilating idempotents which provide an orthogonal decomposition of the unity 1 in Cℓp,q (see part (c) of the theorem and notation therein). Then, for semisimple Clifford algebras Cℓp,q we have obviously m = 2. Furthermore, we have the following corollary results. Corollary 3 (Rotman [26]). (1) The simple components B1, . . . , Bm of a semisimple ring R do not depend on a de- composition of R as a direct sum of minimal left ideals; (2) Let A be a simple artinian ring. Then, (i) A ∼= Mat(n, D) for some division ring D. If L is a minimal left ideal in A, then every simple left A-module is isomorphic to L; moreover, Dop ∼= EndA(L).5 (ii) Two finitely generated left A-modules M and N are isomorphic if and only if dimD(M) = dimD(N). As we can see, part (1) of this last corollary gives a certain invariance in the decomposition of a semisimple ring into a direct sum of simple components. Part (2i), for the left artinian Clifford algebras Cℓp,q implies that simple Clifford algebras (p − q 6= 1 mod 4) are simple algebras isomorphic to a matrix algebra over a suitable division ring D. From the Structure Theorem we know that D is one of R, C, or H, depending on the value of p − q mod 8. Part (2ii) tells us that any two spinor ideals S and S ′, which are simple right K-modules (due the right action of the division ring K = f Cℓp,qf on each of them) are isomorphic since their dimensions over K are the same. We conclude this introduction to the theory of semisimple rings with the following uniqueness theorem. Theorem 3 (Wedderburn-Artin II). Every semisimple ring R is a direct product, (5) R ∼= Mat(n1, D1) × · · · × Mat(nm, Dm), where ni ≥ 1, and Di is a division ring, and the numbers m and ni, as well as the division rings Di, are uniquely determined by R. Thus, the above results, and especially the Wedderburn-Artin Theorem (parts I and In II), shed a new light on the main Structure Theorem given in the following section. particular, we see it as a special case of the theory of semisimple rings, including the left artinian rings, applied to the finite dimensional Clifford algebras Cℓp,q. We remark that the above theory applies to the group algebras kG where k is an algebraically closed field and G is a finite group. 5By Dop we mean the opposite ring of D: It is defined as Dop = {aop a ∈ D} with multiplication defined as aop · bop = (ba)op. ON THE STRUCTURE THEOREM OF CLIFFORD ALGEBRAS 7 3. The Main Structure Theorem on Real Clifford Algebras Cℓp,q We have the following main theorem that describes the structure of Clifford algebras Cℓp,q and their spinorial representations. In the following, we will analyze statements in that theorem. The same information is encoded in the well–known Table 1 in [19, Page 217]. Structure Theorem. Let Cℓp,q be the universal real Clifford algebra over (V, Q), Q is non- degenerate of signature (p, q). (a) When p − q 6= 1 mod 4 then Cℓp,q is a simple algebra of dimension 2p+q isomorphic with a full matrix algebra Mat(2k, K) over a division ring K where k = q − rq−p and ri is the Radon-Hurwitz number.6 Here K is one of R, C or H when (p − q) mod 8 is 0, 2, or 3, 7, or 4, 6. (6) (7) (8) (b) When p−q = 1 mod 4 then Cℓp,q is a semisimple algebra of dimension 2p+q isomorphic to Mat(2k−1, K) ⊕ Mat(2k−1, K), k = q − rq−p, and K is isomorphic to R or H depending whether (p − q) mod 8 is 1 or 5. Each of the two simple direct components of Cℓp,q is projected out by one of the two central idempotents 1 2 (1 ± e12...n). (c) Any element f in Cℓp,q expressible as a product f = 1 2 (1 ± ei1) 1 2 (1 ± ei2) · · · 1 2 (1 ± eik) , j = 1, . . . , k, are commuting basis monomials in B with square 1 and where eij k = q − rq−p generating a group of order 2k, is a primitive idempotent in Cℓp,q. Furthermore, Cℓp,q has a complete set of 2k such primitive mutually annihilating idempotents which add up to the unity 1 of Cℓp,q. (d) When (p − q) mod 8 is 0, 1, 2, or 3, 7, or 4, 5, 6, then the division ring K = f Cℓp,qf is isomorphic to R or C or H, and the map S × K → S, (ψ, λ) 7→ ψλ defines a right K-module structure on the minimal left ideal S = Cℓp,qf. (e) When Cℓp,q is simple, then the map Cℓp,q γ −→ EndK(S), u 7→ γ(u), γ(u)ψ = uψ gives an irreducible and faithful representation of Cℓp,q in S. (f) When Cℓp,q is semisimple, then the map Cℓp,q γ −→ EndK⊕ K(S ⊕ S), u 7→ γ(u), γ(u)ψ = uψ gives a faithful but reducible representation of Cℓp,q in the double spinor space S ⊕ S where S = {uf u ∈ Cℓp,q}, S = {u f u ∈ Cℓp,q} and stands for the grade involution in Cℓp,q. In this case, the ideal S ⊕ S is a right K ⊕ K-module structure, K = {λ λ ∈ K}, and K ⊕ K is isomorphic to R ⊕ R when p − q = 1 mod 8 or to H ⊕ H when p − q = 5 mod 8. Parts (a) and (b) address simple and semisimple Clifford algebras Cℓp,q which are distinguished by the value of p − q mod 4 while the dimension of Cℓp,q is 2p+q. For simple algebras, the Radon-Hurwitz number ri defined recursively as shown, determines the value of the exponent k = q − rq−p such that (9) Cℓp,q ∼= Mat(2k, K) when p − q 6= 1 mod 4. 6The Radon-Hurwitz number is defined by recursion as ri+8 = ri + 4 and these initial values: r0 = 0, r1 = 1, r2 = r3 = 2, r4 = r5 = r6 = r7 = 3. 8 ON THE STRUCTURE THEOREM OF CLIFFORD ALGEBRAS Then, the value of p−q mod 8 ("Periodicity of Eight" cf. [8,19]) determines whether K ∼= R, C or H. Furthermore, this automatically tells us, based on the theory outlined above, that (10) Cℓp,q = L1 ⊕ · · · ⊕ LN , N = 2k, that is, that the Clifford algebras decomposes into a direct sum of N = 2k minimal left ideals (simple left Cℓp,q-modules) Li, each of which is generated by a primitive idempotent. How to find these primitive mutually annihilating idempotents, is determined in Part (c). In Part (b) we also learn that the Clifford algebra Cℓp,q is semisimple as it is the direct sum of two simple algebras: (11) Cℓp,q ∼= Mat(2k−1, K) ⊕ Mat(2k−1, K) when p − q = 1 mod 4. Thus, we have two simple components in the algebra, each of which is a subalgebra. Notice that the two algebra elements 1 2 (1 − e12...n) and c2 = 1 2 (12) c1 = (1 + e12...n) are central, that is, each belongs to the center Z(Cℓp,q) of the algebra.7 This requires that n = p+q be odd, so that the unit pseudoscalar e12...n would commute with each generator ei, and that (e12...n)2 = 1, so that expressions (12) would truly be idempotents. Notice, that the idempotents c1, c2 provide an orthogonal decomposition of the unity 1 since c1 + c2 = 1, and they are mutually annihilating since c1c2 = c2c1 = 0. Thus, (13) Cℓp,q = Cℓp,qc1 ⊕ Cℓp,qc2 where each Cℓp,qci is a simple subalgebra of Cℓp,q. Hence, by Part (a), each subalgebra is isomorphic to Mat(2k−1, K) where K is either R or H depending on the value of p − q mod 8, as indicated. Part (c) tells us how to find a complete set of 2k primitive mutually annihilating idem- potents, obtained by independently varying signs ± in each factor in (6), provide an orthog- onal decomposition of the unity. The set of k commuting basis monomials ei1, . . . , eik , which square to 1, is not unique. Stabilizer groups of these 2k primitive idempotents f1, . . . , fN (N = 2k) under the conjugate action of Salingaros vee groups are discussed in [6, 7]. It should be remarked, that each idempotent in (6) must have exactly k factors in order to be primitive. Thus, we conclude from Part (c) that (14) Cℓp,q = Cℓp,qf1 ⊕ · · · ⊕ Cℓp,qfN , N = 2k, is a decomposition of the Clifford algebra Cℓp,q into a direct sum of minimal left ideals, or, simple left Cℓp,q-modules. Part (d) determines the unique division ring K = f Cℓp,qf , where f is any primitive idempotent, prescribed by the Wedderburn-Artin Theorem, such that the decomposition (9) or (11) is valid, depending whether the algebra is simple or not. This part also reminds us that the left spinor ideals, while remaining left Cℓp,q modules, are right K-modules. This is important when computing actual matrices in spinor representations (faithful and irre- ducible). Detailed computations of these representations in both simple and semisimple 7The center Z(A) of an k-algebra A contains all elements in A which commute with every element in A. In particular, from the definition of the k-algebra, λ1 ∈ Z(Cℓp,q) for every λ ∈ k. ON THE STRUCTURE THEOREM OF CLIFFORD ALGEBRAS 9 cases are shown in [2]. Furthermore, package CLIFFORD has a built-in database which dis- plays matrices representing generators of Cℓp,q, namely e1, . . . , en, n = p + q, for a certain choice of a primitive idempotent f . Then, the matrix representing any element u ∈ Cℓp,q can the be found using the fact that the maps γ shown on Parts (e) and (f), are algebra maps. Finally, we should remark, that while for simple Clifford algebras the spinor minimal left ideal carries a faithful (and irreducible) representation, that is, ker γ = {1}, in the case 2 spinor space S and S carries an irreducible but not faithful of semisimple algebras, each 1 representation. Only in the double spinor space S ⊕ S, one can realize the semisimple algebra faithfully. For all practical purposes, this means that each element u in a semisimple algebra must be represented by a pair of matrices, according to the isomorphism (11). In practice, the two matrices can then be considered as a single matrix, but over K ⊕ K which is isomorphic to R ⊕ R or H ⊕ H, depending whether p − q = 1 mod 8, or p − q = 5 mod 8. We have already remarked earlier that while K is a division ring, K ⊕ K is not. 4. Conclusions In this paper, the author has tried to show how the Structure Theorem on Clifford algebras Cℓp,q is related to the theory of semisimple rings, and, especially of left artinian rings. Detailed computations of spinor representations, which were distributed at the conference, came from [2]. 5. Acknowledgments Author of this paper is grateful to Dr. habil. Bertfried Fauser for his remarks and comments which have helped improve this paper. References [1] R. Ab lamowicz: "On Clifford Algebras and the Related Finite Groups and Group Algebras", in Early Proceedings of Alterman Conference on Geometric Algebra and Summer School on Kahler Calculus, Bra¸sov, Romania, August 1–9, 2016, Ramon Gonz´alez Calvet, ed., (2016) (to appear) [2] R. Ab lamowicz: "Spinor representations of Clifford: a symbolic approach",Computer Physics Commu- nications 115 (1998) 510–535. [3] R. Ab lamowicz and B. Fauser: "Mathematics of CLIFFORD: A Maple package for Clifford and Grassmann algebras", Adv. Appl. Clifford Algebr. 15 (2) (2005) 157–181. [4] R. Ab lamowicz and B. Fauser: CLIFFORD: A Maple package for Clifford and Grassmann algebras, http://math.tntech.edu/rafal/, 2016. [5] Ab lamowicz, R. and B. Fauser: "On the transposition anti-involution in real Clifford algebras I: The transposition map", Linear and Multilinear Algebra 59 (12) (2011) 1331–1358. [6] R. Ab lamowicz and B. Fauser: "On the transposition anti-involution in real Clifford algebras II: Stabi- lizer groups of primitive idempotents", Linear and Multilinear Algebra 59 (12) (2011) 1359–1381. [7] R. Ab lamowicz and B. Fauser: "On the transposition anti-involution in real Clifford algebras III: The automorphism group of the transposition scalar product on spinor spaces", Linear and Multilinear Algebra 60 (6) (2012) 621–644. [8] R. Ab lamowicz and B. Fauser: "Using periodicity theorems for computations in higher dimensional Clifford algebras", Adv. Appl. Clifford Algebr. 24 (2) (2014) 569–587. [9] H. Albuquerque and S. Majid: "Clifford algebras obtained by twisting of group algebras", J. Pure Appl. Algebra 171 (2002) 133–148. 10 ON THE STRUCTURE THEOREM OF CLIFFORD ALGEBRAS [10] Z. Brown: Group Extensions, Semidirect Products, and Central Products Applied to Salingaros Vee Groups Seen As 2-Groups, Master Thesis, Department of Mathematics, TTU (Cookeville, TN, December 2015). [11] C. Chevalley: The Algebraic Theory of Spinors, Columbia University Press (New York, 1954). [12] L. L. Dornhoff, Group Representation Theory: Ordinary Representation Theory, Marcel Dekker, Inc. (New York, 1971). [13] H. B. Downs: Clifford Algebras as Hopf Algebras and the Connection Between Cocycles and Walsh Functions, Master Thesis (in progress), Department of Mathematics, TTU, Cookeville, TN (May 2017, expected). [14] D. Gorenstein, Finite Groups, 2nd ed., Chelsea Publishing Company (New York, 1980). [15] I. N. Herstein, Noncommutative Rings, The Carus Mathematical Monographs 15, The Mathematical Association of America (Chicago, 1968). [16] N. Jacobson, Structure of Rings, Colloquium Publications 37, American Mathematical Society, Provi- dence (1956). [17] G. James and M. Liebeck, Representations and Characters of Groups, Cambridge Univ. Press, 2nd ed. (2010). [18] T.Y. Lam, The Algebraic Theory of Quadratic Forms, Benjamin (London, 1980). [19] P. Lounesto: Clifford Algebras and Spinors, 2nd ed., Cambridge Univ. Press (2001). [20] K. D. G. Maduranga, Representations and Characters of Salingaros' Vee Groups, Master Thesis, De- partment of Mathematics, TTU (May 2013). [21] K. D. G. Maduranga and R. Ab lamowicz: "Representations and characters of Salingaros' vee groups of low order", Bull. Soc. Sci. Lettres L´od´z S´er. Rech. D´eform. 66 (1) (2016) 43–75. [22] C. R. Leedham-Green and S. McKay, The Structure of Groups of Prime Power Order, Oxford Univ. Press (2002). [23] S. Majid, Foundations of Quantum Group Theory, Cambridge University Press (1995). [24] I. R. Porteous, Clifford Algebras and the Classical Groups, Cambridge Studies in Advanced Mathematics 50, Cambridge University Press (1995). [25] D. S. Passman, The Algebraic Structure of Group Rings, Robert E. Krieger Publishing Company (1985). [26] J. J. Rotman, Advanced Modern Algebra, 2nd ed., American Mathematical Society (Providence, 2002). [27] A. M. Walley: Clifford Algebras as Images of Group Algebras of Certain 2-Groups, Master Thesis (in progress), Department of Mathematics, TTU, Cookeville, TN (May 2017, expected).
1502.07988	1	1502	2015-02-18T20:48:50	Introduction to Noncommutative Algebraic Geometry (First Draft)	[ "math.RA" ]	This Lecture Notes is meant to introduce noncommutative algebraic geometry tools (which were invented by M. Artin, W. Schelter, J. Tate, and M. Van den Bergh in the late 1980s) and also graded skew Clifford algebras (which were introduced by T. Cassidy and M. Vancliff).	math.RA	math	Introduction to Noncommutative Algebraic Geometry Manizheh Nafari Abstract This Lecture Notes is meant to introduce noncommutative algebraic geometry tools (which were invented by M. Artin, W. Schelter, J. Tate, and M. Van den Bergh in the late 1980s) and also graded skew Clifford algebras (which were introduced by T. Cassidy and M. Vancliff). 5 1 0 2 b e F 8 1 ] . A R h t a m [ 1 v 8 8 9 7 0 . 2 0 5 1 : v i X r a 1 Contents 1 Introduction 2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Definition of Graded Algebras [2] 2.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Nonexamples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Definition of Quadratic K-Algebra . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6 Nonexample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7 Global Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.8 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.9 Definition of Polynomial Growth (c.f.,[7]) . . . . . . . . . . . . . . . . . . . . . 2.10 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.11 Definition of Gorenstein [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.12 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.13 Definition of Regular Algebras [2] . . . . . . . . . . . . . . . . . . . . . . . . . . 2.14 Definition of Normalizing Sequence . . . . . . . . . . . . . . . . . . . . . . . . . 3 Graded Skew Clifford Algebras . . . . . . . . . . . . . . . . . . 3.1 Definition of Graded Skew Clifford Algebras [4] 3.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Definition of Quadric System [4] . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Definition of Normalizing Quadric System . . . . . . . . . . . . . . . . . . . . . 3.6 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.7 Definition of Zero Locus [4] 3.8 Definition of Base-Point Free [4] . . . . . . . . . . . . . . . . . . . . . . . . . . 3.9 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 3 3 3 4 4 4 4 4 5 5 5 5 5 6 6 6 6 6 7 7 7 7 7 8 8 2 1 Introduction M. Artin, W. Schelter, J. Tate, and M. Van den Bergh introduced the notion of non- commutative regular algebras and invented new methods in algebraic geometry in the late 1980s to study them ([1], [2], [3]). Such algebras are viewed as non-commutative analogues of polynomial rings; indeed, polynomial rings are examples of regular algebras. By the 1980s, a lot of algebras had arisen in quantum physics, specifically quantum groups, and many traditional algebraic techniques failed on these new algebras. In physics, quantum groups are viewed as algebras of non-commuting functions acting on some “non-commutative space”([5]). In the early 1980s, E. K. Sklyanin, a physicist, constructed a family of graded algebras on four generators ([9]). These algebras were later proved to depend on an elliptic curve and an automorphism ([6]). By the late 1980s, it was known that many of the algebras in quantum physics are regular algebras; in particular, the family of algebras constructed by Sklyanin consists of regular algebras. The main results in [1], [2], and [3] concern the classification of regular algebras of global dimension 3 on degree-one generators. The quadratic regular algebras of global dimension 3 can be described using geometry, i.e. the point scheme E ⊆ P2. These algebras, where E contains a line as well as those that are “generic”, are given in [2], and [3], and entail: P2, elliptic curve, conic union a line, triangle, (triple) line, a union of n lines where n ∈ {2, 3} with one intersection point. It should be noted that the cases where E is a nodal cubic curve or a cuspidal cubic curve are not discussed in [2] or [3] as such algebras are not generic. T. Cassidy and M. Vancliff introduced a class of algebras that provide an “easy” way to write down some quadratic regular algebras of global dimension n where n ∈ N ([4]). In fact, they generalized the notion of a graded Clifford algebra and called it a graded skew Clifford algebra (see Definition 3.1). 2 Definitions 2.1 Definition of Graded Algebras [2] Throughout this lecture notes, K denotes an algebraically closed field, char(K) 6= 2, and K× denotes K \ {0}. A K-algebra A is called a graded algebra if: (1) A =Li≥0 Ai where the Ai are vector spaces over K, (2) dimA1 < ∞, (3) AiAj ⊆ Ai+j for all i, j, (4) A0 = K, (5) A generated by A1 only. For each i, Ai is the span of the homogeneous elements of degree i. 2.2 Examples (1) The polynomial ring A = K[x1, . . . , xd] where x1, . . . , xd have degree 1. Here, A1 = Kx1 ⊕ Kx2 ⊕ · · · ⊕ Kxd, 3 and dimKAi = i + d − 1 d − 1 ! for all i (c.f., [7]). (2) The free algebra A = Khx1, . . . , xdi where xi, for all i, have degree ni ∈ Z. Here, A is a non-commutative analogue of the algebra A in (1). 2.3 Nonexamples (1) The algebra A = K[x, y] hx2 − yi , where x and y have degree 1, is not graded. The relation x2 = y in A is not homogeneous and so A1 ∩ A2 6= {0} which violates (1) in Definition 2.1.1. (2) The algebra A = K[x, y] hx2 − yi , where x has degree 1 and y has degree 2, is graded but not generated by A1 since y ∈ A2. 2.4 Definition of Quadratic K-Algebra A K-algebra A is called quadratic if: (1) A is graded (as defined above), (2) A is a quotient of the free algebra by homogeneous relations of degree 2. 2.5 Example The algebra K[x1, . . . , xd] = Khx1, . . . , xdi hxixj − xjxi; 1 ≤ i, j ≤ di , deg(xi) = 1 for all i is quadratic. 2.6 Nonexample The algebra A = K[x] hx3i , where x has degree 1, is graded but is not quadratic. The relation x3 = 0 has degree 3. In order to define a regular algebra, we first need the concepts of polynomial growth, global dimension, and Gorenstein, which we now define. 2.7 Global Dimension The algebra A has global dimension d < ∞ if every A-module M has projective dimension ≤ d and there exists at least one module M with projective dimension d. 4 2.8 Example The polynomial ring, K[x1, . . . , xd], has global dimension d by Hilbert’s syzygy theorem (c.f., [8]). 2.9 Definition of Polynomial Growth (c.f.,[7]) A graded algebra A, as above, is said to have polynomial growth if there exists positive real numbers c, δ such that dimKAn ≤ cnδ for all n ≫ 0. For all known quadratic regular algebras of global dimension d, the minimal such δ is d − 1 ([2, §2]). 2.10 Example Let A = K[x1, x2], then dimKAn = n + 1 1 ! = n + 1 ≤ n1+ǫ, for all ǫ > 0 where n ≫ 0. Thus A has polynomial growth. 2.11 Definition of Gorenstein [1] By [2, §2], for a graded algebra A as in Definition 2.1.1, the global dimension of A equals the projective dimension of the graded left module AK (and projective dimension of the right module KA). The algebra A is Gorenstein if (1) the projective modules P i appearing in a minimal resolution 0 → P d → ... → P 1 → P 0 →A K → 0 of AK are finitely generated, and if (2) applying the functor M M ∗ := HomA(M, A) = {graded homomorphisms : M → A} to the resolution in (1) yields a projective resolution 0 → P 0∗ → P 1∗ → ... → P d∗ → KA → 0 of the graded right A-module KA. 2.12 Example The algebra is Gorenstein ([1, §0]). A = Khx, yi hxy − qyxi , where q ∈ K×, 5 2.13 Definition of Regular Algebras [2] A graded K-algebra A is called a regular algebra if (1) A has polynomial growth, (2) A has finite global dimension, (3) A is Gorenstein. 2.14 Definition of Normalizing Sequence A sequence a1, . . . , an of elements of a ring R with identity is called a normalizing sequence if a1 is normal element in R (i.e. a1R = Ra1) and for each j ∈ {1, . . . , n − 1}, aj+1 is a normal element in R/Pj 3 Graded Skew Clifford Algebras i=1 aiR and also Pn i=1 aiR 6= R. T. Cassidy and M. Vancliff defined a class of algebras in [4] that provide an “easy” way to write down some quadratic regular algebras of global dimension d for all d ∈ N. 3.1 Definition of Graded Skew Clifford Algebras [4] For {i, j} ⊂ {1, . . . , n}, let µij ∈ K× satisfy µij µji = 1 for all i 6= j, and write µ = (µij ) ∈ M (n, K). A matrix M ∈ M (n, K) is called µ-symmetric if Mij = µij Mji for all i, j = 1, . . . , n. Henceforth, suppose µii = 1 for all i, and fix µ-symmetric matrices M1, . . . , Mn ∈ M (n, K). A graded skew Clifford algebra associated to µ and M1, . . . , Mn is a graded K-algebra on degree- one generators x1, . . . , xn and on degree-two generators y1, . . . , yn with defining relations given by: (b) the existence of a normalizing sequence {r1, . . . , rn} of homogeneous elements that span k=1(Mk)ijyk for all i, j = 1, . . . , n, and (a) xixj + µij xjxi =Pn Ky1 + · · · + Kyn. 3.2 Example Let µ21, λ ∈ K×. If then any graded skew Clifford algebra A associated to M1, M2 satisfies M1 =(cid:20) 0 µ21 1 0 (cid:21) , M2 =(cid:20) 2 0 0 2λ (cid:21) , Khx1, x2i 2 − λx1 hx2 2i ։ A since x1x2 + µ12x2x1 = y1, y2 = x1 2, λy2 = x2 2. 6 3.3 Definition of Quadric System [4] Let S be the K-algebra on generators z1, . . . , zn with defining relations zj zi = µij zizj, for all i, j and let qk :=(cid:2) z1 . . . We say {q1, . . . , qn} is a quadric system. zn (cid:3) Mk   z1 ... zn   ∈ S. 3.4 Example For the algebra A in Example 2.2.2, we have Moreover, S = Khz1, z2i hz2z1 − µ12z1z2i . q1 = 2z1z2, q2 = 2z1 2 + 2λz2 2. However, since char(K) 6= 2, we consider: q1 = z1z2, q2 = z1 2 + λz2 2. 3.5 Definition of Normalizing Quadric System A quadric system {q1, . . . , qn} is normalizing if Pn sequence of S. k=1 Kqk ⊂ S is spanned by a normalizing 3.6 Example Referring to Example 2.2.4, in S, zi is normal for all i, and q1z1 = µ12z1q1, q1z2 = µ21z2q1. Therefore q1 is normal in S. In S hq1i , we have q2z1 = z1(z1 2 + λµ2 12z2 2), q2z2 = µ21 2z2(z1 2 + λµ12 2z2 2). So q2 is normal in S hq1i if λ = 0 or if λ 6= 0 and µ12 2 = 1. 3.7 Definition of Zero Locus [4] Suppose A = Khx1, . . . , xni and f ∈ A2. We define the zero locus V(f ) of f to be V(f ) = {p ∈ Pn−1 × Pn−1 : f (p) = 0}, where Pn−1 is identified with P(A∗ 1). Similarly if f1, . . . , fm ∈ A2, then V(f1, . . . , fm) = {p ∈ Pn−1 × Pn−1 : fi(p) = 0 for all i}. 7 3.8 Definition of Base-Point Free [4] Let Z be the zero locus in Pn−1 × Pn−1 of the defining relations of S, i.e. V(zj zi − µij zizj) ⊂ Pn−1 × Pn−1. Z =\i,j The quadric system {q1, . . . , qn} is said to be base-point free (BPF) if Z ∩ V(q1, . . . , qn) is empty. 3.9 Example Referring to Example 2.2.4, let and let Therefore, we have p = ((α1, α2), (β1, β2)) ∈ P1 × P1, (z2z1 − µ12z1z2)(p) = 0. α2β1 − µ12α1β2 = 0. If α2 = 0, then β2 = 0. So ((1, 0), (1, 0)) ∈ P1 × P1. If α2 6= 0, i.e., α2 = 1, then β1 = µ12α1β2. So, ((α1, 1), (µ12α1, 1)) ∈ P1 × P1. Therefore, Z = {((α1, α2), (µ12α1, α2)) : (α1, α2) ∈ P1}. Let p ∈ Z. We have 0 = q1(p) = α1α2, 0 = q2(p) = µ12α1 2 + λα2 2. Thus α1 = α2 = 0 which is contradiction. Therefore {q1, q2} is BPF. References [1] Artin, M. and Schelter, W., Graded Algebras of Global Dimension 3, Adv. Math., 66 (1987), 171-216. [2] M. Artin, J. Tate and M. Van den Bergh, Some Algebras Associated to Automorphisms of Elliptic Curves, The Grothendieck Festschrift 1, Eds. P. Cartier et al. Birkhauser (1990), 33-85. [3] Artin, M., Tate, J., and Van den Bergh, M., Modules Over Regular Algebras of Dimension 3, Invent. Math., 106 (1991), 335-388. [4] Cassidy, T. and Vancliff, M., Generalizations of Graded Clifford Algebras and of Complete Intersections, Journal of the London Mathematical Society 81 (2010), 91-112. [5] Drinfel’d, V. G., Quantum Groups, Proc. Int. Cong. Math., Berkeley 1 (1986), 798-820. [6] Feigin, B. L., and Odesskii, A. B., Elliptic Sklyanin Algebras, Func. Anal. Appl. 23 (1989), 45-54. [7] McConnell, J. C. and Robson, J. C., Noncommutative Noetherian Rings, Graduate Studies in Mathematics, American Mathematical Society, 2001. [8] Rotman, Joseph J., An Introduction to Homological Algebra, Second edition, Universitext, Springer, New York, 2009, xiv+709 pp. [9] Sklyanin, E. K., Some Algebraic Structures Connected to the Yang-Baxter Equation, Func. Anal. Appl. 16 (1982), no. 4, 27-34. 8
1612.02919	2	1612	2017-09-21T08:50:26	A Dedekind Domain with Nontrivial Class Group	[ "math.RA", "math.AC" ]	Analytic properties of function spaces over the real and the complex fields are different in some ways. This reflects in algebraic properties which are different at times and similar in some other respects. For instance, the ring of real-valued continuous functions on a closed interval like $[0,1]$ behaves similarly to the corresponding ring of complex-valued functions; they depend only on the topology of $[0,1]$. The ring $\mathbf{R}[X,Y]/(X^2+Y^2-1)$ of real-valued polynomial functions on the unit circle is not a unique factorization domain - witness the equation $$\cos^2(t) = (1+ \sin(t))(1- \sin(t)).$$ On the other hand, the ring $\mathbf{C}[X,Y]/(X^2+Y^2-1) \cong \mathbf{C}[X+iY, 1/(X+iY)]$ is a principal ideal domain. Again, the rings of convergent power series (over either of these fields) with radius of convergence larger than some number $\rho$ is a Euclidean domain (and hence, a principal ideal domain) - this can be seen by using for a Euclidean "norm" function, the function which counts zeroes (with multiplicity) in the disc $\|z\| \leq \rho$. In this note, we consider the rings $C_{an}(S^1;\mathbf{R})$ of real-analytic functions on the unit circle $\mathit{S}^1$ which are real-valued and the corresponding ring $C_{an}(S^1; \mathbf{C})$ of analytic functions that are complex-valued. We will see that the latter is a principal ideal domain while the former is a Dedekind domain which is not a principal ideal domain - the class group having order $2$.	math.RA	math	A Dedekind Domain with Nontrivial Class Group Vaibhav Pandey, Sagar Shrivastava, and Balasubramanian Sury We show that the ring of real-analytic functions on the unit circle is a Dedekind domain with class number two. Abstract 1 Rings that engage analysts. Analytic properties of function spaces over the real and the complex fields are in some ways different. This is strongly reflected in these spaces' algebraic properties. For instance, the ring of real-valued continuous functions on a closed interval such as [0, 1] behaves similarly to the corresponding ring of complex-valued functions; they depend only on the topology of [0, 1]. The ring of real-valued polynomial functions on the unit circle can be identified with the ring of all real trigonometric polynomials. It is not a unique factor- ization domain as is demonstrated by the equation cos2(t) = (1 + sin(t))(1 − sin(t)). In fact, the above ring is R[X, Y ]/(X 2 + Y 2 − 1) and the equation Y 2 = (1 + X)(1 − X) that holds in the quotient ring gives two different decompositions of Y 2 into irreducible elements Y, 1 + X, 1 − X. On the other hand, the ring C[X, Y ]/(X 2 + Y 2 − 1) ∼= C[X + iY, 1/(X + iY )] is a principal ideal domain. Again, the rings of convergent power series (over either of these fields) with radius of convergence larger than some positive real number ρ is a Euclidean domain (and hence, a principal ideal domain)-this can be seen by using the function that counts zeroes (with multiplicity) in the disk z ≤ ρ as a Euclidean "norm" function (see [3]). In this note, we consider the rings Can(S1; R) of analytic functions on the unit circle S 1 that are real-valued and the corresponding ring Can(S1; C) of analytic functions that are complex-valued. We will see that the latter is a principal ideal domain while the former is a Dedekind domain which is not a principal ideal domain-the class group having order 2. 1 2 Maximal ideals are points. The proof of the fact alluded to is exactly the same as the corresponding proof (that is well known) for the ring of continuous functions on a closed interval. Lemma 1. Maximal ideals of Can(S1, R) and of Can(S1, C) are points. Proof. This is a consequence of the compactness of S 1. Indeed, for each point p ∈ S1, the ideal mp := {f : f (p) = 0} is maximal as the quotient is isomorphic to a field. Let us observe that every maximal ideal m is of the form mp for some p in S 1. If not, then we can find functions fi in m that do not vanish in a neighborhood of pi for each pi in S 1 by continuity. These neighborhoods cover S 1. By compactness of S 1, finitely many of these neighborhoods cover it. Call these f1, . . . , fn. Then the i=1 fifi lies in m and is a unit (as it does not vanish anywhere). This is a contradiction as maximal ideals are proper. function Pn Lemma 2. The ring Can(S1, C) of complex-valued analytic functions on S 1 is a PID. Proof. Clearly, the maximal ideal mp is the principal ideal generated by z −p. Hence, every finite product m a1 p1 · · · m ak pk is principal. We show that every nonzero ideal is such a finite product. Any nonzero analytic function has only finitely many zeroes as zeroes are isolated and S1 is compact. Hence, if I is any nonzero ideal, it has only finitely many common zeroes, say p1, . . . , pk. Let ai be the smallest positive integer such that every element of I has a zero of order at least ai at pi. Hence, for each 0 6= f ∈ I, we have f = (cid:18)Qk other words, I is contained in m i=1(z − pi)ai(cid:19)g for some analytic function g. In a1 p1 · · · m ak pk . Then J := {f / k Yi=1 (z − pi)ai : f ∈ I} is an ideal. If J were a proper ideal, it would be contained in some mp. If p 6∈ {p1, . . . , pk}, then I ⊂ mp which contradicts the fact that p1, . . . , pk are the only common roots of I. Hence p = pi for some 1 ≤ i ≤ k. But if 2 fi ∈ I has order exactly ai at pi, then fi/Qk a contradiction. Hence J is the unit ideal and so j=1(z − pj)aj cannot vanish at pi, I = m a1 p1 · · · m ak pk . Hence I is principal. So, Can(S1, C) is a PID (and hence a Dedekind domain). 3 Ideals in Can(S1; R). Let us recall that a real-analytic function in Can(S1; R) is a function such that f ◦g1 and f ◦g−1 are analytic where g1(x) = e2iπx on (0, 2π) and g−1(x) = e2iπx on (−π, π). Recall we observed that maximal ideals are points. Lemma 3. The product of any two maximal ideals mp1, mp2 (including the case p1 = p2), is principal. ) − cos( p1−p2 mp2 can be generated by the analytic Proof. In fact, it is easy to see that mp1 function fp1,p2(x) = cos(x − (p1+p2) ). To clarify this further, note that when p1 6= p2, the function fp1,p2 has simple zeroes at p1 and p2 and no other zeroes (consider the derivative). If p1 = p2 = p, then the function fp,p(x) = 2 sin2( x−p 2 ) has a double root at p and no other roots. In either case, it follows that any element f ∈ mp1 is analytic. This completes the proof. mp2 satisfies the property that fp1 ,p2 2 2 f This immediately implies the following corollary. Corollary 1. An ideal I = m even. a1 p1 m a2 p2 · · · m an pn is principal if a1 + · · · + an is Lemma 4. An ideal I = m odd. a1 p1 m a2 p2 · · · m an pn is not principal if a1 + · · · + an is Proof. We first show that maximal ideals in the ring Can(S1; R) are not principal. This is obvious because identifying S1 with R/2πZ, the number of zeroes of any analytic function on S1 counted with multiplicity is even-this is an simply because of the intermediate value theorem. Now, if I = m pn with a1 + · · · + an odd, then I = mp1(g) by the even case. If I = (f ), then f ∈ gmp1 ⊂ (g) so that f /g is analytic. But then mp1 = (f /g), which is a contradiction, as f /g has an even number of zeroes counting multiplicity, while mp1 has only a common zero at p1. a2 p2 · · · m a1 p1 m 3 Finally, we have the following factorization result. Theorem 1. Every nonzero proper ideal in the ring Can(S1; R) is of the form m an pn for points p1, · · · , pn. a2 p2 . . . m a1 p1 m Before proving this theorem, we observe the following very interesting fact. Corollary 2. Can(S1; R) is a Dedekind domain which has class number 2. Proof. This immediately follows from Theorem 1, Corollary 1 and Lemma 4. Remarks on Dedekind domains and class groups. Let us recall briefly the role of Dedekind domains in number theory. Dedekind domains are precisely the class of integral domains in which the fractional ideals are in- vertible. The rings of algebraic integers in finite extension fields of Q are natural examples of Dedekind domains. Moreover, any PID is a Dedekind domain. The class group of a Dedekind domain is the group of fractional ideals modulo the principal fractional ideals. A Dedekind domain is a unique factorization domain if and only if the class group of fractional ideals is triv- ial. Many subtleties involved in solving Diophantine equations arise from the fact that many rings of algebraic integers arising in their study have non- trivial class group. The Fermat equation cannot be studied by elementary algebraic methods due to the (amazing) fact that the ring of integers in the field generated by the pth roots of unity for a prime p is not a UFD for any prime p ≥ 23. By a theorem of L. Claborn (see [2]), every abelian group can be realized as the class group of some Dedekind domain; the analogous problem is open for rings of algebraic integers. In other words, it is expected but still unknown whether every finite abelian group can be realized as the class group of a ring of integers in an algebraic number field. Finally, let us prove Theorem 1. Proof. Consider any proper, nonzero ideal I. Let {p1, . . . , pn} be the common zeros of I-as we observed above, this is finite, as every nonzero analytic function on S1 has only finitely many zeroes. For each k ≤ n, let ak be minimal among the orders of zeroes of elements of I at pk. Then, it is clear that I ⊂ m a1 p1 We will show that I = Qn k=1 m ak pk . m a2 p2 · · · m an pn . 4 Let us first assume that a1 + · · · + an is even. Let f be an element of I whose order of zero at pk is ak for 1 ≤ k ≤ n. Such an f exists since I contains elements fk vanishing at pk with order ak, and we may consider a suitable linear combination g1f1 + · · · + gnfn. This function f may have other zeroes different from pk; we wish to change f such that the new element is in I, has zeroes of order ak at pk, and has no other zeroes. This is accomplished as follows. As we observed in the beginning, every analytic function changes signs an even number of times by the intermediate value theorem. Let us write 0 ≤ p1 < p2 < · · · < pn < 2π. In some of the intervals [pi, pi+1] (among [p1, p2], [p2, p3], . . . , [pn, p1]), the function f changes sign an even number of times and in others, it changes sign an odd number of times. The latter happens in an even number of intervals [pi, pi+1]. If we select some analytic function g that has simple zeroes at some interior point of each of these latter intervals and no other zeroes, then the function f g ∈ I and has the property that f g vanishes at each pi exactly to the order ai and has an even number of sign changes in each interval (p1, p2), (p2, p3), . . . , (pn, p1). It also changes signs at an even number of the pi's. At these even number of pi's, there is an analytic function h with simple zeroes and no other roots. We may multiply the analytic function h by an element φ ∈ I that has no zeroes in any of the open intervals (pi, pi+1) (we may square and assume the value of φ is positive in each of these open intervals). By changing the sign of h if necessary, we may assume it has the same sign as f around each pi. Then hφ ∈ I has simple zeroes at the pi's, no other zeroes, and has the sign of f in each open interval (pi, pi+1). Since continuous (hence analytic) functions are bounded on a compact set, therefore, for a large constant c, the function f g + chφ is in ai pi I and has zeroes of order ai at pi and no other zeroes. Hence I ⊇ Qn which shows that these ideals are equal. i=1 m m a1 p1 an Now, assume that a1 + · · · + an is odd. Let f ∈ m pn . Since it must have an even number of zeros (counting multiplicity), it must have a zero q 6∈ {p1, . . . , pn} or it has a zero of order greater than ak at some pk (in which case we put q = pk). Let J = {g ∈ Ig(q) = 0} (in case q = pk, we take g to have zeroes at pk with multiplicity greater than ak). By the even case treated already, J = m mq. Thus f belongs to the right-hand side. Thus f ∈ J ⊂ I. This completes the proof. a2 p2 · · · m a2 p2 · · · m an pn a1 p1 m We end with a remark which is relevant to the fact that the ring of real analytic functions on S1 has class number 2. L. Carlitz proved (in [1]) that Dedekind domains with class number 2 are half-factorial domains; viz., differ- 5 ent irreducible factorizations of elements must have the same length. Finally, we mention that we are interested in generalizations of the above result to compact manifolds. ACKNOWLEDGMENTS. The second author was a project student at the Tata Insti- tute of Fundamental Research, Mumbai, India when this note was written. The authors would like to thank the two referees for suggestions to rewrite some parts more clearly. Interestingly, one of the referees brought the authors' attention to a link https://math. stackexchange.com/questions/ 154605/ring-of-analytic-functions-on-the-circle which con- tains a discussion of the same example studied in this note. References [1] L. Carlitz, A characterization of algebraic number fields with class num- ber two, Proc. Amer. Math. Soc., 11 (1960) 391–392. [2] L. Claborn, Every abelian group is a class group, Pacific J. Math. 18 (1966) 219–222. [3] C. U. Jensen, Some curiosities of rings of analytic functions, J. Pure Appl. Algebra, 38 (1985) 277–283. University of Utah, 1015, Barbara Place, Apt-3, Salt Lake City - 84102, USA. Email: [email protected] School of Mathematics, Tata Institute of Fundamental Research, Homi Bhabha Road, Mumbai, 400005, India. Email: [email protected] Statistics & Mathematics Unit, Indian Statistical Institute, 8th Mile Mysore Road, Ban- galore 560059, India. Email: [email protected] 6
1003.4108	1	1003	2010-03-22T09:59:45	Irreducible actions and compressible modules	[ "math.RA" ]	Any finite set of linear operators on an algebra $A$ yields an operator algebra $B$ and a module structure on A, whose endomorphism ring is isomorphic to a subring $A^B$ of certain invariant elements of $A$. We show that if $A$ is a critically compressible left $B$-module, then the dimension of its self-injective hull $A$ over the ring of fractions of $A^B$ is bounded by the uniform dimension of $A$ and the number of linear operators generating $B$. This extends a known result on irreducible Hopf actions and applies in particular to weak Hopf action. Furthermore we prove necessary and sufficient conditions for an algebra A to be critically compressible in the case of group actions, group gradings and Lie actions.	math.RA	math	IRREDUCIBLE ACTIONS AND COMPRESSIBLE MODULES INES BORGES AND CHRISTIAN LOMP dedicated to the memory of John Dauns Abstract. Any finite set of linear operators on an algebra A yields an operator algebra B and a module structure on A, whose endomorphism ring is isomorphic to a subring AB of certain invariant elements of A. We show that if A is a critically compressible left B-module, then the dimension of its self-injective hull bA over the ring of fractions of AB is bounded by the uniform dimension of A and the number of linear operators generating B. This extends a known result on irreducible Hopf actions and applies in particular to weak Hopf action. Furthermore we prove necessary and sufficient conditions for an algebra A to be critically compressible in the case of group actions, group gradings and Lie actions. 1. Introduction The starting point of this note is a theorem by Bergen et al. that says that for a Hopf algebra H acting on an algebra A with subring of invariants AH , such that A is an irreducible A#H-module, the dimension dim(AAH ) is bound by the uniform dimension of AA and the number of generators of A#H/AnnA#H (A) as A-module. We will generalize their result in two directions: replacing the Hopf action by a subalgebra of linear operators of A which contains the regular left action of A on itself and by weakening the irreducible assumption of A to A being a monoform respectively critically compressible module. Actions of Hopf algebras include group actions, Lie actions and finite group gradings. Several generalizations of Hopf algebras have emerged in recent years, like weak Hopf algebras (or quantum groupoids) introduced by Bohm et al. [5]. Examples of actions of weak Hopf algebras are given by groupoid actions on C ∗-algebras (see [18]) and by quantum groupoids arising from Jones towers (see [13]). Our point of view is that such action on an algebra A should be studied by looking at the module structure on A given by certain algebras of linear operators, i.e. subalgebras of Endk(A). Let k be a commutative ring and let A be an associative unital k-algebra. For any a ∈ A define two linear operators La and Ra in Endk(A) given by La(x) = ax and Ra(x) = xa for all x ∈ A. We identify A with the subalgebra L(A) of Endk(A) generated by all left multiplications La and denote the subalgebra generated by all operators La and Ra by M (A), which is also sometimes referred to as the multiplication algebra of A. As a left L(A)-module, A is isomorphic to L(A) since we assume A to be unital. We will be interested in certain actions on an algebra A that may stem from a bialgebra or more generally a bialgebroid. Usually we are interested in extensions A ⊆ B where B act on A through a ring homomorphism φ : B → Endk(A) such that φ(a) = La for all a ∈ A (like for instance if B = A#H is a smash product). To study the intrinsic properties of A under this action it is enough to look at the subalgebra φ(B) in Endk(A) generated by this action. Therefore we will consider subalgebras B of Endk(A) that contain L(A) and act on A by evaluation. Thus A becomes a cyclic faithful left B-module by ϕ · a := ϕ(a), ∀ϕ ∈ B, a ∈ A. Note that La′ · a = a′a for any a, a′ ∈ A. Key words and phrases. irreducible modules, Ore domains, critically compressible modules, weak Hopf actions. The first author was supported by grant SFRH/PROTEC/49857/2009. The second author was partially sup- ported by Centro de Matemtica da Universidade do Porto (CMUP), financed by FCT (Portugal) through the programs POCTI (Programa Operacional Cincia, Tecnologia, Inovao) and POSI (Programa Operacional Sociedade da Informao), with national and European community structural funds. 1 2 IN ES BORGES AND CHRISTIAN LOMP Since we assume A to be unital, the map Ψ : EndB(A) → A with Ψ(f ) = (1)f , for all f ∈ EndB(A), is an injective ring homomorphism, since ∀f, g ∈ EndB(A) : Ψ(f ◦ g) = ((1)f )g = ((1)f · 1)g = (1)f · (1)g = Ψ(f )Ψ(g). Note that we will write homomorphisms opposite to scalars. Moreover if Ψ(f ) = (1)f = 0, then (a)f = a(1)f = 0 and f = 0. The subalgebra Ψ(EndB(A)), which we will denote by AB, can be described as the set of elements a ∈ A such that for any b ∈ B : b · a = (b · 1)a. A subset I of A is called B-stable if B · I ⊆ I. The B-stable left ideals are precisely the (left) B-submodules of A. In particular, by restricting Ψ, we have HomB(A, I) ≃ I ∩ AB, for any B-stable left ideal of A. Examples 1.1. Let H be a weak Hopf algebra over k (see Definition 2.6) and let A be a left H-module algebra. Denote the action of an element h ∈ H on A by λh ∈ Endk(A) and define B as the subalgebra of Endk(A) generated by L(A) and all operators λh. The left B-submodules of A are precisely the H-stable left ideals of A. Moreover B is a factor of the smash product A#H. A particular case of this are group actions given by a group G and a group homomorphism η : G → Autk(A). Set B = hL(A) ∪ η(G)i ⊆ Endk(A). The B-submodules of A are precisely the G-stable left ideals of A and HomB(A, I) ≃ I ∩ AG, where AG = {a ∈ A ∀g ∈ G : η(g)(a) = a} is the fix ring of A. B is a quotient of the skew group ring A ∗ G. Another example is given by Lie algebras acting as derivations on A, e.g. if δ ∈ Derk(A) is a k-linear derivation of A then consider B = hL(A) ∪ {δ}i ⊆ Endk(A). The left B-submodules of A are the left ideals I that satisfy δ(I) ⊆ I. The operator algebra B is a factor of the ring of differential operator A[z; δ], which as a left A-module is equal to A[z] and its multiplication is given by za = az +δ(a). The map i=0 Lai ◦ δi ∈ B is a surjective k-algebra homomorphism and for any left A[z; δ]-module M we have HomA[z;δ](A, M ) = HomB(A, M ) = M δ = {m ∈ M zm = 0}. In particular EndA[z;δ](A) ≃ Aδ = Ker(δ). A[z; δ] → B withPn i=0 aizi 7→Pn Examples 1.2. For B = M (A), the B-submodules of A are the two-sided ideals of A, and HomB(A, I) ≃ Z(A) ∩ I, where Z(A) denotes the centre of A. The algebra M (A) is a quotient of the enveloping algebra Ae = A ⊗ Aop of A. Let ∗ be an involution of A. Set B = hL(A) ∪ {∗}i. The left B-submodules of A are the twosided ∗-ideals and AB = Z(A, ∗) is the subring of central symmetric elements of A. Note that B can be seen as the factor ring of the skew-group ring Ae#G where G = {id, ∗} is the cyclic group of order two and ∗ ∈ Aut(Ae) is given by (a ⊗ b)∗ := b∗ ⊗ a∗. 2. Irreducible actions of linear Operators. The left uniform dimension of A is denoted by udim(A) and is the supremum of the cardinalities of the index sets of direct sums of left ideals contained in A. A result by Bergen et al. [4, Theorem 2.2] says that if a Hopf algebra H acts finitely on a module algebra A with finite uniform dimension, such that A is a simple A#H-module, then A has finite dimension over AH . The argument of [4, Theorem 2.2] uses the Jacobson Density Theorem which had been gener- alized by Zelmanowitz in [21, 23]. The hypotheses of Zelmanowitz' density theorems are weaker than assuming the existence of a faithful simple module: a non-zero left R-module M is called compressible if it can be embedded in each of its non-zero submodules and it is called critically compressible if it is compressible and cannot be embedded in any of its proper factor modules. A left R-module M is called monoform if any non-zero partial endomorphism f : N → M from a submodule N to M is injective. It is easy to see that the critically compressible modules are precisely those that are compressible and monoform. Zelmanowitz proved the following two weak Density Theorem for rings which have a faithful critically compressible module respectively faithful monoform module. Theorem 2.1 (Zelmanowitz, [22, Theorem 2.2],[23, Proposition 2.1]). Let M be a faithful left R-module with self-injective hull cM and ∆ = EndR(cM ). IRREDUCIBLE ACTIONS AND COMPRESSIBLE MODULES 3 and ru1 6= 0. A ring R with a faithful critically compressible left R-module is called left weakly primitive. independent over ∆ there exists 0 6= f ∈ ∆ such that for any elements u1, u2, . . . , uk ∈ M there exists r ∈ R with rvi = (ui)f , for each i = 1, . . . k. (1) If RM is critically compressible, then for any elements v1, v2, . . . , vk ∈ cM that are lineary (2) If RM is monoform left R-module, then for any v1, . . . , vk ∈ cM , linearly independent over ∆ and u1, . . . , uk ∈ cM with u1 6= 0, there exist r, s ∈ R with ruj = svj for j = 1, . . . , k The self-injective hull of a left R-module M is the submodule cM = Tr(M, E(M )), where E(M ) is the trace of X in Y . Hence any element of cM can be written as a finite sum Pn images of homomorphisms fi : M → E(M ). If M is a simple module, then cM = M . Tr(X, Y ) =X{Imf f ∈ HomR(X, Y )} Using the self-injective hull of BA, we are able of extending [4, Theorem 2.2]. Theorem 2.2. Suppose L(A) ⊆ B ⊆ Endk(A) such that AB is generated by n elements. If one of the following conditions hold is the injective hull of M and i=1(mi)fi of (1) A is a domain and A is a monoform left B-module or (2) A is a critically compressible left B-module or (3) A is a simple left B-module then dim( A∆) ≤ n · udim(A), where A is the self-injective hull of BA and ∆ = EndB( A). Note that in case (3) A = A and in case (2) ∆ is isomorphic to the division ring of fractions of AB. Proof. We want to show that the dimension of bA∆ is bounded by n·udim(A). Since bA is generated as ∆-vector space by the elements of A (any element is a linear combinationPn i=1(ai)fi with fi ∈ ∆ and ai ∈ A) it is enough to consider linearly independent elements in A. Let v1, . . . , vk ∈ A be linearly independent over ∆. Under the condition (1) or (2), we will show that for each 1 ≤ i ≤ k there exist si ∈ B and ai ∈ A such that si · vj = δijxi, where xi is a right non-zero divisor of A. Once the existence of such elements is guaranteed, we can proceed as follows: since B is n−generated over A, there exist an epimorphism of left A-modules ϕ : An → B. Hence for each 1 ≤ i ≤ k, there exist an element ti ∈ An with (ti)ϕ = si. i=1 Ati is direct. Suppose that there are elements a1, . . . , ak ∈ A such that i=1 aiti = 0, then for all 1 ≤ j ≤ k: We will show thatPk Pk 0 = nXi=1 And since xj is a right non-zero divisor, aj = 0. Thus Ln aiti! ϕ · vj = result follows since ai (ti) ϕ · vj = nXi=1 nXi=1 dim( A∆) ≤ udim(An) = n · udim(A). nXi=1 ai (si · vj) = aiδij xi = ajxj . i=1 Ati is a direct sum in An and the To guarantee the existence of such elements si and xi we use Zelmanowitz' weak density theo- rems: Given v1, . . . , vk ∈ A be linearly independent over ∆, define uij = δij1A, for all 1 ≤ i, j ≤ k. (1) Suppose that A is a domain and BA is monoform. Since BA is faithful, Theorem 2.1 shows that for each 1 ≤ i ≤ k, there exist si, ri ∈ B with si · vj = ri · uij = δij(ri · 1A), for all 1 ≤ j ≤ k, and xi = ri · 1A 6= 0. Since A is a domain, xi is a right non-zero divisor. (2) Suppose that BA is critically compressible, then Theorem 2.1 shows that for each 1 ≤ i ≤ k, there exist fi ∈ ∆ and si ∈ B such that si · vj = (uij )fi = δij(1)fi, for all 1 ≤ j ≤ k. 4 IN ES BORGES AND CHRISTIAN LOMP Since any f ∈ ∆ is injective, a(1)f = (a)f 6= 0, for any non-zero a ∈ A. Hence xi = (1)fi is a right non-zero divisor in A. By [22, 11.5(2)], ∆ is isomorphic to Frac(AB). (3) Follows from (2). (cid:3) Theorem 2.2 makes no statement if the uniform dimension of A is infinite. However a domain with finite uniform dimension has uniform dimension 1 and is called left Ore domain (see [11]). If R is a left Ore domain and f : I → R is a partial non-trivial left R-linear endomorphism, then there exists x ∈ I such that (x)f 6= 0. For any y = rx ∈ Ker(f ) ∩ Rx we have r(x)f = (y)f = 0. Since R is a domain and (x)f 6= 0, y = 0 and Ker(f ) ∩ Rx = 0. Since RR is uniform Ker(f ) = 0, i.e. f is injective. Hence any left Ore domain R is itself a monoform left R-module. On the other hand if R is a monform left R-module, then it is also a uniform module by [19, 11.3]. Moreover for any a ∈ R, the endomorphism Ra : [x 7→ xa] is injective. Hence ba = 0 implies b = 0 or a = 0, i.e. R is a domain. This shows the following Lemma: Lemma 2.3. Let A be k-algebra. Then A is a left Ore domain if and only if A is a monoform left A-module. In this case A is a monoform left B-module, for all L(A) ⊆ B ⊆ Endk(A). The last statement is clear, because if f : I → A is a B-linear partial endomorphism, then it is also A-linear and hence injective if A was monoform as left A-module. Corollary 2.4. Let A be a left Ore domain and L(A) ⊆ B ⊆ Endk(A) with AB being generated by n elements, then dim( A∆) ≤ n, where A is the self-injective hull of A and ∆ = EndB( A). Before we apply this result to quantum groupoid actions, we first generalize another result from the theory of Hopf actions to a more general setting. Let A ⊆ B be an extension of k-algebras such that there exists a ring homomorphism Ψ : B → Endk(A) with Ψ(a) = La for all a ∈ A. If B ′ = Im(Ψ), then EndB(A) ≃ ∆ := AB ′ . The extension A ⊆ A#H with H being a (weak) Hopf algebra acting on A is an example of such an extension. Proposition 2.5. Let A ⊆ B be as above and assume that B is a finitely generated left and right A-module. Let ∆ = EndB(A). If A has finite left Goldie dimension and A is a simple left B-module, then the following statements are equivalent: (a) B is a simple ring; (b) B ≃ End(A∆) and A∆ is finitely generated projective; (c) A is a faithful left B-module and HomB(A, B) 6= 0; If BA is free of rank n, then (a − c) are also equivalent to: (d) dim(A∆) = n and HomB(A, B) 6= 0; Proof. By Theorem 2.2, A has finite dimension over ∆ and since B is finitely generated as a left and right A-module, B is an Artinian ring. (a) ⇒ (b) Suppose B is simple, then it is semisimple Artinian and the epimorphism b 7→ b · 1A splits as left B-module by some B-linear map φ : A → B, i.e. BA is projective. Since B is simple and the trace ideal of A in B is non-zero, Tr(A, B) = B, i.e. A is a generator in B-Mod. By a standard module theoretic argument B ≃ End(A∆). (b) ⇒ (c) is clear since A is a generator. (c) ⇒ (a) HomB(A, B) 6= 0 implies that A is isomorphic to a minimal left ideal of B (since A is a simple B-module). As BA is faithful, B is a left primitive ring having a faithful minimal left ideal, i.e. B ≃ Mm(∆) with m = dim(A∆). If BA is free of rank n, then n · dim(A∆) = dim(B∆) holds. (c) ⇒ (d) As seen in the last step (c) implies B ≃ Mm(∆) with m = dim(A∆). On the other hand n · m = dim(B∆) holds, yields n = m. (d) ⇒ (b) if m = dim(A∆) = n, then dim(B∆) = m2 = dim(End(A∆)). Thus B ≃ End(A∆). (cid:3) We intend to apply Theorem 2.2 to weak Hopf algebra actions and recall therefore its definition from [5]. IRREDUCIBLE ACTIONS AND COMPRESSIBLE MODULES 5 Definition 2.6. An associative k-algebra H with multiplication m and unit 1 which is also a coassociative coalgebra with comultiplication ∆ and counit ǫ is called a quantum groupoid or a weak Hopf algebra if it satisfies the following properties: (1) the comultiplication is multiplicative, i.e. for all g, h ∈ H: ∆(gh) = ∆(g)∆(h). (2) the unit and counit satisfy ǫ(f gh) = ǫ(f g1)ǫ(g2h) = ǫ(f g2)ǫ(g1h) and (∆ ⊗ id)∆(1) = (∆(1) ⊗ 1)(1 ⊗ ∆(1)) = (1 ⊗ ∆(1))(∆(1) ⊗ 1) (3) there exists a linear map S : A → A, called antipode, such that for all h ∈ H: S(h) = S(h1)h2S(h3), h1S(h2) = (ǫ ⊗ id)(∆(1)(h ⊗ 1)) =: ǫt(h), S(h1)h2 = (id ⊗ ǫ)((1 ⊗ h)∆(1)) =: ǫs(h). Note that we will use Sweedler's notation for the comultiplication with suppressed summation symbol. The image of ǫt and ǫs are subalgebras Ht and Hs of H which are separable over k ([15, 2.3.4]). Those subalgebras are also characterized by Ht = {h ∈ H : ∆(h) = 11h ⊗ 12} respectively Hs = {h ∈ H : ∆(h) = 11 ⊗ 12h}. A left H-module algebra A over a quantum groupoid H is an associative unital algebra A such that A is a left H-module and for all a, b ∈ A, h ∈ H: (1) h · (ab) = (h1 · a)(h2 · b) and h · 1A = ǫt(h) · 1A Let A be a left H-module algebra over a quantum groupoid H and let λ be the ring homomorphism from H to Endk(A) that defines the left module structure on A, i.e. λh(a) := h · a for all h ∈ H, a ∈ A. Property (1) of the definition above can be interpreted as an intertwining relation λh ◦ La = Lh1·a ◦ λh2 of left multiplications La and left H-actions λh. Recall that the smash product A#H of a left H-module algebra A and a quantum groupoid H is defined on the tensor product A ⊗Ht H where A is considered a right Ht-module by a · z = a(z · 1A) for a ∈ A, z ∈ Ht. Consider the algebra extension A ⊆ A#H =: B and the subalgebra of Endk(A) generated by the H-action and L(A), we can apply Theorem 2.2 and Proposition 2.5 to obtain an quantum groupoid analog of results by Bergen et al. [4, Theorem 2.2] and Cohen et al. [8, Theorem 3.3]. Recall, from [7] that a H-comodule algebra A is an H-Galois extension of AcoH if the canonical map A ⊗AcoH A → A ⊗ H is an isomorphism. If H is finite dimensional and A a left H-module algebra such that AH is divison ring, then by [7, Proposition 2.3] A is a projective generator in A#H-Mod if and only if AH ⊆ A is H ∗-Galois. Corollary 2.7. Let A be a left H-module algebra over a finite dimensional quantum groupoid H. (1) If A is a domain and monoform as left A#H-module or if A is critically compressible as left A#H-module, then dim(AAH ) ≤ dim(H) · udim(A). (2) If A has finite left uniform dimension and is a simple left A#H-module, then the following statements are equivalent: (a) A#H is simple; (b) AH ⊆ A is an H ∗-Galois extension; (c) A is a faithful left A#H-module; If A#HA is free of rank dim(H), then (a − c) are also equivalent to: (d) dim(AAH ) = dim(H). The associativity of A is not needed to prove Theorem 2.2. Non-associative examples of module algebras are given by module algebras over quasi-Hopf algebras. Let H be a quasi-Hopf algebra, that is H is an associative algebra which is a not necessarily coassociative coalgebra satisfying some compatibility conditions (see [9]). H acts on an algebra A if A is a unital algebra in the category of left H-modules (see [6]). In particular its multiplication satisfies (ab)c =P(x1 · a)[(x2 · b)(x3 · c)], for all a, b, c ∈ A, where φ−1 = x1 ⊗ x2 ⊗ x3 ∈ H ⊗ H ⊗ H is the inverse of the Drinfeld reassociator of H. We say that H acts finitely on A if Im(H → Endk(A)) is finite dimensional. By the proof of Theorem 2.2, substituting A by L(A) we get the following 6 IN ES BORGES AND CHRISTIAN LOMP Corollary 2.8. Let A be a left H-module algebra over a quasi-Hopf algebra H which acts finitely on it. If A has finite left uniform dimension and is a simple left A#H-module, then it is finite dimensional over AH . This applies in particular to finite quasi-Hopf action on non-associative division rings, which are now seen to be finite extensions of their (associative) subring of invariants. Quasi-Hopf actions on non-associative division rings were considered for example by Albuquerque and Majid in [1]. 3. Critically Compressible Actions The compressible condition on A in Theorem 2.2(2) is stronger than the monoform condition in 2.2(1). In the sequel of this section, we will examine how far they actually differ. This implies that we have to consider some more module theory. The following notions will be need: a left R-module M is called retractable if HomR(M, N ) 6= 0 for any non-zero submodule N of M , while M is called a fully retractable left R-module if for any g : N → M , HomR(M, N )g 6= 0 for any non-zero submodule N of M . Proposition 3.1. Let M be a left R-module. 1. If M is retractable the following statements are equivalent: (a) M is uniform and EndR(M ) is a domain; (b) M is uniform and any non-zero endomorphism is injective; (c) EndR(M ) is a left Ore domain. 2. Any retractable module M with EndR(M ) being a domain is compressible. 3. The following properties are equivalent: (a) M is critically compressible; (b) M is compressible and monoform; (c) M is fully retractable and EndR(M ) is a left Ore domain; (d) M is retractable, EndR(M ) is a left Ore domain and EndR(cM ) is isomorphic to the division ring of fractions of EndR(M ). Proof. (1.) (a) ⇔ (b) follows from [17, Theorem 1.4]. (a) ⇔ (c) Suppose 0 6= f, g ∈ EndR(M ). M is a uniform module and by definition the intersection of any two non-zero submodules of M have non-zero intersection which in particular results in N := Im(f ) ∩ Im(g) 6= 0. Since M is retractable there exists an endomorphism 0 6= h : M → N . Moreover, EndR(M ) is a domain which implies that any endomorphism belonging to EndR(M ) is injective. In particular h, f and g are monomorphisms. Let α : M → (N )f −1 such that (m)α = m′ if and only if (m)h = (m′)f . As f is injective, α is well-defined. Moreover, for all m ∈ M such that (m)h ∈ Im(f ) there exists an unique m′ ∈ M such that (m′)f = (m)h. The following composition satisfies (m)αf = (m′)f = (m)h, ∀m ∈ M . In an analogous way there exists a map β : M → (N )g−1 such that (m)β = m′ if and only if (m)h = (m′)g. The composition M α→ (N )f −1 f → N satisfy (m)βg = (m′)g = (m)h, ∀m ∈ M . For non-zero maps f, g ∈ EndR(M ) we proved the existence of non-zero endomorphisms α, β ∈ EndR(M ) such that the diagram M β → (N )g−1 g → N M β α / ____ JJ hJJJ (N )f −1 f (N )g−1 %JJJJJJ g / N commutes, i.e. αf = βg and hence EndR(M ) is a left Ore domain. Conversely assume that EndR(M ) is a left Ore domain and let N, L be a non-zero submodules of M such that N ∩ L = 0. Then Hom(M, N ) ∩ Hom(M, L) = Hom(M, N ∩ L) = 0. Since Hom(M, N ) and Hom(M, L) are left ideals of EndR(M ) and since EndR(M ) is uniform as left EndR(M )-module we have / % / IRREDUCIBLE ACTIONS AND COMPRESSIBLE MODULES 7 Hom(M, N ) = 0 or Hom(M, L)=0. But M is retractable which implies that N = 0 or L = 0. Hence M is uniform. (2.) Suppose that M is retractable and EndR(M ) is a domain. Then for any non-zero submodule N ⊆ M , there exists 0 6= f : M → N . Since HomR(M, Ker(f ))f = 0 and EndR(M ) a domain, HomR(M, Ker(f )) = 0. Hence Ker(f ) = 0 as M is retractable. (3.) (a) ⇔ (b) follows from [22, Proposition 1.1], while (a) ⇔ (c) follows from [17, Proposition 2.4] and (1). (a) ⇒ (d) follows from [22, 11.5(2)], while (d) ⇒ (a) holds since if f : N → M is any Hence also f is injective. (cid:3) non-zero endomorphism, then f can be extended to an endomorphism g of cM which is injective. A ring R is a compressible left R-module if and only if R is a domain, while R is a critically compressible left R-module if and only if it is a left Ore domain. Hence the free algebra K hx, yi in two indeterminantes over a field K is an example of a compressible module which is not critically compressible. Zelmanowitz claimed in [21] that any retractable uniform module whose non-zero endomor- phisms are injective would be critically compressible, but confirmed in [22] that he had no proof for this claim. As far as the authors know this claim has not been confirmed up to today. Some partial results on this question have been obtained in [12] and [17]. However we see from Proposi- tion 3.1 using Rodrigues and Sant'Ana's result [17] that a retractable module M is uniform with all non-zero endomorphisms injective if and only if EndR(M ) is a left Ore domain, while M being critically compressible is equivalent to M being fully retractable and EndR(M ) being a left Ore domain. Hence Zelmanowitz' claim is equivalent to the question whether a retractable module M with EndR(M ) being a left Ore domain is actually fully retractable. The following Lemma will be needed: Lemma 3.2 (Garcia Hernandes - Gomes Pardo). Let M be a left R-module. 1. M is fully retractable if and only if Hom(N/T r(M, N ), M ) = 0, for all N ⊆ M ; 2. A retractable self-projective module is fully retractable. Proof. [10, Proposition 1.1 and 1.2] (cid:3) Lemma 3.3. A retractable module M with EndR(M ) being a left Ore domain is critically com- pressible if M is self-projective or if M is a self-generator. Proof. Let M be a self-projective module. From Lemma 3.2 it follows that a self-projective retractable module is fully retractable. Since EndR(M ) is a left Ore domain, we might use Propo- sition 3.1 3.(c) ⇒ (a) to conclude that M is critically compressible. On the other hand, if M is a self-generator module then N = T r(M, N ), for all N ⊆ M . Thus Hom(N/T r(M, N ), M ) = 0 and by Lemma 3.2 it follows that M is fully retractable. Hence, from Proposition 3.1 3.(c) ⇒ (a) it follows that M is critically compressible. (cid:3) 3.1. Compressible operator actions. Let us suppose again that we have an intermediate al- gebra L(A) ⊆ B ⊆ End(A), where A is a k-algebra. Using the identification HomB(A, −) ∼= (−)B we see that A is a retractable left B-module if and only if AB ∩ I 6= 0 for any non-zero B-stable left ideal I of A. In the literature this property is also referred to as AB being large in A. Thus we reencounter a property that is for example known for semiprime PI-algebras A which have a large center, i.e. A is a retractable Ae-module. Another instance is Bergman and Isaacs theorem on finite group action which says that if a finite group G act on a semiprime ring A which is G-torsionfree, then AG is large in A, i.e. A is a retractable left A ∗ G-module. Recall that a multiplicatively closed subset S of right non-zero divisors of a ring R is called left permutable if Sa ∩ Rs is not empty, for every a ∈ R and s ∈ S. Additional to the module theoretic conditions that imply BA to be critically compressible we have the following: Lemma 3.4. Let L(A) ⊆ B ⊆ End(A). Then A is a critically compressible left B-module if AB is a domain and large in A such that (1) udim(A) = 1 or 8 IN ES BORGES AND CHRISTIAN LOMP (2) AB \ {0} is a left permutable set in A and consists of left non-zero divisors in A. Proof. (1) In order to prove that A is a critically compressible B-module it is enough to show that any partial endomorphism of A is injective. Let f : I → A be a non-zero B-linear map with 0 6= BI ⊆ BA and denote K := ker(f ) and J := (I)f . Now suppose K 6= 0. Since BA is retractable there exists an element 0 6= x ∈ AB ∩ K and an element 0 6= y ∈ AB ∩ J which implies that there exists i ∈ I such that y = (i)f . As A is uniform, Ax ∩ Ai 6= 0 and hence there exists z, t ∈ A such that zx = ti 6= 0. But ty = t(i)f = (ti)f = (zx)f = z(x)f = 0. Since 0 = ty = (t)Ry and Ry is a B-linear endomorphism of A, we have by Proposition 3.1 that Ry is injective and hence t = 0. Thus zx = ti = 0 which is a contradiction. Thus K = 0 and we conclude that f is injective. (2) Let f : I → A be a B-linear map with 0 6= BI ⊆ BA and denote K := ker(f ). In order to prove that A is a critically compressible B -module it is enough to show that f is a monomorphism. Since BA is retractable then for all 0 6= L ⊆ A, AB ∩ L 6= 0. In particular if K is a non-zero submodule of A, there exists an element 0 6= x ∈ AB ∩ K. Since AB \ {0} is left permutable set in A, we have that for all a ∈ I there exists y ∈ AB \ {0} and b ∈ A such that bx = ya. But 0 = b(x)f = (bx)f = (ya)f = y(a)f . By the hypthesis AB \ {0} consists of left non-zero divisor set in A. Thus (a)f = 0 or K = 0. Hence we conclude that f is injective. (cid:3) As an application we get a characterisation of the critically compressibility of A for central invariants AB ⊆ Z(A). Corollary 3.5. Let L(A) ⊆ B ⊆ End(A) such that AB ⊆ Z(A). Then A is a critically compress- ible B-module if and only if AB is an integral domain and large in A. Proof. By Proposition 3.1 the elements of AB are non-zero divisors on the right if BA is retractable and AB ≃ EndB(A) is a domain. Since AB ⊆ Z(A), the non-zero elements of AB are left and right non-zero divisors and form a left permutable set in A. By Lemma 3.4, A is critically compressible. The converse is clear by 3.1. (cid:3) 3.2. Group Actions and Group gradings. In this section we will apply our earlier results to group actions and group gradings. Proposition 3.6. Let G be a finite group acting as automorphism on A such that A is G- torsionfree. Then A is a critically compressible left A ∗ G-module if and only if AG is a left Ore domain and large in A. In this case udim(A) ≤ G and dim( A∆) ≤ G2 where bA is the self-injective hull of A and ∆ is isomorphic to the division ring of fractions of AG. Proof. We may assume that G is invertible in A, otherwise we localize A by the powers of G, If G−1 ∈ A then A a self-projective A ∗ G- obtaining an algebra in which G is invertible. module (see [19]). Since A is a retractable left A ∗ G-module and AG ∼= EndA∗G(A) is a left Ore domain, it follows from Lemma 3.3 that A is a critically compressible left A ∗ G-module. Moreover, from Proposition 3.1 1.c) ⇒ b) we know that A is an uniform A ∗ G-module which implies that udim(A∗GA) = 1. By [16, Lemma 1.3] we have udim(A) ≤ G · udim(A∗GA) and hence udim(A) ≤ G. Now using this fact, the relation dim( A∆) ≤ G2 follows directly from Theorem 2.2. Let G be a group. A G-graded k-algebra A is a k-algebra with a decomposition A =Lg∈G Ag into subspaces Ag, g ∈ G, such that AgAh ⊆ Agh, for all g, h ∈ G. For each g ∈ G let πg : A → Ag be the projection onto the g-component. Define B = hL(A) ∪ {πg g ∈ G}i ⊆ Endk(A). The left B-submodules are precisely the G-graded left ideals of A and if e denotes the neutral element of G, then Ae = AB is a subring of A isomorphic to EndB(A). Proposition 3.7. Let G be a group and A be a G-graded k-algebra. Then A is a critically compressible left B-module if and only if Ae is a left Ore domain and large in A. (cid:3) IRREDUCIBLE ACTIONS AND COMPRESSIBLE MODULES 9 Proof. We start by showing that a G-graded k-algebra A is a self-projective B-module. In order to prove that we will show that for any G-graded left ideal I of A we have HomB(A, A/I) = EndB(A)pI , where pI : A → A/I is the projection map. Since (Ae)pI = (Ae + I) /I ∼= Ae/ (I ∩ Ae) = Ae/Ie ∼= (A/I)e = (A/I)B and EndB(A) ∼= Ae, we have EndB(A)pI ∼= (Ae)pI = (A/I)B ∼= HomB(A, A/I). Since A is a self-projective B-module it follows from Lemma 3.3 that A is a critically compressible module. The converse follows from 3.1(3) and EndB(A) ≃ Ae. (cid:3) 3.3. Lie actions. A derivation δ of A is called locally nilpotent if for any a ∈ A there exists n > 0 such that δn(a) = 0. Recall Aδ = Ker(δ). For each non-zero element a ∈ A one defines then its degree (with respect to the derivation δ) as deg(a) = min{n ∈ N0 : δn+1(a) = 0}. Lemma 3.8. Let A be a k-algebra and let δ be a locally nilpotent derivation of A. Then A is a retractable A[z, δ]-module. Proof. By the correspondence HomA[z:δ](A, −) ≃ (−)δ we have to show that every non-zero δ- stable left ideal I of A contains a non-zero constant element, i.e. I ∩ Aδ 6= 0. Since δ is locally nilpotent, for any 0 6= a ∈ I with n = deg(a), we have 0 6= δn(a) ∈ Aδ ∩ I. (cid:3) We will show that if A is an algebra over a field k with a locally nilpotent derivation δ having an element x ∈ A such that δ(x) = 1, then A is a self-projective A[z; δ]-module if char(k) = 0 or δp = 0 in case char(k) = p > 0. In order to do so we first show that any element of A can be written as polynomial in x with coefficients in AB (see also [2]). Lemma 3.9. Let A be an algebra over a field k and δ a locally nilpotent derivation of A or δp = 0 in case of char(k) = p such that δ(x) = 1, for some x in A. Then any element a ∈ A of degree n can be written as a =Pn i=0 cixi, for some ci ∈ Aδ. Proof. Let 0 6= a ∈ A and let n = deg(a). Note that if char(k) = p and δp = 0 then n < p. We will use induction on n: If n = 0 then a ∈ Aδ and there is nothing to prove. Suppose that n > 0. We claim that deg(a − 1 δn(a) − 1 Moreover, for any y ∈ A and c ∈ Aδ we have δ(cy) = cδ(y). In particular, n! δn(a)xn) < n. To see this note that δn(cid:0)a − 1 n! δn(δn(a)xn). Since deg(a) = n, δ(δn(a)) = δn+1(a) = 0 which implies that δn(a) ∈ Aδ. n! δn(a)xn(cid:1) = 1 n! δn(cid:18)a − δn(a)xn(cid:19) = δn(a) − exist c0, c1, . . . , cm ∈ Aδ such that a′ =Pm Hence deg(a − 1 n! δn(a)xn) < n. Now suppose that for all a′ ∈ A such that m = deg(a′) < n, there n! δn(a)xn) < n, by the induction i=0 cixi, equivalently (cid:3) hypothesis there are elements c0, . . . , cn−1 ∈ Aδ with a − 1 a = 1 i=0 cixi. As deg(a − 1 n! δn(a)xn = Pm 1 n! δn(a)δn(xn) = δn(a) − 1 n! δn(a)n! = 0. i=0 cixi, where 1 n! δn(a) ∈ Aδ. Proposition 3.10. Let A be an algebra over a field k and δ a locally nilpotent derivation of A or δp = 0 in case of char(k) = p such that δ(x) = 1, for some x in A. Then A is a self-projective A[z, δ]-module. n! δn(a)xn +Pm Proof. We have to prove that A is a self projective A[z, δ]-module which is equivalent to prove that (A/I)δ = Aδ +I , for all δ-stable left ideals I of A. Thus we need to show that I ∀a ∈ A : a + I ∈ (A/I)δ ⇒ ∃y ∈ Aδ : a + I = y + I. Since (A/I)δ = {a + I : δ(a) + I = 0 + I} = {a + I : δ(a) ∈ I} the above condition can be rewrit- ten as (2) ∀a ∈ A : δ(a) ∈ I ⇒ ∃y ∈ Aδ : a − y ∈ I. We wil prove this by induction on the degree of a. Let a ∈ A be such that δ(a) ∈ I. If deg(a) = 0 then a ∈ Aδ and thus y = a is a solution for (2). Suppose that deg(a) = n > 0 and that for all 10 IN ES BORGES AND CHRISTIAN LOMP b ∈ A such that deg(b) < n and δ(b) ∈ I, there exists a solution y ∈ Aδ such that b − y ∈ I. i=0 cixi with ci ∈ Aδ and n = deg(a). The derivation of a is given by i=0 ciixi−1. In particular, if we consider b = a − 1 n xδ(a) it follows that By Lemma 3.9, a = Pn i=0 ciδ(xi) =Pn δ(a) =Pn δ(xδ(a)) (δ(x)δ(a) + xδ2(a)) δ(b) = δ(a) − = δ(a) − = δ(a) − 1 n 1 n 1 n δ(a) − xδ2(a) 1 n 1 xδ2(a) n = (1 − 1 n )δ(a) − which belongs to I since δ(a) ∈ I and I is δ-stable. In order to show that deg(b) < n we will write b using the above equalities for a and δ(a). Denote the commutator of two elements r, s ∈ A by [r, s] = rs − sr. Thus xi−1(cid:19) − xcnxn−1 (cix + [x, ci]) i n xi−1 − cnxxn−1 − [x, cn]xn−1 b = = = = = cixi − 1 n x nXi=1 cixi + cnxn − cixi + cnxn − ciixi−1! n−1Xi=1(cid:18)xci n−1Xi=1 i n ci i n cixi − n−1Xi=1 n(cid:19) xi − ci(cid:18)1 − i n−1Xi=1 nXi=1 i n nXi=0 n−1Xi=0 n−1Xi=0 n−1Xi=0 n−1Xi=0 xi − [x, ci] i n xi−1 − [x, cn]xn−1 [x, ci]xi−1. Hence deg(b) < n since [x, ci] ∈ Aδ for all ci ∈ Aδ. By induction, there exist y ∈ Aδ such that b − y ∈ I ⇔ a − 1 (cid:3) n xδ(a) ∈ I, a − y ∈ I, as required. n xδ(a) − y ∈ I. Since 1 Corollary 3.11. Let A be an algebra over a field k and δ a locally nilpotent derivation of A or δp = 0 in case of char(k) = p and suppose that there exists x ∈ A such that δ(x) = 1. Then Aδ is a left Ore domain if and only if A is a critically compressible A[z, δ]-module. Proof. Suppose that Aδ is a left Ore domain. From Proposition 3.10, A is a self-projective A[z, δ]- module. Since δ is locally nilpotent, A is a rectratable A[z, δ]-module. From Proposition 3.2 2. it follows that A is fully retractable as A[z, δ]-module. Hence, by Proposition 3.1 3., A is a critically compressible A[z, δ]-module if and only if Aδ is a left Ore domain. (cid:3) Actually A is a domain in case char(k) = 0 and A[z,δ]A is critically compressible. Hence in this case condition (2) of Theorem 2.2 implies condition (1). Proposition 3.12. Let A be an algebra over a field k of characteristic zero and δ be a locally nilpotent derivation of A such that δ(x) = 1, for some x in A. Then the following statements are equivalent: (a) Aδ is a left Ore domain; (b) A is a left Ore domain; (c) A is a critically compressible A[z, δ]-module. Proof. (a) ⇔ (c) follows from Corollary 3.11. (a) ⇒ (b) Let Aδ be a left Ore domain. By Lemma 2.3, Aδ is a monoform left Aδ-module. Now from [2, Lemma 2.1.] it follows that A = Aδ [x, d] is the Ore extension with coefficients from the algebra Aδ and where the derivation d of the algebra Aδ is the restriction of the inner derivation of the algebra A to its subalgebra Aδ defined by d(a) := [x, a], for all a ∈ A. Thus by [3, Proposition IRREDUCIBLE ACTIONS AND COMPRESSIBLE MODULES 11 1.3.] we have that A = Aδ[x, d] is a monoform left A-module. Finally, by Lemma 2.3 A is a left Ore domain. (b) ⇒ (c) Let A be a left Ore domain, then by Lemma 2.3, A is a monoform A[z, δ]-module. In particular A is a uniform A[z, δ]-module. By 3.8, A is a retractable A[z, δ]-module and hence by Proposition 3.1 EndA[z,δ](A) ≃ Aδ is a left Ore domain. (cid:3) However in positiv characteristic A might not be even reduced. Examples 3.13. Let A = k [x] be the commutative polynomial ring in one variable, where k is a field of characteristic p > 0. Consider the locally nilpotent derivation δ of A defined by δ = ∂ ∂x . Now let I = hxpi. Since char(k) = p we have δ(xp) = pxp−1 = 0. Moreover, for all f (x) ∈ A, δ(f (x) · xp) = δ(f (x)) · xp ∈ I. Hence I is δ-stable and we can define a derivation ¯δ in ¯A = A/I such that ¯δ( ¯f (x)) = δ(f (x)) + I. In particular, for ¯x we have ¯δ(¯x) = δ(x) + I = ¯1. On the other ¯f (x) ∈ ¯A. Since ¯A¯δ = k + I ∼= k is a field, ¯A is a simple left ¯A[z; δ]-module hand ¯δp(f ) = 0, for all by Lemma 3.8, but ¯A is not a domain (not even reduced). References [1] H.Albuquerque and S.Majid, Clifford Algebras Obtained by Twisting of Group Algebras, J. Pure Appl. Alg. 171 (2002) [2] V.V.Bavula, The inversion formula for automorphisms of the Weyl algebras and polynomial algebras, J. Pure Appl. Alg. 210, 147-159 (2007) [3] A.D.Bell and K.R.Goodearl, Uniform Rank over differencial operator rings and Poincar?-Birkhoff-Witt ex- tensions, Pac. J. Math. 131, 1 (1988) [4] J.Bergen, M.Cohen and D.Fishman, Irreducible Actions and faithful actions of Hopf algebras, Israel J. Math. 72(1-2), 5-18 (1990) [5] G.Bohm, F.Nill and K.Szlach´anyi, Weak Hopf algebras. I. Integral theory and C ∗-structure, J. Alg. 221 (2), 385 -- 438 (1999) [6] D.Bulacu, F.Panaite and F.Van Oystaeyen, Quasi-Hopf algebra actions and smash products, Comm. Algebra 28, 631-651 (2000) [7] S.Caenepeel and E.De Groot, Galois Theory for weak Hopf Algebras, Rev. Roumaine Math. Pures. Appl., 52 (2), 151?176 (2007) [8] M.Cohen, D.Fishman and S.Montgomery, Hopf Galois extensions, smash products and Morita equivalence., J. Alg. 133 (2), 351-372 (1990) [9] V.Drinfeld, Quasi-Hopf algebras, Leningrad Math. J. 1, 1419-1457 (1989) [10] J.L. Garcia Hernandez and J.L. Gomez Pardo, S elf-injective and PF endomorphism rings, Israel J. Math. 58, 324-350(1987) [11] K.Goodearl and R.B.Warfield, An Introduction to Noncommutative Noetherian Rings, London Math. Soc, Student Texts 61, Cambridge University Press (2004) [12] J-W.Jeong, On critically compressible modules, Master thesis, Kyungpook National University, 1998. [13] L. Kadison and D. Nikshych, Frobenius extensions and weak Hopf algebras, J. Alg. 244 (2001), 312-342. [14] D.Nikshych, Semisimple weak Hopf algebras, J. Alg. 275(2), 639 -- 667 (2004) [15] D.Nikshych and L.Vainerman, Finite quantum groupoids and their applications, in New directions in Hopf algebras. Montgomery, Susan (ed.) et al., Cambridge University Press. Math. Sci. Res. Inst. Publ. 43, 211-262 (2002) [16] D.S.Passman, It is Essentially Maschke's Theorem, Rocky Montain J. Math. 13(1), (1983) [17] V.Rodrigues and A.Sant'Ana, A note on a problem due to Zelmanowitz, Alg. Discr. Math. 3, 85-93 (2009) [18] J-M. Vallin, Actions and coactions of finite quantum groupoids on von Neumann algebras, extensions of the matched pair procedure, J. Alg. 314(2) 789-816 (2007) [19] R.Wisbauer, Modules and algebras: Bimodule structure and group actions on algebras, Pitman Monographs and Surveys in pure and applied Mathematics 81, Longman (1996) [20] R.Wisbauer, Foundations of Module and Ring Theory, Gordon and Breach Science Publishers, (1991) [21] J.M. Zelmanowitz, An extension of the Jacobson Density Theorem, Bull. Amer. Math. Soc. 82(4), 551-553 (1976) [22] J.M. Zelmanowitz, Weakly primitive rings, Comm. Alg., 9(1), 23-45 (1981) [23] J.M. Zelmanowitz, Representations of rings with faithful monoform modules, J. London Math. Soc. 29(2), 237-248 (1984) Instituto Superior De Contabilidade e Administrac¸ ao de Coimbra, Quinta Agricola - Bencanta, 3040-316 Coimbra, Portugal Centro de Matemtica de Universidade do Porto, Rua Campo Alegre 687, 4169-007 Porto, Portugal
1009.1218	3	1009	2011-04-13T03:56:50	Gradings on the exceptional Lie algebras $F_4$ and $G_2$ revisited	[ "math.RA" ]	All gradings by abelian groups are classified on the following algebras over an algebraically closed field of characteristic not 2: the simple Lie algebra of type $G_2$ (characteristic not 3), the exceptional simple Jordan algebra, and the simple Lie algebra of type $F_4$.	math.RA	math	GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED ALBERTO ELDUQUE⋆ AND MIKHAIL KOCHETOV⋆⋆ Abstract. All gradings by abelian groups are classified on the following al- gebras over an algebraically closed field F: the simple Lie algebra of type G2 (char F 6= 2, 3), the exceptional simple Jordan algebra (char F 6= 2), and the simple Lie algebra of type F4 (char F 6= 2). 1. Introduction Gradings on Lie algebras have been extensively used since the beginning of Lie theory: the Cartan grading on a complex semisimple Lie algebra is the Zr-grading (r being the rank) whose homogeneous components are the root spaces relative to a Cartan subalgebra (which is the zero component); symmetric spaces are related to Z2-gradings, Kac -- Moody Lie algebras to gradings by a finite cyclic group, the theory of Jordan algebras and pairs to 3-gradings on Lie algebras, etc. In 1989, a systematic study of gradings on Lie algebras was started by Patera and Zassenhaus [PZ89]. Fine gradings (i.e., those that cannot be refined) on the clas- sical simple complex Lie algebras other than D4 by arbitrary abelian groups were considered in [HPP98]. The arguments there are computational and the problem of classification of fine gradings is not completely settled. The complete classification, up to equivalence, of fine gradings on all classical simple Lie algebras (including D4) over algebraically closed fields of characteristic 0 has recently been obtained in [Eld10]. For any abelian group G, the classification of all G-gradings, up to isomorphism, on the classical simple Lie algebras other than D4 over algebraically closed fields of characteristic different from 2 has been achieved in [BK10] using methods developed in [BSZ01, BZ02, BZ03, BShZ05, BZ06, BZ07, BKM09]. As to the exceptional simple Lie algebras, the classification of all gradings (up to equivalence) for type G2 over an algebraically closed field of characteristic 0 was obtained independently in [DM06] and [BT09], using the results on gradings on the Cayley algebras in [Eld98]. Also, the classification of fine gradings (up to equivalence) for type F4 over an algebraically closed field of characteristic 0 has recently been obtained in [DM09] (see also [Dra]). The method used in that work relies on the fact that, under the stated assumptions on the ground field, any abelian group grading on an algebra is the decomposition into common eigenspaces for some diagonalizable subgroup of the automorphism group of the algebra. It is 2000 Mathematics Subject Classification. Primary 17B70, secondary 17D05, 17C40. Key words and phrases. Graded algebra, fine grading, Weyl group, octonions, Albert algebra. ⋆ Supported by the Spanish Ministerio de Educaci´on y Ciencia and FEDER (MTM 2007- 67884-C04-02) and by the Diputaci´on General de Arag´on (Grupo de Investigaci´on de ´Algebra). ⋆⋆Supported by the Natural Sciences and Engineering Research Council (NSERC) of Canada, Discovery Grant # 341792-07. 1 2 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV shown that any such subgroup is contained in the normalizer of a maximal torus of the automorphism group. Starting from this point, the argument is quite technical, and some computer-aided case-by-case analysis is used. Since the automorphism groups of the simple Lie algebra of type F4 and of the exceptional simple Jordan algebra (the Albert algebra) are isomorphic, in [DM09] the fine gradings on the Albert algebra are computed as well. These methods are being currently used by C. Draper and C. Mart´ın-Gonz´alez to study gradings on the simple Lie algebra of type E6. The purpose of this paper is the classification of gradings on the simple Lie algebras of types G2 and F4 over algebraically closed fields of characteristic different from 2 (and different from 3 for type G2, as there is no simple Lie algebra of type G2 in characteristic 3). Actually, for G2 the situation is simple enough to obtain a description of gradings without assuming the ground field algebraically closed. Our arguments will differ essentially from the arguments in [DM06, BT09, DM09], which depend heavily on the characteristic being 0. The idea is to classify gradings on the Cayley algebra and on the Albert algebra first, and then use automorphism group schemes to transfer the classification to the corresponding Lie algebras. All gradings on the Cayley algebras over an arbitrary field were described in [Eld98], using, essentially, only the properties of the norm and trace. All gradings on the Albert algebra over an algebraically closed field of characteristic different from 2 will be described here, using the well-known properties of this exceptional Jordan algebra. In this way, not only the results on the gradings on the Albert algebra in [DM09] will be extended to positive characteristic, but also the gradings will be described intrinsically, according to structural properties of the Albert algebra and the identity component of the grading. In particular, we obtain an interesting model of the Albert algebra based on the fine Z×Z3 2-grading and the Cayley algebra and another model based on the fine Z3 3-grading and the Okubo algebra -- see (10) and (12), respectively. Once this is done, general arguments with morphisms of affine group schemes (already used in [BK10]) will be applied to show that any grading on the simple Lie algebra of type G2 or F4 is induced from a grading on the Cayley or the Albert algebra, respectively. Our desire to cover characteristic 3 for type F4 has forced us to extend some classical results which, to the best of our knowledge, have appeared in the literature only assuming characteristic different from 2 and 3 (see Propositions 8.1 and 8.2). In Section 2, we collect the basic definitions and properties related to gradings, including their relationship with automorphism group schemes. Section 3 is devoted to a review of the description of gradings on the Cayley algebras in [Eld98] in a way suitable for our purposes; we also obtain, for any abelian group G, a classification of G-gradings up to isomorphism (over an algebraically closed field). These results are applied in Section 4 to describe all gradings on central simple Lie algebras of type G2 over an arbitrary field of characteristic different from 2 and 3, and to classify the gradings up to equivalence and up to isomorphism, assuming the field algebraically closed. Then, in Section 5, the Albert algebra is described, and some subgroups of its automorphism group are considered. Section 6 gives constructions of four fine gradings on the Albert algebra over an algebraically closed field of characteristic different from 2 (one of them does not exist in characteristic 3). In Section 7, these gradings are shown to exhaust the list of fine gradings, up to equivalence. We also obtain, for any abelian group G, a classification of G-gradings up to isomorphism. GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 3 Finally, in Section 8, all gradings on the simple Lie algebra of type F4 are classified under the same assumptions on the ground field. 2. Gradings In this section, we state some basic definitions and facts concerning gradings on (nonassociative) algebras. We also fix the notation that will be used throughout the paper. The reader may consult [Koc09] for a survey of results on gradings on Lie algebras. 2.1. Some definitions. Let A be an algebra over a ground field F. A grading on A is a decomposition Γ : A =Ms∈S As of A into a direct sum of subspaces, called the homogeneous components, such that As2 ⊂ As3 . If a ∈ As, we will say for any s1, s2 ∈ S there exists s3 ∈ S with As1 that a is homogeneous of degree s and write deg a = s. Then: • If A is finite-dimensional, let ni be the number of homogeneous components of dimension i, i = 1, . . . , r, where r is the highest dimension that occurs. (Hence i=1 ini.) The type of Γ is the sequence (n1, n2, . . . , nr). dim A =Pr • Two gradings Γ : A = Ls∈S s′ are said to be equivalent if there exist an isomorphism ψ : A → A′ and a bijection α : S → S′ such that for any s ∈ S we have ψ(As) = A′ As and Γ′ : A′ = Ls′∈S′ A′ α(s). • Let Γ and Γ′ be two gradings on A. The grading Γ is said to be a refinement of Γ′ (or Γ′ a coarsening of Γ) if, for any s ∈ S, there exists s′ ∈ S′ such that As ⊂ As′ . In other words, each homogeneous component of Γ′ is a (direct) sum of some homogeneous components of Γ. A grading is called fine if it admits no proper refinement. • The grading Γ is said to be a group grading (respectively, an abelian group grading) if there is a group (respectively, abelian group) G containing S such that, As2 ⊂ As1s2 , with the multiplication of s1 and s2 in for all s1, s2 ∈ S, we have As1 G. Setting Ag := 0 if g /∈ S, we have Γ : A =Mg∈G Ag where Ag Ah ⊂ Agh for all g, h ∈ G. This is what is called a G-grading on A. A group grading (respectively, abelian group grading) is said to be fine if it admits no proper refinement in the class of group gradings (respectively, abelian group gradings). We will also consider G- gradings on a vector space V , which are just direct sum decompositions of the form be called the support of Γ and denoted by Supp Γ (or Supp V if the grading is clear V =Lg∈G Vg. • Given a G-grading Γ : V = Lg∈G Vg, the subset {g ∈ G Ag 6= 0} of G will from the context). A subspace W ⊂ V is said to be graded if W =Lg∈G Wg where • Given a grading Γ : A =Ls∈S As3. Then we obtain a G-grading: A = Lg∈G0 As, we define the (abelian) group G0 generated As2 ⊂ Ag where Ag is the sum of the by {s ∈ S As 6= 0} subject only to the relations s1s2 = s3 whenever 0 6= As1 Wg = Vg ∩ W . Then we can speak of the support of W . 4 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV homogeneous components As such that the class of s in G is g. In general, this is a coarsening of Γ. If Γ is a group grading (respectively, an abelian group grading), Ag coincides with Γ. The group G0 has the following universal property: given any (abelian) group grading Ah that is a coarsening of Γ, there exists a unique homomorphism of Ag. The group G is called the universal (abelian) group of Γ and denoted U (Γ). The universal (abelian) groups of two equivalent gradings are isomorphic. then S imbeds in G0 and the grading A = Lg∈G0 A = Lh∈H groups α : G0 → H such that Ah = Lg∈α−1(h) • Given a G-grading Γ : A =Lg∈G we obtain an H-grading A =Lh∈H Ah where Ah =Lg∈α−1(h) • Two G-gradings over the same group, Γ : A = Lg∈G Lg∈G Ag and a group homomorphism α : G → H, Ag. This H-grading will be denoted by αΓ and said to be induced by α from Γ. Clearly, αΓ is coarsening of Γ (not necessarily proper). A′ Ag and Γ′ : A′ = g, are said to be isomorphic if there is an isomorphism ψ : A → A′ such that ψ(Ag) = A′ g for all g ∈ G. A G-grading Γ : A = ⊕g∈GAg and an H-grading Γ′ : A′ = ⊕h∈H A′ h are said to be weakly isomorphic if there are isomorphisms α : G → H and ψ : A → A′ such that, for all g ∈ G, we have ψ(Ag) = A′ α(g). This is equivalent to saying that Γ′ is isomorphic to αΓ. It is clear that weakly isomorphic gradings are equivalent, but the converse does not hold in general. However, two equivalent (abelian) group gradings are weakly isomorphic when considered as gradings by their universal (abelian) groups. • The automorphism group of Γ, denoted Aut(Γ), consists of all self-equivalences of Γ, i.e., automorphisms of A that permute the components of Γ. The stabilizer of Γ, denoted Stab(Γ), consists of all automorphisms of the graded algebra A, i.e., automorphisms of A that leave each component of Γ invariant. The diagonal group of Γ, denoted Diag(Γ), is the subgroup of the stabilizer consisting of all automorphisms ϕ such that the restriction of ϕ to any homogeneous component of Γ is the multiplication by a (nonzero) scalar. The quotient group Aut(Γ)/ Stab(Γ), which is a subgroup of Sym(Supp Γ), will be called the Weyl group of Γ and denoted by W (Γ). Each element of W (Γ) extends to a unique automorphism of U (Γ), so W (Γ) can be regarded as a subgroup of Aut(U (Γ)). For example, suppose A is a finite-dimensional algebra over an algebraically closed field and T is a maximal torus in the algebraic group Aut(A). Then the eigenspace decomposition Γ of A relative to T is a X(T )-grading on A where X(T ) is the group of regular characters of T . Let N (T ) be the normalizer of T in Aut(A) and let C(T ) be the centralizer. It is easy to see that T is the connected component of Diag(Γ), Aut(Γ) is N (T ), and Stab(Γ) is C(T ). Hence W (Γ) is W (T ) := N (T )/C(T ). This justifies our use of the term "Weyl group" for W (Γ). Unless stated otherwise, the term grading in this paper will always refer to an abelian group grading, and universal group to universal abelian group. 2.2. Gradings and automorphism group schemes. For background on group schemes the reader may consult [Wat79] or [KMRT98, Chapter VI]. It is well-known that a G-grading Γ on a vector space V is equivalent to a comodule structure ρΓ : V → V ⊗ FG, which is defined by setting ρΓ(v) = v ⊗ g for all v ∈ Vg and g ∈ G. Since G is abelian, the Hopf algebra FG is commutative GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 5 and thus represents an affine group scheme, which we denote by GD. Affine group schemes of this form are called diagonalizable. G can be identified with the group of characters of GD, i.e., morphisms from GD to GL1. If V is finite-dimensional, then ρΓ is equivalent to a morphism ηΓ : GD → GL(V ), i.e., a linear representation of GD on V . If we pick a homogeneous basis {v1, . . . , vn} in V , degΓ(vi) = gi, Γ : F[Xij, det(Xij )−1] → FG can be then the comorphism of representing objects η∗ written explicitly as follows: Xij 7→ δijgi, i, j = 1, . . . , n. In particular, ηΓ is a closed imbedding if and only if η∗ Γ is onto if and only if Supp Γ generates G. If A is a finite-dimensional (nonassociative) algebra, then the automorphism group scheme Aut(A) is defined as follows. For any unital commutative associative F-algebra R, the tensor product A ⊗ R is an R-algebra, and we set Aut(R) := AutR(A ⊗ R). Equivalently, Aut(A) is the subgroupscheme StabGL(A)(µ) where µ : A ⊗ A → A is the multiplication map, which is to be regarded as an element of Hom(A ⊗ A, A) where GL(A) acts in the standard way. If Γ is a G-grading on an algebra A, then the multiplication map µ : A ⊗ A → A is a morphism of GD-representations, which is equivalent to saying that GD stabilizes µ, or that the image of ηΓ : GD → GL(A) is a subgroupscheme of Aut(A). Conversely, a morphism η : GD → Aut(A) gives rise to a G-grading Γ on the algebra A such that ηΓ = η. For any unital commutative associative F-algebra R, the action of R-points of GD by automorphisms of the R-algebra A ⊗ R can be written explicitly: (1) (ηΓ)R(f )(x ⊗ r) = x ⊗ f (g)r for all x ∈ Ag, r ∈ R, g ∈ G, f ∈ Alg(FG, R). A group homomorphism α : G → H gives rise to a morphism αD : H D → GD. Then ραΓ = (id ⊗ α) ◦ ρΓ implies that η αΓ = ηΓ ◦ αD. Now if B is another algebra and we have a morphism θ : Aut(A) → Aut(B), then any G-grading Γ on A induces a G-grading on B via the morphism θ◦ηΓ : GD → Aut(B). We will denote the induced grading by θ(Γ). Clearly, θ(αΓ) = α(θ(Γ)). The group Aut(A) of the F-points of Aut(A) acts by automorphisms of Aut(A) via conjugation. Namely, ϕ ∈ Aut(A) defines a morphism Ad ϕ : Aut(A) → Aut(A) as follows: (2) (Ad ϕ)R(f ) := (ϕ ⊗ id) ◦ f ◦ (ϕ−1 ⊗ id) Comparing (1) and (2), we see that Ad ϕ(Γ) is the grading A = Lg∈G ϕ(Ag). To summarize: for all f ∈ AutR(A ⊗ R). Proposition 2.1. The G-gradings on A are in one-to-one correspondence with the morphisms of affine group schemes GD → Aut(A). Two G-gradings are isomorphic if and only if the corresponding morphisms are conjugate by an element of Aut(A). The weak isomorphism classes of gradings on A with the property that the support generates the grading group are in one-to-one correspondence with the Aut(A)-orbits of diagonalizable subgroupschemes in Aut(A). (cid:3) Let Γ be an abelian group grading on A. Define the subgroupscheme Diag(Γ) of Aut(A) as follows: Diag(Γ)(R) := {f ∈ AutR(A ⊗ R) f Ag ⊗ R ∈ R×idAg ⊗ R for all g ∈ G}. Since Diag(Γ) is a subgroupscheme of a torus in GL(A), it is diagonalizable, so Diag(Γ) = U D for some finitely generated abelian group U . If Γ is realized as a 6 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV G-grading, then (1) shows that the image of the imbedding ηΓ : GD → Aut(A) is a subgroupscheme of Diag(Γ). The imbedding GD → Diag(Γ) corresponds to an epimorphism U → G. We conclude that U satisfies the definition of the universal abelian group of Γ and hence Diag(Γ) = U (Γ)D. Let Γ and Γ′ be two abelian group gradings on A and let Q = Diag(Γ) and Q′ = Diag(Γ′). Now Γ is a refinement of Γ′ if and only if Γ′ = αΓ for some epimorphism α : U (Γ) → U (Γ′) if and only if ηΓ′ = ηΓ ◦ αD. Hence we obtain Γ′ is a coarsening of Γ ⇔ Q′ is a subgroupscheme of Q. It follows that fine gradings correspond to maximal diagonalizable subgroupschemes of Aut(A). To summarize: Proposition 2.2. The equivalence classes of fine gradings on A are in one-to-one correspondence with the Aut(A)-orbits of maximal diagonalizable subgroupschemes in Aut(A). (cid:3) As a consequence of the descriptions in Propositions 2.1 and 2.2, we obtain the following results, which will be used to transfer the classification of gradings from the algebra of octonions to the simple Lie algebra of type G2 and from the Albert algebra to the simple Lie algebra of type F4. Theorem 2.3. Let A and B be finite-dimensional (nonassociative) algebras. As- sume we have a morphism θ : Aut(A) → Aut(B). Then, for any abelian group G, we have a mapping, Γ → θ(Γ), from G-gradings on A to G-gradings on B. If Γ and Γ′ are isomorphic (respectively, weakly isomorphic), then θ(Γ) and θ(Γ′) are isomorphic (respectively, weakly isomorphic). Proof. We have already defined θ(Γ). Let ϕ ∈ Aut(A) and ψ = θF(ϕ). Then the following diagram commutes: Aut(A) Ad ϕ Aut(A) θ θ Aut(B) Ad ψ / Aut(B) This follows immediately from (2) and the equation θR(ϕ ⊗ id) = ψ ⊗ id, which is a consequence of the naturality of θ. Now if ϕ sends Γ to Γ′ (respectively, αΓ to Γ′), then ψ sends θ(Γ) to θ(Γ′) (cid:3) (respectively, θ(αΓ) = α(θ(Γ)) to θ(Γ′)). Theorem 2.4. Let A and B be finite-dimensional (nonassociative) algebras. As- sume we have an isomorphism θ : Aut(A) → Aut(B). Let Γ be a G-grading on A such that G is its universal abelian group. Then Γ is a fine abelian group grading if and only if so is θ(Γ). Also, two such fine abelian group gradings, Γ and Γ′, are equivalent if and only if θ(Γ) and θ(Γ′) are equivalent. Proof. If Γ is fine, then the image of ηΓ : GD → Aut(A) is a maximal diagonalizable subgroupscheme of Aut(A). Hence the image of ηθ(Γ) = θ ◦ ηΓ is a maximal diagonalizable subgroupscheme of Aut(B) and so θ(Γ) is fine. It remains to recall that, if universal groups are used, two fine gradings are equivalent if and only if they are weakly isomorphic, so we can apply Theorem 2.3. (cid:3) / / / GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 7 2.3. Gradings on Lie algebras of derivations. Recall that, for any algebraic affine group scheme G, we have the adjoint rep- [KMRT98, §21]. The differential of resentation Ad : G → GL(cid:0)Lie(G)(cid:1), see e.g. Ad is ad : Lie(G) → gl(cid:0)Lie(G)(cid:1), the adjoint representation of Lie(G). The image of Ad is contained in the subgroupscheme Aut(Lie(G)) of GL(cid:0)Lie(G)(cid:1), and the image of ad is contained in Der(cid:0)Lie(G)(cid:1). For G = Aut(A), we have Lie(G) = Der(A). Hence, given a G-grading Γ on A, we get an induced G-grading Ad (Γ) on Der(A) by Theorem 2.3. Since Ad in this case is the composition of the closed imbedding Aut(A) → GL(A) and the standard action GL(A) → GL(Hom(A, A)), the grading Ad (Γ) is given by the standard FG- comodule structure on Hom(A, A), which is determined by the requirement that the evaluation map, ev : Hom(A, A) ⊗ A → A, be a homomorphism of FG-comodules. This implies that the induced G-grading Ad (Γ) on Der(A) is the natural one: Der(A) =Lg∈G Der(A)g where Der(A)g = {d ∈ Der(A) d(Ah) ⊂ Agh for all h ∈ G}. If we know that Ad : Aut(A) → Aut(L) is an isomorphism, Let L = Der(A). then every G-grading on L is induced from a unique G-grading on A in this way, and we can transfer the classification of gradings from A to L via Theorems 2.3 and 2.4. Let F be the algebraic closure of the ground field F. In order for Ad to be an isomorphism of affine group schemes, the following conditions are necessary: 1) Ad F : AutF(A ⊗ F) → AutF(L ⊗ F) is a bijection; 2) ad : L → Der(L) is a bijection. If char F = 0, then condition 1) alone is sufficient. If char F = p, even the com- bination of both conditions does not imply, in general, that Ad is an isomor- phism. Recall that an algebraic affine group scheme G is smooth if and only if dim Lie(G) = dim G (see e.g. [KMRT98, §21]). The dimension of G coincides with the dimension of the algebraic group G(F). Hence, for G = Aut(A), smoothness is equivalent to the condition dim Der(A) = dim AutF(A ⊗ F). If Aut(A) is smooth, then the combination of 1) and 2) does imply that Ad is an isomorphism of affine group schemes -- see e.g. [KMRT98, (22.5)] and observe that, under conditions 1) and 2), the smoothness of Aut(A) implies the smoothness of Aut(L). 3. Gradings on Cayley algebras The aim of this section is to present the known results about gradings on Cayley algebras in a way that will be convenient for our study of gradings on the Albert algebra. We also obtain, for an arbitrary abelian group G, a classification of G- gradings up to isomorphism on the (unique) Cayley algebra over an algebraically closed field. Throughout this section, the ground field F will be arbitrary, unless stated otherwise. A Cayley algebra C over F is an eight-dimensional unital composition algebra. Then, there exists a nondegenerate quadratic form (the norm) n : C → F such that n(xy) = n(x)n(y) for any x, y ∈ C. Here the norm being nondegenerate means that its polar form: n(x, y) = n(x + y) − n(x) − n(y) is a nondegenerate symmetric bilinear form. 8 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV e1 e1 0 0 0 0 v1 v2 v3 e1 e2 u1 u2 u3 v1 v2 v3 e2 0 e2 u1 u1 0 u2 u2 0 u3 u3 0 v1 0 v1 0 u1 u2 −v3 u3 v2 −v1 v3 −v2 −e1 0 0 v2 0 v2 0 −e1 v3 0 v3 0 0 0 −e1 v1 0 0 0 0 −e2 0 0 0 −e2 0 0 0 0 0 −e2 u3 −u2 u1 0 0 −u3 u2 −u1 Figure 1. Multiplication table of the Cayley algebra The next result summarizes some of the well-known properties of these algebras (see [KMRT98, Chapter VIII] and [ZSSS82, Chapter 2]): Proposition 3.1. Let C be a Cayley algebra over F. Then: 1) Any x ∈ C satisfies the degree 2 Cayley-Hamilton equation: (3) x2 − n(x, 1)x + n(x)1 = 0. 2) The map x 7→ ¯x = n(x, 1)1 − x is an involution, called the standard con- jugation, of C and for any x, y, z ∈ C, x¯x = ¯xx = n(x)1 and n(xy, z) = n(y, ¯xz) = n(x, z ¯y) hold. 3) If the norm represents 0 -- which is always the case if F is quadratically closed -- then there is a "good basis" {e1, e2, u1, u2, u3, v1, v2, v3} of C con- sisting of isotropic elements, such that n(e1, e2) = n(ui, vi) = 1 for any i = 1, 2, 3 and n(er, ui) = n(er, vi) = n(ui, uj) = n(ui, vj) = n(vi, vj) for any r = 1, 2 and 1 ≤ i 6= j ≤ 3, whose multiplication table is shown in Fig- ure 1. In particular, up to isomorphism, there is a unique Cayley algebra whose norm represents 0, which is called the split Cayley algebra. (cid:3) A "good basis" {e1, e2, u1, u2, u3, v1, v2, v3} of the split Cayley algebra C gives a Z2-grading with C(0,0) = Fe1 ⊕ Fe2, C(1,0) = Fu1, C(0,1) = Fu2, C(1,1) = Fv3, C(−1,0) = Fv1, C(0,−1) = Fv2, C(−1,−1) = Fu3. This is called the Cartan grading on the split Cayley algebra, and Z2 is its universal grading group. Remark 3.2. The Cartan grading is fine as a group grading, but it is not so as a general grading, because the decomposition C = Fe1 ⊕ Fe2 ⊕ Fu1 ⊕ Fu2 ⊕ Fu3 ⊕ Fv1 ⊕ Fv2 ⊕ Fv3 is a proper refinement. This refinement is not even a semigroup grading (because (u1u2)u3 = −e2 and u1(u2u3) = −e1 are in different homogeneous subspaces). GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 9 Let Q be a proper four-dimensional subalgebra of the Cayley algebra C such that nQ is nondegenerate, and let u be any element in C \ Q with n(u) = α 6= 0. Then C = Q ⊕ Qu and we get: n(a + bu) = n(a) + αn(b), (a + bu)(c + du) = (ac − α ¯db) + (da + b¯c)u, for any a, b, c, d ∈ Q. Then C is said to be obtained from Q by means of the Cayley -- Dickson doubling process and we write C = CD(Q, α). This gives a Z2-grading on C with C¯0 = Q and C¯1 = Qu. The subalgebra Q above is a quaternion subalgebra which in turn can be obtained from a quadratic subalgebra K through the same process Q = CD(K, β) = K ⊕ Kv, and this gives a Z2-grading of Q and hence a Z2 2-grading of C = K⊕Kv⊕Ku⊕(Kv)u. We write here Q = CD(K, β, α). If char F 6= 2, then K can be obtained in turn from the ground field: K = CD(F, γ), and a Z3 2-grading of C appears. Here we write C = CD(F, γ, β, α). These gradings by Zr 2, r = 1, 2, 3, will be called gradings induced by the Cayley -- Dickson doubling process. The groups Zr 2 are their universal grading groups. The following result describes all possible gradings on Cayley algebras: Theorem 3.3 ([Eld98]). Any abelian group grading on a Cayley algebra is, up to equivalence, either a grading induced by the Cayley -- Dickson doubling process or a coarsening of the Cartan grading on the split Cayley algebra. (cid:3) Remark 3.4. The number of non-equivalent gradings induced by the Cayley -- Dickson doubling process depends on the ground field. Actually, the number of non equiv- alent Z2-gradings coincides with the number of isomorphism classes of quaternion subalgebras Q of the Cayley algebra. For an algebraically closed field F, this is one. Over R there are two non isomorphic Cayley algebras, the classical division algebra of the octonions O = CD(R, −1, −1, −1) and the split Cayley algebra Os = CD(R, 1, 1, 1). Any quater- nion subalgebra of O is isomorphic to H = CD(R, −1, −1), while Os contains quater- nion subalgebras isomorphic to H and to M2(R). On the other hand, for p, q prime numbers congruent to 3 modulo 4, it is easy to check that the quaternion subalgebras Qp = CD(cid:0)Q(i), p(cid:1) and Qq = CD(cid:0)Q(i), q(cid:1) are not isomorphic (i2 = −1). Consider the division algebra Q = CD(cid:0)Q(i), −1(cid:1). The split Cayley algebra over Q is isomorphic to C = CD(Q, 1), and by the classical Four Squares Theorem, Q⊥ contains elements whose norm is −p for any prime number p. Therefore C contains a quaternion subalgebra isomorphic to Qp for any prime number p, and hence the split Cayley algebra over Q is endowed with infinitely many non-equivalent Z2-gradings. Over an algebraically closed field there is a unique Zr for any r = 1, 2, 3. Over R, O is endowed with a unique Zr to equivalence, while Os is endowed with two non equivalent Z2 and Z2 but a unique Z3 2-grading. 2-grading, up to equivalence, 2-grading (r = 1, 2, 3) up 2-gradings, (cid:3) Up to symmetry, any coarsening of the Cartan grading is obtained as follows (with gi = deg(ui), i = 1, 2, 3): g1 = 0 : Then we obtain a "3-grading" by Z: C = C−1 ⊕ C0 ⊕ C1, with C0 = span {e1, e2, u1, v1}, C1 = span {u2, v3}, C−1 = span {u3, v2}. All proper coarsenings have a 2-elementary grading group. 10 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV g1 = g2 : Here we obtain a "5-grading" by Z, with C−2 = Fu3, C−1 = span {v1, v2}, C0 = span {e1, e2}, C1 = span {u1, u2} and C2 = Fv3, which has two proper coarsenings whose grading groups are not 2-elementary: g1 = g2 = g3 : This gives a Z3-grading with C¯0 = span {e1, e2}, C¯1 = span {u1, u2, u3}, C¯2 = span {v1, v2, v3}. g3 = −g3 : This gives a Z4-grading. g1 = −g1 : Here we get a Z × Z2-grading C = C (0,¯0) ⊕ C q (1,¯0) ⊕ C q span {e1, e2} Fu2 q Fv2 (−1,¯0) ⊕ C (0,¯1) ⊕ C q (−1,¯1) ⊕ C (1,¯1) q Fu3 q Fv3 span {u1, v1} Any of its coarsenings is a coarsening of the previous gradings. g1 = −g2 : In this case g3 = 0, and this is equivalent to the grading obtained with g1 = 0. Thus the next result follows: Theorem 3.5 ([Eld98]). Up to equivalence, the nontrivial abelian group gradings on the split Cayley algebra are: 2-gradings induced by the Cayley -- Dickson doubling process, r = 1, 2, 3. (1) The Zr (2) The Cartan grading by Z2. (3) The 3-grading: C0 = span {e1, e2, u3, v3}, C1 = span {u1, v2}, and C−1 = span {u2, v1}. (4) The 5-grading: C0 = span {e1, e2}, C1 = span {u1, u2}, C2 = span {v3}, C−1 = span {v1, v2}, and C−2 = span {u3}. (5) The Z3-grading: C¯0 = span {e1, e2}, C¯1 = span {u1, u2, u3}, and C¯2 = span {v1, v2, v3}. (6) The Z4-grading: C¯0 = span {e1, e2}, C¯1 = span {u1, u2}, C¯2 = span {u3, v3}, and C¯3 = span {v1, v2}. (7) The Z × Z2-grading. (cid:3) In particular, over an algebraically closed field, there are 9 equivalence classes of nontrivial gradings on the (unique) Cayley algebra. Corollary 3.6. Let Γ be a fine abelian group grading on the Cayley algebra C over an algebraically closed field F. Then Γ is equivalent either to the Cartan grading or to the Z3 2-grading induced by the Cayley -- Dickson doubling process. The latter grading does not occur if char F = 2. (cid:3) Let G be an abelian group. Assuming F algebraically closed, we can classify all C be 2-grading induced by the Cayley -- Dickson doubling process. We will need the G-gradings on C up to isomorphism. Let Γ1 the Z3 following result: C be the Cartan grading and let Γ2 Theorem 3.7 ([EK]). Identifying Supp Γ1 Φ of type G2, we have W (Γ1 C) = Aut Φ, W (Γ2 C with the short roots of the root system (cid:3) C) = Aut(Z3 2). To state our classification theorem, we introduce the following notation: • Let γ = (g1, g2, g3) be a triple of elements in G with g1g2g3 = e. Denote by Γ1 C by the homomorphism Z2 → G sending (1, 0) to g1 and (0, 1) to g2. In other words, we set deg ej = e, j = 1, 2, C(G, γ) the G-grading on C induced from Γ1 GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 11 deg ui = gi and deg vi = g−1 triples, γ and γ′, we will write γ ∼ γ′ if there exists π ∈ Sym(3) such that g′ for all i = 1, 2, 3 or g′ , i = 1, 2, 3, for some "good basis" of C. For two such i = gπ(i) π(i) for all i = 1, 2, 3. i = g−1 i • Let H ⊂ G be a subgroup isomorphic to Z3 2. Then Γ2 as a G-grading with support H. We denote this G-grading by Γ2 W (Γ2 are isomorphic, so Γ2 2), all induced gradings αΓ2 C(G, H) is well-defined.) C) = Aut(Z3 C for various isomorphisms α : Z3 C may be regarded C(G, H). (Since 2 → H Theorem 3.8. Let C be the Cayley algebra over an algebraically closed field and let G be an abelian group. Then any G-grading on C is isomorphic to some Γ1 C(G, γ) or Γ2 C(G, H), but not both. Also, • Γ1 • Γ2 C(G, γ) is isomorphic to Γ1 C(G, H) is isomorphic to Γ2 C(G, γ′) if and only if γ ∼ γ′; C(G, H ′) if and only if H = H ′. C. Γ1 C(G, γ) and Γ2 C) that sends Γ1 C for some α : Z3 C is isomorphic to some βΓ1 If γ ∼ γ′, then there is an automorphism in Aut(Γ1 C(G, γ′). Conversely, if ϕ is an automorphism of C sending Γ1 Proof. It follows from Corollary 3.6 that any G-grading is isomorphic to αΓ1 C for some α : Z2 → G or to αΓ2 2 → G. In the second case, if α is not one-to-one, then αΓ2 C(G, T ) cannot be isomorphic, because in the first case dim Ce ≥ 2 and in the second case dim Ce = 1. C(G, γ) to Γ1 C(G, γ′), then, in particular, ϕ maps Ce onto C′ If Ce = C, there is nothing to prove. e. Otherwise Ce is isomorphic to M2(F) or F×F, because it is a composition subalgebra of C (alternatively, one may examine the cases in Theorem 3.5). If Ce is isomorphic to M2(F), then one of gi is e. Say, g3 = e and hence g2 = g−1 1 . The support of the grading then consists of e and g±1 i, we see that γ ∼ γ′. Finally, consider the case dim Ce = 2. Then Ce = C′ e, since both are spanned by the idempotents e1 and e2. Hence ϕ either fixes e1 and e2 or swaps them. In the first case, ϕ preserves the subspaces U and V. Looking at the support of U and the dimensions of the homogeneous components in U, we conclude that (g′ 3) must be a permutation of (g1, g2, g3). In the second case, ϕ swaps U and V and we conclude that (g′ 1 . Applying the same argument to g′ C(G, γ) to Γ1 1, g′ 2, g′ 1 , g−1 3) must be a permutation of (g−1 C(G, H), an isomorphism between Γ2 2, g′ Since H is the support of Γ2 C(G, H ′) forces H = H ′. 1, g′ Γ2 3 ). 2 , g−1 C(G, H) and (cid:3) Note that γ ∼ γ′ if and only if the corresponding homomorphisms Z2 → G are C) = Z2. This is a C) = Aut Φ in its action on the group U (Γ1 conjugate by W (Γ1 special case of the following general result.1 Proposition 3.9. Let A be a finite-dimensional algebra over an algebraically closed field. Let T be a maximal torus in Aut(A). Let G be an abelian group and let Γ and Γ′ be G-gradings induced by homomorphisms α : X(T ) → G and α′ : X(T ) → G, respectively. Then Γ′ is isomorphic to Γ if and only if there exists w ∈ W (T ) such that α′(λ) = α(λw) for all λ ∈ X(T ). Lg∈G Proof. The "if" part is clear. To prove the "only if" part, suppose Γ : A = g = ϕ(Ag) for all g ∈ G. Let T ′ = ϕT ϕ−1. It is a maximal torus in Aut(A). Let H = Stab(Γ′). Then both T and T ′ are contained in H and thus are maximal tori in H. Therefore, Ag, Γ′ : A =Lg∈G g, and there exists ϕ ∈ Aut(A) such that A′ A′ 1The authors would like to thank Prof. Reichstein, University of British Columbia, Canada, for a discussion that was instrumental in proving this result. 12 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV T and T ′ are conjugate in H, i.e., there exists ψ ∈ H such that ψT ′ψ−1 = T . Let eϕ = ψϕ. Then, by construction, we have eϕTeϕ−1 = T and A′ g = eϕ(Ag) for all g ∈ G. Hence we can take w to be the image of the element eϕ ∈ N (T ) in the quotient group W (T ) = N (T )/C(T ). (cid:3) 4. Gradings on G2 The central simple Lie algebras of type G2 appear as the algebras of derivations of the Cayley algebras. The gradings on the simple Lie algebra of type G2 over an algebraically closed field of characteristic 0 were obtained independently in [DM06] and [BT09], using the results on gradings on the (unique) Cayley algebra in [Eld98]. In this section the gradings on the simple Lie algebras of type G2 will be ob- tained over arbitrary fields of characteristic different from 2 and 3. Note that, in characteristic 3, the Lie algebra of derivations of a Cayley algebra is not simple (see e.g. [BEMN02]). So let C be a Cayley algebra over a field F, char F 6= 2, 3, and let g = Der(C). Then we have the affine group scheme Aut(C) and the morphism Ad : Aut(C) → Aut(g). Let F be the algebraic closure of F. Then AutF(C ⊗ F) is the simple algebraic group of type G2. It is well-known that Ad F : AutF(C ⊗ F) → AutF(g ⊗ F) is bijective. Since any derivation of g is inner (see [Sel67]), the differential is also bijective. Finally, since ad : g → Der(g) dim AutF(C ⊗ F) = 14 = dim Der(C), we conclude that Aut(C) is smooth. It follows that Ad : Aut(C) → Aut(g) is an isomorphism of affine group schemes and hence Theorems 2.3 and 2.4 yield the following result: Theorem 4.1. Let C be a Cayley algebra over a field F, char F 6= 2, 3. Then the abelian group gradings on Der(C) are those induced by such gradings on C. The algebras C and Der(C) have the same classification of fine gradings up to equiv- alence and, for any abelian group G, the same classification of G-gradings up to isomorphism. (cid:3) If C is split, then g = Der(C) is the split simple Lie algebra of type G2, and the Cartan grading on C induces the Cartan decomposition of g relative to a split Cartan subalgebra. The latter will be called the Cartan grading on g. Corollary 4.2. Let C be a Cayley algebra over a field F, char F 6= 2, 3. Then any abelian group grading on the simple Lie algebra g = Der(C) is, up to equivalence, either a Zr 2-grading, r = 1, 2, 3, induced by the Cayley -- Dickson doubling process on C, or a coarsening of the Cartan grading on the split algebra g. In particular, if F is algebraically closed, then there are, up to equivalence, exactly two fine abelian g with universal group Z2 and the Cayley -- group gradings on g: the Cartan grading Γ1 g with universal group Z3 Dickson grading Γ2 2. (cid:3) GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 13 Let Γ1 g(G, γ) and Γ2 g(G, H) be the G-gradings induced by Γ1 g and Γ2 C(G, γ) and Γ2 C(G, H) are induced from Γ1 tively, in the same way as Γ1 Theorem 3.8). g, respec- C (see C and Γ2 Corollary 4.3. Let g be the simple Lie algebra of type G2 over an algebraically closed field F, char F 6= 2, 3. Let G be an abelian group. Then any G-grading on g is isomorphic to some Γ1 g(G, γ) or Γ2 g(G, γ) is isomorphic to Γ1 g(G, H) is isomorphic to Γ2 g(G, H), but not both. Also, g(G, γ′) if and only if γ ∼ γ′; g(G, H ′) if and only if H = H ′. • Γ1 • Γ2 (cid:3) If one wants to obtain a classification of all abelian group gradings on Der(C) up to equivalence, then one should be careful when applying Theorem 4.1, because each grading on our list in Theorem 3.5 can be realized as a G-grading for many different groups G. For example, consider the 3-grading on the split Cayley algebra C in Theorem 3.5(3): C0 = span {e1, e2, u3, v3}, C1 = span {u1, v2}, C−1 = span {u2, v1}. As a Z- grading it induces a 5-grading on Der(C), with Der(C)2 = span {Du1,v2 } 6= 0, where Da,b : c 7→ [[a, b], c] + 3(cid:0)(ac)b − a(cb)(cid:1) is the inner derivation defined by a, b ∈ C (the linear span of the inner derivations fills Der(C)), so it has 5 different nonzero homogeneous components. Its type is (2, 0, 0, 3). However, up to equivalence, this grading on C is also a Z3-grading, and as such it induces a Z3-grading on Der(C) of type (0, 0, 0, 1, 2). As a further example, the Cartan grading on the split Cayley algebra C can be realized as a G-grading for any abelian group G containing two elements g1 and g2 such that the elements e, g1, g2, g1g2, g−1 2 , (g1g2)−1 are all different. In 3-grading, with g1 = (¯1, ¯0) and g2 = (¯0, ¯1). particular, it can be obtained as a Z2 However, the induced Z2 3-grading on Der(C) is not equivalent to the Cartan grading, as some of the nonzero root spaces coalesce in the Z2 1 , g−1 3-grading. Easy combinatorial arguments give all the gradings on Der(C) in terms of the gradings on the Cayley algebra C in Theorem 3.5 (see [Koc09, Figure 1]): Theorem 4.4. Let C be a split Cayley algebra over a field of characteristic different from 2 and 3. Up to equivalence, the nontrivial abelian group gradings on Der(C) are: (1) The Zr 2-gradings induced by the Cayley -- Dickson doubling process, r = 1, 2, 3. (2) Eleven gradings induced by the Cartan grading on C with universal groups: Z2, Z7, Z8, Z9, Z10, Z, Z6 × Z2, Z × Z2, Z12, Z × Z3 and Z2 3. (3) Three gradings induced by the 3-grading on C with universal groups Z, Z3 and Z4. (4) Three gradings induced by the 5-grading on C with universal groups Z, Z5 and Z6. (5) The Z3-grading induced by the Z3-grading on C. (6) The Z4-grading induced by the Z4-grading on C. (7) Three gradings induced by the Z × Z2-grading on C with universal groups (cid:3) Z × Z2, Z3 × Z2 and Z4 × Z2. In particular, over an algebraically closed field of characteristic different from 2 and 3, there are exactly 25 equivalence classes of nontrivial gradings on the simple Lie algebra of type G2. 14 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV 5. The Albert algebra Let C be the Cayley algebra over an algebraically closed field F of characteristic different from 2. The Albert algebra is the algebra of Hermitian 3 × 3-matrices over C: A = H3(C, ∗) =    ¯a3 α1 a3 α2 ¯a2 a2 ¯a1 a1 α3  : α1, α2, α3 ∈ F, a1, a2, a3 ∈ C   = FE1 ⊕ FE2 ⊕ FE3 ⊕ ι1(C) ⊕ ι2(C) ⊕ ι3(C), (4) where 1 0 0 0 0 0 E1 =  ι1(a) = 2  0 0 0 0 0 ¯a 0 a 0 0 0 0   ,   , 0 0 0 1 0 0 E2 =  ι2(a) = 2  0 a 0 0 0 0 ¯a 0 0 0 0 0   ,   , 1 0 0 E3 =  ι3(a) = 2  0 0 0 0 0 1 0 a 0 0 0 ¯a 0 0 0   ,   , with (commutative) multiplication given by XY = 1 2 (X ·Y +Y ·X), where X ·Y de- notes the usual product of matrices X and Y . Then Ei are orthogonal idempotents with E1 + E2 + E3 = 1. The rest of the products are as follows: (5) Eiιi(a) = 0, Ei+1ιi(a) = ιi(a)ιi+1(b) = ιi+2(¯a¯b), 1 2 ιi(a) = Ei+2ιi(a), ιi(a)ιi(b) = 2n(a, b)(Ei+1 + Ei+2), for any a, b ∈ C, with i = 1, 2, 3 taken modulo 3. (This convention about indices will be used without further mention.) For the main properties of the Albert algebra the reader may consult [Jac68]. This is the only exceptional simple Jordan algebra over F. Any element X ∈ A satisfies the generic degree 3 equation (6) X 3 − T (X)X 2 + S(X)X − N (X)1 = 0, for the linear form T (the trace), the quadratic form S, and the cubic form N (the norm) given by: T (X) = α1 + α2 + α3, S(X) = 1 2(cid:0)T (X)2 − T (X 2)(cid:1) = 3Xi=1(cid:0)αi+1αi+2 − 4n(ai)(cid:1), N (X) = α1α2α3 + 8n(a1, ¯a2¯a3) − 4 αin(ai), 3Xi=1 for X =P3 i=1(cid:0)αiEi + ιi(ai)(cid:1). We note that the trace T is associative: for all X, Y, Z ∈ A T(cid:0)(XY )Z(cid:1) = T(cid:0)X(Y Z)(cid:1) and symmetric: T (XY ) = T (Y X) for all X, Y ∈ A. The next result shows the good behavior of the trace form T (X, Y ) := T (XY ) of the Albert algebra with respect to gradings. It will be crucial in what follows. GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 15 Theorem 5.1. Let G be an abelian group and let A =Lg∈G Ag be a G-grading on the Albert algebra over an algebraically closed field of characteristic different from 2. Then T (AgAh) = 0 unless gh = e. Proof. If the characteristic of the ground field F is not 3, the result is very easy to prove, because T (X) = 1 9 trace(LX ) for any X ∈ A, where LX denotes the multiplication by X. Let us give a proof that includes the case of characteristic 3. We may assume, without loss of generality, that G is generated by the support of the grading, and hence it is finitely generated. It is sufficient to prove T (Ag) = 0 for all g 6= e. If the order of g is ≥ 3, then equation (6) shows that for any X ∈ Ag, S(X) = 0 and either T (X) = 0 or X 2 = 0. In the latter case, T (X)2 = 2S(X) + T (X 2) = 0, so again T (X) = 0. Hence T (Ag) = 0 for any g ∈ G of order ≥ 3. But G = G1G2 ∼= G1 × G2 where G2 is the 2-torsion subgroup of G and G1 is 2-torsion free. Then G1 has no elements of order 2, and hence the trace of any non-identity homogeneous component of the G1-grading induced by the projection G → G1 is 0. In other words, T (Agh) = 0 for any e 6= g ∈ G1 and any h ∈ G2. Now consider the G2-grading induced by the projection G → G2. Since the characteristic is not 2, the homogeneous components are the common eigenspaces for a family of commuting automorphisms. But for ϕ ∈ Aut(A) and X ∈ A with ϕ(X) = λX, 1 6= λ ∈ F, we get T (X) = T (ϕ(X)) = λT (X), so T (X) = 0. Therefore, T (Agh) = 0 for any g ∈ G1 and e 6= h ∈ G2. The result follows. (cid:3) Corollary 5.2. Under the assumtions of Theorem 5.1, Ae is a semisimple Jordan algebra. Moreover, if the degree of Ae is 2, then Ae is isomorphic to F × F. Proof. The restriction T Ae is nondegenerate by Theorem 5.1, and if I is an ideal of Ae with I2 = 0, then for any X ∈ I, T (X Ae) = 0, as any element in X Ae is nilpotent (see [Jac68, p. 226]). Then Dieudonn´e's Lemma [Jac68, p. 239] proves that Ae is semisimple. If the degree of Ae is 2, then either Ae is isomorphic to F × F (a direct sum of two copies of the degree one simple Jordan algebra), or it is a simple Jordan algebra of degree 2. In the latter case let mX (λ) = λ2 − T ′(X)λ + S′(X) be the generic minimal polynomial of Ae. With mX (λ) = λ3 − T (X)λ2 + S(X)λ− N (X) being the generic minimal polynomial in A, it follows that there is a linear form T ′′ : Ae → F such that mX (λ) = (λ − T ′′(X)) mX (λ) for any X ∈ Ae (see [Jac68, §VI.3]). Then N (X) = S′(X)T ′′(X) for any X ∈ Ae and since S′ and N are multiplicative, so is T ′′. It follows that ker T ′′ is a codimension one ideal of Ae, a contradiction. (cid:3) We will make use of some subgroups of the automorphism group Aut(A). First we will consider StabAut A(E1, E2, E3), the stabilizer of the three orthogonal idem- potents E1, E2 and E3. The orthogonal group of C relative to its norm will be denoted by O(C, n), and the special orthogonal group by SO(C, n). Definition 5.3. A triple (f1, f2, f3) ∈ O(C, n)3 is said to be related if f1(¯x¯y) = f2(x) f3(y) for all x, y ∈ C. To simplify the notation, consider the para-Hurwitz product x • y = ¯x¯y on C -- see [KMRT98, Chapter VIII]. Note that, for any x, y, z ∈ C, n(x • y, z) = n(¯x¯y, z) = n(¯x, zy) = n(x, ¯y ¯z) = n(x, y • z), and (x• y)• x = ¯x¯y ¯x = (yx)¯x = n(x)y = x• (y • x). In other words, (7) n(x • y, z) = n(x, y • z), (x • y) • x = n(x)y = x • (y • x), 16 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV for all x, y, z ∈ C. Consider the trilinear form on C given by hx, y, zi = n(x • y, z). Equation (7) shows that hx, y, zi = hy, z, xi for any x, y, z ∈ C. Lemma 5.4. Let f1, f2, f3 be three elements in O(C, n), then: • (f1, f2, f3) is a related triple if and only if hf1(x), f2(y), f3(z)i = hx, y, zi for any x, y, z ∈ C. • (f1, f2, f3) is related if and only if so is (f2, f3, f1). Proof. The triple (f1, f2, f3) is related if and only if f1(x • y) = f2(x) • f3(y) for any x, y ∈ C, and this happens if and only if n(cid:0)f1(x • y), f1(z)(cid:1) = n(cid:0)f2(x) • f3(y), f1(z)(cid:1) for any x, y, z ∈ C. But f1 is orthogonal, so n(cid:0)f1(x • y), f1(z)(cid:1) = n(x • y, z), and this is equivalent to hf2(x), f3(y), f1(z)i = hx, y, zi. The cyclic symmetry of hx, y, zi completes the proof. (cid:3) Denote by lx and rx the left and right multiplications in the para-Cayley algebra (C, •): lx(y) = x • y = ¯x¯y, rx(y) = y • x = ¯y ¯x. Then equation (7) shows that l∗ x = rx and lxrx = n(x)id = rxlx for any x ∈ C, where ∗ denotes the adjoint relative to the norm n. Let Cl(C, n) be the Clifford algebra of the space C relative to the norm. The linear map C −→ EndF(C ⊕ C), x 7→(cid:18) 0 rx lx 0(cid:19) , extends to an algebra isomorphism (see [KMRT98, §35] or [Eld00]) Φ : Cl(C, n) → EndF(C ⊕ C), which is in fact an isomorphism of Z2-graded algebras, where the Clifford algebra Cl(C, n) is Z2-graded with deg x = ¯1 for all x ∈ C, and EndF(C ⊕ C) is Z2-graded with the ¯0-component being the endomorphisms that preserve the two copies of C, and the ¯1-component being the endomorphisms that swap these copies. The standard involution τ on Cl(C, n) is defined by setting τ (x) = x for all x ∈ C. We define an involution on EndF(C⊕ C) as the adjoint relative to the quadratic form n ⊥ n on C ⊕ C. Since l∗ x = rx for any x ∈ C, it follows that Φ is an isomorphism of algebras with involution. Consider now the corresponding spin group: Spin(C, n) = {u ∈ Cl(C, n) : u · τ (u) = 1 and u · C · u−1 ⊂ C}, = {x1 · x2 · . . . · x2r : r ≥ 0, xi ∈ C and n(x1)n(x2) · · · n(x2r) = 1}, where the multiplication in Cl(C, n) is denoted u · v. For any u ∈ Spin(C, n), the map χu : C → C, x 7→ u · x · u−1 is in SO(C, n), and the map χ : Spin(C, n) → SO(C, n), u 7→ χu is a group homomorphism, which is onto and whose kernel is just the cyclic group of two elements {±1}. Besides, for any u ∈ Spin(C, n), Φ(u) is an even endomorphism of C ⊕ C, so there are linear maps ρ± u ∈ EndF(C) with Φ(u) =(cid:18)ρ− u 0 u(cid:19) . 0 ρ+ Theorem 5.5. Let A be the Albert algebra over an algebraically closed field of characteristic different from 2. Then the map Spin(C, n) −→ GL(C)3, u 7→ (χu, ρ+ u , ρ− u ), GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 17 is a one-to-one group homomorphism whose image coincides with the set of related triples in O(C, n)3. In particular, any related triple is contained in SO(C, n)3. Proof. The map is one-to-one because so is Φ. For u ∈ Spin(C, n), we have u·τ (u) = 1, so ρ± u ∈ O(C, n), as Φ is an isomorphism of algebras with involution. Also, for any x ∈ C, u · x = χu(x) · u. Applying Φ to both sides, we obtain: Thus ρ− (ρ− u (x • y) = χu(x) • ρ+ u (y), for all x, y ∈ C. Hence u ) is related, and so is (χu, ρ+ u , χu, ρ+ Conversely, let (f1, f2, f3) be a related triple, and let u be the (even) element in u ) by Lemma 5.4. u , ρ− Cl(C, n) such that Φ(u) =(cid:18)f3 0 0 f2(cid:19). Then u · τ (u) = 1 since Φ is an isomorphism of algebras with involution. For any x ∈ C, (cid:18)ρ− u(cid:19)(cid:18) 0 0 ρ+ rx u lx = lχu(x)ρ+ u , or ρ− u 0 lx 0(cid:19) =(cid:18) 0 rχu (x) lχu(x) 0 (cid:19)(cid:18)ρ− u 0 u(cid:19) . 0 ρ+ 0 f2(cid:19)(cid:18) 0 lx 3 0 0(cid:19)(cid:18)f −1 0 (cid:19) rx f3lxf −1 2 2 (cid:19) 0 f −1 0 0 Φ(u · x · u−1) =(cid:18)f3 =(cid:18) =(cid:18) 0 = Φ(cid:0)f1(x)(cid:1), f2rxf −1 rf1(x) 2 lf1(x) 0 (cid:19) where we have used the equations f3(x • y) = f1(x) • f2(y) and f2(y • x) = f3(y) • f1(x)). It follows that u ∈ Spin(C, n), χu = f1 and hence (f1, f2, f3) = (χu, ρ+ u ). The last assertion follows because if (f1, f2, f3) is related, then there is an element u ∈ Spin(C, n) such that f1 = χu ∈ SO(C, n). But (f2, f3, f1) and (f3, f1, f2) are also related, so f2, f3 ∈ SO(C, n) as well. (cid:3) u , ρ− Corollary 5.6. The group StabAut A(E1, E2, E3) is isomorphic to Spin(C, n). Proof. Any automorphism ϕ ∈ StabAut A(E1, E2, E3) stabilizes each of the sub- spaces ιi(C) = {X ∈ A : Ei+1X = 1 2 X = Ei+2X}, and hence there are linear automorphisms fi ∈ GL(C) such that ϕ(cid:0)ιi(x)(cid:1) = ιi(fi(x)) for any i = 1, 2, 3 and x ∈ C. But ιi(x)2 = 4n(x)(cid:0)Ei+1 + Ei+2(cid:1), so we obtain fi ∈ O(C, n) for any i, and ι2(x)ι3(y) = ι1(x • y) for any x, y ∈ C, whence it follows that (f1, f2, f3) is a related triple. It remains to apply Theorem 5.5. (cid:3) Corollary 5.7. The group StabAut A(E1, E2, E3, ι1(1)) is isomorphic to Spin(C0, n), where C0 denotes the space of trace zero octonions, i.e., the orthogonal complement to 1 in C. Proof. Corollary 5.6 provides identifications: StabAut A(E1, E2, E3, ι1(1)) ∼= {(χu, ρ+ u , ρ− ∼= {(χc, ρ+ c , ρ− ∼= Spin(C0, n). u ) : u ∈ Spin(C, n), χu(1) = 1}, c ) : c ∈ Spin(C0, n)} (cid:3) Note that for x1, x2 ∈ C, we have lx1 Φ(x1 · x2) =(cid:18) 0 0 (cid:19)(cid:18) 0 rx1 rx2 lx2 0 (cid:19) =(cid:18)lx1rx2 0 0 rx1 lx2(cid:19) . 18 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV If x1, x2 ∈ C0, then, for any y ∈ C, we compute: x1 • (y • x2) = ¯x1 ¯y ¯x2 = ¯x1(x2y) = −x1(x2y). Similarly, (x2 • y) • x1 = −(yx2)x1. Hence, for c = x1 · x2 · . . . · x2r ∈ Spin(C0, n), we have (8) ρ+ x1·x2·...·x2r = (−1)rRx1Rx2 · · · Rx2r , ρ− x1·x2·...·x2r = (−1)rLx1Lx2 · · · Lx2r , where Lx and Rx denote the left and right multiplications by x in C. 6. Construction of fine gradings on the Albert algebra We continue to assume that the ground field F is algebraically closed of charac- teristic different from 2. The aim of this section is to construct four fine gradings on the Albert algebra (the fourth one will exist only for char F 6= 3). If char F = 0, these gradings (although presented in a somewhat different form) are known to be the only fine gradings, up to equivalence [DM06]. The next section will be devoted to proving the same result for char F 6= 2. 6.1. Cartan grading. Let G = Z4 and use additive notation. Consider the fol- lowing elements in G: a1 = (1, 0, 0, 0), g1 = (0, 0, 1, 0), a2 = (0, 1, 0, 0), g2 = (0, 0, 0, 1), a3 = (−1, −1, 0, 0), g3 = (0, 0, −1, −1). Then a1+a2+a3 = 0 = g1+g2+g3. Take a "good basis" {e1, e2, u1, u2, u3, v1, v2, v3} of the Cayley algebra. The assignment deg e1 = deg e2 = 0, deg ui = gi = − deg vi gives the Cartan grading of the Cayley algebra C. Now the assignment deg Ei = 0, deg ιi(e1) = ai = − deg ιi(e2), deg ιi(ui) = gi = − deg ιi(vi), deg ιi(ui+1) = ai+2 + gi+1 = − deg ιi(vi+1), deg ιi(ui+2) = −ai+1 + gi+2 = − deg ιi(vi+2). for any i = 1, 2, 3, gives a Z4-grading on the Albert algebra A. Indeed, since C is graded by the second component of Z2 × Z2, it suffices to look at the first component, and by the cyclic symmetry of the product, it is enough to check that deg(cid:0)ι3(¯x¯y)(cid:1) = deg ι1(x) + deg ι2(y) for any x, y in the "good basis" of C, and this is straightforward. This grading will be called the Cartan grading on A. Its type is (24, 0, 1). Note that ιi(e1)ιi(e2) = 2(Ei+1 + Ei+2) is homogeneous in any refinement of the Cartan grading. Then Ei = (Ei + Ei+1)(Ei−1 + Ei) is homogeneous too in any refinement, and it follows that E1, E2, E3 must be homogeneous of the same degree in any refinement. Hence the Cartan grading is fine. (Actually, this proves that it is fine not just as an abelian group grading, but as a general grading.) Also, the elements (9) ι1(e1), ι1(e2), ι2(e1), ι2(e2), ι1(u1), ι1(v1), ι2(u2), ι2(v2) GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 19 constitute a set of generators of A. In any grading Γ : A = Lg∈G Ag in which these elements are homogeneous, as ι1(e1)ι1(e2) = 2(E2 + E3), we obtain that E2 + E3 is homogeneous. But this is an idempotent, so its degree must be e, and we have deg ι1(e1) deg ι1(e2) = e. In the same vein, deg ι2(e1) deg ι2(e2) = deg ι1(u1) deg ι1(v1) = deg ι2(u2) deg ι2(v2) = e. Therefore the assignment a1 7→ deg ι1(e1), a2 7→ deg ι2(e1), g1 7→ deg ι1(u1) and g2 7→ deg ι2(u2) determines a group homomorphism α : Z4 → G. This proves the following result: Theorem 6.1. Let Γ : A =Lg∈G Ag be a grading of the Albert algebra in which the elements in (9) are homogeneous. Then there is a group homomorphism α : Z4 → G such that Γ is the grading induced by α from the Cartan grading. In particular, Z4 is the universal group of the Cartan grading. (cid:3) 6.2. Z5 repeated application of the Cayley -- Dickson doubling process: 2-grading. As discussed in Section 3, the Cayley algebra C is obtained by K = F ⊕ Fw1, H = K ⊕ Kw2, C = H ⊕ Hw3, with w2 w3 = u2−v2), and this gives a (uniquely determined up to isomorphism) Z3 of C by setting deg w1 = (¯1, ¯0, ¯0), deg w2 = (¯0, ¯1, ¯0), deg w3 = (¯0, ¯0, ¯1). i = 1 for i = 1, 2, 3 (one may take w1 = e1 − e2, w2 = u1 − v1 and 2-grading Then A is obviously Z5 2-graded as follows: deg Ei = (¯0, ¯0, ¯0, ¯0, ¯0), i = 1, 2, 3 deg ι1(x) = (¯1, ¯0, deg x), deg ι2(x) = (¯0, ¯1, deg x), deg ι3(x) = (¯1, ¯1, deg x), for homogeneous elements x ∈ C. The type of this grading is (24, 0, 1). This grading will be referred to as the Z5 With the same arguments as for the Cartan grading, this grading is fine (even 2-grading on A. as a general grading). Theorem 6.2. Let Γ : A = Lg∈G the elements Ag be a grading of the Albert algebra in which ι1(1), ι2(1), ι3(wj), j = 1, 2, 3, are homogeneous. Then there is a group homomorphism α : Z5 the grading induced by α from the Z5 2-grading. 2 → G such that Γ is In particular Z5 2 is the universal group of the Z5 2-grading. Proof. Since ι1(1) is homogeneous for Γ, so is ι1(1)2 = 4(E2 + E3). But E2 + E3 is an idempotent, so its degree must be e, and hence the degree of ι1(1) has order ≤ 2. The same happens to all the homogeneous elements above, and since these elements constitute a set of generators of A, the result follows. (cid:3) 20 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV 6.3. Z × Z3 ing elements in A: 2-grading. Take an element i ∈ F with i2 = −1 and consider the follow- E = E1, eE = 1 − E = E2 + E3, ν(a) = iι1(a) for all a ∈ C0, ν±(x) = ι2(x) ± iι3(¯x) i 2 S± = E3 − E2 ± ι1(1). for all x ∈ C, These elements span A, and the multiplication is given by: (10) 1 2 ν±(x), EeE = 0, ES± = 0, Eν(a) = 0, Eν±(x) = eES± = S±, S+S− = 2eE, S±ν(a) = 0, S±ν∓(x) = ν±(x), S±ν±(x) = 0, ν(a)ν(b) = −2n(a, b)eE, eEν(a) = ν(a), ν±(x)ν±(y) = 2n(x, y)S±, eEν±(x) = ν(a)ν±(x) = ±ν±(xa), ν±(x), 1 2 for any x, y ∈ C and a, b ∈ C0. There appears a Z-grading on A: ν+(x)ν−(y) = 2n(x, y)(2E + eE) + ν(¯xy − ¯yx), (11) A = A−2 ⊕ A−1 ⊕ A0 ⊕ A1 ⊕ A2, with A±2 = FS±, A±1 = ν±(C), and A0 = FE ⊕(cid:16)FeE ⊕ ν(C0)(cid:17). Note that the subspace FeE ⊕ν(C0) is the Jordan algebra of the quadratic form −4nC0, with unity eE. 2-grading on C considered previously combines with this Z-grading to give 2-grading as follows: The Z3 a Z × Z3 deg S± = (±2, ¯0, ¯0, ¯0), deg ν±(x) = (±1, deg x), deg E = 0 = deg eE, deg ν(a) = (0, deg a), for homogeneous elements x ∈ C and a ∈ C0. This grading will be referred to as the Z × Z3 2-grading on A. Its type is (25, 1) and again it is fine (even as a general grading). Theorem 6.3. Let Γ : A = Lg∈G the elements Ag be a grading of the Albert algebra in which ν±(1), ν(wj ), j = 1, 2, 3, are homogeneous. Then there is a group homomorphism α : Z × Z3 Γ is the grading induced by α from the Z × Z3 2-grading. In particular Z × Z3 2 is the universal group of the Z × Z3 2-grading. 2 → G such that Proof. As in Theorem 6.2, if ν(wj ) is homogeneous for Γ, then its degree has order ≤ 2, and as in Theorem 6.1, if ν±(1) is homogeneous, then deg ν+(1) deg ν−(1) = e. Since the elements above constitute a set of generators of A, the result follows. (cid:3) GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 21 Remark 6.4. Note that the stabilizer StabAut A(E1, E2, E3, ι1(1)), which is isomor- phic to Spin(C0, n) by Corollary 5.7, coincides with StabAut A(E, S+, S−). Also, relative to the Z-grading in equation (11): A±1 = {X ∈ A S∓X = 0, EX = 1 2 X}, ν(C0) = {X ∈ A S±X = 0 = EX}. Hence StabAut A(E1, E2, E3, ι1(1)) stabilizes the Z-grading. Moreover, given any c = x1·x2·. . .·x2r ∈ Spin(C0, n), i.e., xj ∈ C0 for any j and n(x1)n(x2) · · · n(x2r) = 1, the corresponding automorphism ϕc in StabAut A(E1, E2, E3, ι1(1)) fixes Ei, i = 1, 2, 3, acts as χc on ι1(C), as ρ+ c = (−1)rLx1Lx2 · · · Lx2r on ι3(C) -- see (8). But ν±(x) = ι2(x) ± iι3(¯x), so for all x ∈ C, we have: c = (−1)rRx1Rx2 · · · Rx2r on ι2(C) and as ρ− ϕc(ν±(x)) = (−1)r(cid:16)ι2(cid:0)((xx2r) · · · )x1(cid:1) ± iι3(cid:0)x1(· · · (x2r ¯x))(cid:1)(cid:17) = (−1)r(cid:16)ι2(cid:0)((xx2r) · · · )x1(cid:1) ± iι3(cid:0)((xx2r) · · · )x1(cid:1)(cid:17) = ν±(ρ+ c (x)). 6.4. Z3 of a "good basis" of C as follows: 3-grading. Define an order 3 automorphism τ of C that acts on the elements τ (ei) = ei, τ (uj ) = uj+1, τ (vj ) = vj+1 for i = 1, 2 and j = 1, 2, 3, and a new multiplication on C: x ∗ y = τ (¯x)τ 2(¯y), for all x, y ∈ C. Then n(x ∗ y) = n(x)n(y) for any x, y, since τ preserves the norm. Moreover, for any x, y, z ∈ C: n(x ∗ y, z) = n(τ (¯x)τ 2(¯y), z) = n(τ (¯x), zτ 2(y)) = n(¯x, τ 2(z)τ (y)) = n(x, τ (¯y)τ 2(¯z)) = n(x, y ∗ z). Hence (C, ∗, n) is a symmetric composition algebra (see [Eld09] or [KMRT98, Chap- ter VIII]). Actually, (C, ∗) is the Okubo algebra over F. Its multiplication table is shown in Figure 2. This Okubo algebra is Z2 3-graded by setting deg e1 = (¯1, ¯0) and deg u1 = (¯0, ¯1), with the degrees of the remaining elements being uniquely determined. Assume now that char F 6= 3. Then this Z2 3-grading is determined by two com- muting order 3 automorphisms ϕ1, ϕ2 ∈ Aut(C, ∗): ϕ1(e1) = ωe1, ϕ2(e1) = e1, ϕ1(u1) = u1, ϕ2(u1) = ωu1, where ω is a primitive third root of unity in F. 22 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV e1 e2 0 e1 e2 u1 −u2 v1 0 u2 −u3 v2 0 u3 −u1 v3 0 e2 0 u1 0 e1 −u3 0 −v2 0 v1 0 0 −v3 −e2 0 −v2 v1 −v3 0 0 u1 −e1 0 0 u2 0 −u1 −v3 0 v2 0 0 v2 −v1 0 0 u3 0 v3 −v2 −u2 0 0 −e1 −u3 −e2 0 u2 −e1 −v1 0 v3 0 0 0 −u1 0 u3 −v1 0 −u2 −e2 0 Figure 2. Multiplication table of the Okubo algebra Define now ιi(x) = ιi(τ i(x)) for all i = 1, 2, 3 and x ∈ C. Then the multiplication in the Albert algebra A = ⊕3 E2 i = Ei, EiEi+1 = 0, i=1(cid:0)FEi ⊕ ιi(C)(cid:1) is given by: (12) Eiιi(x) = 0, Ei+1ιi(x) = ιi(x)ιi+1(y) = ιi+2(x ∗ y), ιi(x) = Ei+2ιi(x), 1 2 ιi(x)ιi(y) = 2n(x, y)(Ei+1 + Ei+2), for i = 1, 2, 3 and x, y ∈ C. The commuting order 3 automorphisms ϕ1, ϕ2 of (C, ∗) extend to commuting order 3 automorphisms of A (which will be denoted by the same symbols) as follows: ϕj(Ei) = Ei, ϕj(cid:0)ιi(x)(cid:1) = ιi(ϕj (x)) for all i = 1, 2, 3, j = 1, 2 and x ∈ C. On the other hand, the linear map ϕ3 ∈ EndF(A) defined by ϕ3(Ei) = Ei+1, ϕ3(cid:0)ιi(x)(cid:1) = ιi+1(x), for all i = 1, 2, 3 and x ∈ C, is another order 3 automorphism, which commutes with ϕ1 and ϕ2. The subgroup of Aut(A) generated by ϕ1, ϕ2, ϕ3 is isomorphic to Z3 3-grading on A of type (27). This grading is obviously fine, and Z3 3 and induces a Z3 3 is its universal group. This grading will be referred to as the Z3 3-grading on A (char F 6= 3). Remark 6.5. We may define the elements ρ¯0(x) = ι1(x) + ι2(x) + ι3(x), ρ¯1(x) = ι1(x) + ω2ι2(x) + ωι3(x), ρ¯2(x) = ι1(x) + ωι2(x) + ω2ι3(x), for any x ∈ C. Then the eigenspaces of ϕ3 are: (1 = E1 + E2 + E3), A¯0 = F1 ⊕ ρ¯0(C) A¯1 = F(E1 + ω2E2 + ωE3) ⊕ ρ¯1(C), A¯2 = F(E1 + ωE2 + ω2E3) ⊕ ρ¯2(C). GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 23 The subalgebra A¯0 is isomorphic to the Jordan algebra M3(F)+, the 3 × 3 matrices with the symmetrized product, and the decomposition A = A¯0 ⊕ A¯1 ⊕ A¯2 gives the First Tits Construction of A (see [Jac68, p. 412]). 7. Classification of gradings on the Albert algebra The aim of this section is to classify the fine gradings on the Albert algebra A up to equivalence and then, for any abelian group G, all G-gradings on A up to isomorphism. Throughout this section, we will assume that the ground field F is algebraically closed of characteristic different from 2. Theorem 7.1. Let A be the Albert algebra over an algebraically closed field F, char F 6= 2. Then, up to equivalence, the fine abelian group gradings on A, their universal groups and types are the following: A defined in §6.1; universal group Z4; type (24, 0, 1). • The Cartan grading Γ1 • The grading Γ2 • The grading Γ3 • If char F 6= 3, then also the grading Γ4 A defined in §6.2; universal group Z5 A defined in §6.3; universal group Z × Z3 2; type (24, 0, 1). 2; type (25, 1). A defined in §6.4; universal group Z3 3; type (27). We already know that the gradings Γj and 6.3 for j = 1, 2, 3, obvious for Γ4 Γ : A = Lg∈G A, j = 1, 2, 3, 4, are fine (Theorems 6.1, 6.2 A). So it will suffice to show that any grading A for some j = 1, 2, 3, 4 (j 6= 4 if char F = 3), by a homomorphism U (Γj A) → G. The proof will be divided into cases according to the degree of the semisimple subalgebra Ae, which can be 1, 2 or 3 (see Corollary 5.2). Ag of the Albert algebra is induced from Γj 7.1. Degree 3. In case the degree of Ae is 3, Ae contains three orthogonal primitive idempotents, and the coordinatization results in [Jac68, §III.2 and §IX.1] show that we may assume that E1, E2, E3 are in Ae. Hence the subspaces ιi(C) = {X ∈ A : Ei+1X = Ei+2X = 1 2 X} are graded subspaces of A, i = 1, 2, 3. Assume first that for some i there is a basis of ιi(C) consisting of homogeneous elements: {ιi(xj), ιi(yj) : j = 1, 2, 3, 4} such that n(xj , yk) = δij , n(xj, xk) = 0 = n(yj, yk) (a basis consisting of four orthogonal hyperbolic pairs). This is the case if all the homogeneous components of ιi(C) are isotropic for the trace form (recall T (ιi(x)ιi(y)) = 4n(x, y) for any x, y ∈ C and any i = 1, 2, 3). We may assume i = 1. There is an element f1 ∈ SO(C, n) which takes this basis to our "good basis" B = {e1, e2, u1, u2, u3, v1, v2, v3} of C. Take c ∈ Spin(C, n) such that f1 = χc and consider the automorphism in StabAut A(E1, E2, E3) determined by the related triple (χc, ρ+ c ) (see Corollary 5.6). c , ρ− Therefore we may assume, through this automorphism, that all the elements ι1(ej), ι1(ui) and ι1(vi), for j = 1, 2 and i = 1, 2, 3, are homogeneous. Then ι1(v1)(cid:0)ι1(v2)(cid:0)ι1(v3)ι3(C)(cid:1)(cid:1) = ι2(((Cv3)v2)v1) = Fι2(e1), and this proves, since ι3(C) is a graded subspace, that ι2(e1) is homogeneous. In the same vein, we get that ι2(e2), ι3(e1) and ι3(e2) are homogeneous. Finally, ι2(u2) = −ι3(e2)ι1(u2) and ι2(v2) = −ι3(e1)ι1(v2) are homogeneous too. Theorem 6.1 finishes the proof in this case. Otherwise, in each ιi(C) we may find some homogeneous element ιi(xi) with n(xi) 6= 0, and we may scale it to get n(xi) = 1. 24 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV Lemma 7.2. Let x1, x2 ∈ C be elements of norm 1, then there is an automorphism ϕ ∈ StabAut A(E1, E2, E3) such that ϕ(ιi(xi)) = ιi(1), for i = 1, 2. Proof. First take an element f1 ∈ SO(C, n) which takes x1 to 1, and extend it as before to find a related triple (f1, f2, f3). The associated automorphism in StabAut A(E1, E2, E3) takes ι1(x1) to ι1(1) and ι2(x2) to some ι2(y2) with n(y2) = 1. Thus we may assume x1 = 1. Assuming x1 = 1, take an element a ∈ C0 with n(a) = 1, n(a, x2) = 0. Then n(x2a, 1) = n(x2, ¯a) = −n(x2, a) = 0, so x2a ∈ C0, and n(x2a) = n(x2)n(a) = 1. Consider the element c = (x2a)·a ∈ Spin(C0, n). Then (χc, ρ+ c ) is a related triple inducing an automorphism ϕ in StabAut A(E1, E2, E3) with ϕ(ι1(1)) = ι1(χc(1)) = ι1(1) and ϕ(ι2(x2)) = ι2(ρ+ (cid:3) c (x2)) = −ι2((x2a)(x2a)) = ι2(1), as required. c , ρ− Therefore, in this situation we may assume that ι1(1) and ι2(1) are homogeneous elements. Let a = deg ι1(1) and b = deg ι2(1). Since ιi(1)2 = 4(Ei+1 + Ei+2) is an idempotent, we get a2 = b2 = e. For x, y ∈ C, ι3(xy) = ι1(¯x)ι2(¯y) = (cid:0)ι2(1)ι3(x)(cid:1)(cid:0)ι3(y)ι1(1)(cid:1), so if we define Cg = {x ∈ C : ι3(x) ∈ Aabg} we get that for x ∈ Cg and y ∈ Ch, ι3(xy) ∈ (AbAabg)(AabhAa) ⊂ Aabgh, so Cg Ch ⊂ Cgh and C = ⊕g∈GCg is a G-grading on C. Hence either there is a good basis of C consisting of homogeneous elements, but then ι3(C) has a basis consisting of homogeneous elements forming four orthogonal hyperbolic pairs, and this case has already been treated, or this grading in C is equivalent to the Z3 2-grading on C, and Theorem 6.2 shows that our grading Γ is induced by the Z5 2-grading of A. In fact, we obtain more than what we need for the proof of Theorem 7.1: Proposition 7.3. Let Γ : A = Lg∈G Ag be a grading of the Albert algebra with E1, E2, E3 ∈ Ae. If there exists i = 1, 2, 3 and an element x ∈ C with n(x) = 0 and ιi(x) homogeneous, then Γ is induced from the Cartan grading. Otherwise Γ is induced from the Z5 2-grading and all homogeneous components in each ιj(C), j = 1, 2, 3, are one-dimensional and orthogonal relative to the trace form. Moreover, in the latter case, up to equivalence there are three different gradings 2 and (7, 8, 0, 1), 2 and (0, 0, 7, 0, 0, 1). The homogeneous component of highest dimension is whose universal grading groups and types are Z5 and Z3 Ae in all cases. 2 and (24, 0, 1), Z4 Proof. If ιi(x) is a nonzero homogeneous element with n(x) = 0, then since the trace form is nondegenerate and T (ιj(a)ιj (b)) = 4n(a, b) for any j = 1, 2, 3 and a, b ∈ C, there is another homogeneous element ιi(y) with n(y) = 0 and n(x, y) = 1. Then n(x + y) = 1 so C = (¯x + ¯y)C = ¯xC + ¯yC. As ¯xC and ¯yC are isotropic spaces, its dimension is at most 4. We get C = ¯xC ⊕ ¯yC, so ιi+2(C) = ιi+2(¯xC) ⊕ ιi+2(¯yC) = ιi(x)ιi+1(C) ⊕ ιi(y)ιi+1(C) is the direct sum of two isotropic graded subspaces (for the trace form). Therefore, ιi+2(C) has a basis consisting of homogeneous elements forming four orthogonal hyperbolic pairs, and hence Γ is induced from the Cartan grading. Otherwise all the homogeneous components in each graded subspace ιj(C) are one dimensional and not isotropic, and hence orthogonal relative to the trace form, because of Theorem 5.1. The arguments preceding this proposition show that we may assume deg ι1(1) = a, deg ι2(1) = b and deg ι3(wj ) = abcj, j = 1, 2, 3, with all the elements a, b, c1, c2, c3 having order 2, and that C is graded with deg wj = cj, GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 25 j = 1, 2, 3, so the subgroup H generated by c1, c2, c3 is isomorphic to Z3 2. If a, b ∈ H, Lemma 7.2 allows us to assume a = b = e and we get Supp Γ = H ∼= Z3 2. If only one of a, b or ab are in H, by symmetry we may assume a ∈ H, and again we may assume a = e, thus getting Supp Γ = hb, Hi ∼= Z4 2, and Γ is equivalent to the fine Z5 (cid:3) 2. Otherwise Supp Γ ∼= Z5 2-grading. The types are easily computed. 7.2. Degree 2. If the degree of Ae is 2, Corollary 5.2 shows that Ae = FE⊕F(1−E) for an idempotent E with T (E) = 1 (and hence T (1−E) = 2). We may assume that E = E1, so that eE = 1 − E = E2 + E3. The grading on A restricts to a grading on {X ∈ A EX = 0} = FE2 ⊕ FE3 ⊕ ι1(C) = FeE ⊕ V, where V = F(E2 − E3) ⊕ ι1(C), which is the Jordan algebra of a quadratic form with unity eE, because (E2 −E3)2 = 2 T (ι1(x)ι1(y))eE. E2 + E3 = eE, (E2 − E3)ι1(C) = 0 and ι1(x)ι1(y) = 2n(x, y)eE = 1 2 T (XY )eE for any X, Y ∈ V. But the gradings on the Jordan Hence XY = 1 algebras of quadratic forms are quite easy to describe: the unity is always in the identity component, and the restriction of the grading to the vector space V is just a decomposition into subspaces: V = ⊕g∈GVg, with T (VgVh) = 0 unless gh = e. Then either: 1) For any g ∈ Supp(V), g2 = e and dim Vg = 1, or 2) There are homogeneous elements X, Y ∈ V with T (X 2) = T (Y 2) = 0 and T (X, Y ) = 1. Let us prove that the first case is not possible. Assume that for any g ∈ Supp(V), g2 = e and dim Vg = 1. Let H be the subgroup of G generated by Supp(V), which is 2-elementary: H ∼= Zr 2, with r ≥ 4 as dim V = 9. Since {e} ∪ Supp(V) has 10 elements, it is not a subgroup of H, and hence there are elements g 6= h ∈ Supp(V) such that gh 6∈ Supp(V). Then Vg = FX for some X with X 2 = E. Hence 2 (eE + X) and eE3 = 1 2 (eE − X) are nonzero orthogonal idempotents whose eE2 = 1 sum is eE = 1 − E1. Thus E1, eE2 and eE3 are orthogonal primitive idempotents and 2 (eE + X) and E3 = 1 2 (eE − X), so that X = E2 − E3. we may assume that E2 = 1 Then we have V = F(E2 − E3) ⊕ ι1(C) and g 6∈ Supp(ι1(C)). Let G = G/hgi and consider the induced G-grading on A, denoting by ¯a the class of a ∈ G modulo hgi. Then E1, E2, E3 ∈ A¯e, so that each ιi(C) are graded subspaces. Besides, ι1(C) is already a graded subspace of the original G-grading whose homogeneous components are all one-dimensional and non isotropic (relative to the norm of C). Moreover, since ι1(C)gh = Vgh = 0, ι1(C)¯h = ι1(C)h ⊕ ι1(C)gh = ι1(C)h is one-dimensional and not isotropic. Proposition 7.3 gives that each homogeneous component of the G-grading on each ιi(C) is one-dimensional and not isotropic. Take a ∈ G such that ι2(C)¯a 6= 0, so that there is an element x ∈ C with n(x) 6= 0 such that ι2(C)¯a = Fι2(x). Then: (cid:0)ι2(C) ⊕ ι3(C)(cid:1)¯a = ι2(C)¯a ⊕ ι3(C)¯a. If ι3(C)¯a = 0, then ι2(x) is homogeneous for the G-grading, and so is ι2(x)2 = 4n(x)(E1 + E3), a contradiction with Ae = FE1 ⊕ F(E2 + E3). Hence we have ι3(C)¯a 6= 0. We conclude that the supports, for the G-grading, of both ι2(C) and ι3(C) coin- cide. But since n(x) 6= 0, we have ι3(C) = ι1(C)ι2(C)¯a. Since ι3(C)¯a 6= 0, it follows that ι1(C)¯e 6= 0, which means ι1(C)g 6= 0, a contradiction with g 6∈ Supp(ι1(C)). 26 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV We are left with the second case, i.e., there are homogeneous elements X ∈ Vg, Y ∈ Vg−1 with T (X 2) = T (Y 2) = 0 and T (XY ) = 1, and g 6= e because Ae = 2 (eE+X+Y ) FE⊕FeE. Then (X+Y )2 = T (XY )eE = eE and hence 1 are nonzero idempotents with sum eE, so we may assume X + Y = E3 − E2. Then X − Y is an element of {Z ∈ A E1Z = 0 = (E2 − E3)Z} = ι1(C), and T ((X − Y )2) = −2. By Lemma 7.2, we may assume X − Y = i 2 ι1(1). In other words, we may assume that the elements S+ = X = (E3 − E2) + i 2 ι1(1) and 2 ι1(1) are homogeneous, say S+ ∈ Ag and S− ∈ Ag−1 S− = Y = (E3 − E2) − i 2 (eE−X−Y ) and 1 Consider the Z-grading of A in (11). The subspaces A±1 = {Z ∈ A EZ = (because S+S− = 2eE ∈ Ae). 2 Z, S±Z = 0} are then graded subspaces as well as A0 = FE ⊕ FeE ⊕ ν(C0), since ν(C0) = {Z ∈ A EZ = 0 = S±Z}. Assume now that there is an element 0 6= x ∈ C with n(x) = 0 such that ν+(x) is homogeneous: ν+(x) ∈ (A1)h1 . The nondegeneracy of the trace form shows with T (ν+(x)ν−(y)) = that there is an homogeneous element ν−(y) ∈ (A−1)h 1 −1 1 8n(x, y) 6= 0. Then ν+(x)ν−(y) = 2n(x, y)(2E + eE) + ν(¯xy − ¯yx) ∈ Ae = FE ⊕ FeE. Hence ¯xy = ¯yx. But then n(x, y)1 = ¯xy + ¯yx = 2¯xy, a contradiction, since n(¯xy) = n(x)n(y) = 0 while n(x, y) 6= 0 and n(1) = 1 6= 0. Therefore, all the homogeneous components in A1 are one-dimensional and not isotropic (relative to the norm of C once we identify A1 = ν+(C) with C). Fix an homogeneous element ν+(x) ∈ (A1)a, with n(x) = 1. Then ν+(x)2 = 4n(x)S+, so a2 = g. The proof of Lemma 7.2 shows that there is an element c ∈ Spin(C0, n) such that ρ+ c (x) = 1, so Remark 6.4 allows us to assume that x = 1. Thus we have ν+(1) ∈ (A1)a, a2 = g, and hence ν−(1) = S−ν+(1) ∈ (A−1)a−1 . In this situation, for any x, y ∈ C such that ν+(x) ∈ (A1)h1 , ν+(y) ∈ (A1)h2, we have: (ν+(x)ν−(1))ν+(y) =(cid:16)2n(x, 1)(2E + E) + ν(¯x − x)(cid:17)ν+(y) = 3n(x, 1)ν+(y) + ν+(y(¯x − x)) = 4n(x, 1)ν+(y) − ν+(yx), as x + ¯x = n(x, 1)1. If n(x, 1) 6= 0, then 0 6= ν+(x)ν+(1) ∈ FS+, so that h1a = g = a2, so h1 = a and (ν+(x)ν−(1))ν+(y) ∈ (A1)aa−1h2 = (A1)h2 , and ν+(yx) ∈ (A)a−1h1h2. On the other hand, if n(x, 1) = 0, then ν+(yx) = −(ν+(x)ν−(1))ν+(y) ∈ (A1)a−1h1h2 too. Thus, consider the subspaces Ch = {x ∈ C ν+(x) ∈ (A1)ah} for h ∈ G. Then Ch2 ⊂ Ch1h2 and we get a grading of C in which all the homogeneous components Ch1 are one-dimensional. Hence this is isomorphic to the Z3 2-grading of C. Since 1 ∈ Ce, we have ν+(1) ∈ Aa, ν−(1) ∈ Aa−1, and ν(wj ) = ν+(wj)ν−(1) are homogeneous too, for w1, w2 and w3 as in Theorem 6.3. This Theorem shows that Γ is induced from the Z × Z3 2-grading. In fact, we can say more. Let a = deg ν+(1) and bj = deg ν(wj ), j = 1, 2, 3. Then the subgroup H = hb1, b2, b3i is isomorphic to Z3 2 and a2 = g 6= e as dim Ae = 2. Then Supp Γ = ha, Hi, and the homogeneous components of the 5-grading in (11) have supports Supp A±2 = {a±2}, Supp A±1 = a±1H, Supp A0 = H. If this subsets are disjoint, Γ is equivalent to the Z × Z3 2-grading. Otherwise we have one of the following possibilities: • a4 = e but a2 6∈ H, thus getting a Z4 × Z3 2-grading of type (23, 2). GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 27 • a2 ∈ a−1H. In this case a3 = b ∈ H, and hence (ab)3 = 1 and (ab)2 = a2. As before we may change a by ab and hence assume a3 = e. We get a Z3 × Z3 2-grading of type (21, 3). • a2 ∈ H (recall a2 6= e). Since all the homogeneous components of the Z3 2- grading of C, with the exception of the neutral component, play the same role we may assume a2 = b1 and we obtain a unique, up to equivalence, grading by Z4 × Z2 2 of type (6, 9, 1). We summarize our arguments: Proposition 7.4. Let Γ : A = ⊕g∈GAg be a grading of the Albert algebra with dim Ae = 2. Then Γ is induced from the Z×Z3 2-grading. Moreover, up to equivalence there are four such different gradings whose universal grading groups and types are Z × Z3 2 and (6, 9, 1). (cid:3) 2 and (21, 3), and Z4 × Z2 2 and (23, 2), Z3 × Z3 2 and (25, 1), Z4 × Z3 7.3. Degree 1. Finally, consider the case of a grading Γ : A = Lg∈G Albert algebra with dim Ae = 1, or Ae = F1. Ag of the Let g ∈ Supp Γ be an element of order 2. Let G = G/hgi and consider the induced G-grading. Then A¯e = Ae ⊕ Ag is a degree two Jordan algebra, so dim Ag = 1 by Corollary 5.2, and A¯e = FE ⊕ F(1 − E) for an idempotent E with T (E) = 1. But Ag = {X ∈ A¯e X 6∈ F1, X 2 ∈ F1} ∪ {0} = F(1 − 2E), and T (1 − 2E) = 3 − 2 = 1, while T (Ag) = T (AgAe) = 0 by Theorem 5.1, a contradiction. Therefore, for any element g ∈ Supp Γ, we have g = e or the order of g is at least 3. Take now an element g ∈ Supp Γ, g 6= e (so its order is at least 3), and take X ∈ Ag and Y ∈ Ag−1 with T (XY ) 6= 0 Hence 0 6= XY ∈ Ae = F1 and we may take XY = 1. This implies T (1) 6= 0, which shows that char F 6= 3. The first linearization of equation (6) gives X 2Y + 2(XY )X − T (Y )X 2 − 2T (X)XY + S(X)Y + S(X, Y )X − N (X; Y )1 = 0 (N (X; Y ) being quadratic on X and linear on Y ). But T (X) = T (X 2) = T (Y ) = 0, so the component in Ag of the above equation gives X 2Y + 2(XY )X + S(X, Y )X = 0, and S(X, Y ) = −T (XY ) = −3, so that X 2Y + 2X − 3X = 0, or X 2Y = X. Then X is invertible in the Jordan sense [Jac68, p. 51] with inverse Y . Since T (X) = 0 = S(X), we have X 3 − N (X)1 = 0, so 0 6= X 3 ∈ Ae, which forces g3 = e. Therefore, any element of Supp Γ different from e has order 3. Since we may assume that G is generated by Supp Γ, we conclude that G is an elementary 3-group. Moreover, with X as above, the quadratic operator UX is invertible and takes any Ah to Ag2h. In particular Ae = UX (Ag), which forces dim Ag = 1. Also, for any other h ∈ Supp Γ, UX (Ah) = Ag2h, so we get that for any g, h ∈ Supp Γ, g−1h ∈ Supp Γ. It follows that Supp Γ is a group, isomorphic to Z3 3. Since we have shown that char F 6= 3, the grading Γ is given by three commuting order 3 automorphisms ϕ1, ϕ2, ϕ3 of A. Let S be the subalgebra of elements fixed by ϕ1 and ϕ2. Then dim S = 3, S = Ae ⊕ Ag ⊕ Ag2 for some g ∈ Supp Γ. Take X ∈ Ag with X 3 = 1. Thus S is isomorphic to F × F × F, and we may assume that S = FE1 ⊕ FE2 ⊕ FE3 with ϕ3(Ei) = Ei+1 for any i = 1, 2, 3. For each i, the subspace ιi(C) = {X ∈ A Ei+1X = 1 2 X = Ei+2X} is invariant under ϕ1 and ϕ2, while ϕ3(ιi(C)) = ιi+1(C). 28 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV For x, y ∈ C define x ∗ y by ι3(x ∗ y) = ϕ3(ι3(x))ϕ2 3(ι3(y)). Then: ι3((x ∗ y) ∗ x) = ϕ3(ι3(x ∗ y))ϕ2 3(ι3(x)) =(cid:0)ϕ2 = ϕ2 3(ι3(x))ι3(y)(cid:1)ϕ2 3(ι3(x))(cid:0)ϕ2 3(ι3(x))ι3(y)(cid:1). 3(ι3(x)) 3(ι3(x)) = ι2(x′) for some x′ ∈ C with n(x) = n(x′) (since T (ιi(x)2) = 8n(x) But ϕ2 and T is invariant under ϕ3), and ι2(x′)(cid:0)ι2(x′)ι3(y)(cid:1) = ι2(x′)ι1(¯x′ ¯y) = ι3(¯x′ ¯y ¯x′) = ι3((yx′)¯x′) = n(x′)ι3(y) = n(x)ι3(y). Hence (x ∗ y) ∗ x = n(x)y and, in the same vein, we get x ∗ (y ∗ x) = n(x)y. It follows that (C, ∗) is a symmetric composition algebra (see [KMRT98, Chapter VIII]), and ϕ1 and ϕ2 give, by restriction to ι3(C), two commuting order 3 automorphisms of (C, ∗), and hence a grading of (C, ∗) by Z2 3. We obtain that (C, ∗) is the Okubo algebra over F and the grading is the unique, up to equivalence, Z2 3-grading on (C, ∗) [Eld09]. Moreover, setting ιi(x) = ϕi 3(ι3(x)), we recover exactly the multiplication in A in equations (12). This shows that Γ is equivalent to the Z3 3-grading of A. The proof of Theorem 7.1 is complete. 7.4. Classification of G-gradings up to isomorphism. Now we obtain, for any abelian group G, a classification of G-gradings on A up to isomorphism. We will need the following result describing the Weyl groups of the fine gradings Γj A, j = 1, 2, 3, 4. Theorem 7.5 ([EK]). Identifying Supp Γ1 Φ of type F4, we have W (Γ1 of the subgroup Z3 subgroup of Aut(Z3 2 (as a set). W (Γ3 3). A) = Aut Φ. W (Γ2 A) is the stabilizer in Aut(Z2 A with the short roots of the root system 2 × Z3 2) A) is the commutator (cid:3) 2). W (Γ4 A) = Aut(Z × Z3 1 bw2j 2 bw3j 3 To state our classification theorem, we introduce the following notation: • Let γ = (b1, b2, b3, b4) be a quadruple of elements in G. Denote by Γ1 A(G, γ) A by the homomorphism Z4 → G sending the G-grading on A induced from Γ1 the i-th element of the standard basis of Z4 to bi, i = 1, 2, 3, 4. For two such quadruples, γ and γ′, we will write γ ∼ γ′ if there exists w ∈ Aut Φ such that j = bw1j b′ bw4j 4 where w = (wij ) is considered as an element of GL4(Z). • Let γ = (b1, b2, b3) be a triple of elements in G with b1b2b3 = e and b2 i = e, i = 1, 2, 3. Let H ⊂ G be a subgroup isomorphic to Z3 2. Fix an isomorphism α : Z3 A by the 2 → G sending the i-th element of the standard basis of Z2 homomorphism Z2 2 to bi, i = 1, 2, and restricting to α on Z3 2. It follows from Theorem 7.5 that the isomorphism class of the induced grading does not depend on the choice of α. For two such triples, γ and γ′, we will write γ ∼ γ′ if there exists π ∈ Sym(3) such that b′ i ≡ bπ(i) (mod H) for all i = 1, 2, 3. A(G, H, γ) the G-grading induced from Γ2 2 → H and denote by Γ2 2 × Z3 • Let g be an element of G such that g2 6= e. Let H ⊂ G be a subgroup A(G, H, g) 2 → G sending the isomorphic to Z3 the G-grading induced from Γ3 A by the homomorphism Z × Z3 2. Fix an isomorphism α : Z3 2 → H and denote by Γ3 GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 29 element 1 in Z to g and restricting to α on Z3 2. It follows from Theorem 7.5 that the isomorphism class of the induced grading does not depend on the choice of α. For two elements, g and g′, we will write g ∼ g′ if g′ ≡ g (mod H) or g′ ≡ g−1 (mod H). 3. Then Γ4 A) has index 2 in Aut(Z3 • Let H ⊂ G be a subgroup isomorphic to Z3 A may be regarded as a G-grading with support H. Since W (Γ4 3), there are two isomorphism classes among the induced gradings αΓ4 A for various isomorphisms α : Z3 3 → H. They can be distinguished as follows: fix a primitive third root of unity ω and a generating set {g1, g2, g3} for H, then in one isomorphism class we will have (X1X2)X3 = ωX1(X2X3) and in the other (X1X2)X3 = ω−1X1(X2X3) where Xi are nonzero elements with deg Xi = gi, i = 1, 2, 3 -- see [EK, §4.5]. We denote these two (isomorphism classes of) G-gradings by Γ4 A(G, H, δ) where δ ∈ {+, −}. Theorem 7.6. Let A be the Albert algebra over an algebraically closed field of characteristic different from 2. Let G be an abelian group. Then any G-grading on A is isomorphic to some Γ1 A(G, H, δ) (characteristic 6= 3 in this latter case), but not two from this list. Also, A(G, H, g) or Γ4 A(G, H, γ), Γ3 A(G, γ), Γ2 • Γ1 • Γ2 A(G, γ) is isomorphic to Γ1 A(G, H, γ) is isomorphic to Γ2 A(G, γ′) if and only if γ ∼ γ′; A(G, H ′, γ′) if and only if H = H ′ and γ ∼ γ′; • Γ3 A(G, H, g) is isomorphic to Γ3 A(G, H ′, g′) if and only if H = H ′ and g ∼ g′; A(G, H, δ) is isomorphic to Γ4 A(G, H ′, δ′) if and only if H = H ′ and δ = • Γ4 δ′. A) → G. In the case j = 2, if the restriction αZ3 Proof. By Theorem 7.1, we know that any G-grading Γ : A = Lg∈G Ag is iso- morphic to αΓj A for some j = 1, 2, 3, 4 (j 6= 4 if char F = 3) and a homomorphism α : U (Γj is not one-to-one, then Propositions 7.3 tells us that Γ can also be induced from Γ1 A by a homomorphism Z4 → G. In the case j = 3, if the restriction αZ3 is not one-to-one or 1 ∈ Z is sent to an element of order ≤ 2, then Propositions 7.4 implies that the degree of the algebra Ae is 3 and hence, by Propositions 7.3, Γ is isomorphic to a grading induced from Γ1 A. In the case j = 4, if α is not one-to-one, then Ae has degree 3 and the same argument applies. We have shown that Γ is isomorphic to a grading from our list. A or Γ2 2 2 Now, two gradings on our list that have different j's cannot be isomorphic, because the degree of Ae is 1 for j = 4, it is 2 for j = 3, and 3 for j = 1, 2; in the latter case the gradings can be distinguished as follows: for any grading induced from Γ1 A by a homomorphism Z4 → G where G is an elementary 2-group, every homogeneous component Ag, g 6= e, has even dimension, whereas the gradings Γ2 A(G, H, γ) possess homogeneous components of odd dimension other than Ae (see their types in Proposition 7.3). It remains to consider isomorphisms between two gradings with the same j. The "if" part follows from Theorem 7.5, which shows that one grading can be mapped to the other by an automorphism in Aut(Γj A). The proof of the "only if" part will be divided into cases according to the value of j. 30 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV 1) Since Γ1 A is the eigenspace decomposition relative to a 4-dimensional torus in Aut(A), and the latter is the simple algebraic group of type F4, this case is covered by Proposition 3.9. A(G, H, γ) to Γ′ = Γ2 2) Suppose ϕ ∈ Aut(A) sends Γ = Γ2 A(G, H ′, γ′). Then, in particular, it maps Ae to A′ e. If bi ∈ H for all i, then Supp Γ = H and hence Supp Γ′ = H, which forces H ′ = H and b′ i ∈ H for all i. Suppose that at least one of the bi is not in H. Then, in fact, at least two of them, say b2 and b3, are not in H. Hence Ae is not simple -- precisely, FE1 is a factor of Ae. Then Fϕ(E1) is a factor of A′ e and hence the idempotent ϕ(E1) is one of Ei, i = 1, 2, 3. The automorphism of A defined by Ei 7→ Ei+1, ιi(x) 7→ ιi+1(x), for all x ∈ C and i = 1, 2, 3, belongs to Aut(Γ2 It follows that ϕ leaves the subspace FE2 ⊕ FE3 ⊕ ι1(C) invariant. The support of this subspace is, on the one hand, b1H and, on the other hand, b′ 1H ′. It follows that H = H ′ and b1 ≡ b′ 1 (mod H). Also, ϕ leaves the subspace ι2(C) ⊕ ι3(C) invariant, and the support of this subspace is, on the one hand, b2H ∪ b3H and, on the other 3 (mod H), or b2 ≡ b′ hand, b′ 3 (mod H) and b3 ≡ b′ A), so we may assume without loss of generality that ϕ(E1) = E1. 3H. It follows that b2 ≡ b′ 2 (mod H) and b3 ≡ b′ 2H ∪b′ 2 (mod H). 3) Suppose ϕ ∈ Aut(A) sends Γ3 A(G, H ′, g′). Since E = E1 is the unique idempotent of trace 1 in Ae and in A′ e, we have ϕ(E1) = E1. Hence the subspaces FE2 ⊕ FE3 ⊕ ι1(C) and ι2(C) ⊕ ι3(C) are invariant under ϕ. Looking at the supports, we get: A(G, H, g) to Γ3 H ∪ {g±2} = H ′ ∪ {(g′)±2} and gH ∪ g−1H = g′H ′ ∪ (g′)−1H ′. The first condition shows that the intersection H ∩ H ′ has at least 6 elements, and hence it generates both H and H ′. Therefore, H = H ′. Now the second condition gives that g′ ≡ g (mod H) or g′ ≡ g−1 (mod H). 4) This case is clear from the definition of Γ4 A(G, H, δ). (cid:3) Corollary 7.7. Let A be the Albert algebra over an algebraically closed field of characteristic different from 2. Then any abelian group grading on A is either induced from the Cartan grading or is equivalent to one of the following: • a Z5 • a Z× Z3 2-grading of type (24, 0, 1), a Z4 2-grading of type (7, 8, 0, 1), or a Z3 2- grading of type (0, 0, 7, 0, 0, 1), if the degree of the neutral component is 3; 2-grading of type (23, 2), a Z3 × Z3 2- 2-grading of type (6, 9, 1), if the degree 2-grading of type (25, 1), a Z4 × Z3 grading of type (21, 3), or a Z4 × Z2 of the neutral component is 2; • a Z3 3-grading of type (27) if the degree of the neutral component is 1 and (cid:3) the characteristic is not 3. 8. Gradings on F4 We continue to assume that the ground field F is algebraically closed and char F 6= 2. The simple Lie algebra of type F4 appears as the algebra of derivations of the Albert algebra. In order to describe it, consider first the local version of Definition 5.3. Let C be the Cayley algebra over F. Its triality Lie algebra is defined as tri(C) = {(d1, d2, d3) ∈ so(C, n)3 d1(x • y) = d2(x) • y + x • d3(y) ∀x, y ∈ C}. (Recall x•y = ¯x¯y and lx(y) = ry(x) = x•y.) As in Lemma 5.4, if (d1, d2, d3) belongs to tri(C), so does (d3, d1, d2). The Lie bracket in tri(C) is the componentwise bracket, GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 31 and we get the order 3 automorphisms θ (triality automorphism): (13) θ : (d1, d2, d3) 7→ (d3, d1, d2). Each triple (d1, d2, d3) ∈ tri(C) induces a derivation of the Albert algebra A: (14) D(d1,d2,d3) : Ei 7→ 0, ιi(x) 7→ ιi(di(x)), for any i = 1, 2, 3 and x ∈ C. Also, for any x ∈ C and i = 1, 2, 3, consider the derivation Di(x) = 2[Lιi(x), LEi+1]: (15) Di(x) : Ei ιi(y) ιi+1(y) ιi+2(y) 2 ιi(x), Ei+2 7→ − 1 2 ιi(x), 7→ 0, Ei+1 7→ 1 7→ 2n(x, y)(−Ei+1 + Ei+2), 7→ −ιi+2(x • y), 7→ ιi+1(y • x), for all y ∈ C. Then we get ([Jac68, Theorem IX.17]): (16) Der(A) = Dtri(C) ⊕(cid:0) 3Mi=1 Di(C)(cid:1). One verifies at once the following properties (see [Eld09, §5.3]): (17) [D(d1,d2,d3), Di(x)] = Di(di(x)), [Di(x), Di+1(y)] = Di+2(x • y), [Di(x), Di(y)] = 2θi(Dx,y) for all x, y ∈ C, (d1, d2, d3) ∈ tri(C) and i = 1, 2, 3, where θ is the triality automor- phism in (13) and where Dx,y = Dtx,y with tx,y =(cid:0)σx,y, 1 2 n(x, y)id − rxly, 1 2 n(x, y)id − lxry(cid:1) ∈ tri(C), σx,y(z) = n(x, z)y − n(y, z)x ∈ so(C, n). Moreover, the projection of tri(C) onto any of its components gives an isomorphism tri(C) → so(C, n). Take a "good basis" B = {e1, e2, u1, u2, u3, v1, v2, v3} of C and consider the sub- space h of g = Der(A) spanned by De1,e2 and Dui,vi for i = 1, 2, 3. This is an abelian subalgebra of g. Actually, the image of h in so(C, n) under the projection of tri(C) onto its first component is the span of σe1 ,e2 and σui,vi , i = 1, 2, 3, so it is a Cartan subalgebra of so(C, n). Consider the linear maps ǫj : h → F, j = 0, 1, 2, 3, that constitute the dual basis to Duj ,vj , j = 0, 1, 2, 3, where u0 := e1 and v0 := e2. Since we have: σe1 ,e2 : e1 7→ −e1, σui,vi : ui 7→ −ui, e2 7→ e2, vi 7→ vi, ui, vi 7→ 0, e1, e2, uj, vj 7→ 0 (j 6= i) 1 2 id − re1 le2 : 1 2 id − rui lvi : 1 2 id − le1 re2 : 1 2 id − luirvi : e1 7→ 1 2 e1, e1 7→ 1 2 e1, uj 7→ 1 2 uj, e1 7→ 1 2 e1, e1 7→ − 1 2 e1, uj 7→ 1 2 uj, e2 7→ − 1 e2 7→ − 1 2 e2, ui 7→ − 1 2 e2, ui 7→ − 1 2 ui, vi 7→ 1 2 ui, vi 7→ 1 2 vi, 2 vi, vj 7→ − 1 e2 7→ − 1 (j 6= i), 2 vj 2 e2, ui 7→ 1 e2 7→ 1 2 e2, vj 7→ − 1 2 vj ui 7→ − 1 (j 6= i), 2 ui, vi 7→ − 1 2 ui, vi 7→ 1 2 vi, 2 vi, 32 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV we obtain that the weights of h in ι1(C), and hence the roots in D1(C), are ±ǫj, j = 0, 1, 2, 3, the weights in ι2(C), and hence the roots in D2(C), are 1 2 (±ǫ0 ± ǫ1 ± ǫ2 ± ǫ3) with an even number of + signs, and the weights in ι3(C), and hence the roots in D3(C), are 1 2 (±ǫ0 ± ǫ1 ± ǫ2 ± ǫ3) with an odd number of + signs. From [σa,b, σx,y] = σσa,b(x),y + σx,σa,b(y) for any a, b, x, y ∈ C we obtain that the roots in Dtri(C) are ±ǫr ± ǫs, 0 ≤ r 6= s ≤ 3. Hence h is a Cartan subalgebra of g with the following set of roots: Φ = {±ǫr ± ǫs 0 ≤ r 6= s ≤ 3} ∪ {±ǫr 0 ≤ r ≤ 3} ∪ { 1 2 (±ǫ0 ± ǫ1 ± ǫ2 ± ǫ3)}. Note that the root spaces in Dtri(C) are the subspaces FDui,uj , FDui,vj and FDvi,vj for 0 ≤ i 6= j ≤ 3, while in Di(C), i = 1, 2, 3, the root spaces are the subspaces FDi(x) for x ∈ B. It follows at once that for any α ∈ Φ and Xα ∈ gα, the linear maps X 3 on g are zero. Consider the Z4-grading on g induced by the Cartan grading on A. Its ho- mogeneous components are precisely the root spaces above, i.e., it is the Cartan decomposition of g relative to h. We will call it the Cartan grading on g. α on A, and ad 3 Xα Proposition 8.1. Let A be the Albert algebra over an algebraically closed field of characteristic different from 2 and let g = Der(A). Then any derivation of g is inner. Proof. This is well-known for char F 6= 2, 3 (see [Sel67]). We include a proof that is valid also in characteristic 3, where the Killing form is trivial. The Cartan grading on g induces a grading on Der(g). It suffices to consider homogeneous elements D ∈ Der(g). Suppose a ∈ Z4 and D ∈ Der(g)a. If ga = 0, then D(h) = D(g0) = 0. If a = 0, then the subspaces g0 and gα, α ∈ Φ, are invariant under D and hence D(h) = 0 again. Finally, suppose ga = FXα for some α ∈ Φ. Then there is a linear map λ : h → F such that D(H) = λ(H)Xα for all H ∈ h. Hence for any H1, H2 ∈ h, we have 0 = D([H1, H2]) = [D(H1), H2] + [H1, D(H2)], which gives λ(H1)α(H2) = λ(H2)α(H1). Therefore, either λ = 0 or the linear maps λ and α have the same kernel, so λ = µα for some µ ∈ F. Hence the derivation D + µad Xα annihilates h. We have shown that Der(g) = ad (g) + {D ∈ Der(g) D(h) = 0}. Now take a system ∆ of simple roots. For instance, ∆ = {α1, α2, α3, α4} (18) where α1 = 1 2 (ǫ0 − ǫ1 − ǫ2 − ǫ3), α2 = ǫ3, α3 = ǫ2 − ǫ3 and α4 = ǫ1 − ǫ2. Any derivation D ∈ Der(g) which annihilates h preserves the root spaces, so there are scalars µi ∈ F such that D(Xαi ) = µiXαi , and hence D(X−αi) = −µiX−αi. Take H ∈ h such that αi(H) = µi for i = 1, 2, 3, 4. The multiplication rules in (17) show that the elements X±αi, i = 1, 2, 3, 4, generate g. It follows that D = ad H , which completes the proof. (cid:3) Proposition 8.2. Let A be the Albert algebra over an algebraically closed field of characteristic different from 2 and let g = Der(A). Then the map Ad : Aut(A) → Aut(g), ϕ 7→ (D 7→ ϕ ◦ D ◦ ϕ−1), is a group isomorphism. Proof. Again, this is well-known for char F 6= 2, 3 (see [Sel67, p. 71]). We include a proof that works also in characteristic 3. Since ϕ ◦ ad X ◦ ϕ−1 = ad ϕ(X) for all X ∈ g, we see that Ad is one-to-one. The following argument will show that it is onto. GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 33 Consider the order 2 automorphism of C given by: (19) σ : e1 ↔ e2, ui ↔ vi, for all i = 1, 2, 3. This automorphism σ extends to an order 2 automorphism of A by means of σ(Ei) = Ei, σ(ιi(x)) = ιi(σ(x)), for all i = 1, 2, 3 and x ∈ C, and hence it induces an order 2 automorphism of g, which will be denoted by σ as well. Note that the restriction of σ to h is −id, and σ takes any root space gα to g−α. Given x, y, x′, y′ ∈ C with n(x, x′) = 1 = n(y, y′) and n(Fx + Fx′, Fy + Fy′) = 0, we get [[σx,y, σx′,y′], σx,y] = [σx,x′ + σy,y′ , σx,y] = −2σx,y. Hence, in particular, for i 6= j, we obtain: [[Dui,uj , −σ(Dui,uj )], Dui,uj ] = 2Dui,uj , where, as before, u0 = e1 and v0 = e2. It follows that {[Dui,uj , −σ(Dui,uj )], Dui,uj , −σ(Dui,uj )} is an sl2-triple in g, i.e., a triple {E, F, H} satisfying [H, E] = 2E, [H, F ] − 2F and [E, F ] = H, and thus spanning a subalgebra isomorphic to sl2(F). With the same arguments we get sl2-triples starting with Dui,vj or Dvi,vj , 0 ≤ i 6= j ≤ 3. In a similar vein, for x in the "good basis" B of C: [[Di(x), Di(σ(x))], Di(x)] = 2[θi(Dx,σ(x)), Di(x)] = 2Di(cid:0)σx,σ(x)(x)(cid:1) = −2Di(x), so {[Di(x), −σ(Di(x))], Di(x), −σ(Di(x))} is an sl2-triple. Take the system ∆ of simple roots in (18), and the corresponding set of positive roots: Φ+ = {ǫr, ǫr ± ǫs, 1 2 (ǫ0 ± ǫ1 ± ǫ2 ± ǫ3) 0 ≤ r < s ≤ 3}. For each α ∈ Φ+, choose the nonzero element Xα in the root spaces gα to be of the form Dx,y or Di(x) for some x, y ∈ B and i = 1, 2, 3. In particular, Xα1 = D3(e1), Xα2 = D1(v3), Xα3 = Dv2,v3, Xα4 = Dv1,v2 . Take Xα = −σ(X−α) for α ∈ Φ− = −Φ+. With Hi = [Xαi , −σ(Xαi)], the basis BCh = {Hi, Xα 1 ≤ i ≤ 4, α ∈ Φ} is a Chevalley basis of g (see [Hum72, proof of Proposition 25.2]) whose structure constants lie in Z if char F = 0 and in the field Z/pZ if char F = p. Moreover, the structure constants of the action of the elements Xα on A are in 1 2 Z if char F = 0 and in Z/pZ if char F = p. Let AC be the complex Albert algebra, so that gC = Der(AC) is the simple Lie 2n a ∈ Z, n ∈ N}. In AC, 2 ] and 2 ]. Then our Albert algebra A 2 ] F, and its Lie algebra of derivations g = Der(A) algebra of type F4 over C. Consider the ring Z[ 1 let AZ[ 1 let gZ[ 1 over F is isomorphic to AZ[ 1 is isomorphic to gZ[ 1 2 ] be the linear span of the basis {Ei, ιi(x) i = 1, 2, 3, x ∈ B} over Z[ 1 2 ] be the linear span of the basis BCh over Z[ 1 2 ] = { a 2 ] ⊗Z[ 1 2 ] F. 2 ] ⊗Z[ 1 According to Steinberg [Ste61, 4.1], the automorphism group of g is generated by the operators exp(µ ad Xα ), α ∈ Φ, µ ∈ F×. These are indeed automorphisms, even in characteristic 3, since they are obtained by specialization from the automorphism exp(t ad Xα ) in gZ[ 1 2 , t], which is a subalgebra of gC if we identify t with a transcendental element in C. (Here we are using the same symbol Xα to denote an 2 ] ⊗Z[ 1 2 ] Z[ 1 34 ALBERTO ELDUQUE AND MIKHAIL KOCHETOV element in gC = Der(AC) and in g = Der(A), but this should cause no confusion.) Now, exp(t ad Xα)(Y ) = exp(tXα)Y exp(−tXα) = (exp tXα)Y (exp tXα)−1, for all Y ∈ AC, i.e., we have exp(t ad Xα) = Ad (exp tXα). The operator exp µXα on A is an automorphism of A, since it is obtained by spe- 2 , t]. We also have exp(µ ad Xα ) = (cid:3) cialization from an automorphism in AZ[ 1 Ad (exp µXα), which completes the proof. 2 ]⊗Z[ 1 2 ]Z[ 1 Corollary 8.3. Let C be a Cayley algebra over a field F, char F 6= 2. Let A = H3(C) and g = Der(A). Then Ad : Aut(A) → Aut(g) is an isomorphism of affine group schemes. Proof. Let F be the algebraic closure of F. Since Der(g) ⊗ F = DerF(g ⊗ F), g ⊗ F = DerF(A ⊗ F), and A ⊗ F = H3(C ⊗ F), we may pass from F to F and thus assume that F is algebraically closed. Then Aut(A) is the simple algebraic group of type F4 (see [KMRT98, (25.13)] and the references therein) and hence dim Aut(A) = 52 = dim Der(A), which means that Aut(A) is smooth. Now, the maps Ad F : Aut(A) → Aut(g) and ad : g → Der(g) are both bijective, by Propositions 8.2 and 8.1, respectively. The result follows. (cid:3) Now Theorems 2.3 and 2.4 yield the following result: Theorem 8.4. Let A be the Albert algebra over an algebraically closed field F, char F 6= 2. Then the abelian group gradings on Der(A) are those induced by such gradings on A. The algebras A and Der(A) have the same classification of fine gradings up to equivalence and, for any abelian group G, the same classification of G-gradings up to isomorphism. (cid:3) Corollary 8.5. We use the notation of Theorems 8.4 and 7.1. Then, up to equiv- alence, the fine abelian group gradings on the simple Lie algebra g = Der(A), their universal groups and types are the following: g induced by Γ1 A; universal group Z4; type (48, 0, 0, 1). • The Cartan grading Γ1 • The grading Γ2 • The grading Γ3 • The grading Γ4 g induced by Γ2 g induced by Γ3 g induced by Γ4 A; universal group Z5 A; universal group Z × Z3 A; universal group Z3 2; type (24, 0, 0, 7). 2; type (31, 0, 7). 3; type (0, 26) -- this one exists only if char F 6= 3. Proof. Only the type of these gradings has to be checked and this is straightforward. The most difficult case is for the Z × Z3 2-grading. Since g = [LA, LA] (see [Jac68, Corollary IX.11]), we obtain: g = g−3 ⊕ g−2 ⊕ g−1 ⊕ g0 ⊕ g1 ⊕ g2 ⊕ g3 where gn = Pr+s=n[LAr , LAs] and Ar as in (11). But [LS±, Lν±(C)] = 0, so g±3 = 0. The local version of Remark 6.4 shows that g0 contains a subalgebra isomorphic to so(C0, n). Also, [LE, LA±1] = [LE, Lν±(C)] is an 8-dimensional subspace of g±1, since [LE, Lν±(x)](S∓) = 1 2 ν±(x), and [LS±, Lν(C0)] is a 7-dimensional subspace of g±2, since [LS±, Lν(a)](ν±(1)) = −2ν±(a). It follows that dim g0 = 22, dim g±1 = 8 and dim g±2 = 7 (actually, g±1 = [LE, Lν±(C)] and g±2 = [LS±, Lν(C0)]). Hence the type of the Z × Z3 2-grading, which is obtained by refining the Z-grading on g above using the Z3 2-grading on C, is (31, 0, 7), where the seven 3-dimensional homogeneous components are in so(C0, n), which is contained in g0. (cid:3) GRADINGS ON THE EXCEPTIONAL LIE ALGEBRAS F4 AND G2 REVISITED 35 Let Γ1 g(G, γ), Γ2 g(G, H, γ), Γ3 g(G, H, g), and Γ4 g, j = 1, 2, 3, 4, respectively, in the same way as for Γj g(G, H, δ) be the G-gradings in- A (see Theorem duced by Γj 7.6). Corollary 8.6. Let g be the simple Lie algebra of type F4 over an algebraically closed field F, char F 6= 2. Let G be an abelian group. Then any G-grading on A is isomorphic to some Γ1 g(G, H, g) or Γ4 g(G, H, δ) (characteris- tic 6= 3 in this latter case), but not two from this list. Also, g(G, H, γ), Γ3 g(G, γ), Γ2 g(G, γ) is isomorphic to Γ1 • Γ1 g(G, H, γ) is isomorphic to Γ2 • Γ2 g(G, H, g) is isomorphic to Γ3 • Γ3 • Γ4 g(G, H, δ) is isomorphic to Γ4 δ′. g(G, γ′) if and only if γ ∼ γ′; g(G, H ′, γ′) if and only if H = H ′ and γ ∼ γ′; g(G, H ′, g′) if and only if H = H ′ and g ∼ g′; g(G, H ′, δ′) if and only if H = H ′ and δ = (cid:3) Corollary 8.7. Using the notation of Corollary 8.6, any abelian group grading on g is either induced from the Cartan grading or equivalent to one of the following: 2-grading of type (1, 8, 0, 0, 7), or a Z3 2- 2-grading of type (24, 0, 7), a Z4 • a Z5 grading of type (0, 0, 1, 0, 0, 0, 7); • a Z × Z3 Z4 × Z3 Z4 × Z2 2-grading of type (31, 0, 7), a Z8 × Z2 2-grading of type (17, 7, 7), a Z3 × Z3 2-grading of type (0, 8, 2, 0, 6); 2-grading of type (19, 6, 7), a 2-grading of type (3, 14, 7), or a • a Z3 supports in each of the components of the Z-grading g = L2 3-grading of type (0, 26) if char F 6= 3. Proof. Consider, for example, the gradings Γ = Γ2 g(G, H, g) and the corresponding grading on A. The homogeneous components of the 5-grading in (11) have supports Supp A±2 = {g±2}, Supp A±1 = g±1H, Supp A0 = H. Hence Γ has the following r=−2 gr: Supp g±2 = g±2(H \ {e}) (as g±2 = [LE, Lν(C0)]), Supp g±1 = g±1H, and Supp g0 = H. If these subsets are disjoint, then Γ is equivalent to the fine Z × Z3 2-grading. Otherwise we have several possibilities where some homogeneous components of this fine grading coalesce as in the arguments preceding Proposition 7.4, plus a new possibility where g4 ∈ H \{e} and hence Supp Γ is a group isomorphic to Z8×Z2 2. With combinatorial arguments of this kind, one completes the proof. (cid:3) References [BK10] Y.A. Bahturin and M. Kochetov, Classification of group gradings on simple Lie alge- bras of types A, B, C, and D, J. Algebra 324 (2010), no. 11, 2971 -- 2989. [BKM09] Y.A. Bahturin, M. Kochetov, and S. Montgomery, Group gradings on simple Lie algebras in positive characteristic, Proc. Amer. Math. Soc., 137 (2009), no. 4, 1245 -- 1254. Y.A. Bahturin, S. Sehgal, and M. Zaicev, Group gradings on associative algebras, J. Algebra, 241 (2001), no. 2, 677 -- 698. [BSZ01] [BShZ05] Y.A. Bahturin, I. Shestakov, and M. Zaicev, Gradings on simple Jordan and Lie [BT09] [BZ02] [BZ03] algebras, J. Algebra, 283 (2005), no. 2, 849 -- 868. Y.A. Bahturin and M.V. Tvalavadze, Group gradings on G2, Communications in Al- gebra 37 (2009), no. 3, 885 -- 893. Y.A. Bahturin and M. Zaicev, Group gradings on matrix algebras, Canad. Math. Bull., 45 (2002), no. 4, 499 -- 508. Y.A. Bahturin and M. Zaicev, Graded algebras and graded identities, Polynomial iden- tities and combinatorial methods (Pantelleria, 2001), 101 -- 139, Lecture Notes in Pure and Appl. Math., 235, Dekker, New York, 2003. 36 [BZ06] [BZ07] ALBERTO ELDUQUE AND MIKHAIL KOCHETOV Y.A. Bahturin and M. Zaicev, Gradings on Simple Lie Algebras of Type "A", J. Lie Theory, 16 (2006), no. 4, 719 -- 742. Y.A. Bahturin and M. Zaicev, Involutions on graded matrix algebras, J. Algebra, 315 (2007), no. 2, 527 -- 540. [BEMN02] P.A. Bjerregaard, A. Elduque, C. Mart´ın-Gonz´alez, and F.J. Navarro-M´arquez, On [Dra] [DM06] [DM09] [Eld98] [Eld00] [Eld09] [Eld10] [EK] the Cartan-Jacobson theorem, J. Algebra 250 (2002), no. 2, 397 -- 407. C. Draper, A non computational approach to gradings on f4, preprint, Jordan Theory Preprint Archives 282. C. Draper and C. Mart´ın-Gonz´alez, Gradings on g2, Linear Algebra Appl. 418 (2006), no. 1, 85 -- 111. C. Draper and C. Mart´ın-Gonz´alez, Gradings on the Albert algebra and on f4, Rev. Mat. Iberoam. 25 (2009), no. 3, 841 -- 908. A. Elduque, Gradings on octonions, J. Algebra 207 (1998), no. 1, 342 -- 354. A. Elduque, On triality and automorphisms and derivations of composition algebras, Linear Algebra Appl. 314 (2000), no. 1-3, 49 -- 74. A. Elduque, Gradings on symmetric composition algebras, J. Algebra 322 (2009), no. 10, 3542 -- 3579. A. Elduque, Fine gradings on simple classical Lie algebras, J. Algebra 324 (2010), no. 12, 3532 -- 3571. A. Elduque and M. Kochetov, Weyl groups of fine gradings on matrix algebras, octo- nions and the Albert algebra, preprint, arXiv: 1009.1462 [math.RA]. [HPP98] M. Havl´ıcek, J. Patera, and E. Pelantova, On Lie gradings. II, Linear Algebra Appl. [Hum72] [Jac68] 277 (1988), no. 1-3, 97 -- 125. J.E. Humphreys, Introduction to Lie algebras and representation theory, Graduate Texts in Mathematics, Vol. 9, Springer-Verlag, New York, 1972. N. Jacobson, Structure and representations of Jordan algebras, American Mathemat- ical Society Colloquium Publications, Vol. XXXIX, American Mathematical Society, Providence, RI, 1968. [KMRT98] M-A. Knus, A. Merkurjev, M. Rost, and J-P. Tignol, The book of involutions, Ameri- can Mathematical Society Colloquium Publications, vol. 44, American Mathematical Society, Providence, RI, 1998. J. Patera and H. Zassenhaus, On Lie gradings. I, Linear Algebra Appl. 112 (1989), 87 -- 159. [PZ89] [Koc09] M. Kochetov, Gradings on finite-dimensional simple Lie algebras, Acta Appl. Math. [Sel67] [Ste61] 108 (2009), no. 1, 101 -- 127., G.B. Seligman, Modular Lie algebras, Ergebnisse der Mathematik und ihrer Grenzge- biete, Band 40, Springer-Verlag New York, Inc., New York, 1967. R. Steinberg, Automorphims of classical Lie algebras, Pacific J. Math. 11 (1961), 1119 -- 1129. [Wat79] W.C. Waterhouse, Introduction to affine group schemes, Graduate Texts in Mathe- matics 66, Springer-Verlag, New York, 1979. [ZSSS82] K.A. Zhevlakov, A.M. Slin'ko, I.P. Shestakov, and A.I. Shirshov, Rings that are nearly associative, Pure and Applied Mathematics, vol. 104, Academic Press Inc. [Harcourt Brace Jovanovich Publishers], New York, 1982. Departamento de Matem´aticas e Instituto Universitario de Matem´aticas y Aplica- ciones, Universidad de Zaragoza, 50009 Zaragoza, Spain E-mail address: [email protected] Department of Mathematics and Statistics, Memorial University of Newfoundland, St. John's, NL, A1C5S7, Canada E-mail address: [email protected]
1805.09041	1	1805	2018-05-23T10:21:00	A Note on the Primary Decomposition of k-Ideals in Semirings	[ "math.RA" ]	We establish the primary decomposition and uniqueness of primary decomposition for k-ideals in commutative Noetherian semirings.	math.RA	math	A Note on the Primary Decomposition of 𝒌 −Ideals in Semirings Ram Parkash Sharma1,𝑎, Ricah Sharma1,𝑏, S. Kar2,𝑐 and Madhu1,∗ ¹Department of Mathematics and Statistics, Himachal Pradesh University, Summer Hill, Shimla-171005, India ²Department of Mathematics, Jadavpur University, West Bengal, Kolkata-700032, India *Corresponding author E-mail address: [email protected] a. [email protected] b. [email protected] c. [email protected] ABSTRACT: We establish the primary decomposition and uniqueness of primary decomposition for 𝑘 −ideals in commutative Noetherian semirings. KEYWORDS: Semiring, 𝑘 −ideals, Primary decomposition. ideals, Irreducible ideals, Primary MSC: 16Y60, 16Y99. 1. Introduction Ideals play an important role in both semiring theory and ring theory, but in the absence of additive inverses in semirings, their structure differs from that of ring theory. Due to this difference in both the theories, the role of 𝑘 −ideals (if 𝑥 + 𝑦 ∈ 𝐼, 𝑥 ∈ 𝐼, then 𝑦 ∈ 𝐼) becomes significant in semirings. It is pertinent to note here that many results which are true for ideals in rings have been established, by many authors, for 𝑘 −ideals in semirings (c.f. [5], [6], [7], [8]). This fact has motivated different researchers to settle the primary decomposition for k- ideals in semirings analogous to the primary decomposition theorem in rings (Lasker Noether theorem) which states that in a commutative Noetherian ring, every ideal can be described as a finite intersection of primary ideals. The above result of ring theory is not true for arbitrary ideals in semirings as noticed in [1]. R. E. Atani and S. E. Atani first proved that in a commutative Noetherian semiring, every proper 𝑘 −ideal can be represented as a finite intersection of 𝑘 −primary ideals [1, Theorem 4]. As observed in [4], there are some errors in the results used to prove this theorem. For example, 𝐼 + 𝑅𝑎ⁿ is not a 𝑘 −ideal, even if 𝐼 is a 𝑘 −ideal. But the authors of [1] took it for granted that the ideal 𝐼 + 𝑅𝑎ⁿ is a 𝑘 −ideal. P. Lescot [4] found these errors after observing in Example 6.2 that {0} ideal may not be a finite intersection of 𝑘 −primary ideals in a commutative Noetherian semiring. With these observation, P. Lescot developed the theory of weak primary decomposition for semirings of characteristic 1. But still the question of settling the primary decomposition for a proper 𝑘 −ideal (other than {0} and semiring 𝑅) remained unsolved. In this direction, S. Kar et. al. [3, Theorem 4.4] proved that every proper 𝑘 − ideal of a commutative Noetherian semiring 𝑅 can be expressed as a finite intersection of 𝑘 −irreducible ideals, where an ideal 𝐼 is 𝑘 −irreducible if for any two 𝑘 −ideals 𝐽, 𝐾 of 𝑅, 𝐼 = 𝐽 ∩ 𝐾 implies that either 𝐼 = 𝐽 or 𝐼 = 𝐾. They tried to prove the primary decomposition by observing that in a commutative Noetherian semiring, every 𝑘 −irreducible ideal is a 𝑘 −primary ideal [3, Theorem 4.6]. But the proof of this result has the same errors again and the problem still remained unsolved. In this paper, we provide a correct proof of Theorem 4.6 of [3], which settle the primary decomposition for 𝑘 −ideals in commutative Noetherian semirings. In order to establish the uniqueness of primary decomposition for semirings, first we prove some basic results required for uniqueness and finally also establish the uniqueness of primary decomposition for 𝑘 −ideals of commutative Noetherian semirings, that is, if 𝐼 = ⋂ 𝑄𝑖 , where each 𝑄𝑖 is a primary and √𝑄𝑖 = 𝑃𝑖, then the set {𝑃1, 𝑃2, … . 𝑃𝑛} is independent of particular choice of decomposition of 𝐼. 𝑛 𝑖=1 2. Primary Decomposition of 𝒌 −Ideals in Semirings In this section, we prove the primary decomposition and uniqueness theorem for 𝑘 −ideals in commutative Noetherian semirings. First, we recall some definitions and results from [2] and [3] which are necessary to prove the main results of this section. Definition 2.1 [2]. A semiring is a nonempty set 𝑅 on which operations of addition and multiplication have been defined such that for 𝑟 ∈ 𝑅, the following conditions are satisfied: (1) (𝑅, +) is a commutative monoid with identity element 0𝑅; (2) (𝑅, . ) is a monoid with identity element 1𝑅; (3) Multiplication distributes over addition from either side; (4) 0𝑅. 𝑟 = 0𝑅 = 𝑟. 0𝑅; (5) 1𝑅 ≠ 0𝑅. A semiring 𝑅 is said to be commutative, if (𝑅, . ) is a commutative monoid. Throughout this paper, 𝑅 is a commutative semiring with identity. Definition 2.2 [2]. Let 𝐼 be an ideal of a semiring 𝑅. Then the radical of 𝐼, denoted by √𝐼 is defined as √𝐼 = {𝑎 ∈ 𝑅 𝑎ⁿ ∈ 𝐼 for some 𝑛 ≥ 1}. Definition 2.3 [2]. A proper ideal 𝐼 of a semiring 𝑅 is called a prime ideal, if 𝑎𝑏 ∈ 𝐼 for 𝑎, 𝑏 ∈ 𝑅 implies that either 𝑎 ∈ 𝐼 or 𝑏 ∈ 𝐼. Definition 2.4 [2]. A proper ideal 𝐼 of a semiring 𝑅 is said to be a primary ideal of 𝑅 if for any 𝑥, 𝑦 ∈ 𝐼, 𝑥𝑦 ∈ 𝐼 implies that either 𝑥 ∈ 𝐼 or 𝑦 ∈ √𝐼. If √𝐼 = 𝑃, then 𝐼 is called 𝑃 −primary. Definition 2.5 [3]. A proper 𝑘 −ideal 𝐼 of a semiring 𝑅 is called a 𝑘 −irreducible ideal of 𝑅 if for any two 𝑘 −ideals 𝐽, 𝐾 of 𝑅, 𝐼 = 𝐽 ∩ 𝐾 implies that either 𝐼 = 𝐽 or 𝐼 = 𝐾. We now present the notion of primary decomposition of k-ideals as follows: Definition 2.6. Let 𝐼 be a 𝑘 −ideal of a semiring 𝑅. Then 𝐼 is said to have a primary decomposition if 𝐼 can be expressed as 𝐼 = ⋂ 𝑄𝑖 , where each 𝑄𝑖 is a primary ideal of 𝑅. 𝑛 𝑖=1 Also, a primary decomposition of the type 𝐼 = ⋂ 𝑄𝑖 with √𝑄𝑖 = 𝑃𝑖, is called a reduced primary decomposition of 𝐼, if 𝑃𝑖′𝑠 are distinct and 𝐼 cannot be expressed as an intersection of a proper subset of ideals 𝑄𝑖 in the primary decomposition of 𝐼. A reduced primary decomposition can be obtained from any primary decomposition by deleting those 𝑄𝑗 that contains ⋂ 𝑄𝑖 and grouping together all distinct √𝑄𝑖′𝑠. 𝑛 𝑖=1 𝑛 𝑖=1 𝑖≠𝑗 The main aim of this paper is to prove the existence and uniqueness of primary decomposition for 𝑘 −ideals in a commutative Noetherian semiring. First, we prove the existence part as stated below: Theorem 2.7 (Primary Decomposition of 𝒌 −Ideals). Let 𝐼 be a 𝑘 −ideal of a Noetherian semiring 𝑅. Then 𝐼 can be represented as a finite intersection of primary ideals of 𝑅, that is, 𝐼 has a reduced primary decomposition. The decomposition of 𝑘 −ideals in terms of 𝑘 −irreducible ideals has already been proved in [3] as follows: Theorem 2.8 [3, Theorem 4.4]. Every proper 𝑘 −ideal of a commutative Noetherian semiring 𝑅 can be expressed as a finite intersection of 𝑘 −irreducible ideals. Therefore, primary decomposition for 𝑘 −ideals in semirings follows immediately if we prove Theorem 2.9. Let 𝑅 be a commutative Noetherian semiring. Then every 𝑘 −irreducible ideal of 𝑅 is a primary ideal of 𝑅. Kar et. al. [3, Theorem 4.6] made an attempt to prove the same, but the proof of the result has some errors as mentioned in the introduction. Before we prove the required result, we analyse some common errors committed by many authors regarding 𝑘 −ideals, and prove some facts about k-ideals. First, we note that any ideal 𝐼 =< 𝑎 > is not a 𝑘 −ideal in semirings as observed by J. S. Golan in [2, Example 6.17]. That is, if 𝐼 = < (1 + 𝑥) > is an ideal of ℕ[𝑥] (semiring of polynomials over non-negative integers ℕ in the indeterminate 𝑥), then (1 + 𝑥)³ = (𝑥³ + 1) + 3𝑥(1 + 𝑥) ∈ 𝐼, 3𝑥(1 + 𝑥) ∈ 𝐼, but (𝑥³ + 1) ∉ 𝐼 implies that 𝐼 is not a 𝑘 −ideal of ℕ[𝑥]. Any ideal generated by a single element becomes a 𝑘 −ideal, if we impose some conditions on a semiring as shown below: Lemma 2.10. Let 𝑅 be an additively cancellative, yoked and zerosumfree semiring. Then for any 𝑎 ∈ 𝑅, the ideal 𝐼 =< 𝑎 > is a 𝑘 −ideal of 𝑅. Proof. Let 𝑎₁ + 𝑎₂, 𝑎₁ ∈ 𝐼. Then there exist some 𝑟₁, 𝑟₂ ∈ 𝑅 such that 𝑎₁ + 𝑎₂ = 𝑎𝑟₁ and 𝑎₁ = 𝑎𝑟₂. As 𝑅 yoked, so there exists some 𝑟₃ ∈ 𝑅 such that 𝑟₁ + 𝑟₃ = 𝑟₂ or 𝑟₂ + 𝑟₃ = 𝑟₁. If 𝑟₁ + 𝑟₃ = 𝑟₂, then 𝑎𝑟₁ = 𝑎𝑟₂ + 𝑎₂ = 𝑎𝑟₁ + 𝑎𝑟₃ + 𝑎₂ implies that 𝑎𝑟₃ + 𝑎₂ = 0, as 𝑅 is additively cancellative. Also, 𝑎₂ = 𝑎𝑟₃ = 0 ∈ 𝐼, as 𝑅 is zerosumfree. Further, if 𝑟₂ + 𝑟₃ = 𝑟₁, then 𝑎𝑟₂ + 𝑎₂ = 𝑎𝑟₁ = 𝑎(𝑟₂ + 𝑟₃) = 𝑎𝑟₂ + 𝑎𝑟₃ implies that 𝑎₂ = 𝑎𝑟₃ ∈ 𝐼, since 𝑅 is additively cancellative. Thus, 𝐼 =< 𝑎 > is a 𝑘 −ideal of 𝑅.■ The semiring considered in above example is additively cancellative, zerosumfree, but it is not yoked, for let 𝑓(𝑥) = 5𝑥² + 9𝑥 + 2 and 𝑔(𝑥) = 11𝑥² + 3𝑥 + 5 be two polynomials in ℕ[𝑥], then there exists no ℎ(𝑥) ∈ ℕ[𝑥] such that either 𝑓(𝑥) + ℎ(𝑥) = 𝑔(𝑥) 𝑜𝑟 𝑔(𝑥) + ℎ(𝑥) = 𝑓(𝑥). Similar to an ideal generated by a single element, the sum of two 𝑘 −ideals may not be a 𝑘 −ideal in a semiring. There are plenty of 𝑘 −ideals in the semiring ℕ, but their sum is not a 𝑘 −ideal. However, the sum of two 𝑘 −ideals is a 𝑘 −ideal in a lattice ordered semiring (c.f. [2], Corollary 21.22). While proving Theorem 2.9, the authors wrongly used that the ideals (𝑄+< 𝑎𝑘 >) and (𝑄+< 𝑏 >) are 𝑘 −ideals. In view of the above observations, Theorem 2.9 follows verbatim as proved in [3, Theorem 4.6] for additively cancellative, yoked and zerosumfree semirings, because in this case, both an ideal generated by a single element and sum of two 𝑘 −ideals is a 𝑘 −ideal. Here, we give a proof of Theorem 2.9 without resorting to these restrictions. Proof of Theorem 2.9. Let 𝑄 be a 𝑘 −irreducible ideal of a Noetherian semiring 𝑅. Let 𝑎𝑏 ∈ 𝑄 be such that 𝑏 ∉ 𝑄. Now, we construct two ideals 𝐼 and 𝐽 of 𝑅 as follows : 𝐼 =< 𝑎𝑘 > +𝑄 and 𝐽 =< 𝑏 > +𝑄. Then, clearly, 𝑄 ⊆ 𝐼 ∩ 𝐽. Let 𝑦 ∈ 𝐼 ∩ 𝐽. Then 𝑦 = 𝑎ⁿ𝑧 + 𝑞 for some 𝑧 ∈ 𝑅 and 𝑞 ∈ 𝑄. Again 𝑎𝐽 ⊆ 𝑄 (since 𝑎𝑏 ∈ 𝑄) and so 𝑎𝑦 ∈ 𝑄 (since 𝑦 ∈ 𝐽). Therefore, 𝑎𝑦 = 𝑎ⁿ⁺¹𝑧 + 𝑎𝑞. Thus, we get that 𝑎ⁿ⁺¹𝑧 + 𝑎𝑞 ∈ 𝑄. Also, 𝑎𝑞 ∈ 𝑄, since 𝑞 ∈ 𝑄 and 𝑄 is an ideal of 𝑅. Thus, it follows that 𝑎ⁿ⁺¹𝑧 ∈ 𝑄, since 𝑄 is a 𝑘 −ideal of 𝑅. Construct a set 𝐴𝑛 = {𝑥 ∈ 𝑅 𝑎ⁿ𝑥 ∈ 𝑄}. It is easy to check that 𝐴𝑛 is an ideal of 𝑅 and 𝐴₁ ⊆ 𝐴₂. . .. is an ascending chain of ideals. Since 𝑅 is Noetherian, 𝐴𝑛=𝐴𝑛+1=.... for some 𝑛 ∈ ℤ⁺. Again 𝑎ⁿ⁺¹𝑧 ∈ 𝑄 implies that 𝑧 ∈ 𝐴𝑛+1 = 𝐴𝑛. It demonstrates that 𝑎ⁿ𝑧 ∈ 𝑄 which implies that 𝑦 ∈ 𝑄. Thus, 𝐼 ∩ 𝐽 = 𝑄. Let 𝐽𝑎𝑐(𝑅) denote the Jacobson radical of semiring 𝑅, that is, the intersection of all maximal 𝑘 −ideals of 𝑅. Assume that 𝐴 is an ideal of 𝑅 and 𝐽𝑎𝑐(𝐴) denotes the intersection of all maximal 𝑘 −ideals of 𝑅 containing 𝐴. If 𝐴 denotes the 𝑘 −closure (i.e. 𝐴 = {𝑎 ∈ 𝐴} 𝑎 + 𝑏 = 𝑐 for some 𝑏, 𝑐 ∈ 𝐴) of an ideal ) = √𝐼 ∩ √𝐽, where √𝑃 denotes the nil radical of an ideal 𝑃 (c.f. [4], ). ) ⊆ 𝐽𝑎𝑐(𝐼 ∩ 𝐽). Thus, 𝐴 of 𝑅, we have (√𝐼 ∩ 𝐽 Lemma 2.2). We know that √𝑃 ⊆ 𝐽𝑎𝑐(𝑃). Thus, 𝐼 ∩ 𝐽 ⊆ √𝐼 ∩ √𝐽 = √𝐼 ∩ 𝐽 Therefore, 𝐽𝑎𝑐(𝐼 ∩ 𝐽) ⊆ 𝐽𝑎𝑐(𝐼 ∩ 𝐽 𝐽𝑎𝑐(𝐼 ∩ 𝐽 ) = 𝐽𝑎𝑐(𝐼 ∩ 𝐽) = 𝐽𝑎𝑐(𝐼) ∩ 𝐽𝑎𝑐(𝐽). So we have 𝐽𝑎𝑐(𝐼 ∩ 𝐽 ) = 𝐽𝑎𝑐(𝐼) ∩ 𝐽𝑎𝑐(𝐽). ). Again 𝐼 ∩ 𝐽 ⊆ 𝐼 ∩ 𝐽 ⇒ 𝐽𝑎𝑐(𝐼 ∩ 𝐽 ⊆ 𝐽𝑎𝑐(𝐼 ∩ 𝐽 Now, 𝑄 = 𝐼 ∩ 𝐽 ⇒ 𝑄 = 𝐼 ∩ 𝐽 𝐽𝑎𝑐(𝑄) = 𝐽𝑎𝑐(𝐼 ∩ 𝐽 ) = 𝐽𝑎𝑐(𝐼) ∩ 𝐽𝑎𝑐(𝐽), that is, 𝐽𝑎𝑐(𝑄) = 𝐽𝑎𝑐(𝐼) ∩ 𝐽𝑎𝑐(𝐽). ⇒ 𝑄 = 𝐼 ∩ 𝐽 , since 𝑄 is a 𝑘 −ideal of 𝑅. This shows that Again 𝑄 is a 𝑘 −irreducible ideal, which implies that 𝐽𝑎𝑐(𝑄) is 𝑘 −irreducible. 𝐽𝑎𝑐(𝑄) = 𝑄 ∩ 𝐽𝑎𝑐(𝑅) but 𝐽𝑎𝑐(𝑄) ≠ 𝐽𝑎𝑐(𝑅). Thus, 𝐽𝑎𝑐(𝑄) = 𝑄, since 𝐽𝑎𝑐(𝑄) is 𝑘 −irreducible. Accordingly, 𝑄 = 𝐽𝑎𝑐(𝐼) ∩ 𝐽𝑎𝑐(𝐽), where each of 𝐽𝑎𝑐(𝐼) and 𝐽𝑎𝑐(𝐽) are 𝑘 −ideals of 𝑅. Now 𝑏 ∈ 𝐽 implies that 𝑏 ∈ 𝐽𝑎𝑐(𝐽), but 𝑏 ∉ 𝑄, that is, 𝑄 ≠ 𝐽𝑎𝑐(𝐽). So 𝑄 = 𝐽𝑎𝑐(𝐼), since 𝑄 is 𝑘 −irreducible. Now 𝑎ⁿ ∈ 𝐼 ⇒ 𝑎ⁿ ∈ 𝐽𝑎𝑐(𝐼) = 𝑄. Hence 𝑄 is 𝑘 −primary.■ Remark 2.11. Now Theorem 2.7 follows by combining [3, Theorem 4.4] and Theorem 2.9. It is important to note here that all 𝑄𝑖′𝑠 in the primary decomposition of a 𝑘 −ideal 𝐼 have an additional property that these are also 𝑘 −ideals. Now it only remains to prove the uniqueness of primary decomposition of a 𝑘 −ideal 𝐼. The following lemma will be used to prove the uniqueness of reduced primary decomposition of 𝑘 −ideals in semirings. Lemma 2.12. Let 𝑅 be a semiring and 𝑄 a 𝑃 −primary ideal of 𝑅. Then we have (i) If 𝑥 ∈ 𝑅, (𝑄: 𝑥) = {𝑟 ∈ 𝑅 𝑟𝑥 ∈ 𝑄} is an ideal of 𝑅; (ii) If 𝑥 ∈ 𝑄, then (𝑄: 𝑥) = 𝑅; (iii) If 𝑥 ∉ 𝑃, then (𝑄: 𝑥) = 𝑄; (iv) If 𝑥 ∉ 𝑄, then (𝑄: 𝑥) is 𝑃 −primary. Proof. (i) Let 𝑟₁, 𝑟₂ ∈ (𝑄: 𝑥) and 𝑎 ∈ 𝑅. Then 𝑟₁𝑥, 𝑟₂𝑥 ∈ 𝑄 implies that (𝑟₁ + 𝑟₂)𝑥 ∈ 𝑄 and 𝑟₁𝑎𝑥 ∈ 𝑄 as 𝑄 is an ideal of 𝑅. Thus, (𝑄: 𝑥) is an ideal of 𝑅. (ii) If 𝑥 ∈ 𝑄, then 𝑅𝑥 ⊆ 𝑄, as 𝑄 is an ideal of 𝑅 which implies that 𝑅 ⊆ (𝑄: 𝑥). Also, (𝑄: 𝑥) is an ideal of R and so (𝑄: 𝑥) = 𝑅. (iii) Assume that 𝑥 ∉ 𝑃. If 𝑎 ∈ 𝑄, then 𝑎𝑥 ∈ 𝑄 implies that 𝑎 ∈ (𝑄: 𝑥). For the converse part, suppose that 𝑏 ∉ 𝑄 and 𝑥𝑏 ∈ 𝑄. Then, 𝑥 ∈ √𝑄 = 𝑃 as 𝑄 is 𝑃 −primary ideal of 𝑅 which contradicts that 𝑥 ∉ 𝑃. Thus, 𝑥𝑏 ∉ 𝑄 and so 𝑏 ∉ (𝑄: 𝑥). (iv) Suppose that 𝑥 ∉ 𝑄. If 𝑦 ∈ (𝑄: 𝑥), then 𝑥𝑦 ∈ 𝑄 implies that 𝑦 ∈ √𝑄 = 𝑃 as 𝑄 is 𝑃 −primary ideal of 𝑅. Thus, 𝑄 ⊆ (𝑄: 𝑥) ⊆ 𝑃 implies that 𝑃 = √𝑄 ⊆ √(𝑄: 𝑥) ⊆ √𝑃. Also, √𝑃 = 𝑃 , as 𝑃 is a prime ideal of 𝑅 and so 𝑃 = √(𝑄: 𝑥). Now, we show that (𝑄: 𝑥) is a primary ideal of 𝑅. The ideal (𝑄: 𝑥) is proper as 𝑥 ∉ 𝑄 and so, 1 ∉ (𝑄: 𝑥). Assume that 𝑎𝑏 ∈ (𝑄: 𝑥) and 𝑏 ∉ √(𝑄: 𝑥) for 𝑎, 𝑏 ∈ 𝑅. Then, 𝑎𝑏𝑥 ∈ 𝑄 and 𝑄 is a 𝑃 −primary ideal of 𝑅 which implies that either 𝑎𝑥 ∈ 𝑄 or 𝑏 ∈ 𝑃 = √(𝑄: 𝑥). Thus, 𝑎 ∈ (𝑄: 𝑥) as 𝑏 ∉ √(𝑄: 𝑥).■ We now prove the uniqueness of the reduced primary decomposition of a 𝑘 −ideal of a semiring as follows: Theorem 2.13 (Uniqueness of Primary Decomposition). Let 𝑅 be a commutative Noetherian semiring and 𝐼 a 𝑘 −ideal of 𝑅. If 𝐼 = ⋂ 𝑄𝑖 is a reduced primary decomposition of 𝐼 with √𝑄𝑖 = 𝑃𝑖 for 𝑖 = 1,2. . . . 𝑛, then {𝑃₁, 𝑃₂. . . . 𝑃𝑛} = {Prime ideals 𝑃 ∃ 𝑥 ∈ 𝑅 such that 𝑃 = √(𝐼: 𝑥)}. The set {𝑃₁, 𝑃₂. . . . 𝑃𝑛} is independent of the particular reduced primary decomposition chosen for 𝐼. 𝑛 𝑖=1 Proof. Let 𝑥 ∈ 𝑅. Then √(𝐼: 𝑥) = √(⋂ 𝑄𝑖 𝑛 𝑖=1 : 𝑥) = ⋂ √(𝑄𝑖: 𝑥) 𝑛 𝑖=1 𝑛 = ⋂ 𝑖=1 𝑥∉𝑄𝑖 𝑃𝑖 by Lemma 2.12 𝑛 (iv) and therefore, √(𝐼: 𝑥) ⊆ 𝑃𝑖 for all 𝑖 = 1,2. . . . 𝑛. Also, if √(𝐼: 𝑥) is prime, then ∏ 𝑖=1 𝑥∉𝑄𝑖 𝑃𝑖 ⊆ 𝑛 ⋂ 𝑖=1 𝑥∉𝑄𝑖 𝑃𝑖 = √(𝐼: 𝑥) implies that 𝑃𝑖 ⊆ √(𝐼: 𝑥) for some 𝑖 = 1,2. . . 𝑛. Thus, we have {Prime ideals 𝑃 ∃ 𝑥 ∈ 𝑅 such that 𝑃 = √(𝐼: 𝑥)} ⊆ {𝑃₁, 𝑃₂. . . . 𝑃𝑛}. On the other hand, for 𝑖 ∈ {1,2. . . . 𝑛}, we have ⋂ 𝑄𝑗 𝑛 𝑗=1 𝑗≠𝑖 ⊈𝑄𝑖, as the primary decomposition is reduced. So there exists some 𝑥𝑖 ∈ ⋂ 𝑄𝑗 𝑛 𝑗=1 𝑗≠𝑖 and 𝑥𝑖 ∈ 𝑄𝑖. If 𝑦 ∈ (𝑄𝑖: 𝑥𝑖), then 𝑦𝑥𝑖 ∈ 𝑄𝑖, and 𝑦𝑥𝑖 ∈ (⋂ 𝑄𝑗 𝑛 𝑗=1 𝑗≠𝑖 ) ∩ 𝑄𝑖 = 𝐼 which implies that 𝑦 ∈ (𝐼: 𝑥𝑖),. Thus, (𝑄𝑖: 𝑥𝑖) ⊆ (𝐼: 𝑥𝑖) ⊆ (𝑄𝑖: 𝑥𝑖), as 𝐼 ⊆ 𝑄𝑖. So (𝑄𝑖: 𝑥𝑖) = (𝐼: 𝑥𝑖) implies that √(𝑄𝑖: 𝑥𝑖) = √(𝐼: 𝑥𝑖) =𝑃𝑖 by Lemma 2.12 (iv). Hence {𝑃₁, 𝑃₂. . . . 𝑃𝑛} = {Prime ideals 𝑃 ∃ 𝑥 ∈ 𝑅 such that 𝑃 = √(𝐼: 𝑥𝑖)}.■ References 1. R. E. Atani, S. E. Atani, Ideal theory in commutative semirings, Buletinal Academiei De Stiinte 2 (57) (2008) 14-23. 2. J. S. Golan, Semirings and Their Applications, Kluwer Academic Publishers, Dordrecht, 1999. 3. S. Kar, S. Purkait, B. Davvaz, Fuzzy k-primary decomposition of fuzzy k-ideals in a semiring, Fuzzy information and Engineering 7 (2015) 405-422. 4. P. Lescot, Prime and primary ideals in semirings, Osaka J. Math 52 (2015) 721-736. 5. R. P. Sharma, Madhu, On strong radical theory of semirings, Recent Advances on Continuum Mechanics and Algebra, Regal Publications, New Delhi, 2011, pp. 121-131. 6. R. P. Sharma, Madhu, On Connes subgroups and graded semirings, Vietnam Journal of Mathematics 38 (3) (2010) 287-298. 7. R. P. Sharma, R. Joseph, Prime ideal of group graded semirings and their smash products, Vietnam Journal of Mathematics (Springer- Verlag) 36 (4) (2008) 415-426. 8. R. P. Sharma, T. R. Sharma, G-prime ideals in semirings and their skew group semirings, Comm. Algebra 34 (2006) 4459-4465.
1409.3605	2	1409	2016-01-05T15:16:44	Stable homology over associative rings	[ "math.RA", "math.AC" ]	We analyze stable homology over associative rings and obtain results over Artin algebras and commutative noetherian rings. Our study develops similarly for these classes; for simplicity we only discuss the latter here. Stable homology is a broad generalization of Tate homology. Vanishing of stable homology detects classes of rings---among them Gorenstein rings, the original domain of Tate homology. Closely related to gorensteinness of rings is Auslander's G-dimension for modules. We show that vanishing of stable homology detects modules of finite G-dimension. This is the first characterization of such modules in terms of vanishing of (co)homology alone. Stable homology, like absolute homology, Tor, is a theory in two variables. It can be computed from a flat resolution of one module together with an injective resolution of the other. This betrays that stable homology is not balanced in the way Tor is balanced. In fact, we prove that a ring is Gorenstein if and only if stable homology is balanced.	math.RA	math	STABLE HOMOLOGY OVER ASSOCIATIVE RINGS OLGUR CELIKBAS, LARS WINTHER CHRISTENSEN, LI LIANG, AND GREG PIEPMEYER Abstract. We analyze stable homology over associative rings and obtain re- sults over Artin algebras and commutative noetherian rings. Our study devel- ops similarly for these classes; for simplicity we only discuss the latter here. Stable homology is a broad generalization of Tate homology. Vanishing of stable homology detects classes of rings -- among them Gorenstein rings, the original domain of Tate homology. Closely related to gorensteinness of rings is Auslander's G-dimension for modules. We show that vanishing of stable homo- logy detects modules of finite G-dimension. This is the first characterization of such modules in terms of vanishing of (co)homology alone. Stable homology, like absolute homology, Tor, is a theory in two variables. It can be computed from a flat resolution of one module together with an injective resolution of the other. This betrays that stable homology is not balanced in the way Tor is balanced. In fact, we prove that a ring is Gorenstein if and only if stable homology is balanced. Introduction The homology theory studied in this paper was introduced by P. Vogel in the 1980s. Vogel did not publish his work, but the theory appeared in print in a 1992 paper by Goichot [18], who called it Tate -- Vogel homology. As the name suggests, the theory is a generalization of Tate homology for modules over finite group rings. Vogel and Goichot also considered a generalization of Tate cohomology, which was studied in detail by Avramov and Veliche [5]. In that paper, the theory was called stable cohomology, to emphasize a relation to stabilization of module categories. To align terminology, we henceforth refer to the homology theory as stable homology. For modules M and N over a ring R, stable homology is a Z-indexed family of i (M, N ). These fit into an exact sequence abelian groups gTorR (†) · · · → gTorR i (M, N ) → TorR i (M, N ) → TorR i (M, N ) → gTorR i−1(M, N ) → · · · i (M, N ) and TorR where the groups TorR i (M, N ) form, respectively, the unbounded homology and the standard absolute homology of M and N . We are thus led to study stable and unbounded homology simultaneously. Our investigation takes cues from the studies of stable cohomology [5] and absolute homology, but our results look quite different. This comes down to an inherent asymmetry in the definition Date: 29 December 2015. 2010 Mathematics Subject Classification. 16E05, 16E30, 16E10, 13H10. Key words and phrases. Stable homology, Tate homology, Gorenstein ring, G-dimension. This research was partly supported by a Simons Foundation Collaboration Grant for Math- ematicians, award no. 281886 (L.W.C.), NSA grant H98230-14-0140 (L.W.C.) and NSFC grant 11301240 (L.L.). Part of the work was done during L.L.'s stay at Texas Tech University with support from the China Scholarship Council; he thanks the Department of Mathematics and Statistics at Texas Tech for its kind hospitality. 2 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER of stable homology that is not present in either of these precursors. It manifests itself in different ways, but it is apparent in most of our results. ∗ ∗ ∗ When we consider stable homology gTorR(M, N ), the ring acts on M from the right and on N from the left. In this paper an R-module is a left R-module; to distinguish right R-modules, we speak of modules over the opposite ring R◦. We study stable homology over associative rings and obtain conclusive results for Artin algebras and commutative noetherian rings. In the following overview, R denotes an Artin alge- bra or a commutative noetherian local ring. The results we discuss are special cases of results obtained within the paper; internal references are given in parentheses. One expects a homology theory to detect finiteness of homological dimensions. Our first two results reflect the asymmetry in the definition of stable homology. Right vanishing (3.1, 3.2). For a finitely generated R◦-module M , the following conditions are equivalent. (i) M has finite projective dimension. i (M, −) = 0 for all i ∈ Z. (ii) gTorR (iii) There is an i > 0 with gTorR i (M, −) = 0. Left vanishing (5.1, 5.12). For a finitely generated R-module N , the following conditions are equivalent. (i) N has finite injective dimension. i (−, N ) = 0 for all i ∈ Z. (ii) gTorR (iii) There is an i 6 0 with gTorR i (−, N ) = 0. These two vanishing results reveal that stable homology cannot be balanced in the way absolute homology, Tor, is balanced. Balancedness (4.5, 4.6, 4.7). The following conditions on R are equivalent. (i) R has finite injective dimension over R and over R◦. (ii) For all finitely generated R◦-modules M , all finitely generated R-modules N , (iii) For all R◦-modules M , all R-modules N , and all i ∈ Z there are isomorphisms and all i ∈ Z there are isomorphisms gTorR gTorR i (M, N ) ∼= gTorR◦ (N, M ). i i (M, N ) ∼= gTorR◦ i (N, M ). Another way to phrase part (i) above is to say that R is Iwanaga-Gorenstein. On that topic: a commutative noetherian ring A is Gorenstein (regular or Cohen -- Macaulay) if the local ring Ap is Iwanaga-Gorenstein (regular or Cohen -- Macaulay) for every prime ideal p in A. It follows that a commutative noetherian ring of finite self-injective dimension is Gorenstein, but a Gorenstein ring need not have finite self-injective dimension; consider, for example, Nagata's regular ring of infinite Krull dimension [28, appn. exa. 1]. It came as a surprise to us that balancedness of stable homology detects goren- steinness outside of the local situation (4.6): A commutative noetherian ring A is i (N, M ) are isomorphic for all finitely generated A-modules M and N and all i ∈ Z. This turns out to be only one of several ways in which stable homology captures global properties of rings. Indeed, Gorenstein if and only if gTorA i (M, N ) and gTorA STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 3 vanishing of stable homology detects if a commutative noetherian (not necessarily local) ring is regular (3.3), Gorenstein (3.12), or Cohen -- Macaulay (5.11). For finitely generated modules over a noetherian ring, Auslander and Bridger [1] introduced a homological invariant called the G-dimension. A noetherian ring is Iwanaga-Gorenstein if and only if there is an integer d such that every finitely generated left or right module has G-dimension at most d; see Avramov and Martsinkovsky [4, 3.2]. In [18, sec. 3] Tate homology was defined for modules over Iwanaga-Gorenstein rings and shown to agree with stable homology. Iacob [24] generalized Tate homology further to a setting that includes the case where M is a finitely generated R◦-module of finite G-dimension. We prove: Tate homology (6.4). Let M be a finitely generated R◦-module of finite G- i (M, −) is isomorphic dimension. For every i ∈ Z the stable homology functor gTorR to the Tate homology functor dTorR There are two primary generalizations of G-dimension to modules that are not finitely generated: Gorenstein projective dimension and Gorenstein flat dimension. They are defined in terms of resolutions by Gorenstein projective (flat) modules, notions which were introduced by Enochs, Jenda, and collaborators; see Holm [21]1. In general, it is not known if Gorenstein projective modules are Gorenstein flat. i (M, −). Gorenstein projective dimension. We show (6.7) that Tate homology agrees with Iacob's definition of Tate homology dTorR(M, −) is for R◦-modules M of finite stable homology gTorR(M, −) for all such R◦-modules if and only if every Gorenstein projective R◦-module is Gorenstein flat. Here is an example to illustrate how widely the various homology theories differ. Example. Let (R, m) be a commutative artinian local ring with m2 = 0, and assume that R is not Gorenstein. If k denotes a field, then k[x, y]/(x2, xy, y2) is an example of such a ring. Let E be the injective hull of the residue field R/m. Then: (a) Absolute homology TorR i (E, E) is non-zero for every i ≥ 2. (c) Unstable homology TorR (b) Stable homology gTorR (d) Tate homology "dTorR(E, E)" is not defined. i (E, E) is zero for every i. i (E, E) agrees with TorR i (E, E) for every i. Since R is not Gorenstein, E has infinite G-dimension; see [22, thm. 3.2]. That It also explains (a), as the first syzygy of E is a explains (d); see 6.1 and 6.2. k-vector space, whence vanishing of TorR i (E, E) for some i > 2 would imply that E has finite projective dimension and hence finite G-dimension; see [4, thm. 4.9]. Left vanishing, see p. 2, yields (b), and then the exact sequence (†) on p. 1 gives (c). Finally we return to the topic of vanishing. A much studied question in Goren- stein homological algebra is how to detect finiteness of G-dimension. A finitely generated module G over a commutative noetherian ring A has G-dimension zero if it satisfies Exti A(HomA(G, A), A) for all i > 0 and the canonical map G → HomA(HomA(G, A), A) is an isomorphism. This is the original defini- tion [1], and the last requirement cannot be dispensed with, as shown by Jorgensen A(G, A) = 0 = Exti 1 Enochs and Jenda consolidated their work in [16, chaps. 10 -- 12], which deal almost exclusively with Gorenstein rings. In [21] Holm does Gorenstein homological algebra over associative rings, so that is our standard reference for this topic. Basic definitions are recalled in 6.6. 4 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER and S¸ega [26]. There are other characterizations of modules of finite G-dimension in the literature, but none that can be expressed solely in terms of vanishing of (co)homology functors. Hence, we were excited to discover: G-dimension (3.11). Let A be a commutative noetherian ring. A finitely gener- i (M, A) = 0 holds for ated A-module M has finite G-dimension if and only if gTorA all i ∈ Z. 1. Tensor products of complexes All rings in this paper are assumed to be associative algebras over a commutative ring k, where k = Z is always possible, and k = R works when R is commutative. For an Artin algebra R, one can take k to be an artinian ring such that R is finitely generated as a k-module. We recall that in this situation the functor D(−) = Homk(−, E), where E is the injective hull of k/Jac(k), yields a duality between the categories of finitely generated R-modules and finitely generated R◦-modules. By M(R) we denote the category of R-modules. We work with complexes, index these homologically, and follow standard notation (see the appendix in [9] for what is not covered below). 1.1. For a complex X with differential ∂X and an integer n ∈ Z the symbol Cn(X) denotes the cokernel of ∂X n+1, and Hn(X) denotes the homology of X in degree n, i.e., Ker ∂X n+1. A complex with H(X) = 0 is called acyclic, and a morphism of complexes X → Y that induces an isomorphism H(X) → H(Y ) is called a quasi- isomorphism. The symbol ≃ is used to decorate quasi-isomorphisms; it is also used for isomorphisms in derived categories. n /Im ∂X 1.2. For an R◦-complex X and an R-complex Y , the tensor product X ⊗R Y is the k-complex with degree n term (X ⊗R Y )n = `i∈Z(Xi ⊗R Yn−i) and differential given by ∂(x ⊗ y) = ∂X i (x) ⊗ y + (−1)ix ⊗ ∂Y n−i(y) for x ∈ Xi and y ∈ Yn−i. Contrast the standard tensor product complex in 1.2 with the construction in 1.3, which first appeared in [18]. Similar constructions for Hom were also treated in [18] and in great detail by Avramov and Veliche [5]. We recall the Hom constructions in the Appendix, where we also study their interactions with the one below. 1.3 Definition. For an R◦-complex X and an R-complex Y , consider the graded k-module X ⊗R Y with degree n component (X ⊗R Y )n = Y i∈Z Xi ⊗R Yn−i . Endowed with the degree −1 homomorphism ∂ defined on elementary tensors as in 1.2, it becomes a complex called the unbounded tensor product. It contains the ten- sor product, X ⊗R Y , as a subcomplex. The quotient complex (X ⊗R Y )/(X ⊗R Y ) is denoted X e⊗R Y , and it is called the stable tensor product. 1.4. The definition above yields an exact sequence of k-complexes, (1.4.1) 0 −→ X ⊗R Y −→ X ⊗R Y −→ X e⊗R Y −→ 0 . Notice that if X or Y is bounded, or if both complexes are bounded on the same side (above or below), then the unbounded tensor product coincides with the usual tensor product, and the stable tensor product X e⊗R Y is zero. STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 5 Note that the unbounded tensor product X ⊗R Y is the product totalization of the double complex (Xi ⊗R Yj). 1.5. We collect some basic results about the unbounded and stable tensor prod- ucts. For the unbounded tensor product the proofs mimic the proofs for the tensor product, and one obtains the results for the stable tensor product via (1.4.1). As above X is an R◦-complex and Y is an R-complex. (a) There are isomorphisms of k-complexes and X ⊗R Y ∼= Y ⊗R◦ X X e⊗R Y ∼= Y e⊗R◦ X . (b) The functors X ⊗R − and X e⊗R − are additive and right exact. (c) The functors X ⊗R − and X e⊗R − preserve degree-wise split exact sequences. (d) The functors X ⊗R − and X e⊗R − preserve homotopy. (e) A morphism α : Y → Y ′ of R-complexes yields isomorphisms of k-complexes It follows that X ⊗R α is a quasi-isomorphism if and only if X ⊗R Cone α is Cone(X ⊗R α) ∼= X ⊗R Cone α and Cone(X e⊗R α) ∼= X e⊗R Cone α . acyclic, and similarly for X e⊗R α. 1.6 Lemma. Let D = (Di,j) be a double k-complex. Let z be a cycle in the product totalization of D. Assume z contains a component zm,n that satisfies zm,n = ∂h(x′) + ∂v(x′′) for some x′ ∈ Dm+1,n and x′′ ∈ Dm,n+1. If both (1) Hh (2) Hv m+k(D∗,n−k) = 0 for every k > 0 and n+k(Dm−k,∗) = 0 for every k > 0, then z is a boundary in the product totalization of D. Proof. The goal is to prove the existence of a sequence (xi) with xm+1 = x′ and xm = x′′ such that zm+k,n−k = ∂h(xm+k+1) + ∂v(xm+k) holds for all k ∈ Z; i.e., one has z = ∂(x) in the product totalization of D. Assume k is positive and that xm, . . . , xm+k have been constructed. Then zm+k−1,n−k+1 = ∂h(xm+k) + ∂v(xm+k−1) holds. As z is a cycle in the product totalization of D, one has 0 = ∂h(zm+k,n−k) + ∂v(zm+k−1,n−k+1) = ∂h(zm+k,n−k) + ∂v(∂h(xm+k) + ∂v(xm+k−1)) = ∂h(zm+k,n−k) − ∂h(∂v(xm+k)) . Thus, zm+k,n−k − ∂v(xm+k) is a horizontal cycle, and hence by (1) it is a hori- zontal boundary. There exists, therefore, an element xm+k+1 with ∂h(xm+k+1) + ∂v(xm+k) = zm+k,m−k. By induction, this argument yields the elements xm+k for k > 1; a symmetric argument using (2) yields xm+k for k < 0. (cid:3) 1.7 Proposition. Let X be an R◦-complex and let Y be an R-complex. (a) If X is bounded above and Xi ⊗R Y is acyclic for all i, then X ⊗R Y is acyclic. (b) If Y is bounded below and Xi ⊗R Y is acyclic for all i, then X ⊗R Y is acyclic. 6 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER Proof. The product totalization of the double complex (Xi ⊗ Yj) is X ⊗R Y . (a) Let m be an upper bound for X, let n be an integer, and let z = (zm+k,n−k)k be a cycle in X ⊗R Y of degree m + n. The component zm,n is a cycle in Xm ⊗R Y and hence a boundary as Xm ⊗R Y is assumed to be acyclic. Moreover, the assump- tion that Xi ⊗R Y is acyclic for every i ensures that condition (2) in Lemma 1.6 is met. Condition (1) is satisfied due to the boundedness of X; thus z is a boundary in X ⊗R Y . As n is arbitrary, it follows that X ⊗R Y is acyclic. (b) A similar argument with the roles of m and n exchanged handles this case. (cid:3) 2. Homology Now we consider the homology of unbounded and stable tensor product complexes. Our notation differs slightly from the one employed by Goichot [18]. 2.1 Definition. Let M be an R◦-module and N be an R-module. Let P ≃−−→ M be −−→ I be an injective resolution. The k-modules a projective resolution and let N ≃ TorR i (M, N ) = Hi(P ⊗R I) are the unbounded homology modules of M and N over R, and the stable homology modules of M and N over R are gTorR i (M, N ) = Hi+1(P e⊗R I) . 2.2. As any two projective resolutions of M , and similarly any two injective res- olutions of N , are homotopy equivalent, it follows from 1.5(d) that the definitions in 2.1 are independent of the choices of resolutions. 2.3. Notice from 1.4 that if M has finite projective dimension, or if N has finite injective dimension, then for every i ∈ Z one has TorR i (M, N ) and i (M, N ) = TorR hence gTorR i (M, N ) = 0. 2.4. The standard liftings of module homomorphisms to morphisms of resolutions imply that the definitions of TorR i (M, N ) are functorial in either argument; that is, for every i ∈ Z there are functors i (M, N ) and gTorR TorR i (−, −), gTorR i (−, −) : M(R◦) × M(R) −→ M(k) . These functors are homological in the sense that every short exact sequence of R◦- modules 0 → M ′ → M → M ′′ → 0 and every R-module N give rise to a connected exact sequence of stable homology modules (2.4.1) · · · → gTorR i+1(M ′′, N ) → gTorR i (M ′, N ) → gTorR i (M, N ) → gTorR i (M ′′, N ) → · · · , and to an analogous sequence of Tor modules. Similarly, every short exact sequence 0 → N ′ → N → N ′′ → 0 of R-modules and every R◦-module M give rise to a connected exact sequence, (2.4.2) · · · → gTorR i+1(M, N ′′) → gTorR i (M, N ′) → gTorR i (M, N ) → gTorR and an analogous sequence of Tor modules. For details see [18, sec. 1]. i (M, N ′′) → · · · , Stable and unbounded homology entwine with absolute homology. STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 7 2.5. For every R◦-module M and every R-module N the exact sequence of com- plexes (1.4.1) yields an exact sequence of homology modules: (2.5.1) · · · → gTorR i (M, N ) −→ TorR i (M, N ) −→ TorR i (M, N ) −→ gTorR i−1(M, N ) → · · · Flat resolutions. Just like absolute homology, unbounded and stable homology can be computed using flat resolutions in place of projective resolutions. 2.6 Proposition. Let M be an R◦-module and N be an R-module. If F ≃−−→ M is a flat resolution and N ≃−−→ I is an injective resolution, then there are isomorphisms, TorR i (M, N ) ∼= Hi(F ⊗R I) and gTorR i (M, N ) ∼= Hi+1(F e⊗R I) for every i ∈ Z, and they are natural in either argument. Proof. Let P ≃−−→ M be a projective resolution, then there is a quasi-isomorphism α : P → F . For every R-complex E the complex Cone(α) ⊗R E is acyclic by [11, cor. 7.5]. In particular, Cone(α) ⊗R I is acyclic, and it follows from Proposition 1.7 that Cone(α) ⊗R I is acyclic; now Cone(α) e⊗R I is acyclic in view of (1.4.1). Thus, the morphisms α ⊗R I and α e⊗R I are quasi-isomorphisms by 1.5(e), and the de- sired isomorphisms follow from 2.1; it is standard to verify that they are natural. (cid:3) In view of 1.4 the next result is now immediate. 2.7 Corollary. For every R◦-module M of finite flat dimension, gTorR holds for every i ∈ Z. i (M, −) = 0 (cid:3) 2.8 Remark. If every flat R◦-module has finite projective dimension, then Corol- lary 2.7 is covered by 2.3. There are wide classes of rings over which flat modules have finite projective dimension: Iwanaga-Gorenstein rings, see [16, thm. 9.1.10], and more generally rings of finite finitistic projective dimension, see Jensen [25, prop. 6]; in a different direction, rings of cardinality at most ℵn for some n ∈ N, see Gruson and Jensen [20, thm. 7.10]. On the other hand, recall that any product of fields is a von Neumann regular ring, i.e., every module over such a ring is flat. Osofsky [29, 3.1] shows that a sufficiently large product of fields has infinite global dimension and hence must have flat modules of infinite projective dimension. The next result is an analogue for stable homology of [5, thm. 2.2]. 2.9 Proposition. Let M be an R◦-module and let n ∈ Z. The following conditions are equivalent. (i) The connecting morphism gTorR every i > n. i (M, −) → TorR i (M, −) is an isomorphism for (ii) TorR (iii) TorR i (M, E) = 0 for every injective R-module E and every i > n. i (M, −) = 0 for every i > n. Proof. For every injective R-module E and for every i ∈ Z one has gTorR by 2.3. Thus (i) implies (ii). i (M, E) = 0 In the following we use the notation (−)⊃s for the soft truncation below at s. (ii) =⇒ (iii): Let N be an R-module and N ≃−−→ I be an injective resolution. Let P ≃−−→ M be a projective resolution and let P + denote its mapping cone. Let E be an injective R-module; by assumption one has TorR i (M, E) = 0 for all i > n. By 8 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER right exactness of the tensor product, P + is acyclic by 1.5(a) and Proposition 1.7. Thus, for every i > n one has ⊃n−2 ⊗R E is acyclic and hence P + ⊃n−2 ⊗R I TorR i (M, N ) = Hi(P ⊗R I) = Hi(P + ⊃n−2 ⊗R I) = 0. (Notice that for all n 6 0 one has P + ⊃n−2 = P +.) (iii) =⇒ (i): This is immediate from the exact sequence (2.5.1). (cid:3) As vanishing of TorR ≫0(M, −) detects finite flat dimension of M , the next result is an immediate consequence of Proposition 2.9. In the parlance of [15] it says that the flat and copure flat dimensions agree for R◦-modules M with gTorR(M, −) = 0. 2.10 Corollary. Let M be an R◦-module. If gTorR i (M, −) = 0 for all i ∈ Z, then i (M, E) 6= 0 for some injective R-module E } . (cid:3) fdR◦ M = sup{i ∈ Z TorR Dimension shifting is a useful tool in computations with stable homology. 2.11. For a module X, we denote by ΩmX the mth syzygy in a projective resolution of X and by ΩmX the mth cosyzygy in an injective resolution of X. By Schanuel's lemma a syzygy/cosyzygy is uniquely determined up to a projective/injective sum- mand. In view of 2.3 and the exact sequences in 2.4 one gets (2.11.2) Moreover, if F ≃−−→ M is a flat resolution of M , then there are isomorphisms, for m > 0 . i−m(ΩmM, N ) i+m(M, ΩmN ) for m > 0 and (2.11.1) (2.11.3) gTorR i (M, N ) ∼= gTorR gTorR i (M, N ) ∼= gTorR gTorR i (M, N ) ∼= gTorR this follows from (2.4.1) and Corollary 2.7. i−m(Cm(F ), N ) for m > 0 ; 2.12. Suppose R is commutative and S is a flat R-algebra. For every S ◦-module M , a projective resolution P ≃−−→ M over S ◦ is a flat resolution of M as an R- i (M, −) is a functor from M(R) to M(S ◦). Similarly, an injective resolution N ≃−−→ I of an i (−, N ) module. Thus, it follows from Proposition 2.6 that stable homology gTorR S-module is also an injective resolution of N as an R-module, whence gTorR is a functor from M(R) to M(S). 3. Vanishing of stable homology gTor(M, −) We open with a partial converse to Corollary 2.7; recall, for example from [16, prop. 3.2.12], that a finitely presented module is flat if and only if it is projective. 3.1 Theorem. Let M be an R◦-module that has a degree-wise finitely generated projective resolution. The following conditions are equivalent. (i) pdR◦ M is finite. i (M, −) = 0 for all i ∈ Z. i (M, −) = 0 for some i > 0. 0 (M, Homk(M, E)) = 0 for some faithfully injective k-module E. Moreover, if R is left noetherian with idR R finite, then (i) -- (iv ) are equivalent to i (M, −) = 0 for some i ∈ Z. (ii) gTorR (iii) gTorR (iv ) gTorR (iii' ) gTorR STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 9 Proof. The implications (i) =⇒ (ii) =⇒ (iii) =⇒ (iii' ) are clear; see 2.3. Part (iv ) follows from (iii) by dimension shifting (2.11.2). For (iv ) =⇒ (i) use Theorem A.8 to get 0 = gTorR 0 (M, Homk(M, E)) ∼= Homk(gExt0 R◦ (M, M ), E) . −i(M, −) = 0 for some i > 0 implies gTorR R◦ (M, M ) is zero, hence M has finite projective dimension by [5, prop. 2.2]. Finally, assume that R is left noetherian with idR R finite. Let E be a faithfully injective k-module; in view of what has already been proved, it is sufficient to show 0 (M, Homk(M, E)) = 0. It follows from the assumptions on R that every free R-module has finite injective dimension. Thus every projective R-module P has finite injective dimension, and 2.3 yields i (−, P ) = 0 for all i ∈ Z. From the exact sequence (2.4.2) it follows that there −i(M, Ωi Homk(M, E)) for i > 0, (cid:3) Thus gExt0 that gTorR gTorR are isomorphisms gTorR so vanishing of gTorR 0 (M, Homk(M, E)) ∼= gTorR −i(M, −) implies gTorR 0 (M, Homk(M, E)) = 0. 3.2 Corollary. Let R be an Artin algebra with duality functor D(−). For a finitely generated R◦-module M the following conditions are equivalent. (i) pdR◦ M is finite. i (M, N ) = 0 for all finitely generated R-modules N . (ii) There is an i > 0 with gTorR (iii) gTorR 0 (M, D(M )) = 0. Proof. Part (ii) follows from (i) in view of 2.3. Part (iii) follows from (ii) by dimension shifting (2.11.2) in a degree-wise finitely generated injective resolution of D(M ). Finally, one has D(M ) = Homk(M, E), see the first paragraph in Section 1, so (i) follows from (iii) by Theorem 3.1, as the injective hull E of k/Jac(k) is faithfully injective. (cid:3) By a result of Bass and Murthy [6, lem. 4.5], a commutative noetherian ring is regular if and only if every finitely generated module has finite projective dimension. 3.3 Corollary. For a commutative noetherian ring R, the following are equivalent: (i) R is regular. i (M, −) = 0 for all finitely generated R-modules M and all i ∈ Z. i (M, −) = 0 for all finitely generated R-modules M . (ii) gTorR (iii) There is an i ∈ Z with gTorR Proof. The implications (i) =⇒ (ii) =⇒ (iii) are clear; see 2.3. (iii) =⇒ (i): If i > 0, then every finitely generated R-module has finite projective dimension by Theorem 3.1; i.e., R is regular. If i < 0, then for any finitely generated i (Ω−iM, −) = 0 by dimension shifting (cid:3) R-module M one has gTorR (2.11.1), and as above it follows that R is regular. 0 (M, −) ∼= gTorR Tate flat resolutions. We show that under extra assumptions on M one can compute gTorR(M, N ) without resolving the second argument. 3.4. Let N be an R-module; choose an injective resolution N ≃−−→ I and a projective resolution P ≃−−→ N . The composite P → N → I is a quasi-isomorphism, so the complex CN P I = Cone(P → N → I) is acyclic, and up to homotopy equivalence it is independent of the choice of resolutions. This construction is functorial in N . 10 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER 3.5 Lemma. Let F be a bounded below complex of flat R◦-modules; let N be an R-module with a projective resolution P ≃−−→ N and an injective resolution N ≃−−→ I. P I in the derived category D(k), and it There is an isomorphism F e⊗R I ≃ F ⊗R CN is functorial in the second argument; cf. 3.4. Proof. There is by 1.5(c) a short exact sequences of k-complexes 0 −→ F ⊗R I −→ F ⊗R CN P I −→ F ⊗R Σ P −→ 0 . With (1.4.1) it forms a commutative diagram whose rows are triangles in D(k), F ⊗R P F ⊗R I / F ⊗R CN P I / Σ(F ⊗R P ) ≃ F ⊗R I / F ⊗R I ≃ / Σ(F ⊗R I) , / F e⊗R I where the top triangle is rotated back once. The isomorphism is induced by the composite quasi-isomorphism P → N → I, as one has F ⊗R P = F ⊗R P ; cf. 1.4. Completion of the morphism of triangles yields the desired isomorphism via the Triangulated Five Lemma; see [23, prop. 4.3]. Functoriality in the second argument is straightforward to verify. (cid:3) 3.6 Lemma. Let N be an R-module with a projective resolution P ≃−−→ N and an injective resolution N ≃−−→ I. Let T be a complex of flat R◦-modules such that T ⊗R E is acyclic for every injective R-module E. There is then an isomorphism T ⊗R CN P I ≃ Σ(T ⊗R N ) in D(k) which is functorial in the second argument; cf. 3.4. Proof. By the assumptions and Proposition 1.7(a) the complex T ⊗R I is acyclic, so application of T ⊗R − to the exact sequence 0 → I → CN P I → Σ P → 0 yields a quasi-isomorphism T ⊗R CN P I → T ⊗R Σ P ; cf. 1.5(c). Now, let P + denote the mapping cone of the quasi-isomorphism P → N . The complex T ⊗R P + is acyclic by Proposition 1.7(b), so application of T ⊗R − to the mapping cone sequence 0 → N → P + → Σ P → 0 yields an isomorphism T ⊗R Σ P ≃ Σ(T ⊗R N ) in D(k), and the desired isomorphism follows as N is a module; cf. 1.4. Functoriality in the second argument is straightforward to verify; cf. 3.4. (cid:3) 3.7 Remark. The quasi-isomorphisms in Lemmas 3.5 and 3.6 hold with CN placed by CN F I where F ≃−−→ N is a flat resolution. P I re- The objects discussed in the next paragraph appear in the literature under a variety of names; here we stick with the terminology from [27]. 3.8. An acyclic complex T of flat R◦-modules is called totally acyclic if T ⊗R E is acyclic for every injective R-module E. A Tate flat resolution of an R◦-module M is a pair (T, F ), where F ≃−−→ M is a flat resolution and T is a totally acyclic complex of flat R◦-modules with T>n ∼= F>n for some n > 0. Tate's name is invoked here because these resolutions can be used to compute Tate homology; see [27, thm. A]. 3.9. Over a left coherent ring R, an R◦-module M has a Tate flat resolution if and only if it has finite Gorenstein flat dimension; see [27, prop. 3.4]. If R is noetherian, then every finitely generated R◦-module M of finite G- dimension has a Tate flat resolution. Here is a direct argument: By [4, thm. 3.1] there exists a pair (T, P ) where P ≃−−→ M is a projective resolution and T is an / / / / / / / STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 11 acyclic complex of finitely generated projective R◦-modules with the following prop- erties: the complex HomR◦ (T, R) is acyclic, and there is an n > 0 with T>n ∼= P>n. For every injective R-module E one now has (3.9.1) T ⊗R E ∼= T ⊗R HomR(R, E) ∼= HomR(HomR◦ (T, R), E) where the last isomorphism holds because T is degree-wise finitely generated; see [7, prop. VI.5.2]. Thus T is a totally acyclic complex of flat R◦-modules and (T, P ) is a Tate flat resolution of M . The next result shows that stable homology can be computed via Tate flat res- olutions. We apply it in Section 6 to compare stable homology to Tate homology, but before that it is used in the proofs of Theorems 3.11 and 4.2. 3.10 Theorem. Let M be an R◦-module that has a Tate flat resolution (T, F ). For every R-module N and every i ∈ Z there is an isomorphism, and it is functorial in the second argument. i (M, N ) ∼= Hi(T ⊗R N ) , gTorR Proof. Choose n > 0 such that T>n and F>n are isomorphic, and consider the degree-wise split exact sequence 0 −→ T6n−1 −→ T −→ F>n −→ 0 . Let P ≃−−→ N be a projective resolution and N ≃−−→ I be an injective resolution; set C = CN P I ; cf. 3.4. There is, by 1.5(c), an exact sequence of k-complexes, 0 −→ T6n−1 ⊗R C −→ T ⊗R C −→ F>n ⊗R C −→ 0 . As C is acyclic, it follows from Proposition 1.7(a) that T6n−1 ⊗R C is acyclic, whence the surjective morphism above is a quasi-isomorphism. Since T is a to- tally acyclic complex of flat modules, there is by Lemma 3.6 an isomorphism T ⊗R C ≃ Σ(T ⊗R N ) in D(k). Moreover, Lemma 3.5 yields an isomorphism F>n ⊗R C ≃ F>n e⊗R I in D(k). Thus one has Σ(T ⊗R N ) ≃ F>n e⊗R I in D(k). This explains the second isomorphism in the next computation. gTorR i (M, N ) ∼= gTorR i−n(Cn(F ), N ) = Hi−n+1((Σ−n F>n) e⊗R I) = Hi+1(F>n e⊗R I) ∼= Hi+1(Σ(T ⊗R N )) = Hi(T ⊗R N ) The first isomorphism holds by (2.11.3); the first equality follows from Proposi- tion 2.6 as the canonical surjection Σ−n F>n → Cn(F ) is a flat resolution. (cid:3) As discussed in the introduction, the G-dimension is a homological invariant for finitely generated modules over noetherian rings. Characterizations of modules of finite G-dimension have traditionally involved both vanishing of (co)homology and invertibility of a certain morphism; see for example [9, (2.1.6), (2.2.3), (3.1.5), and (3.1.11)]. More recently, Avramov, Iyengar, and Lipman [3] showed that a finitely generated module M over a commutative noetherian ring R has finite G-dimension if and only M is isomorphic to the complex RHomR(RHomR(M, R), R) in the 12 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER derive category D(R). The crucial step in our proof of the next theorem is to show that vanishing of stable homology gTorR(M, R) implies such an isomorphism. 3.11 Theorem. Let R be a commutative noetherian ring. For a finitely generated R-module M , the following conditions are equivalent. (i) G-dimR M < ∞. i (M, N ) = 0 for every R-module N of finite flat dimension and all i ∈ Z. i (M, R) = 0 for all i ∈ Z. (ii) gTorR (iii) gTorR Proof. The implication (ii) =⇒ (iii) is trivial. rem 3.10 yields isomorphisms gTorR (i) =⇒ (ii): By 3.9 the R◦ module M has a Tate flat resolution (T, F ). Theo- i (M, −) ∼= Hi(T ⊗R −) for all i ∈ Z. It follows by induction on the flat dimension of N that T ⊗R N is acyclic; cf. [10, lem. 2.3]. (iii) =⇒ (i): Let P ≃−−→ M be a degree-wise finitely generated projective reso- lution, and let R ≃−−→ I be an injective resolution. By assumption, the complex P e⊗R I is acyclic, which explains the third ≃ below: M ≃ P ⊗R R ≃ P ⊗R I ≃ P ⊗R I ∼= P ⊗R HomR(R, I) ∼= HomR(HomR(P, R), I). The last isomorphism holds by Proposition A.6 as HomR(P, R) equals HomR(P, R). Now it follows from [3, thm. 2] and [9, thm. (2.2.3)] that G-dimR M is finite. (cid:3) 3.12 Corollary. For a commutative noetherian ring R the following are equivalent. (i) R is Gorenstein. (ii) gTorR (iii) gTorR i (M, R) = 0 for all finitely generated R-modules M and all i ∈ Z. i (M, R) = 0 for all finitely generated R-modules M and all i < 0. Proof. By a result of Goto, R is Gorenstein if and only if every finitely generated R-module has finite G-dimension [19, cor. 2]. Combining this with Theorem 3.11 one gets the the equivalence of (i) and (ii). The implication (ii) =⇒ (iii) is clear. Finally, (ii) follows from (iii) by dimension shifting (2.11.1). (cid:3) 4. Balancedness of stable homology (N, M ). Absolute homology is balanced over any ring: there are always isomorphisms TorR(M, N ) ∼= TorR◦ It follows already from Corollary 3.2 that stable homology can be balanced only over special rings. Indeed, if R is an Artin alge- bra and stable homology is balanced over R, then the dual module of R has finite projective dimension over both R and R◦, whence R is Iwanaga-Gorenstein. The converse is part of Corollary 4.5. We open this section with a technical lemma; it is similar to a result of Enochs, Estrada, and Iacob [14, thm. 3.6]. Recall the notation Cm(−) from 1.1. 4.1 Lemma. Let T and T ′ be acyclic complexes of flat R◦-modules and flat R- modules, respectively. For all integers m and n there are isomorphisms in D(k), T6n−1 ⊗R Cm(T ′) ≃ Σn−m(Cn(T ) ⊗R T ′ T>n ⊗R Cm(T ′) ≃ Σn−m(Cn(T ) ⊗R T ′ 6m−1) >m) T ⊗R Cm(T ′) ≃ Σn−m(Cn(T ) ⊗R T ′) . STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 13 Proof. Because the complexes T and T ′ are acyclic, there are quasi-isomorphisms Cn(T ) → Σ1−n T6n−1 and Cm(T ′) → Σ1−m T ′ 6m−1. Hence [10, prop. 2.14(b)] yields T6n−1 ⊗R Cm(T ′) ≃ T6n−1 ⊗R Σ1−m T ′ 6m−1 ≃ Σn−m(Σ1−n T6n−1 ⊗R T ′ ≃ Σn−m(Cn(T ) ⊗R T ′ 6m−1) . 6m−1) This demonstrates the first isomorphism in the statement, and the second is proved similarly. Finally, these two isomorphisms connect the exact sequences 0 → T6n−1 ⊗R Cm(T ′) → T ⊗R Cm(T ′) → T>n ⊗R Cm(T ′) → 0 and 0 → Cn(T ) ⊗R T ′ 6m−1 → Cn(T ) ⊗R T ′ → Cn(T ) ⊗R T ′ >m → 0 , and one obtains the last isomorphism via the Triangulated Five Lemma in D(k); see [23, prop. 4.3]. (cid:3) 4.2 Theorem. Let M be an R◦-module and N be an R-module. If they both have Tate flat resolutions, then for each i ∈ Z there is an isomorphism gTorR i (M, N ) ∼= gTorR◦ i (N, M ) . Proof. Let (T, F ) and (T ′, F ′) be Tate flat resolutions of M and N , respectively. Choose n ∈ Z such that there are isomorphisms T>n ∼= F>n and T ′ >n. For every i ∈ Z one has ∼= F ′ >n i (M, N ) ∼= Hi(T ⊗R N ) gTorR ∼= Hi−n(T ⊗R Cn(T ′)) ∼= Hi−n(Cn(T ) ⊗R T ′) ∼= Hi(M ⊗R T ′) ∼= gTorR◦ (N, M ) , i where the first and the last isomorphisms follow from Theorem 3.10, the second and fourth isomorphisms follow by dimension shifting, and the third isomorphism holds by the last isomorphism in Lemma 4.1. (cid:3) 4.3 Definition. Let M be an R◦-module and N be an R-module. Stable homology is balanced for M and N if one has gTorR i (M, N ) ∼= gTorR◦ i (N, M ) for all i ∈ Z. 4.4. Theorem 4.2 says that stable homology is balanced for all (pairs of) R◦- and R- modules that have Tate flat resolutions. If R is Iwanaga -- Gorenstein, then every R◦- module and every R-module has a Tate flat resolution, see 3.9 and [16, thm. 12.3.1], so stable homology is balanced for all (pairs of) R◦- and R-modules. 4.5 Corollary. For an Artin algebra R the following conditions are equivalent. (i) R is Iwanaga-Gorenstein. (ii) Stable homology is balanced for all R◦- and R-modules. (iii) Stable homology is balanced for all finitely generated R◦- and R-modules. 14 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER Proof. Per 4.4, part (i) implies (ii), which clearly implies (iii). Let E = D(R) be the dual module of R; it is injective and finitely generated over R◦ and over R. Thus, if stable homology is balanced for finitely generated R◦- and R modules, then it follows from 2.3 and Corollary 3.2 that pdR E as well as pdR◦ E is finite. By duality, both idR R and idR◦ R are then finite. (cid:3) 4.6 Corollary. A commutative noetherian ring R is Gorenstein if and only if stable homology is balanced for all finitely generated R-modules. Proof. Over a Gorenstein ring, the G-dimension of every finitely generated module is finite by [19, cor. 2], so every finitely generated module has a Tate flat resolution; see 3.9. Thus, balancedness of stable homology follows from Theorem 4.2. Con- i (M, R) = 0 for every finitely (cid:3) versely, balancedness of stable homology implies gTorR generated R-module M and all i ∈ Z, so R is Gorenstein by Corollary 3.12. 4.7 Corollary. A commutative noetherian ring R of finite Krull dimension is Gorenstein if and only if stable homology is balanced over R. Proof. If R is Gorenstein of finite Krull dimension, then it is Iwanaga-Gorenstein. Therefore, stable homology is balanced over R per 4.4. The converse holds by Corollary 4.6. (cid:3) 5. Vanishing of stable homology gTor(−, N ) i (−, N ) over an Artin algebra can by duality be Vanishing of stable homology gTorR understood via vanishing of gTorR i (M, −). 5.1 Proposition. Let R be an Artin algebra with duality functor D(−). For a finitely generated R-module N the following conditions are equivalent. (i) idR N is finite. i (−, N ) = 0 for all i ∈ Z. (ii) gTorR (iii) There is an integer i 6 0 with gTorR (iv ) gTorR 0 (D(N ), N ) = 0. R◦-modules M . i (M, N ) = 0 for all finitely generated Proof. The implications (i) =⇒ (ii) =⇒ (iii) are clear; see 2.3. Part (iv ) follows from (iii) by dimension shifting (2.11.1), as D(N ) is finitely generated. Finally, 0 (D(N ), D(D(N ))) implies by Corollary 3.2 that (cid:3) vanishing of gTorR Local rings. To analyze vanishing of stable homology gTorR pdR◦ D(N ) is finite, whence idR N is finite, and so (i) follows from (iv ). 0 (D(N ), N ) ∼= gTorR tative noetherian rings, we start locally. i (−, N ) over commu- 5.2. Let R be a commutative noetherian local ring with residue field k. For an R-module M , the depth invariant can be defined as depthR M = inf{i ∈ Z Exti R(k, M ) 6= 0} , and if M is finitely generated, then its injective dimension can be computed as idR M = sup{i ∈ Z Exti R(k, M ) 6= 0} . STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 15 The depth is finite for M 6= 0. The ring R is Cohen -- Macaulay if there exists a finitely generated module M 6= 0 of finite injective dimension; this is a consequence of the New Intersection Theorem due to Peskine and Szpiro [30] and Roberts [32]. The following is an analogue of [5, thm. 6.1]. 5.3 Lemma. Let R be a commutative noetherian local ring with residue field k. For every finitely generated R-module N and for every i ∈ Z there is an isomorphism TorR i (k, N ) ∼= Y j∈Z Homk(Extj R(k, R), Extj−i R (k, N )) i (k, N ) is a k-vector space. of R-modules, and in particular, TorR Proof. Let P ≃−−→ k be a degree-wise finitely generated projective resolution and let N ≃−−→ I and R ≃−−→ J be injective resolutions. By definition TorR i (k, N ) is the ith homology of the complex P ⊗R I which we compute using Proposition A.6 as follows P ⊗R I ∼= P ⊗R HomR(R, I) ∼= HomR(HomR(P, R), I). Next we simplify using quasi-isomorphisms and Hom-tensor adjointness: HomR(HomR(P, R), I) ≃ HomR(HomR(P, J), I) ≃ HomR(HomR(k, J), I) ∼= HomR(HomR(k, J) ⊗k k, I) ∼= Homk(HomR(k, J), HomR(k, I)) . Finally, pass to homology. (cid:3) 5.4 Remark. Lemma 5.3 suggests that stable homology gTorR i (k, N ) may be a k- vector space, and indeed it is. If R is a ring and if x annihilates the R◦-module M , or the R-module N , then there are two homotopic lifts -- zero and multiplication by x -- to the projective resolution in the case of M or to the injective resolution in the case of N ; see Definition 2.1 and [34, 22.6 and 2.3.7]. Hence multiplication by x is zero on unbounded and on stable homology of M against N . In particular, stable homology gTorR i (k, N ) is a k-vector space. gTorR 5.5 Proposition. Let R be a commutative noetherian local ring with residue field k and let N be a finitely generated R-module. If for some i ∈ Z the k-vector space i (k, N ) has finite rank, then N has finite injective dimension, or R is Gorenstein. j (k, N ) has finite rank, so TorR Proof. Each k-vector space TorR i+1(k, N ) has finite rank by the assumption and the exact sequence (2.5.1). For a finitely generated R-module M 6= 0 the vector spaces Extj R(k, M ) are non-zero for all j between depthR M < ∞ and idR M ; see Roberts [31, thm. 2]. When TorR i+1(k, N ) has finite rank, it follows from Lemma 5.3 that R or N has finite injective dimension. (cid:3) Compared to the characterization of globally Gorenstein rings in Corollary 3.12, condition (iii) below is sharper. 5.6 Theorem. Let R be a commutative noetherian local ring with residue field k. The following conditions are equivalent. (i) R is Gorenstein. (ii) gTorR i (−, R) = 0 for all i ∈ Z. 16 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER i (k, R) has finite rank for some i ∈ Z. (iii) gTorR (iv ) There exists a finitely generated R-module M such that gTorR i (M, R) = 0 holds for all i ∈ Z. rank for some i ∈ Z and gTorR Proof. The implications (i) =⇒ (ii) =⇒ (iii) =⇒ (iv ) are clear; see 2.3. Let M be a module as specified in (iv ); Theorem 3.11 yields G-dimR M < ∞. By a result of Holm [22, thm. 3.2], R is Gorenstein if idR M is also finite. Now it follows from Proposition 5.5 that R is Gorenstein. (cid:3) i (k, M ) has finite 5.7 Remark. Let R be a commutative noetherian local ring with residue field k, and let E denote the injective hull of k; it is a faithfully injective R-module. i (k, k) and A computation based on Theorem A.8 shows that the ranks of gTorR gExti R(k, k) are simultaneously finite: rankk gTorR i (k, k) = rankk gTorR i (k, HomR(k, E)) = rankk HomR(gExti = rankk gExti R(k, k) . R(k, k), E) Combined with this equality of ranks, [5, thm. 6.4, 6.5, and 6.7] yield characteriza- tions of regular, complete intersection, and Gorenstein local rings in terms of the i (k, k). For example, R is regular if and only i (k, k) = 0 holds for some (equivalently, all) i ∈ Z, and R is Gorenstein if size of the stable homology spaces gTorR if gTorR and only if gTorR i (k, k) has finite rank for some (equivalently, all) i ∈ Z. Commutative rings. If R is commutative and p is a prime ideal in R, then the local ring Rp is a flat R-algebra, and for an Rp-module M it follows from 2.12 that the stable homology modules gTorR i (M, N ) are Rp-modules. 5.8 Lemma. Let R be a commutative noetherian ring and let N be an R-module; let p be a prime ideal in R and let M be an Rp-module. For every i ∈ Z there is a natural isomorphism of Rp-modules, gTorR i (M, N ) ∼= gTorRp i (M, Np) . (−, Np) = 0 for all prime ideals p in R. Hence, gTorR i (−, N ) = 0 implies gTorRp i Proof. Let N ≃−−→ I be an injective resolution. Let P ≃−−→ M be a projective reso- lution over Rp; it is a flat resolution of M as an R-module. The second isomorphism below follows from Proposition A.4. P ⊗R I ∼= (P ⊗Rp Rp) ⊗R I ∼= P ⊗Rp (Rp ⊗R I) ∼= P ⊗Rp (Rp ⊗R I) The computation gives P ⊗R I ∼= P ⊗Rp Ip; similarly one gets P ⊗R I ∼= P ⊗Rp Ip. Now (1.4.1) and the Five Lemma yield P e⊗R I ∼= P e⊗Rp Ip, and the desired iso- morphisms follow as Np → Ip is an injective resolution by Matlis Theory. (cid:3) The proof of the next result is similar. Compared to Lemma 5.8 the noetherian hypothesis on R has been dropped, as it was only used to invoke Matlis Theory. STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 17 5.9 Lemma. Let R be a commutative ring and let M be an R-module; let p be a prime ideal in R and let N be an Rp-module. For every i ∈ Z there is a natural isomorphism of Rp-modules, gTorR i (M, N ) ∼= gTorRp i (Mp, N ) . Hence, gTorR i (M, −) = 0 implies gTorRp i (Mp, −) = 0 for all prime ideals p in R. (cid:3) 5.10 Theorem. Let R be a commutative noetherian ring and let N be a finitely i (−, N ) = 0 holds for some i ∈ Z, then idRp Np is generated R-module. finite for every prime ideal p in R. If gTorR Proof. From the hypotheses and Lemma 5.8 one has gTorRp However, if Rp is Gorenstein, then vanishing of gTorRp (−, Np) = 0. It follows from Proposition 5.5 that the local ring Rp is Gorenstein, or idRp Np is finite. (−, Np) = 0 implies that Np is finite by Corollary 4.7 and Theorem 3.1, and then idRp Np is finite. (cid:3) i i pdRp The next corollary is now immediate per the remarks in 5.2. 5.11 Corollary. A commutative noetherian ring R is Cohen -- Macaulay if there is (cid:3) i (−, N ) = 0 for some i ∈ Z. a finitely generated R-module N 6= 0 with gTorR 5.12 Corollary. Let R be a commutative noetherian ring of finite Krull dimension. For a finitely generated R-module N , the following conditions are equivalent. (i) idR N is finite. (ii) gTorR (iii) gTorR i (−, N ) = 0 for all i ∈ Z. i (−, N ) = 0 for some i ∈ Z. Proof. The implications (i) =⇒ (ii) =⇒ (iii) are clear; see 2.3. Part (i) follows from (iii) as idR N equals sup{idRp Np p is a prime ideal in R} 6 dim R. (cid:3) 5.13 Remark. We do not know if the assumption of finite Krull dimension in N is locally of finite injective dimension, but that does not imply finite injective dimension over R: Just consider a Gorenstein ring R of infinite Krull dimension. Corollary 5.12 is necessary. By Theorem 5.10, vanishing of gTorR(−, N ) implies that On the other hand, we do not know if gTorR(−, R) vanishes for such a ring. 6. Comparison to Tate homology In this section we compare stable homology to Tate homology. We parallel some of the findings of Avramov and Veliche [5]. First we recall a few definitions. 6.1. An acyclic complex T of projective R◦-modules is called totally acyclic if HomR◦ (T, P ) is acyclic for every projective R◦-module P ; cf. 3.8. A complete projective resolution of an R◦-module M is a diagram T −−→ P ≃−−→ M , where T is a totally acyclic complex of projective R◦-modules, P ≃−−→ M is a projective resolution, and i is an isomorphism for i ≫ 0; see [33, sec. 2]. Let M be an R◦-module with a complete projective resolution T → P → M . For an R-module N , the Tate homology of M and N over R are the k-modules i (M, N ) = Hi(T ⊗R N ) for i ∈ Z; see Iacob [24, sec. 2]. dTorR 18 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER 6.2. An R◦-module has a complete projective resolution if and only if it has finite Gorenstein projective dimension; see [33, thm. 3.4]. If R is noetherian and M is a finitely generated R◦-module with a complete pro- jective resolution, then M has finite G-dimension and it has a complete projective resolution T → P → M with T and P degree-wise finitely generated and T → P surjective; see [33, 2.4.1] and [4, thm. 3.1]. 6.3 Lemma. Let M be an R◦-module that has a complete projective resolution T → P → M . The following conditions are equivalent. (i) T ⊗R E is acyclic for every injective R-module E. (ii) There are isomorphisms of functors gTorR i (M, −) ∼= dTorR i (M, −) for all i ∈ Z. Proof. Assume that T ⊗R E is acyclic for every injective R-module E. The pair (T, P ) is then a Tate flat resolution of M , see 3.8, and it follows from Theorem 3.10 i (M, −) are isomorphic for all i ∈ Z. For that the functors gTorR i (M, −) and dTorR the converse, let E be an injective R-module. By 2.3 one then has for all i ∈ Z, and hence H(T ⊗R E) = 0. 0 = gTorR i (M, E) ∼= dTorR i (M, E) (cid:3) 6.4 Theorem. Let R be noetherian, and let M be a finitely generated R◦-module that has a complete projective resolution. There are isomorphisms of functors gTorR i (M, −) ∼= dTorR i (M, −) for all i ∈ Z . Proof. The module M has a complete projective resolution T → P → M with T and P degree-wise finitely generated; see 6.2. The isomorphisms (3.9.1) show that the complex T ⊗R E is acyclic for every injective R-module E, and Lemma 6.3 finishes the argument. (cid:3) 6.5 Remark. The isomorphisms of homology modules in Theorem 6.4 actually follow from one isomorphism in D(k), but this is unapparent in the proof, which rests on Theorem 3.10. The finitely generated module M has a complete projective resolution T → L → M with T and L degree-wise finitely generated and T → L surjective; see 6.2. Thus the kernel K = Ker(T → L) is a bounded above complex of projective modules. Given a module N , let C be the cone as one would construct in 3.4. By Proposition 1.7(a) the complex K ⊗R C is acyclic, so that Lemma 3.5 and 1.5(c) yield Σ−1(L e⊗R I) ≃ Σ−1(L ⊗R C) ≃ Σ−1(T ⊗R C). As in the proof above, T ⊗R E is acyclic, so Lemma 3.6 gives Σ−1(T ⊗R C) ≃ T ⊗R N , and combining these isomorphisms in D(k) gives the desired one. Without extra assumptions on the ring, we do not know if stable homology agrees with Tate homology whenever the latter is defined. In general, the relation between stable homology and Tate homology is tied to an unresolved problem in Gorenstein homological algebra. Theorem 6.7 explains how. 6.6. An R◦-module G is called Gorenstein projective if there exists a totally acyclic complex T of projective R◦-modules with C0(T ) ∼= G; see 6.1. Similarly, an R◦- module G is called Gorenstein flat if there exists a totally acyclic complex T of flat R◦-modules with C0(T ) ∼= G; see 3.8. STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 19 6.7 Theorem. The following conditions on R are equivalent. (i) Every Gorenstein projective R◦-module is Gorenstein flat. (ii) For every R◦-module M that has a complete projective resolution there are isomorphisms of functors gTorR i (M, −) ∼= dTorR i (M, −) for all i ∈ Z. Proof. Assume that every Gorenstein projective R◦-module is Gorenstein flat. Let T → P → M be a complete projective resolution. It follows that T is a totally acyclic complex of flat modules; see Emmanouil [13, thm. 2.2]. Thus stable homology and Tate homology coincide by Lemma 6.3. For the converse, let M be a Gorenstein projective R◦-module and let T be a totally acyclic complex of projective R◦-modules with M ∼= C0(T ). Since there are isomorphisms of functors gTorR i (M, −) for all i ∈ Z, it follows from 2.3 that T ⊗R E is acyclic for every injective R-module E. Thus T is a totally acyclic complex of flat R◦-modules, and so M is Gorenstein flat. (cid:3) i (M, −) ∼= dTorR 6.8 Remark. As Holm notes [21, prop. 3.4], the obvious way to achieve that every Gorenstein projective R◦-module is Gorenstein flat is to ensure that (1) the Pon- tryagin dual of every injective R-module is flat, and (2) that every flat R◦-module has finite projective dimension. The first condition is satisfied if R is left coherent, and the second is discussed in Remark 2.8. A description of the rings over which Gorenstein projective modules are Gorenstein flat seems elusive; see [13, sec. 2]. Complete homology . In his thesis, Triulzi considers the J-completion of the ho- mological functor TorR(M, −) = {TorR i (M, −) i ∈ Z}. His construction is similar to Mislin's P-completion of covariant Ext and Nucinkis' I-completion of contravari- ant Ext. The resulting homology theory is called complete homology. Like stable homology, it is a generalization of Tate homology. We compare these two general- izations in [8]. From the point of view of stable homology, it is interesting to know when it agrees with complete homology, because the latter has a universal property. In this direction the main results in [8] are that these two homology theories agree over Iwanaga-Gorenstein rings, and for finitely generated modules over Artin alge- bras and complete commutative local rings. Moreover, the two theories agree with Tate homology, whenever it is defined, under the exact same condition; that is, if and only if every Gorenstein projective module is Gorenstein flat; see Theorem 6.7. We start by recalling the definition of stable cohomology. Appendix A.1. Let X and Y be R-complexes, following [5, 18] we denote by HomR(X, Y ) the subcomplex of HomR(X, Y ) with degree n term HomR(X, Y )n = a i∈Z HomR(Xi, Yi+n) . It is called the bounded Hom complex, and the quotient complex ]HomR(X, Y ) = HomR(X, Y )/HomR(X, Y ) is called the stable Hom complex. For R-modules M and N with projective resolutions PM ≃−−→ M and PN ≃−−→ N , the k-modules Exti R(M, N ) = H−i(HomR(PM , PN )) , 20 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER are called the bounded cohomology of M and N over R, and the stable cohomology modules of M and N over R are R(M, N ) = H−i(]HomR(PM , PN )) . gExti Avramov and Veliche [5] use the notation dExti R(M, N ) for the stable cohomol- ogy; this notation is standard for Tate cohomology, which coincides with stable cohomology whenever the former is defined; see [5, cor. 2.4]. A.2 Proposition. Let X and Y be R-complexes. (a) If X or Y is bounded above, and HomR(Xi, Y ) is acyclic for all i, then the complex HomR(X, Y ) is acyclic. (b) If X or Y is bounded below, and HomR(X, Yi) is acyclic for all i, then the complex HomR(X, Y ) is acyclic. Proof. Similar to the proof of Proposition 1.7. (cid:3) Standard isomorphisms. We study composites of the functors Hom and ⊗. To the extent possible, we establish analogs of the standard isomorphisms for com- posites of Hom and ⊗; see [7, sec. II.5 and VI.5]. There seems to be no analog of Hom-tensor adjunction [7, prop. II.5.2]. The setup is the same for Propositions A.4 -- A.6; namely: A.3. Let X be a complex of R◦-modules, let Y be a complex of (R, S ◦)-bimodules, and let Z be a complex of S-modules. Under finiteness conditions, the unbounded tensor product is associative. A.4 Proposition. For complexes as in A.3, under either of the following conditions • X and Z are complexes of finitely presented modules • Y is a bounded complex there is an isomorphism of k-complexes, (X ⊗R Y ) ⊗S Z −→ X ⊗R (Y ⊗S Z) , and it is functorial in X, Y , and Z. Proof. For every n ∈ Z one has, and ((X ⊗R Y ) ⊗S Z)n = Y (X ⊗R (Y ⊗S Z))n = Y i∈Z (cid:16)Qj∈Z Xj ⊗R Yi−j(cid:17) ⊗S Zn−i Xj ⊗R (cid:0)Qi∈Z Yi−j ⊗S Zn−i(cid:1) , j∈Z and from each of these modules there is a canonical homomorphism to Y i∈Z Y j∈Z (Xj ⊗R Yi−j) ⊗S Zn−i ∼= Y Y i∈Z j∈Z Xj ⊗R (Yi−j ⊗S Zn−i) . If Y is bounded, then these homomorphisms are isomorphisms as the inner most products are finite. Recall, e.g. from [16, thm. 3.2.22], that the functor M ⊗R − commutes with products if M is finitely presented. Thus, the canonical homo- morphisms are isomorphisms when X and Z are complexes of finitely presented modules. It is straightforward to verify that these isomorphisms commute with the differentials and form an isomorphism of complexes. (cid:3) STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 21 The model for the following swap isomorphism is [7, ex. II.4]. The proof is similar to the proof of Proposition A.4 and uses that the functor HomR(M, −) commutes with coproducts if M is a finitely generated R-module. A.5 Proposition. For complexes as in A.3, under either of the following conditions • X and Z are complexes of finitely generated modules • Y is a bounded complex there is an isomorphism of k-complexes, HomR◦ (X, HomS(Z, Y )) −→ HomS(Z, HomR◦ (X, Y )) , and it is functorial in X, Y , and Z. (cid:3) The next result is an analog of [7, prop. VI.5.2 and VI.5.3]; it is used below to establish a duality between stable homology and stable cohomology. A.6 Proposition. Let X, Y , and Z be as in A.3, and assume further that X is a complex of finitely presented R◦-modules. There is a morphism of k-complexes, X ⊗R HomS(Y, Z) −→ HomS(HomR◦ (X, Y ), Z) , and it is functorial in X, Y , and Z. Furthermore, it is an isomorphism if X is a complex of projective modules or Z is a complex of injective modules. Proof. For every n ∈ Z one can compute as follows, (Xi ⊗R HomS(Y, Z)n−i) (X ⊗R HomS(Y, Z))n = Y = Y ∼= Y i∈Z i∈Z (Xi ⊗R Qj∈Z Y i∈Z j∈Z HomS(Yj , Zn−i+j)) (Xi ⊗R HomS(Yj , Zn−i+j)) , where the isomorphism holds as the module Xi is finitely presented for every i ∈ Z; see [16, thm. 3.2.22]. On the other hand, for every n ∈ Z one has HomS(HomR◦ (X, Y )h, Zn+h) h∈Z HomS(HomR◦ (X, Y ), Z)n = Y = Y ∼= Y = Y h∈Z h∈Z HomS(`i∈Z Y Y i∈Z i∈Z j∈Z HomR◦ (Xi, Yi+h), Zn+h) HomS(HomR◦ (Xi, Yi+h), Zn+h) HomS(HomR◦ (Xi, Yj), Zn−i+j) . Now set (θXY Z)n = Qi∈ZQj∈Z(−1)i(n−j)θXiYj Zn−i+j , where θXiYj Zn−i+j : Xi ⊗R HomS(Yj , Zn−i+j) −→ HomS(HomR◦ (Xi, Yj), Zn−i+j) is the homomorphism of k-modules given by θXiYj Zn−i+j (x ⊗ ψ)(φ) = ψφ(x). It is straightforward to verify that θXY Z is a morphism of k-complexes and functorial in X, Y , and Z. 22 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER Finally, if each module Xi is projective or each module Zn−i+j is injective, then it follows from [7, prop. VI.5.2 and VI.5.3] that θXiYj Zn−i+j is invertible for all i, j, n ∈ Z, and so the morphism θXY Z is invertible. (cid:3) A.7 Lemma. Let P be a bounded below complex of finitely generated projective R◦-modules and let X be a complex of R◦-modules with H(X) bounded. For every injective k-module E, there is an isomorphism in the derived category D(k): P e⊗R Homk(X, E) −→ Σ Homk(]HomR◦ (P, X), E) , and it is functorial in P , X, and E. Proof. Set (−)∨ = Homk(−, E). In the commutative square of k-complexes P ⊗R X ∨ τ P ⊗R X ∨ θ ∼= α HomR◦ (P, X)∨ ϑ∨ / HomR◦ (P, X)∨ each horizontal morphism is (the dual of) a canonical embedding. The vertical map α is the isomorphism from Proposition A.6. The morphism θ is the standard evaluation map; it is a quasi-isomorphism by [2, 4.4(I)]. The square induces a morphism of triangles in the homotopy category: P ⊗R X ∨ τ P ⊗R X ∨ / Cone τ / Σ(P ⊗R X ∨) θ≃ ∼= α γ ≃ Σ θ HomR◦ (P, X)∨ ϑ∨ / HomR◦ (P, X)∨ / Cone ϑ∨ / Σ HomR◦ (P, X)∨ The construction of γ is functorial in all three arguments, and it is a quasi-isomorphism because α and θ are quasi-isomorphisms. Recall, say from [17, III.3.4 -- 5], that there are natural quasi-isomorphisms Cone τ ≃ Coker τ = P e⊗R X ∨ and Cone ϑ∨ ∼= Σ(Cone ϑ)∨ ≃ Σ(Coker ϑ)∨ = Σ ]HomR◦ (P, X)∨ . They yield the desired isomorphism in the derived category P e⊗R Homk(X, E) −→ Σ Homk(]HomR◦ (P, X), E) . With regard to functoriality of γ, notice that a morphism between arguments, P → P1 say, induces the solid commutative square in the following diagram. Coker τ /❴❴❴❴❴❴❴❴❴❴❴❴❴❴❴ Coker τ1 Cone τ Cone τ1 γ γ1 Cone ϑ∨ Cone(ϑ∨ 1 ) 8rrrrrrrrr ≃ ≃ %▲▲▲▲▲▲▲▲▲ e❑❑❑❑❑❑❑❑❑ ≃ ≃ zttttttttt Σ(Coker ϑ)∨ /❴❴❴❴❴❴❴❴❴❴❴❴ Σ(Coker ϑ1)∨ / / / / / / / / / / / ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ / / e 8 / / z % / STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 23 Commutativity in the derived category of the dashed square is now a consequence. Functoriality in the other arguments is handles similarly. (cid:3) The Lemma immediately yields a useful duality. A.8 Theorem. Let M and N be R◦-modules and assume that M has a degree- wise finitely generated projective resolution. For every injective k-module E and for every i ∈ Z there is an isomorphism of k-modules, Homk(gExti R◦ (M, N ), E) ∼= gTorR i (M, Homk(N, E)) , and it is functorial in M , N , and E. (cid:3) To prove the next two results one proceeds as in the proof of Proposition A.6. A.9 Proposition. Let X be a complex of finitely generated R-modules and let Y and Z be as in A.3. There is a morphism of k-complexes HomR(X, Y ) ⊗S Z −→ HomR(X, Y ⊗S Z) , and it is functorial in X, Y , and Z. Furthermore, it is an isomorphism under each of the following conditions • Z is a complex of finitely generated projective modules • X is a complex of finitely presented modules and Z is a complex of flat modules • X is a complex of projective modules (cid:3) A.10 Proposition. Let X be a complex of R-modules, let Y be a complex of (R, S ◦)-bimodules, and let Z be a complex of finitely presented S-modules. There is a morphism of k-complexes, HomR(X, Y ) ⊗S Z −→ HomR(X, Y ⊗S Z) , and it is functorial in X, Y , and Z. Furthermore, it is an isomorphism if X or Z is a complex of projective modules. (cid:3) Pinched tensor products. Christensen and Jorgensen devised in [12] a pinched tensor product, − ⊗✶ R −, to compute Tate homology. In view of Theorem 3.10 their proof of [12, thm. 3.5] applies verbatim to yield the next result; we refer the reader to [12] for the definition of the pinched tensor product. A.11 Theorem. Let M be an R◦-module that has a Tate flat resolution (T, F ), let A be an acyclic complex of R-modules and set N = C0(A). For every i ∈ Z, there is an isomorphism of k-modules The next corollary is an analogue of [12, Corollary 4.10]. i (M, N ) ∼= Hi(T ⊗✶ R A) . gTorR (cid:3) A.12 Corollary. Let R be commutative and let M and N be Gorenstein flat R- modules with corresponding totally acyclic complexes of flat modules T and T ′, R T ′ is an acyclic i (M, N ) = 0 holds for all i ∈ Z, then T ⊗✶ complex of flat R-modules, and the following statements are equivalent: respectively. If gTorR (i) T ⊗✶ (ii) gTorR R T ′ is a totally acyclic complex of flat R-modules. i (M, N ⊗R E) = 0 holds for every injective R-module E and all i ∈ Z. 24 O. CELIKBAS, L. W. CHRISTENSEN, L. LIANG, AND G. PIEPMEYER When these conditions hold, M ⊗R N is a Gorenstein flat R-module and T ⊗✶ is a corresponding totally acyclic complex of flat R-modules. R T ′ Proof. It follows from the definition of pinched tensor products that T ⊗✶ then the complex is acyclic by Theorem A.11. if (T ⊗✶ is a complex of flat R-modules, and if gTorR R T ′ i (M, N ) = 0 holds for all i ∈ Z, It is totally acyclic if and only R T ′) ⊗R E ∼= T ⊗✶ R (T ⊗R E) is acyclic for every injective R-module E; that is, if and only if gTorR i (M, N ⊗R E) = 0 holds for every injective R-module E and all i ∈ Z. Finally, it follows from the definition of pinched tensor products that there is an isomorphism M ⊗R N ∼= C0(T ⊗✶ (cid:3) R T ′). References [1] Maurice Auslander and Mark Bridger, Stable module theory, Memoirs of the American Math- ematical Society, No. 94, American Mathematical Society, Providence, R.I., 1969. MR0269685 [2] Luchezar L. Avramov and Hans-Bjørn Foxby, Homological dimensions of unbounded com- plexes, J. Pure Appl. Algebra 71 (1991), no. 2-3, 129 -- 155. MR1117631 [3] Luchezar L. Avramov, Srikanth B. Iyengar, and Joseph Lipman, Reflexivity and rigidity for complexes. I. Commutative rings, Algebra Number Theory 4 (2010), no. 1, 47 -- 86. MR2592013 [4] Luchezar L. Avramov and Alex Martsinkovsky, Absolute, relative, and Tate cohomology of modules of finite Gorenstein dimension, Proc. London Math. Soc. (3) 85 (2002), no. 2, 393 -- 440. MR1912056 [5] Luchezar L. Avramov and Oana Veliche, Stable cohomology over local rings, Adv. Math. 213 (2007), no. 1, 93 -- 139. MR2331239 [6] Hyman Bass and M. Pavaman Murthy, Grothendieck groups and Picard groups of abelian group rings, Ann. of Math. (2) 86 (1967), 16 -- 73. MR0219592 [7] Henri Cartan and Samuel Eilenberg, Homological algebra, Princeton Landmarks in Math- ematics, Princeton University Press, Princeton, NJ, 1999, With an appendix by David A. Buchsbaum, Reprint of the 1956 original. MR1731415 [8] Olgur Celikbas, Lars Winther Christensen, Li Liang, and Greg Piepmeyer, Complete homo- logy over associative rings, Israel J. Math, to appear. Preprint arXiv:1501.00297 [math.RA]. [9] Lars Winther Christensen, Gorenstein dimensions, Lecture Notes in Mathematics, vol. 1747, Springer-Verlag, Berlin, 2000. MR1799866 [10] Lars Winther Christensen, Anders Frankild, and Henrik Holm, On Gorenstein projective, injective and flat dimensions -- A functorial description with applications, J. Algebra 302 (2006), no. 1, 231 -- 279. MR2236602 [11] Lars Winther Christensen and Henrik Holm, The direct limit closure of perfect complexes, J. Pure Appl. Algebra 219 (2015), no. 3, 449 -- 463. MR3279365 [12] Lars Winther Christensen and David A. Jorgensen, Tate (co)homology via pinched complexes, Trans. Amer. Math. Soc. 366 (2014), no. 2, 667 -- 689. MR3130313 [13] Ioannis Emmanouil, On the finiteness of Gorenstein homological dimensions, J. Algebra 372 (2012), 376 -- 396. MR2990016 [14] Edgar E. Enochs, Sergio Estrada, and Alina C. Iacob, Balance with unbounded complexes, Bull. Lond. Math. Soc. 44 (2012), no. 3, 439 -- 442. MR2966988 [15] Edgar E. Enochs and Overtoun M. G. Jenda, Copure injective resolutions, flat resolvents and dimensions, Comment. Math. Univ. Carolin. 34 (1993), no. 2, 203 -- 211. MR1241728 [16] Edgar E. Enochs and Overtoun M. G. Jenda, Relative homological algebra, de Gruyter Ex- positions in Mathematics, vol. 30, Walter de Gruyter & Co., Berlin, 2000. MR1753146 [17] Sergei I. Gelfand and Yuri I. Manin, Methods of homological algebra, second ed., Springer Monographs in Mathematics, Springer-Verlag, Berlin, 2003. MR1950475 [18] Fran¸cois Goichot, Homologie de Tate-Vogel ´equivariante, J. Pure Appl. Algebra 82 (1992), no. 1, 39 -- 64. MR1181092 [19] Shiro Goto, Vanishing of Exti A(M, A), J. Math. Kyoto Univ. 22 (1982/83), no. 3, 481 -- 484. MR0674605 [20] Laurent Gruson and Christian U. Jensen, Dimensions cohomologiques reli´ees aux foncteurs (i), Paul Dubreil and Marie-Paule Malliavin Algebra Seminar, 33rd Year (Paris, 1980), lim ←− Lecture Notes in Math., vol. 867, Springer, Berlin, 1981, pp. 234 -- 294. MR0633523 STABLE HOMOLOGY OVER ASSOCIATIVE RINGS 25 [21] Henrik Holm, Gorenstein homological dimensions, J. Pure Appl. Algebra 189 (2004), no. 1-3, 167 -- 193. MR2038564 [22] Henrik Holm, Rings with finite Gorenstein injective dimension, Proc. Amer. Math. Soc. 132 (2004), no. 5, 1279 -- 1283. MR2053331 [23] Thorsten Holm and Peter Jørgensen, Triangulated categories: definitions, properties, and examples, Triangulated categories, London Math. Soc. Lecture Note Ser., vol. 375, Cambridge Univ. Press, Cambridge, 2010, pp. 1 -- 51. MR2681706 [24] Alina Iacob, Absolute, Gorenstein, and Tate torsion modules, Comm. Algebra 35 (2007), no. 5, 1589 -- 1606. MR2317632 [25] Christian U. Jensen, On the vanishing of lim (i), J. Algebra 15 (1970), 151 -- 166. MR0260839 ←− [26] David A. Jorgensen and Liana M. S¸ega, Independence of the total reflexivity conditions for modules, Algebr. Represent. Theory 9 (2006), no. 2, 217 -- 226. MR2238367 [27] Li Liang, Tate homology of modules of finite Gorenstein flat dimension, Algebr. Represent. Theory 16 (2013), no. 6, 1541 -- 1560. MR3127346 [28] Masayoshi Nagata, Local rings, Interscience Tracts in Pure and Applied Mathematics, No. 13, Interscience Publishers a division of John Wiley & Sons New York-London, 1962. MR0155856 [29] Barbara L. Osofsky, Homological dimension and cardinality, Trans. Amer. Math. Soc. 151 (1970), 641 -- 649. MR0265411 [30] Christian Peskine and Lucien Szpiro, Dimension projective finie et cohomologie locale. Ap- plications `a la d´emonstration de conjectures de M. Auslander, H. Bass et A. Grothendieck, Inst. Hautes ´Etudes Sci. Publ. Math. (1973), no. 42, 47 -- 119. MR0374130 [31] Paul Roberts, Two applications of dualizing complexes over local rings, Ann. Sci. ´Ecole Norm. Sup. (4) 9 (1976), no. 1, 103 -- 106. MR0399075 [32] Paul Roberts, Le th´eor`eme d'intersection, C. R. Acad. Sci. Paris S´er. I Math. 304 (1987), no. 7, 177 -- 180. MR0880574 [33] Oana Veliche, Gorenstein projective dimension for complexes, Trans. Amer. Math. Soc. 358 (2006), no. 3, 1257 -- 1283. MR2187653 [34] Charles A. Weibel, An introduction to homological algebra, Cambridge Studies in Advanced Mathematics, vol. 38, Cambridge University Press, Cambridge, 1994. MR1269324 University of Missouri, Columbia, MO 65211, U.S.A. Current address: University of Connecticut, Storrs, CT 06269, U.S.A E-mail address: [email protected] Texas Tech University, Lubbock, TX 79409, U.S.A. E-mail address: [email protected] URL: http://www.math.ttu.edu/~ lchriste Lanzhou Jiaotong University, Lanzhou 730070, China E-mail address: [email protected] Columbia Basin College, Pasco, WA 99301, U.S.A. E-mail address: [email protected]
1606.03862	3	1606	2017-03-01T18:10:54	Representing finitely generated refinement monoids as graph monoids	[ "math.RA", "math.KT", "math.OA" ]	Graph monoids arise naturally in the study of non-stable K-theory of graph C*-algebras and Leavitt path algebras. They play also an important role in the current approaches to the realization problem for von Neumann regular rings. In this paper, we characterize when a finitely generated conical refinement monoid can be represented as a graph monoid. The characterization is expressed in terms of the behavior of the structural maps of the associated $I$-system at the free primes of the monoid.	math.RA	math	REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS AS GRAPH MONOIDS PERE ARA AND ENRIQUE PARDO Abstract. Graph monoids arise naturally in the study of non-stable K-theory of graph C-algebras and Leavitt path algebras. They play also an important role in the current approaches to the realization problem for von Neumann regular rings. In this paper, we characterize when a finitely generated conical refinement monoid can be represented as a graph monoid. The characterization is expressed in terms of the behavior of the structural maps of the associated I-system at the free primes of the monoid. 1. Introduction The class of commutative monoids satisfying the Riesz refinement property -- refinement monoids for short -- has been largely studied over the last decades in connection with various problems such as non-stable K-Theory of rings and C ∗-algebras (see e.g. [4, 10, 11, 19, 24]), classification of Boolean algebras (see e.g. [20], [25]), or its own structure theory (see e.g. [16, 17, 28]). An important invariant in non-stable K-theory is the commutative monoid V(R) associated to any ring R, consisting of the isomorphism classes of finitely generated projective (left, say) R-modules, with the operation induced from direct sum. If R is a (von Neumann) regular ring or a C-algebra with real rank zero (more generally, an exchange ring), then V(R) is a refinement monoid (e.g., [10, Corollary 1.3, Theorem 7.3]). The realization problem asks which refinement monoids appear as a V(R) for R in one of the above-mentioned classes. Wehrung [29] constructed a conical refinement monoid of cardinality ℵ2 which is not isomorphic to V(R) for any regular ring R, but it is an important open problem, appearing for the first time in [18], to determine whether every countable conical refinement monoid can be realized as V(R) for some regular ring R. See [3] for a survey on this problem, and [9] for some recent progress on the problem, with connections with the Atiyah Problem. An interesting situation in which the answer to the realization problem is affirmative is the following: Theorem 1.1 ([5, Theorem 4.2, Theorem 4.4]). Let E be a row-finite graph, let M(E) be its graph monoid, and let K be any field. Then there exists a (not necessarily unital) von Neumann regular K-algebra QK(E) such that V(QK(E)) ∼= M(E). 2010 Mathematics Subject Classification. Primary 06F20, Secondary 16D70, 19K14, 20K20, 46L05, 46L55. Key words and phrases. Refinement monoid, regular monoid, primely generated monoid, graph monoid. Both authors are partially supported by the DGI-MINECO and European Regional Development Fund, jointly, through Project MTM2014-53644-P. The second author was partially supported by PAI III grant FQM-298 of the Junta de Andaluc´ıa. 1 2 PERE ARA AND ENRIQUE PARDO Thus, an intermediate step that could be helpful to give an answer to the realization prob- lem is to characterize which conical refinement monoids are representable as graph monoids. The first author, Perera and Wehrung gave such a characterization in the concrete case of finitely generated antisymmetric refinement monoids [13, Theorem 5.1]. These monoids are a particular case of primely generated refinement monoids (see e.g. [16]). Recall that an ele- ment p in a monoid M is a prime element if p is not invertible in M, and, whenever p ≤ a + b for a, b ∈ M, then either p ≤ a or p ≤ b (where x ≤ y means that y = x + z for some z ∈ M). The monoid M is primely generated if every non-invertible element of M can be written as a sum of prime elements. Primely generated refinement monoids enjoy important cancella- tion properties, such as separative cancellation and unperforation, as shown by Brookfield in [16, Theorem 4.5, Corollary 5.11(5)]. Moreover, it was shown by Brookfield that any finitely generated refinement monoid is automatically primely generated [16, Corollary 6.8]. In [12], the authors of the present paper showed that any primely generated refinement monoid can be represented, up to isomorphism, as the monoid associated to an I-system (a sort of semilattice of cancellative semigroups defined over a suitable poset I). This result generalizes the representation of two well-known classes of monoids: (1) The class of primitive monoids, i.e. antisymmetric primely generated refinement monoids, see [25]. These monoids are described by means of a set I endowed with an antisymmetric transitive relation. (2) The class of primely generated conical regular refinement monoids. These monoids were characterized by Dobbertin in [17] in terms of partial orders of abelian groups. In the present paper, we will benefit of the picture developed in [12] to state a character- ization of monoids representable as graph monoids, in the case of finitely generated conical refinement monoids (see Theorem 5.6). The condition, that relies on the behavior of free primes in the representation by I-systems, generalizes [13, Theorem 5.1]. A main device used to obtain realization results for refinement monoids by regular rings or C-algebras of real rank zero has been the consideration of algebras associated to graphs. Of particular importance has been the use of graph C-algebras (see e.g. [21], [22]) and [2], [1], [11]). Indeed, the von Neumann regular algebra of Leavitt path algebras (see e.g. QK(E) appearing in Theorem 1.1 is a specific universal localization of the Leavitt path algebra LK(E) associated to the row-finite graph E. Some other algebras associated to graphs are also important in further constructions of von Neumann regular rings, as the first author showed in [4]. In that paper, given any finite poset P, a von Neumann regular algebra QK(P) is constructed so that V(QK(P)) ∼= M(P), where M(P) is the primitive monoid with associated poset of primes P and with all primes being free. In the present paper, Leavitt path algebras play an instrumental role, being essential in determining the condition which characterizes finitely generated graph monoids (see Proposition 5.9). It is interesting to observe that, as shown in that Proposition, the obstruction to realize an arbitrary finitely generated conical refinement monoid as a graph monoid is of K-theoretical nature. The solution of this K- theoretical problem was one of the keys to obtain the main result of [4] (see [4, Theorem 3.2]), and will surely play a vital role in the forthcoming approaches to the resolution of the realization problem for finitely generated refinement monoids. The paper is organized as follows. In Section 2, we recall all the definitions and results that will be necessary to follow the contents of the subsequent sections. In Section 3, we study REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 3 the special case of finitely generated regular refinement monoids, and we prove that every finitely generated conical regular refinement monoid can be represented as a graph monoid (Theorem 3.6). In the short Section 4, we use the techniques developed in Section 3 to offer an easy presentation of the monoid M = Z+ ∪ {∞} as a graph monoid. This was done in a somewhat more involved way in [13, Example 6.5]. In Section 5, we prove the main result of the paper, obtaining the characterization of the finitely generated conical refinement monoids which are graph monoids. The result is stated in terms of the theory of I-systems established in [12], and concerns the behavior of the maps associated to the I-system at the free primes (see Theorem 5.6). Finally we use our main result to recover the characterization of finitely generated primitive graph monoids obtained in [13]. 2. Preliminaries In this section, we will recall the definitions and results necessary to follow the contents of the paper in a self-contained way. We divide this section in four parts. 2.1. Basics on commutative monoids. All semigroups and monoids considered in this paper are commutative. We will denote by N the semigroup of positive integers, and by Z+ the monoid of non-negative integers. Given a commutative monoid M, we set M ∗ := M \ {0}. We say that M is conical if M ∗ is a semigroup, that is, if, for all x, y in M, x + y = 0 only when x = y = 0. Given a monoid M, the antisymmetrization M of M is the quotient monoid of M by the congruence given by x ≡ y if and only if x ≤ y and y ≤ x (see [16, Notation 5.1]). We will denote the class of an element x of M in M by x. We say that a monoid M is separative provided 2x = 2y = x + y always implies x = y; there are a number of equivalent formulations of this property, see e.g. [10, Lemma 2.1]. We say that M is a refinement monoid if, for all a, b, c, d in M such that a + b = c + d, there exist w, x, y, z in M such that a = w + x, b = y + z, c = w + y and d = x + z. It will often be convenient to present this situation in the form of a diagram, as follows: c d a w x b y z . A basic example of refinement monoid is the monoid M(E) associated to a countable row- finite graph E [11, Proposition 4.4]. If x, y ∈ M, we write x ≤ y if there exists z ∈ M such that x + z = y. Note that ≤ is a translation-invariant pre-order on M, called the algebraic pre-order of M. All inequalities in commutative monoids will be with respect to this pre-order. An element p in a monoid M is a prime element if p is not invertible in M, and, whenever p ≤ a + b for a, b ∈ M, then either p ≤ a or p ≤ b. The monoid M is primely generated if every non-invertible element of M can be written as a sum of prime elements. An element x ∈ M is regular if 2x ≤ x. An element x ∈ M is an idempotent if 2x = x. An element x ∈ M is free if nx ≤ mx implies n ≤ m. Any element of a separative monoid is either free or regular. In particular, this is the case for any primely generated refinement monoid, by [16, Theorem 4.5]. 4 PERE ARA AND ENRIQUE PARDO A subset S of a monoid M is called an order-ideal if S is a subset of M containing 0, closed under taking sums and summands within M. An order-ideal can also be described as a submonoid I of M, which is hereditary with respect to the canonical pre-order ≤ on M: x ≤ y and y ∈ I imply x ∈ I. A non-trivial monoid is said to be simple if it has no non-trivial order-ideals. generated by a subset X of a semigroup S. coproduct (resp. the product) of the semigroups Sk, k ∈ Λ, in the category of commutative semigroups. If the semigroups Sk are subsemigroups of a semigroup S, we will denote by If (Sk)k∈Λ is a family of (commutative) semigroups,Lk∈Λ Sk (resp. Qk∈Λ Sk) stands for the Pk∈Λ Sk the subsemigroup of S generated by Sk∈Λ Sk. Note that Pk∈Λ Sk is the image of the canonical map Lk∈Λ Sk → S. We will use the notation hXi to denote the semigroup η : M → H to a group H there is a unique group homomorphism eη : G(M) → H such that eη ◦ ψM = η. G(M) is abelian and it is generated as a group by ψ(M). If M is already a group then G(M) = M. If M is a semigroup of the form N × G, where G is an abelian group, then G(M) = Z × G. In this case, we will view G as a subgroup of Z × G by means of the identification g ↔ (0, g). Given a semigroup M, we will denote by G(M) the Grothendieck group of M. There exists a semigroup homomorphism ψM : M → G(M) such that for any semigroup homomorphism Let M be a conical commutative monoid, and let x ∈ M be any element. The archimedean component of M generated by x is the subsemigroup GM [x] := {a ∈ M : a ≤ nx and x ≤ ma for some n, m ∈ N}. For any x ∈ M, GM [x] is a simple semigroup. If M is separative, then GM [x] is a cancella- tive semigroup; if moreover x is a regular element, then GM [x] is an abelian group. 2.2. Primely generated refinement monoids. The structure of primely generated refine- ment monoids has been recently described in [12]. We recall here some basic facts. Given a poset (I, ≤), we say that a subset A of I is a lower set if x ≤ y in I and y ∈ A implies x ∈ A. For any i ∈ I, we will denote by I ↓ i = {x ∈ I : x ≤ i} the lower subset generated by i. We will write x < y if x ≤ y and x 6= y. The following definition [12, Definition 1.1] is crucial for this work: Definition 2.1 ([12, Definition 1.1]). Let I = (I, ≤) be a poset. An I-system is given by the following data: J = (I, ≤, (Gi)i∈I, ϕji (i < j)) (a) A partition I = If ree ⊔ Ireg (we admit one of the two sets If ree or Ireg to be empty). (b) A family {Gi}i∈I of abelian groups. We adopt the following notation: (c) A family of semigroup homomorphisms ϕji : Mi → Gj for all i < j, to which we (1) For i ∈ Ireg, set Mi = Gi, and bGi = Gi = Mi. (2) For i ∈ If ree, set Mi = N × Gi, and bGi = Z × Gi Observe that, in any case, bGi is the Grothendieck group of Mi. associate, for all i < j, the unique extension bϕji : bGi → Gj of ϕji to a group homo- from bGi to bGj). We require that the family {ϕji} satisfies the following conditions: morphism from the Grothendieck group of Mi to Gj (we look at these maps as maps REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 5 defines a functor from the category I to the category of abelian groups (where (2) For each i ∈ If ree we have that the map (1) The assignment we set bϕii = id bGi is surjective. (cid:26) i (i < j) for all i ∈ I). Mk<i ϕik : Mk<i 7→ bϕji (cid:27) 7→ bGi Mk → Gi We say that an I-system J = (I, ≤, (Gi)i∈I, ϕji (i < j)) is finitely generated in case I is a finite poset and all the groups Gi are finitely generated. Example 2.2. We present a family of I-systems, where I is a fixed poset. Let I = {p, q1, q2, . . . , qr}, i=1 ϕp,qi : Lr Lr i=1 where p > qi for all i, and all qi are pairwise incomparable. We set I = Ifree. Since qi are minimal free primes we must have Gqi = {eqi}. To complete the definition of the I-system J we only need an abelian group Gp and semigroup homomorphisms ϕp,qi : N → Gp such that N → Gp is surjective. We distinguish two cases: (1) If r = 1, then Gp must be a finite cyclic group. (2) If r > 1, then we can take Gp of the form Zs ⊕ Zk1 ⊕ · · · ⊕ Zkr−s, where 0 ≤ s < r, 1 ≤ kj for 1 ≤ j ≤ r − s, and where ϕp,qi(1) = (0, . . . , 0, 1, 0, . . . , 0), with a 1 in the i-th position, for 1 ≤ i ≤ s, and ϕp,qi(1) = (−1, −1, . . . , −1, 0, . . . , 0, 1, 0, . . . , 0), where there are s −1's, and where 1 ∈ Zki−s, for s < i ≤ r. To every I-system J one can associate a primely generated conical refinement monoid M(J ), and conversely to any primely generated conical refinement monoid M, we can asso- ciate an I-system J such that M ∼= M(J ), see Sections 1 and 2 of [12] respectively. Given a poset I, and an I-system J , we construct a semilattice of groups based on the semilattice (under set-theoretic union) of all the finitely generated lower subsets of I. These are precisely the lower subsets a of I such that the set Max(a) of maximal elements of a is finite and each element of a is under some of the maximal ones. In case I is finite, and since the intersection of lower subsets of I is again a lower subset, A(I) is a lattice. For any a (a ⊆ b) to be the canonical embedding partial order of groups (I, ≤,bGi), by following the model introduced in [17]. Let A(I) be the a ∈ A(I), we define bHa =Li∈a bGi, and we define f b of bHa into bHb. Given a ∈ A(I), i ∈ a and u ∈ bGi, we define χ(a, i, u) ∈ bHa by Let Ua be the subgroup of bHa generated by the set Now, for any a ∈ A(I), set eGa = bHa/Ua, and let Φa : bHa → eGa be the natural onto {χ(a, i, u) − χ(a, j,bϕji(u)) : i < j ∈ Max(a), u ∈ bGi}. map. Then, for any a ⊆ b ∈ A(I) we have that f b χ(a, i, u)j =(cid:26) u if j = i, if j 6= i. 0j a(Ua) ⊆ Ub, so that there exists a unique 6 PERE ARA AND ENRIQUE PARDO Φa Φb f b a ef b a a(a ⊂ b)) is a semilattice of groups. Thus, the set homomorphism ef b a : eGa → eGb which makes the diagram bHb bHa eGa / eGb commutative. Hence, (A(I), (eGa)a∈A(I),ef b fM (J ) := Ga∈A(I)eGa, endowed with the operation x + y := ef a∪b (x) + ef a∪b eGa, y ∈ eGb, is a primely generated regular refinement monoid by [17, Proposition 1]. Note that bH∅ = eG∅ = {0}. We refer the reader to [17] for further details on this construction. Let Ha be the subsemigroup of bHa defined by Ha =(cid:26)(zi)i∈a ∈ bHa : zi ∈(cid:26) i ∈ afree \ Max(a)free (cid:27) . In what follows, whenever i < j ∈ I with j a free element, x = (n, g) ∈ N × Gj and y ∈ Mi, we will see x + ϕji(y) as the element (n, g + ϕji(y)) ∈ N × Gj. This is coherent with our (y) for any a, b ∈ A(I) and any x ∈ {(0, 0i)} ∪ (N × Gi) i ∈ Max(a)free a b N × Gi for for identification of Gj as the subgroup {0} × Gj of bGj = Z × Gj. By [12, Lemma 1.3], we can define a semilattice of semigroups (A(I), (Ha)a∈A(I), f b a(a ⊂ b)). Now, we construct a monoid associated to it. For this, consider the congruence ∼ defined on Ha, for a ∈ A(I), given by x ∼ y ⇐⇒ x − y ∈ Ua. Lemma 2.3 ([12, Lemma 1.4]). Let a ∈ A(I). The congruence ∼ on Ha agrees with the congruence ≡, generated by the pairs (x + χ(a, i, α), x + χ(a, j, ϕji(α))), for x ∈ Ha, i < j ∈ Max(a) and α ∈ Mi. Corollary 2.4 ([12, Corollary 1.5]). For every a ∈ A(I), Ma := Ha/∼ = Ha/≡ is a sub- Definition 2.5. Given an I-system J = (I, ≤, Gi, ϕji(i < j)), we denote by M(J ) the set monoid of eGa. Fa∈A(I) Ma. By [12, Lemma 1.3] and Corollary 2.4, M(J ) is a submonoid of fM (J ). Observe that Lemma 2.3 gives: Corollary 2.6 ([12, Corollary 1.6]). M(J ) is the monoid generated by Mi, i ∈ I, with respect to the defining relations x + y = x + ϕji(y), i < j, x ∈ Mj, y ∈ Mi. / / / REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 7 Notation. Assume J is an I-system. For i ∈ I and x ∈ Mi we will denote by χi(x) the element [χ(I ↓ i, i, x)] ∈ M(J ). Note that, by Corollary 2.6, M(J ) is the monoid generated by χi(x), i ∈ I, x ∈ Mi, with the defining relations χj(x) + χi(y) = χj(x + ϕji(y)), i < j, x ∈ Mj, y ∈ Mi. be written as a difference of two elements of Ma. For this, it is enough to show that for each Lemma 2.7. For each a ∈ A(I), the Grothendieck group of Ma is the group eGa. Proof. Since Ma is a submonoid of eGa, we only have to show that every element of eGa can i ∈ a and each x ∈ bGi, the element χ(a, i, x) of bHa can be written as a difference of two elements in Ha. If i ∈ areg, then we can write χ(a, i, x) =(cid:16)χ(a, i, x) + Xj∈afree χ(a, j, (1, ej))(cid:17) −(cid:16) Xj∈afree χ(a, j, (1, ej))(cid:17) ∈ Ha − Ha. If i ∈ afree, then select n ∈ N such that (n, ei) + x ∈ Mi, and write χ(a, i, x) = χ(a, j, (1, ej))(cid:17) −(cid:16)χ(a, i, (n, ei)) + Xj∈afree (cid:16)χ(a, i, (n, ei) + x) + Xj∈afree Lemma 2.8. Let i ∈ I and set a = I ↓ i. Then Ma = Mi and eGa = bGi. Proof. We have a surjective monoid homomorphism φ : Mi → Ma sending x ∈ Mi to χi(x). Define a monoid homomorphism ψ : Ha → Mi by χ(a, j, (1, ej))(cid:17) ∈ Ha − Ha. (cid:3) ψ(Xj≤i χ(xj, j, a)) =Xj≤i ϕij(xj), where xi ∈ Mi, xj ∈ Mj = Gj for j ∈ areg \ {i} and xj ∈ {(0, ej)} ∪ Mj for j ∈ afree \ {i}. (We are setting here ϕii = IdMi.) Then, ψ clearly factors through the congruence ≡ described in Lemma 2.3, and so induces a monoid homomorphism ψ : Ma → Mi, which is clearly the (cid:3) inverse map of φ. A similar proof gives that eGa = bGi. of lower subsets of a poset I. We will denote by L(M) the lattice of order-ideals of a monoid M and by L(I) the lattice Proposition 2.9 ([12, Proposition 1.9]). Let J be an I-system. Then there is a lattice isomorphism L(I) ∼= L(M(J )). More precisely, given a lower subset J of I, the restricted J-system is JJ := (J, ≤, (Gi)i∈J , ϕji, (i < j ∈ J)), and the map J 7→ M(JJ ) defines a lattice isomorphism from L(I) onto L(M(J )). Lemma 2.10. Let I be a poset and let J be an I-system. Let J be a finitely generated lower subset of I and let JJ be the restricted J-system. Then the Grothendieck group of the associated order-ideal M(JJ ) of M(J ) is precisely eGJ. 8 PERE ARA AND ENRIQUE PARDO Proof. We have M(JJ ) =Fa∈A(J) Ma. Let x be an element in MJ , and define a semigroup homomorphism τ : M(JJ ) −→ G(MJ ) by τ (z) = (x + f J a (z)) − x for z ∈ Ma. Then it is easily seen that τ is the canonical map from M(JJ ) to its Grothendieck group, that is, that for every semigroup homomorphism λ : M(JJ ) → G, where G is a group, there is a unique group homomorphism eλ : G(MJ ) → Indeed, given λ as above, eλ is just the canonical map from the G such that λ = eλ ◦ τ . Grothendieck group G(MJ ) of MJ to G induced by the restriction of λ to MJ . Now, we can apply Lemma 2.7 to derive the result. (cid:3) Recall that given a poset I, and an element i ∈ I, the lower cover of i in I is the set L(I, i) = {j ∈ I : j < i and [j, i] = {j, i}}. We now consider the special case of a finitely generated conical regular refinement monoid M. Our goal will be to realize M as the graph monoid M(E) for some row-finite directed graph E. In this case the results from [12] reduce to Dobbertin's results [17]. For the rest of this subsection, let M be a finitely generated conical regular refinement monoid. The antisymmetrization M of M is then an antisymmetric regular refinement monoid (cf. [16, Theorem 5.2]). The set P is defined by taking a representative p for each prime p ∈ P(M ). For each p ∈ P, the archimedean component Mp of p is a finitely generated abelian group, denoted by Gp. Since we have p = x for all x ∈ Gp, we may take p = ep, the neutral element of the group Gp, as a canonical representative of p, so that P = {e ∈ M : e = 2e and e is prime }. With the order induced from M, P is a finite poset. Note that e ≤ f in P if and only if f = e + f . For e ∈ P, the associated group is Ge = {x ∈ M : e ≤ x ≤ e}, which is precisely the archimedean component GM [e] of e. Finally if e ≤ f in P, then the induced map ϕf e : Ge → Gf is defined by ϕf,e(x) = x + f for x ∈ Ge. This structure defines the P-system JM associated to M. 2.3. Graph monoids. Now, we will recall the basic elements about graphs and their monoids that are necessary in the sequel. A (directed) graph E = (E0, E1, r, s) consists of two countable sets E0, E1 and maps r, s : E1 → E0. The elements of E0 are called vertices and the elements of E1 edges. A vertex v ∈ E0 is a sink if s−1(v) = ∅. A graph E is finite if E0 and E1 are finite If s−1(v) is a finite set for every v ∈ E0, then the graph is called row-finite. We sets. will only deal with row-finite graphs in this paper, so we make the convention that all graphs appearing henceforth are row-finite; we will make this assumption explicit in the statements of the main results. A path µ in a graph E is a sequence of edges µ = (µ1, . . . , µn) such that r(µi) = s(µi+1) for i = 1, . . . , n − 1. In such a case, s(µ) := s(µ1) is the source of µ and r(µ) := r(µn) is the range of µ. If s(µ) = r(µ) and s(µi) 6= s(µj) for every i 6= j, then µ is a called a cycle. We say that a cycle µ = (µ1, . . . , µn) has an exit if there is a vertex v = s(µi) and an edge f ∈ s−1(v) \ {µi}. If v = s(µ) = r(µ) and s(µi) 6= v for every i > 1, then µ is a called a closed simple path based at v. For a path µ we denote by µ0 the set of its vertices, i.e., {s(µ1), r(µi) i = 1, . . . , n}. For n ≥ 2 we define En to be the set of paths of length n, and E∗ =Sn≥0 En the set of all paths. REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 9 We define a relation ≥ on E0 by setting v ≥ w if there is a path µ ∈ E∗ with s(µ) = v and r(µ) = w. A subset H of E0 is called hereditary if v ≥ w and v ∈ H imply w ∈ H. A set H is saturated if every vertex which feeds into H and only into H is again in H, that is, if s−1(v) 6= ∅ and r(s−1(v)) ⊆ H imply v ∈ H. Denote by H (or by HE when it is necessary to emphasize the dependence on E) the set of hereditary saturated subsets of E0. The set T (v) = {w ∈ E0 v ≥ w} is the tree of v, and it is the smallest hereditary subset of The hereditary saturated closure of a set X is defined as the smallest hereditary and saturated subset of E0 containing X. It is shown in [11] that the hereditary saturated closure of a set E0 containing v. We extend this definition for an arbitrary set X ⊆ E0 by T (X) =Sx∈X T (x). X is X =S∞ Λ0(X) = T (X), and Λn(X) = {y ∈ E0 s−1(y) 6= ∅ and r(s−1(y)) ⊆ Λn−1(X)} ∪ Λn−1(X), for n ≥ 1. n=0 Λn(X), where We recall here some graph-theoretic constructions which will be of interest. For a hereditary subset of E0, the quotient graph E/H is defined as (E0 \ H, {e ∈ E1 r(e) 6∈ H}, r(E/H)1, s(E/H)1), and the restriction graph is EH = (H, {e ∈ E1 s(e) ∈ H}, r(EH)1, s(EH )1). Definition 2.11. Given a graph E: (1) We say that E is transitive if every two vertices of E0 are connected through a finite path. (2) We say that a nonempty subset S of E0 is strongly connected if the graph (S, s−1(S) ∩ r−1(S), sS, rS) is transitive. In particular, if F is a subgraph of E, we say that F is strongly connected if so does the subset F 0 of E0. For a row-finite graph E, the graph monoid associated to E, denoted by M(E), is the commutative monoid given by the generators {av v ∈ E0}, with the relations: ar(e) for every v ∈ E0 that emits edges. (2.1) av = X{e∈E1s(e)=v} Let F be the free commutative monoid on the set E0. The nonzero elements of F can be i=1 xi, where xi ∈ E0. Now we will give a description of the congruence on F generated by the relations (2.1) on F. It will be convenient to introduce the following notation. For x ∈ E0, write written in a unique form up to permutation asPn r(x) := X{e∈E1s(e)=x} r(e) ∈ F. With this new notation relations (2.1) become x = r(x) for every x ∈ E0 that emits edges. Definition 2.12 ([11, Section 4]). Define a binary relation →1 on F \ {0} as follows. Let i=1 xi be an element in F as above and let j ∈ {1, . . . , n} be an index such that xj emits on F\{0}, that is, α → β if and only if there is a finite string α = α0 →1 α1 →1 · · · →1 αt = β. i=1 xi →1Pi6=j xi + r(xj). Let → be the transitive and reflexive closure of →1 Pn edges. ThenPn 10 PERE ARA AND ENRIQUE PARDO Let ∼ be the congruence on F generated by the relation →1 (or, equivalently, by the relation →). Namely α ∼ α for all α ∈ F and, for α, β 6= 0, we have α ∼ β if and only if there is a finite string α = α0, α1, . . . , αn = β, such that, for each i = 0, . . . , n − 1, either αi →1 αi+1 or αi+1 →1 αi. The number n above will be called the length of the string. (cid:3) It is clear that ∼ is the congruence on F generated by relations (2.1), and so M(E) = F/∼. Lemma 2.13 ([11, Lemma 4.3]). Let α and β be nonzero elements in F. Then α ∼ β if and only if there is γ ∈ F such that α → γ and β → γ. Proposition 2.14 ([11, Proposition 4.4]). The monoid M(E) associated with any row-finite graph E is a refinement monoid. Let E be a graph. For any subset H of E0, we will denote by I(H) the order-ideal of M(E) generated by H. Recall that we denote by L(M) the lattice of order-ideals of a commutative monoid M. Order-ideals of M(E) correspond to hereditary saturated subsets of E0, as follows: Proposition 2.15 ([11, Proposition 5.2]). For any row-finite graph E, there is a natural lattice isomorphism from HE to L(M(E)) sending H ∈ HE to the order-ideal I(H) generated by H. Also, we have: Lemma 2.16 ([11, Proposition 5.2], [15, Lemma 2.1]). Let H be a subset of E0, with hered- itary saturated closure H. Then I(H) = I(H), and H = I(H) ∩ E0. This means that I(H) is generated as a monoid by the set H = I(H)∩E0. We can improve this result, as follows Lemma 2.17. Let E be a graph, and let H be a hereditary subset of E0. Then, the order-ideal I(H) is generated as a monoid by the set {av : v ∈ H}. Proof. Take v ∈ Λ1(H). Then r(s−1(v)) is a finite subset of H, and so av =Pe∈s−1(v) ar(e) a vertex in Λn(H) for all n ≥ 1. Since H =S∞ belongs to the submonoid generated by {aw : w ∈ H}. By induction, the same happens for n=0 Λn(H), and I(H) is generated as a monoid (cid:3) by H, we obtain the result. Thus, we can conclude: Lemma 2.18. Let E be a directed graph and let H be a hereditary subset of E0. Then the order-ideal I(H) generated by H is isomorphic to the graph monoid of the restriction graph EH . Proof. Since H is a hereditary subset of E0, the map φ : M(EH ) → I(H) av 7→ av is a well-defined monoid homomorphism. Moreover, φ is an onto map by Lemma 2.17. To show it is injective, let α and β be elements in the free commutative monoid generated by H, and assume that the elements of M(E) represented by α and β agree. By Lemma 2.13, there exists γ ∈ F such that α → γ and β → γ. By the definition of the relation → and the REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 11 fact that H is hereditary, it follows that the vertices appearing in γ belong to H, and that α → γ and β → γ in M(EH ). Hence, α and β represent the same element of M(EH ). (cid:3) Given an order-ideal S of a monoid M we define a congruence ∼S on M by setting a ∼S b if and only if there exist e, f ∈ S such that a + e = b + f . Let M/S be the factor monoid obtained from the congruence ∼S; see e.g. [10]. Lemma 2.19 ([11, Lemma 6.6]). Let E be a row-finite graph. For a saturated hereditary subset H of E0, consider the order-ideal I(H) associated with H. Then there is a natural monoid isomorphism M(E)/I(H) ∼= M(E/H). 2.4. K-Theory for rings. Here, we recall a few elements on K-Theory for Leavitt path algebras, that will be necessary in the sequel. For a ring R (with local units), let M∞(R) be the directed union of Mn(R) (n ∈ N), where the transition maps Mn(R) → Mn+1(R) are given by x 7→ ( x 0 0 0 ). We define V(R) to be the set of isomorphism classes (denoted [P ]) of finitely generated projective left R-modules, and we endow V(R) with the structure of a commutative monoid by imposing the operation [P ] + [Q] := [P ⊕ Q] for any isomorphism classes [P ] and [Q]. Equivalently [26, Chapter 1], V(R) can be viewed as the set of equivalence classes V(e) of idempotents e in M∞(R) with the operation V(e) + V(f ) := V(cid:0)(cid:0) e 0 0 f(cid:1)(cid:1) for idempotents e, f ∈ M∞(R). The group K0(R) of a ring R with local units is the universal group of V(R). Recall that, as any universal group of a commutative monoid, the group K0(R) has a standard structure of partially pre-ordered abelian group. The set of positive elements in K0(R) is the image of V(R) under the natural monoid homomorphism V(R) → K0(R). Let E = (E0, E1, r, s) be a graph, and let K be a field. We define the Leavitt path algebra LK(E) associated with E as the K-algebra generated by a set {v v ∈ E0} of pairwise orthogonal idempotents, together with a set of variables {e, e∗ e ∈ E1}, which satisfy the following relations: (1) s(e)e = er(e) = e for all e ∈ E1. (2) r(e)e∗ = e∗s(e) = e∗ for all e ∈ E1. (3) e∗e′ = δe,e′r(e) for all e, e′ ∈ E1. Note that the relations above imply that {ee∗ e ∈ E1} is a set of pairwise orthogonal idempotents in LK(E). In general the algebra LK(E) is not unital, but it has a set of local (4) v =P{e∈E1s(e)=v} ee∗ for every v ∈ E0 that emits edges. units given by {Pv∈F v}, where F ranges on all finite subsets of E0. So, the above facts apply. Let E be a graph. For any subset H of E0, with hereditary saturated closure H, we will denote by I(H) the ideal of LK(E) generated by H. Theorem 2.20 ([11, Theorem 3.5]). Let E be a row-finite graph. Then there is a natural monoid isomorphism V(LK(E)) ∼= M(E). Moreover, by [11, Theorem 3.5], [11, Theorem 5.2], [11, Lemma 6.6] and [15, Lemma 2.3(1)], we conclude that 12 PERE ARA AND ENRIQUE PARDO Lemma 2.21. Let E be a row-finite graph, and let H a hereditary subset of E0. Then, for any field K we have that V(LK(E))/V(I(H)) ∼= V(LK(E/H)) ∼= V(LK(E)/I(H)). 3. Representing finitely generated regular refinement monoids Dobbertin showed in [17] that every finitely generated conical regular refinement monoid can be represented as a partial order of finitely generated abelian groups (see also [12]). Thus, in order to represent a finitely generated conical regular refinement monoid as a graph monoid, it suffices to show that some basic diagrams of abelian groups and homomorphisms of groups can be represented as graph monoids. In this section, we will prove that this is possible without further restrictions. In order to make clearer the argument, we will divide this task in several steps. First, we will show that any finitely generated abelian group can be represented using a graph monoid. It is worth to remark here that there are easy constructions of finite graphs realizing the monoids of the form H ∪ {0}, where H is a finitely generated abelian group, see for instance [3, p. 27], where a construction of the second-named author is outlined. We use infinite graphs in Lemma 3.1 because this is definitely needed to realize a monoid coming from a map between two finitely generated abelian groups, and we need to prepare the ground for that result. The second-named author and Fred Wehrung have shown that in the case where G = H = Z2 and ϕ : H → G is the trivial map sending everything to 0, the corresponding regular refinement monoid M(ϕ) cannot be realized by using a finite graph. However we show in Proposition 3.3 that all the monoids of this sort can be realized using infinite graphs. Lemma 3.1. Let H be a finitely generated abelian group. Then, H is representable as the semigroup M(E)∗ associated to a (not necessarily finite) row-finite directed graph E. Proof. We start by fixing notation. Using the Structure Theorem for Finitely Generated Abelian Groups, we can assume that H = Zr ⊕ Zn1 ⊕ · · · ⊕ Zns where r, s ≥ 0, and ni ≥ 1 for all i. We define N := r + s. Now, we will define our graph. The graph E has vertices E0 1 = {v1, . . . , vN +1, vj i (r + 1 ≤ i ≤ N, j ≥ 1)}. Instead of fixing what are the edges of E extensively, we will express the relations that these edges define on the graph monoid M(E1), as follows: (1) vN +1 = 2vN +1 + vi + ni−rvi rPi=1 NPi=r+1 (2) For every r + 1 ≤ i ≤ N: i + v3 i (a) v1 (b) for every j ≥ 1, v2j (c) for every j ≥ 1, v2j+1 i = ni−rvi + v1 i = v2j−1 = v2j i i + v2j i i + v2j+1 i + v2j+3 i (3) For every 1 ≤ i ≤ r, vi = 2vi + vj + nj−rvj + vN +1 rPj=1,j6=i NPj=r+1 REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 13 rPj=1 NPj=r+,1j6=i (4) For every r + 1 ≤ i ≤ N, vi = vj + nj−rvj + (ni−r + 1)vi + vN +1 + v1 i Let us carefully explain which are the properties enjoyed by the graph E: (a) First, we will show a graphical representation of the part of E described in point (2) above, for any vertex vi with r + 1 ≤ i ≤ N. Consider the subgraph Ei: •vi •v1 i •v2 i •v3 i •v4 i •v5 i · · · (ni−r) Notice that Ei is transitive, and moreover every vertex in E0 i is basis point of at least two different simple closed paths. So, the monoid M(Ei) associated to Ei is simple and regular (because in M(Ei) we have 2vi ≤ vi and 2vj i for every j ≥ 1). i ≤ vj (b) By relation (1) above, there are paths connecting vN +1 with every vertex vi (1 ≤ i ≤ i (r + 1 ≤ i ≤ N, j ≥ 1) by relations (2) and (4) N), and thus also with every vertex vj above. Moreover, in M(E) we have 2vN +1 ≤ vN +1. (c) By relation (3) above, for every 1 ≤ k ≤ r, there are paths connecting vk with every i (r + 1 ≤ i ≤ N, j ≥ 1) vertex vi (1 ≤ i ≤ N + 1), and thus also with every vertex vj by relations (2) and (4) above. Moreover, in M(E) we have 2vk ≤ vk. (d) By relation (4) above, for every r+1 ≤ k ≤ N, there are paths connecting vk with every i (r + 1 ≤ i ≤ N, j ≥ 1) vertex vi (1 ≤ i ≤ N + 1), and thus also with every vertex vj by relations (2) and (4) above. Moreover, in M(E) we have 2vk ≤ vk. As a consequence: (e) The graph E is transitive, and thus M(E) is a simple monoid. (f) For every v ∈ E0 1 , 2v ≤ v in M(E), whence M(E) is regular. Thus, since M(E) is simple, we conclude that M(E)∗ is an abelian group. Now, we will identify this group up to isomorphism. By relation (1) above, the neutral element of M(E)∗ is e = vN +1 + vi + rXi=1 NXi=r+1 ni−rvi. (3.1) In particular, relations (1) and (3) above simply say that vi = vi + e for i ∈ {1, . . . , r, N + 1}. Now, by using relation (2) above, we have: i by relation (2a). for every j ≥ 1 by (2b). i • e = ni−rvi + v3 • e = v2j−1 • e = v2j i + v2j+3 previous identity. As a consequence, e = vj says that i for every j ≥ 1 by (2c), and then e = v2j i for every j ≥ 1 by the i for every r + 1 ≤ i ≤ N and every j ≥ 1, and thus relation (2a) Hence, hvr+1, . . . , vN i generates a copy of Zn1 ⊕ · · · ⊕ Zns into M(E)∗. e = ni−rvi for every r + 1 ≤ i ≤ N. (3.2) ) ) W W i i ' ' W W o o W W o o ' ' W W o o W W o o 14 PERE ARA AND ENRIQUE PARDO Replacing equation (3.2) in the corresponding places of equation (3.1), we obtain e = vN +1 + vi. (3.3) rXi=1 Hence, hv1, . . . vr, vN +1i generates a copy of Zr into M(E)∗. Summarizing, the semigroup of nonzero elements of M(E) is isomorphic to H, as desired. (cid:3) Remark 3.2. rPi=1 (a) In Lemma 3.1, we are representing the group H as a group generated by vertices. In order to ease operating with this representation, we need to identify the inverses of the vertices, seen as elements of the group. If we follow the notation of Lemma 3.1, hvr+1, . . . , vN i generates a copy of Zn1 ⊕ · · · ⊕ Zns into M(E)∗, and (ni − 1)vi will be the symmetric of vi in M(E)∗ for r + 1 ≤ i ≤ N. Also, hv1, . . . vr, vN +1i generates a copy of Zr into M(E)∗. In this copy, {v1, . . . , vr} are the free generators of the group, while vN +1 help us to express r-tuples of Zr with negative entries. To be precise, since e = vN +1 + vi, for any 1 ≤ i ≤ r the element v1 + · · · + vi−1 + vi+1 + · · · + vr + vN +1 will be the symmetric of vi in M(E)∗ for 1 ≤ i ≤ r. In order to simplify the notation in the sequel, when we work with a graph monoid M(E), we will denote by v− the symmetric of the vertex v ∈ E0 seen as an element in the archimedean component of M(E) containing v. (b) Note that in the isomorphism M(E)∗ ∼= H = Zr ⊕ Zn1 ⊕ · · · ⊕ Zns, the vertices v1, . . . , vN correspond to the canonical generators of H as an abelian group. We will refer to this fact saying that v1, . . . , vN are the canonical generators of H. (c) We will work later in Section 5 with semigroup generators of a group. Observe that {v1, . . . , vN , vN +1} is indeed a family of semigroup generators of H. Let G and H be finitely generated abelian groups, and let ϕ : H → G be a group homo- morphism. According to [17], there exists a regular refinement monoid M(ϕ) associated to ϕ. The monoid M(ϕ) has exactly two prime idempotents e and f , with e + f = f , GM [e] = H, GM [f ] = G, and the map ϕf,e : GM [e] → GM [f ] given by ϕf,e(x) = x + f is exactly the map ϕ. We will show that M(ϕ) is representable as a graph monoid. Proposition 3.3. Let ϕ : H → G be a homomorphism between finitely generated abelian groups H and G, and let M(ϕ) be the associated regular refinement monoid. Then M(ϕ) is representable as the monoid M(E) of a (not necessarily finite) row-finite directed graph E. Proof. We start by fixing notation. Using the Structure Theorem for Finitely Generated Abelian Groups, we can assume that and H = Zr ⊕ Zn1 ⊕ · · · ⊕ Zns G = Zt ⊕ Zm1 ⊕ · · · ⊕ Zml, REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 15 where r, s, t, l ≥ 0, and ni, mj ≥ 1 for all i, j. We define N := r + s and M := t + l. Moreover, if M < N, we can add ml+1 = ml+2 = · · · = ml+(N −M ) = 1, so that we are thinking G = Zt ⊕ Zm1 ⊕ · · · ⊕ Zml ⊕ Z1 ⊕ · · · ⊕ Z1. Thus, without loss of generality, we can assume that N ≤ M. Under this representation of H and G, the homomorphism ϕ is represented by a block matrix A =(cid:18) A1,1 A2,1 A2,2 (cid:19) 0 with entries labeled ai,j ∈ Z, and without loss of generality we can assume that the entries in blocks A2,l (l = 1, 2) satisfy 0 < ai,j ≤ mi−t for every j. We will define our graph E in two layers, the first one representing the group H, while the second will represent simultaneously both the group G and the homomorphism ϕ. To construct the first layer, we use Lemma 3.1 to build a graph E1 such that M(E1)∗ ∼= H, with E0 1 = {v1, . . . , vN +1, vj i (r + 1 ≤ i ≤ N, j ≥ 1)}, such that v1, . . . , vN are the canonical generators of M(E1)∗ = H (see Remark 3.2(b)). Now, we will construct the layer corresponding to G, and simultaneously the group homo- morphism ϕ : H → G. To this end, we will define a new collection of vertices E0 2, jointly with a family of relations stated in similar terms as those used in Lemma 3.1. Nevertheless, these new relations will involve not only the vertices of E0 1 . The new set of vertices E0 2 , but also those of E0 2 is E0 2 = {w1, . . . , wM +1, wj i (t + 1 ≤ i ≤ M, j ≥ 1)}. As in Lemma 3.1, instead of fixing what are the edges emitted by the vertices of E0 2 extensively, we will express the relations that these edges define on the graph monoid. Let E be the graph defined by taking E0 = E0 2 and where all the relations enjoyed by the vertices are as follows: 1 ⊔ E0 • The relations given by the set of edges of the graph E1. • The relations we list below: (1) wM +1 = 2wM +1 + wi + mi−twi tPi=1 MPi=t+1 (2) For every t + 1 ≤ i ≤ M: i = mi−twi + w1 (a) w1 (b) for every j ≥ 1, w2j (c) for every j ≥ 1, w2j+1 i + w3 i i = w2j−1 = w2j i i (3) For every 1 ≤ i ≤ t, + w2j i i + w2j+1 i + w2j+3 i wi = (ai,i + 2)wi + (aj,i + 1)wj + (aj,i + mj−t)wj + wM +1 + v− i . Notice that whenever aj,i < 0 for some 1 ≤ j ≤ t, we will replace in the above relation aj,iwj by (−aj,i)wk + (−aj,iwM +1). MXj=t+1 tXj=1,j6=i tPk=1,k6=j 16 PERE ARA AND ENRIQUE PARDO (4) For every t + 1 ≤ i ≤ M, wi = tXj=1 (aj,i + 1)wj + MXj=t+1,j6=i (aj,i + mj−t)wj + (ai,i + mi−t + 1)wi + wM +1 + w1 i + v− i . As in the previous relation, whenever aj,i < 0 for some 1 ≤ j ≤ t, we will replace in the above relation aj,iwj by (−aj,i)wk + (−aj,iwM +1). tPk=1,k6=j If N < M, and N < i ≤ M, then the term v− i appearing in (4) (or in (3) if N < i ≤ t) should be interpreted as the neutral element 0 of the graph monoid. So, for these values of i, the relations are of the form (4) (or (3)) in Lemma 3.1. Now, consider the graph E, and let us identify the properties enjoyed by the elements of M(E): • Observe that E0 1 is a hereditary and saturated subset of E0, so that M(E1) is an order- ideal of M(E), by Lemma 2.18. Moreover M(E1)∗ is the archimedean component of e in M(E) and, by construction M(E1)∗ ∼= H. • By similar arguments to those used in Lemma 3.1, every vertex in E0 2 is a regular element. exist paths from wi to wj, and also from wi to wk • By similar arguments to those used in Lemma 3.1, for any 1 ≤ i, j ≤ M + 1 there j to wi for every k ≥ 1. 2 is strongly connected in E, and thus the ver- 2 generate an archimedean component GM (E)[f ] of M(E) with neutral element f . Hence, M(E1)∗ equals GM (E)[e]. Also, E0 j and from wk tices in E0 Moreover, M(E) is a regular refinement monoid. By using the relations defined above, and arguing as in Lemma 3.1, we have that: • By relation (1) above, f = tXi=1 wi + wM +1 + MXi=t+1 mi−twi. • By relation (2) above, f = wj i for every t + 1 ≤ i ≤ M and every j ≥ 1, while f = mi−twi for every t + 1 ≤ i ≤ M. • By replacing equation (3.5) in equation (3.4), we obtain f = w1 + · · · + wt + wM +1 (3.4) (3.5) (3.6) Hence, the monoid hw1, . . . wM , wM +1i generated by w1, . . . , wN +1 is indeed a group, which is a subgroup of GM (E)[f ]. Now, we will state the relation between e and f in M(E). To this end, notice that: 1 ∈ M(E1)∗, we have that e ≤ v− • Since v− 1 . • By relation (3) above (or (4) if t = 0), v− 1 ≤ w1, so that e ≤ w1. • By equation (3.6) (or (3.5) if t = 0), w1 ≤ f . wi + wj + wM +1 + MXj=t+1 tXj=1,j6=i aj,iwj +" tXj=1 MXj=1 tPj=1 i , we have that v− wj + wM +1 + MPj=t+1 mj−twj# + (v− i + vi). MXj=t+1 REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 17 Hence, e ≤ f in M(E). So there are only two archimedean components in M(E), correspond- ing to the idempotents e and f , and e ≤ f in M(E). We want to describe the homomorphism φf e : GM (E)[e] → GM (E)[f ] x 7→ x + f . Observing that v1, . . . , vN are canonical generators of GM (E)[e] ∼= H, we are going to compute φf e (vi). We will compute φf e (vi) for every 1 ≤ i ≤ N, by using relations (3) and (4) above. To be precise, recall that r ≤ r + s = N ≤ M = t + l. So, it can occur either N ≤ t or N > t. In the first case, to determine φf e (vi) we will only need relation (3) above. In the second case, we will use relation (3) above to determine φf e (vi) for 1 ≤ i ≤ t, and relation (4) above to determine φf e (vi) for t + 1 ≤ i ≤ N. Let us suppose that we are in the second case. Take i ∈ {1, . . . , t}, and add vi on both sides of relation (3), as follows: wi + vi = (ai,i + 2)wi + (aj,i + 1)wj + (aj,i + mj−t)wj + wM +1 + v− i + vi = By definition of v− i and w− i + vi = e and w− i + wi = f . Also, e + f = f , and mj−twj. Thus, by adding w− i on both sides of by equation (3.4), f = the previous identity, we have f + vi = f + MXj=1 aj,iwj + f + e = aj,iwj + f = MXj=1 aj,iwj MXj=1 because f is the neutral element of the group GM (E)[f ]. On the other hand, if i ∈ {t + 1, . . . , N}, a similar argument shows that adding vi + w− i on both sides of relation (4) gives us Hence, using (3.5), (3.6) and the above relations, we obtain a monoid homomorphism vi + f = aj,iwj. MXj=1 γ : M(ϕ) → M(E) extending the canonical isomorphism H = GM (ϕ)[e] → GM (E)[e] and sending the canonical generators of G = GM (ϕ)[f ] to w1, . . . , wM . We are going to define an inverse δ : M(E) → M(ϕ) of the map γ. For this, it is enough to define the map on the vertices of E and to show that the defining relations of M(E) are preserved by this assignment. The images of the vertices in E1 are dictated by the inverse map of the isomorphism from H onto GM (E)[e]. Let x1, . . . , xM be the canonical generators of G. We define δ(wi) = xi for 1 ≤ i ≤ M, and δ(wM +1) = −(x1 + · · · + xt). 18 PERE ARA AND ENRIQUE PARDO Finally define δ(wj It is easy to show that all relations (1)-(4) are preserved by δ, so that this assignment gives a well-defined monoid homomorphism δ. It is now clear that δ is the inverse of γ. i ) = f , where f is the neutral element of G. Therefore we obtain that γ is an isomorphism from M(ϕ) onto M(E). Observe that γ sends the canonical set of generators of G = GM (ϕ)[f ] onto w1, . . . , wM , so that the vertices w1, . . . , wM are canonical generators of GM (E)[f ] and the canonical map φf e has associated matrix A with respect to the canonical generators v1, . . . , vN of GM (E)[e] and w1, . . . , wM of GM (E)[f ]. (cid:3) Remark 3.4. In the above proof, we could have assumed that N ≤ l and use only relations (4) to encode the map ϕ in the graph monoid M(E). The proof is then a little bit shorter, since we would not need to distinguish different cases when we deal with the computation of the elements φf e (vi). This is the approach that we will follow in the proof of the general case (see Proposition 5.13). Let us illustrate our results with two concrete applications of Proposition 3.3: (1) For any n ∈ N, we will compute the graph En such that M(En) represents the homo- morphism n· : Z → Z. We will follow the same notation as in the proof, so that r = t = 1, s = l = 0, N = M = 1, the matrix A = (n), and for each n ∈ N the associated graph En is: (2) •w2 •w1 (n + 2) (2) •v1 ) •v2 (2) (2) We will compute the graph F such that M(F ) represents the group homomorphism 0· : Z2 → Z2. As in the previous example, we will follow the same notation as in the proof, so that r = t = 0, s = l = 1, N = M = 1, the matrix A = (2), and the associated graph F is: (2) •w2 (5) •w1 •w1 1 •w2 1 •w3 1 •w4 1 •w5 1 •v2 1 •v3 1 •v4 1 •v5 1 (3) •v1 (2) (2) ) •v1 1 •v2 (2) · · · · · · The next step is to show a result analogous to Proposition 3.3, which allows us to represent confluent maps of groups. This is precisely what we need to use in the inductive step of the 2 2 * * j j l l 3 3 ) i i k k Z Z 2 2 * * W W j j ' ' W W o o W W o o ' ' W W o o W W o o 3 3 ) W W i i ' ' W W o o W W o o ' ' W W o o W W o o Z Z T T W W REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 19 proof of the general result. The proof is very similar to the one of Proposition 3.3, so we will only give a brief sketch of it. Let H1, . . . , Hn, G be finitely generated abelian groups, and let ϕi : Hi → G be group homomorphisms (1 ≤ i ≤ n). Let M(ϕ) be the regular refinement monoid associated to ϕ := (ϕ1, . . . , ϕn), that is, M(ϕ) has exactly n + 1 prime idempotents e1, . . . , en and f , with ei + f = f , and GM [ei] = Hi, GM [f ] = G, and the map ϕf ei : GM [ei] → GM [f ] given by ϕf ei(x) = x + f is exactly the map ϕi for 1 ≤ i ≤ n. Proposition 3.5. Let H1, . . . , Hn, G be finitely generated abelian groups, and let ϕi : Hi → G be group homomorphisms (1 ≤ i ≤ n). Then, these maps can be represented simultaneously as the monoid M(E) associated to a (not necessarily finite) row-finite directed graph E. Proof. We start by fixing notation. Using the Structure Theorem for Finitely Generated Abelian Groups, we can assume that for i = 1, . . . , n, and Hi = Zti ⊕ Z mi 1 ⊕ · · · ⊕ Z mi li , G = Zr ⊕ Zp1 ⊕ · · · ⊕ Zps, where ti, li, r, s ≥ 0, and mj i , pi ≥ 1 for all i, j. We define Ni := ti +li (i = 1, 2) and M := r+s; moreover, as in Proposition 3.3, we can assumePn Now, we will define our graph E essentially as in Proposition 3.3, defining a first layer corresponding to the groups H1, . . . , Hn, and a second layer corresponding to the group G and to the different maps ϕi, which are represented by suitable matrices as in Proposition 3.3. i=1 Ni ≤ M. To deal with the first layer we use Lemma 3.1 to build n mutually disconnected strongly con- nected graphs Ei such that M(Ei)∗ ∼= Hi for i = 1, . . . , n, with corresponding sets of vertices such that the first Ni vertices of graph Ei are a canonical set of generators of GM (Ei)[ei] = Hi. The final step consists of constructing the layer corresponding to the group G, together with the homomorphisms ϕi : Hi → G. We will do that by adding a new collection of vertices T 0, jointly with a family of relations, similar to the ones used in Proposition 3.3. These relations will involve not only vertices in T 0, but also in E0 i . The idea is to apply again the procedure described in the proof of Proposition 3.3. Observe that, just as in Proposition 3.3 we obtain an isomorphism M(ϕ) → M(E) sending the canonical generating sets of each Hi and of G to the canonical generating sets of vertices spanning the corresponding archimedean components GM (E)[ei] and GM (E)[f ] respectively. (cid:3) We can now obtain the main result of this section. Theorem 3.6. If M is a finitely generated conical regular refinement monoid, then there exists a countable row-finite directed graph E such that M ∼= M(E). Proof. By Dobbertin's result [17], M is of the form M(JM ), for the I-system JM of abelian groups {GM [e] e ∈ I}, where e ranges on the poset I of prime idempotents of M, as explained in Section 2. Since M is finitely generated, all the groups GM [e] are finitely generated. We start by fixing a suitable form for these groups, namely we set GM [e] = Zre ⊕ Zne 1 ⊕ · · · ⊕ Zne se , 20 PERE ARA AND ENRIQUE PARDO where re, se ≥ 0, and ne i ≥ 1 for all i. Set Ne = re + se. We can assume that for each f ∈ I, i=1 Nei ≤ Nf , where L(I, f ) = {e1, . . . , en} is the lower cover of f in I. we havePn We can now proceed to show the result by order-induction. So assume that we have a lower e , subset J of I, and that we have built a countable row-finite graph EJ such that E0 where each Ee is a strongly connected graph, with J =Fe∈J E0 E0 e = {ve 1, . . . , ve Ne+1, (ve)j i (re + 1 ≤ i ≤ Ne, j ≥ 1)}. and an isomorphism γJ : M(J) → M(EJ ), where M(J) is the order-ideal of M generated by J, such that γJ sends the canonical gener- ators of GM [e] to ve Ne for all e ∈ J. In case J 6= I, let f be a minimal element of I \ J and write J ′ = J ∪ {f }. We will show that the above statement holds for the lower subset J ′ in place of J. This clearly establishes the result, because I is a finite poset. 1, . . . , ve Let Ef be the graph associated to GM [f ], as in Lemma 3.1. There are two cases to consider: (1) f is a minimal element of I. Set EJ ′ := Ef ⊔ EJ . Then M(J ′) = M({f }) ⊕ M(J) ∼= M(Ef ) ⊕ M(EJ ) = M(EJ ′) in a canonical way, showing the result. (2) f is not a minimal element of I. Let L(I, f ) = {e1, . . . , en} be the lower cover of f in I. Write ϕi = ϕf,ei for i = 1, . . . , n, and consider the graph E associated to the maps ϕi, i = 1, . . . , n, as in Proposition 3.5. The first layer of this graph consists exactly of the disjoint union of the graphs Eei, for i = 1, . . . , n, so it is a subgraph of our graph EJ . Let EJ ′ be the graph with E0 E (Ef ). The graph EJ ′ is a countable row-finite graph with E0 e is a strongly connected subset of E0, e have the desired form for all e ∈ J ′. We have to prove the existence and the sets of edges E1 of the isomorphism γJ ′. J ′ =Fe∈J ′ E0 e , where each E0 f and with E1 J ′ = E1 J ′ = E0 J ⊔ s−1 J ⊔ E0 Since E0 J is a hereditary and saturated subset of E0 J ′ we get from Lemma 2.18 that the J is precisely M(EJ ). Similarly, the order-ideal of M(J ′) order-ideal of M(EJ ′) generated by E0 generated by J is precisely M(J), and M(J ′) coincides with M(JJ ′), which is the monoid associated to the partial order of groups J := JM restricted to J ′ (see e.g. [12, Proposition 1.9]). Define a map γJ ′ : M(J ′) → M(EJ ′) as follows. The map γJ ′ agrees with the isomorphism γJ when restricted to the order-ideal M(J) of M(J ′). The map γJ ′ restricted to GM (J ′)[f ] = GM [f ] is just the map obtained by sending the canonical generators of GM [f ] to the canonical generators vf of GM (EJ ′ )[f ]. In order to prove that γJ ′ gives a well-defined monoid homomorphism, it suf- fices by [12, Corollary 1.6] to show that if e < f and x ∈ GM [e] then γJ ′(x) + γJ ′(f ) = γJ ′(ϕf e(x)) + γJ ′(f ), that is, γJ ′(x) + f = γJ ′(ϕf e(x)). Since e < f and I is finite, there is some i such that e ≤ ei. Using the properties of the map defined in Proposition 3.5 and the 1 , . . . , vf Nf REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 21 induction hypothesis, we get γJ ′(ϕf e(x)) = γJ ′(ϕf,ei(ϕei,e(x))) = φf ei(γJ ′(ϕei,e(x))) = φf = φf ei(φei ei(γJ(ϕei,e(x))) = φf e (γJ ′(x)) = f + γJ ′(x). e (γJ(x))) This shows that the defining relations of M(J ′) are preserved and so the map γJ ′ is a well- defined homomorphism from M(J ′) to M(EJ ′). To build the inverse δJ ′ of γJ ′ we follow the idea in the proof of Proposition 3.3. The image by δJ ′ of the vertices in EJ is determined by the inverse of γJ . Let x1, . . . , xNf be the canonical generators of GM [f ]. We define δJ ′(vf i ) = xi for 1 ≤ i ≤ Nf , and δJ ′(vf Nf +1) = −(x1 + · · · + xrf ). Finally define δJ ′((vf )j i ) = f . It is easily checked that δJ ′ preserves the defining relations of M(EJ ′), and so it gives a well-defined homomorphism from M(EJ ′) to M(J ′), which is clearly the inverse of γJ ′. This concludes the proof of the result. (cid:3) 4. A simple example We are going to represent one of the most simple examples of primitive monoids, using a method similar to the described above. This is the monoid M = Z+ ∪ {∞} which appears in [13]. Note that this example can be described as M = hp, a a = 2a, a = a + pi. Our method is completely different from the method used in [13], and provides a simpler graph than the one given in [13, Example 6.5]. Lemma 4.1. The monoid M = hp, a a = 2a, a = a + pi can be represented by a graph monoid. Proof. The lower component is freely generated by p, so we introduce a vertex v0 with a single loop e0 around it. Now the second component is regular with trivial associated group, so we use the above method setting G = Z1, and we consider vertices E0 = {v0, v1, vj 1(j ≥ 1)}. Instead of fixing what are the edges of E extensively, we will express the relations that these edges define on the graph monoid M(E), as follows: (1) v0 = v0 (2) (a) v1 1 = v1 + v1 1 + v3 1 (b) for every j ≥ 1, v2j (c) for every j ≥ 1, v2j+1 1 = v2j−1 = v2j 1 1 + v2j 1 1 + v2j+1 1 + v2j+3 1 (3) v1 = 2v1 + v1 1 + v0. So, the graph E turns out to be 22 PERE ARA AND ENRIQUE PARDO (2) •v1 •v1 1 •v2 1 •v3 1 •v4 1 •v5 1 · · · •v0 Then one can show as in the previous results that v1 = vj 1 = f for all j ≥ 1, so we obtain from (3) that f = f + v0, so the graph monoid M(E) gives the desired monoid M. Note that M is not representable by a graph monoid of a finite graph by [13, Theorem 6.1]. (cid:3) 5. Representing finitely generated refinement monoids In this section, we will obtain our main result (Theorem 5.6), which gives a characteri- zation of the finitely generated conical refinement monoids which are graph monoids. For this, it is fundamental to use the characterization of these monoids obtained in [12], which generalizes the results of Dobbertin [17] for the regular case, and the results of Pierce [25] for the antisymmetric case. First, we will recall some facts from [12], that will help to understand the statement of the main result and its proof. Let J = (I, ≤, (Gi)i∈I, ϕji (i < j)) be an I-system as in Definition 2.1. For every i ∈ I, Gi is an abelian group. For each i ∈ I, we define a commutative semigroup Mi and an abelian group bGi as follows: Mi = Gi = bGi if i is regular, while Mi = N × Gi and bGi = Z × Gi if i is free. Then, the monoid M(J ) associated to the I-system J is the monoid generated by {Mi}i∈I, with respect to the defining relations x + y = x + ϕji(y), i < j, x ∈ Mj, y ∈ Mi. Conversely, given a primely generated refinement monoid M, we can associate an I-system JM to M ([12, Section 2]). To do this, let M be its antisymmetrization, let P(M ) be its set of prime elements, and let I ⊂ P(M) be a set of representatives of P(M ) in P(M) such that p = 2p for any p ∈ Preg. Then, I is a poset with the ordering defined by q < p if and only if q < p in M . The poset I will be called the poset of primes of M. For each p ∈ I, let Mp be the archimedean component of p in M. We have: (1) If p is regular, then Mp is an abelian group, denoted by Gp ([16, Lemma 2.7]). (2) If p is free, then G′ p = {p + α : α ∈ M and p + α ≤ p} is an abelian group (with respect to the operation ◦ given by (p + α) ◦ (p + β) = p + (α + β)), isomorphic to the subgroup Gp := {(p + α) − p : p + α ∈ G′ p} of G(Mp) ([12, Remark 2.5]). Moreover, ϕ : N × Gp (n, (p + α) − p) → Mp 7→ np + α is a monoid isomorphism ([12, Lemma 2.4]). Thus, given p, q ∈ I such that q < p, the map ϕpq : Mq → Gp x 7→ (p + x) − p 3 3 * * W W i i ' ' W W o o W W o o ' ' W W o o W W o o W W REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 23 is a well-defined semigroup homomorphism, and JM := (I, (Gp)p∈I, ϕpq(q < p)) is the I-system associated to M. Moreover, M(JM ) is naturally isomorphic to M ([12, Theorem 2.7]). Assume that M is a finitely generated conical refinement monoid, with poset of primes I. Let p be a free prime and let L(I, p) = {q1, . . . , qn} be the lower cover of p. Then the archimedean component Mp of p has the form Mp = N × Gp for a finitely generated abelian group Gp. We will assume that q1, . . . , qr are free primes and that qr+1, . . . , qn are regular primes. We will use here the terminology and notation established in Subsection 2.2. Let Jp be the lower subset of I generated by q1, . . . , qn, and let MJp be the associated semigroup (cf. Corollary 2.4). By Lemma 2.10, the Grothendieck group of the order-ideal M(JJp) associated to Jp is precisely eGJp = G(MJp). ϕp : MJp → Gp Lemma 5.1. With the above notation, there exists a surjective semigroup homomorphism induced by the maps ϕp,q for q < p. Proof. Let us suppose that q1, . . . , qr are free primes and qr+1, . . . , qn are regular primes. Write subsemigroup of S. (See Subsection 2.2 for the definitions of these objects.) We have a group S := ⊕q<pMq, and identify S with a subsemigroup of bHJp = ⊕q<pbGq. By a slight abuse of notation, we will write qi for χ(Jp, qi, (1, eqi)) ∈ bHJp. Note that HJp is in general a proper homomorphism ψp := ⊕q<pbϕp,q : bHJp → Gp such that ψp(S) = Gp. In order to show that (ψp)HJp is surjective, it is enough to prove that ep ∈ ψp(HJp). Now, for each i = 1, . . . , r, we have −ϕp,qi(qi) = ψp(si) for some si ∈ S. Therefore ep = rep = ψp(cid:16) rXi=1 (qi + si)(cid:17), and Pr morphism i=1(qi + si) ∈ HJp, showing the surjectivity. Now (ψp)HJp clearly factors through MJp = HJp/∼, giving rise to a surjective semigroup homomorphism ϕp : MJp → Gp, as desired. (cid:3) Now, by the universal property of the Grothendieck group, we get a unique group homo- G(ϕp) : eGJp → Gp such that G(ϕp) ◦ ιJp = ϕp, where ιJp : MJp → G(MJp) is the natural map. Definition 5.2. We define the set G(MJp)++ of strictly positive elements of G(MJp) as the image in G(MJp) of the natural map ιJp : MJp → G(MJp). Note that G(MJp)++ is just a subsemigroup of G(MJp), which does not contain the neutral element of G(MJp) in general. Lemma 5.3. Every element in eG++ formPr i=1 χqi(ni, gi)+Pn Jp = G(MJp)++ can be represented by an element of the i=r+1 χqi(gi) for some ni ∈ N, i = 1, . . . , r and gi ∈ Gqi, i = 1, . . . , n. 24 PERE ARA AND ENRIQUE PARDO Proof. This follows immediately from the description of MJp given in Section 2, using that {q1, . . . , qn} = Max(Jp). (cid:3) Example 5.4. Note that the positive cone G(MJp)++ does not coincide in general with the positive cone obtained by considering the image of M(JJp) in G(M(JJp)) = G(MJp). For instance consider the graph monoid M = ha, b b = b + 2ai, and J the poset of primes {a, b} of M. Then ι(a) is in the image of the canonical map ι : M → G(M), but ι(a) /∈ G(MJb)++, since G(MJb) = Z × Z2 and G(MJb)++ = N × Z2. The following definition is handy to express the conditions charactering finitely generated graph monoids. Definition 5.5. Let G1 be an abelian group with a distinguished subsemigroup G++ of strictly positive elements, and let G2 be an abelian group. We say that a group homomorphism f : G1 → G2 is an almost isomorphism in case f is surjective and the kernel of f is a cyclic subgroup of G1 generated by an element in G++ . 1 1 We can now state the main result of the paper. Theorem 5.6. Let M be a finitely generated conical refinement monoid. Then M is a graph monoid if and only if for each free prime p of M, the map is an almost isomorphism. G(ϕp) : eGJp → Gp In particular, we can apply Theorem 5.6 to obtain, using the results in [5], the following partial affirmative answer to the realization problem for von Neumann regular rings. Corollary 5.7. Let M be a finitely generated conical refinement monoid such that, for all free primes p of M, the map G(ϕp) : eGJp → Gp is an almost isomorphism. Then, there exists a (countable) row-finite graph E such that, for any field K, the von Neumann regular K-algebra QK(E) of the quiver E satisfies V(QK(E)) ∼= M. Theorem 5.6 enables us to build examples 'a la carte' of monoids which are or aren't graph monoids (see also Corollary 5.14). The easiest example of a non-graph monoid is still the following: Example 5.8. ([13, Theorem 4.2]) Consider the monoid M := hp, a, b : p = p + a, p = p + bi. Then M is a finitely generated conical refinement monoid which is not a graph monoid. Indeed, the generators p, a, b are free primes of M, and M is an antisymmetric monoid, so that Gi is the trivial group for i ∈ {p, a, b}. The map G(ϕp) : eGJp → Gp becomes the map Z2 → 0, whose kernel is obviously non-cyclic. Therefore M is not a graph monoid by Theorem 5.6. We will split the proof of Theorem 5.6 in two parts. First, we prove that the condition is necessary. REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 25 Proposition 5.9. Let M be a finitely generated conical refinement monoid, and let p be a free prime of M. If M is a graph monoid then the map is an almost isomorphism. G(ϕp) : eGJp → Gp Proof. Assume that M is a graph monoid. Let E be a (row-finite) countable graph without sinks such that M ∼= M(E). (The condition that E does not have sinks can be assumed because we can add a loop at every sink, without changing the corresponding graph monoid.) Let p be a free prime of M. Let J = JM be the I-system associated to M, so that M = M(J ) [12, Theorem 2.7]. The condition in the statement only depends on the restricted (I ↓ p)- system JI↓p. Moreover, by Proposition 2.9, Proposition 2.15 and Lemma 2.18, we get M(JI↓p) ∼= I(H) ∼= M(EH ), where H is the hereditary and saturated subset of E0 corresponding to the order-ideal M(JI↓p) of M = M(J ). Therefore M(JI↓p) is a graph monoid and, restricting attention to I ↓ p, we may (and will) assume that p is the largest element of I. We have V(Lk(E)) ∼= M(E) ∼= M for any field k [11, Theorem 3.5]. Fix a field k for the rest of the argument. By Proposition 2.9, the order-ideals of M correspond to the lower subsets of the poset I. Moreover, the order-ideals of M correspond to the graded-ideals of Lk(E) and to the hereditary saturated subsets of E0 [11, Theorem 5.3]. Since p is the largest element of I, and p + x = p + ϕp,q(x) for all x ∈ Mq, we get that K0(Lk(E)) = G(M) = G(Mp) = Z × Gp, which is denoted by bGp (see Lemmas 2.8 and 2.10). Now, let I be the graded ideal of Lk(E) corresponding to the lower subset Jp of I generated by the primes q1, . . . , qn in the lower cover of p. Let H be the hereditary and saturated subset of E0 corresponding to I, so that I = I(H). Since Jp is a maximal lower subset of I, it follows that I = I(H) is a maximal graded-ideal of Lk(E), and at the same time it gives rise to the maximal order-ideal S := M(JJp) of M associated to the restricted Jp-system JJp. Observe that M/S ∼= Z+. Moreover, by Lemma 2.21, we have V(Lk(E)/I) ∼= V(Lk(E))/V(I) ∼= M/S ∼= Z+. Since Lk(E)/I = Lk(E)/I(H) ∼= Lk(E/H) is a graded-simple Leavitt path algebra, the tri- cotomy holds for Lk(E/H) (see [1, Proposition 3.1.14]). Since V(Lk(E/H)) ∼= Z+, Lk(E/H) must be Morita-equivalent to either k or k[t, t−1], the k-algebra of Laurent polynomials. As the graph E is row-finite and does not have sinks, the only possibility is that Lk(E/H) is Morita-equivalent to k[t, t−1]. Thus, the graph E/H must have a unique cycle c, without exits, to which all the other vertices of E/H connect. Let E′ be the graph obtained from E by removing all vertices in E0 \ (H ∪ c0) and all the edges emitted by them. Then (E′)0 is hereditary in E0, and its saturation in E is precisely E0. Therefore M(E′) = M(E) by Lemma 2.18. Hence, replacing E with E′, we can assume that E/H consists exactly of a unique cycle c. Further, let v be a vertex in the cycle c. It is easily shown, by a direct computation which only involves the definition of the graph monoid, that replacing the cycle c with a single loop based at v does not change the monoid M(E). (In this construction, we 26 PERE ARA AND ENRIQUE PARDO just re-define the starting vertex of the edges emitted by vertices of the cycle to H to be v.) Thus, we can assume that E/H is a single loop c, based at v. In particular, we have Lk(E)/I(H) ∼= Lk(E/H) ∼= k[t, t−1]. Observe that K0(I) = G(V(I)) = G(MJp) = eGJp by Lemma 2.10. Hence, the map K0(I) → K0(Lk(E)) can be identified with the map ι ◦ G(ϕp), where ι : Gp → bGp = Z × Gp is the canonical inclusion. On the other hand, we have K1(Lk(E)/I) = K1(k[t, t−1]) = K1(k) ⊕ K0(k) ∼= k× ⊕ Z [27, Theorem III.3.8]). Clearly, we have that the factor k×, which is generated by (see e.g. the units of the field k, is contained in the image of the map K1(Lk(E)) → K1(Lk(E)/I), while the factor Z is generated by multiplication by the unit t of k[t, t−1]. Hence, the exact sequence in K0 and K1 corresponding to the short exact sequence 0 −−−→ I −−−→ Lk(E) −−−→ k[t, t−1] −−−→ 0 gives the exact sequence Z ∂−−−→ eGJp ι◦G(ϕp) −−−−→ cGp −−−→ Z −−−→ 0. (5.1) Since we can lift the unit t to the von Neumann regular element c in vLk(E)v, we obtain (cf. [23, Proposition 1.3]) that The map ∂ : Z → eGJp is induced by the connecting homomorphism in algebraic K-theory. Now, c∗c = v and cc∗ +Pe∈s−1(v)\{c} ee∗ = v, so that [ee∗] = Xe∈s−1(v)\{c} ∂([t]) = Xe∈s−1(v)\{c} ∂([t]) = [v − cc∗] − [v − c∗c] ∈ K0(I). [r(e)] ∈ K0(I). K0(I) → G(MJp). Hence, the kernel of the canonical map G(ϕp) : eGJp → Gp is generated by the element in eGJp = G(MJp) corresponding to the image of the elementPe∈s−1(v)\{c}[r(e)] under the isomorphism It remains to show thatPe∈s−1(v)\{c}[r(e)] is a strictly positive element of eGJp. Let γ : M → M(E) be the isomorphism between M and M(E). Then each γ(qi) is a prime element of the graph monoid M(EH ) (see Lemma 2.18). Consider the tree T (v) of v. Then, T (v) is a hereditary subset of E0 containing v, and by hypothesis the order-ideal I(T (v)) generated by T (v) must be M(E). By Lemma 2.17, I(T (v)) (and so M(E)) is generated as a monoid by all the elements of the form aw with w ∈ T (v). Fix an index i ∈ {1, . . . , n}. By the preceding j=1 awj , where wj ∈ T (v) for all j. Since awj ≤ γ(qi) for all j, we see that all wj ∈ H. Since γ(qi) is prime, we get that γ(qi) ≤ awj for some j. So, for some vertex w ∈ T (v) ∩ H, γ(qi) ≡ aw (where ≡ denotes the antisymmetric relation generated by ≤). Thus, we can choose an edge e ∈ s−1(v) \ {c} such that r(e) connects to w, and so γ(qi) ≤ ar(e). This shows that qi ≤ γ−1(ar(e)) ∈ M(JJp). Therefore, there exists some lower subset J ′ of Jp such that γ−1(ar(e)) ∈ MJ ′, and since qi ≤ γ−1(ar(e)), it follows argument, we can write γ(qi) =Pm that qi ∈ J ′. Since this holds for every i = 1, . . . , n, we conclude that γ−1(Pe∈s−1(v)\{c} ar(e)) belongs to the maximal component MJp of M(JJp). Hence,Pe∈s−1(v)\{c}[r(e)] is the image of REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 27 γ−1(Pe∈s−1(v)\{c} ar(e)) under the canonical map MJp → G(MJp), and so it is strictly positive in G(MJp). This shows that G(ϕp) is an almost isomorphism. (cid:3) To show the converse, we will use a method which is similar to the method employed in the proof of Theorem 3.6. 5.10. In order to establish the correct setting for the induction argument, we need to introduce some terminology and notation. Since M is finitely generated, all the groups GM [p], for p a regular prime, are finitely generated. In the case where p is regular, we will assume that p is the neutral element of the group Gp = Mp = GM [p]. If p is a free prime, then the archimedean component GM [p] is of the form GM [p] = N × Gp, where Gp is a finitely generated group. Here the free prime p is identified with the element (1, ep) of N × Gp. We start by fixing a suitable form for these groups, namely we set, for p = e a regular prime, Ge = Zre ⊕ Zne 1 ⊕ · · · ⊕ Zne se , where re, se ≥ 0, and ne i ≥ 1 for all i. For p = e a regular prime, set Ne = re + se. We denote by x1, . . . , xNe the canonical set of group generators of the group Ge = Zre ⊕ Zne . Further, we set xNe+1 = −(x1 +· · ·+xre), and we observe that x1, . . . , xN e, xNe+1 is a family of semigroup generators for Ge, which we will call the canonical family of semigroup generators for Ge. 1 ⊕ · · · ⊕ Zne For p a free prime, we will denote by {gp Np} a family of semigroup generators of the group Gp, that is, every element of Gp is a finite sum of some of the elements in the family {gp Np}. Now, we have the following result: 1, . . . , gp 1, . . . , gp se Lemma 5.11. If p is a free prime, then we can assume that each gp i is of the form ϕp,q(g) for q < p, where g is either one of the canonical semigroup generators of Gq if q is a regular prime, or g is q if q is a free prime. Proof. We will show the result by (order-)induction. If p is a minimal prime which is free, then Gp is a trivial group, and the statement holds vacuously, taking the empty family of generators for Gp. Assume now that p is a free prime, and that the statement holds for all free primes below p. By [12, Definition 1.1(c2)] the map Mq<p ϕp,q : Mq<p Mq → Gp is surjective. Thus, a family of semigroup generators for Gp is obtained by taking ϕp,q(h), where q < p ranges on the set of regular primes below p, and h ranges on the family of canonical semigroup generators of the group Gq, together with the family {ϕp,q(q)}∪{bϕp,q(h)}, where q < p ranges on the set of free primes below p, and h ranges on the family of canonical semigroup generators of Gq. By induction hypothesis, applied to the free prime q, each h can be taken of the form ϕq,q′(h′), where either q′ < q is a regular prime and h′ is a canonical 28 PERE ARA AND ENRIQUE PARDO semigroup generator in Gq′, or it is the form ϕq,q′(q′), where q′ < q is a free prime. In the former case, we get and in the latter case we get bϕp,q(h) = bϕp,q(ϕq,q′(h′)) = bϕp,q(bϕq,q′(h′)) = bϕp,q′(h′) = ϕp,q′(h′), bϕp,q(h) = bϕp,q(bϕq,q′(q′)) = bϕp,q′(q′) = ϕp,q′(q′), which shows the result. Definition 5.12. Let p ∈ Ifree. We say that a family of elements {gp Np} of Gp is the canonical set of semigroup generators if this family consists of all elements of the form ϕp,q(g) for q < p, where g is either one of the canonical semigroup generators of Gq if q is a regular prime, or g is q if q is a free prime. 1, . . . , gp (cid:3) We are now ready to prove the converse of Proposition 5.9. Proposition 5.13. Let M be a finitely generated conical refinement monoid such that the exists a countable row-finite directed graph E such that M ∼= M(E). natural map G(ϕp) : eGJp → Gp is an almost isomorphism for every free prime. Then there generators of Gp for any free prime p ∈ M. We will also require that r +Pn Proof. Notice that, thanks to Lemma 5.11, we can always choose a canonical set of semigroup i=1 Nqi 6 sp for each p ∈ Ireg, where L(I, p) = {q1, . . . , qn} is the lower cover of p, and r is the number of free primes in L(I, P ). We will show the result by order-induction. Assume that we have a lower subset J of I, and that we have built a countable row-finite graph EJ such that E0 q , where each Eq is a strongly connected graph, with J =Fq∈J E0 q = {vq 1, . . . , vq E0 Nq+1, (vq)j i (rq + 1 ≤ i ≤ Nq, j ≥ 1)} for a regular prime q ∈ J, and E0 is an isomorphism q = {vq} if q ∈ J is a free prime. We also assume that there γJ : M(J) → M(EJ ), where M(J) is the order-ideal of M generated by J, such that γJ sends the canonical semi- group generators xq Nq+1 of Gq to vq Nq+1 for all regular q ∈ J, and sends the element q of GM [q] to vq for all free q ∈ J. 1, . . . , xq For q′ < q in J and x ∈ Mq′, we have γJ(ϕq,q′(x)) = φq q′ : M(EJ )[q′] → G(EJ )q is the structural map associated to the JJ-system coming from the finitely generated conical refinement monoid M(EJ ). q′(γJ(x)), where φq 1, . . . , vq In case J 6= I, let p be a minimal element of I \ J and write J ′ = J ∪ {p}. We will show that the above statement holds for the lower subset J ′ in place of J. This clearly establishes the result, because I is a finite poset. There are two cases to consider: (1) p is a minimal element of I: In this case, we will associate a graph Ep to GM [p]: when p is a regular prime, we define Ep using Lemma 3.1, while in case p is a free prime, we take Ep to be the one-loop graph based at the vertex vp. Now, let EJ ′ := Ep ⊔ EJ . Then M(J ′) = M({p}) ⊕ M(J) ∼= M(Ep) ⊕ M(EJ ) = M(EJ ′) REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 29 in a canonical way, showing the result. (2) p is not a minimal element of I: In this case, let L(I, p) = {q1, . . . , qn} be the lower cover of p in I. Here, we will assume that q1, . . . , qr are free primes and qr+1, . . . , qn are regular primes. We will define E0 p will be specified later. The edges in the graph EJ ′ will be the edges coming from EJ and a new family of edges that we will describe. Now, we have two different cases to consider. p , where E0 J ′ = E0 J ⊔ E0 (i) p is a regular prime: In this case, the set of vertices E0 p is defined as in Lemma 3.1, so that E0 p = {vp 1, . . . , vp Np+1, (vp)j i (rp + 1 ≤ i ≤ Np, j ≥ 1)}. The relations R that define the new edges of the graph EJ ′, all departing from the vertices in E0 p, are as follows: (1) vp Np+1 = 2vp Np+1 + vp i + np i−rvp i (2) For every r + 1 ≤ i ≤ N: rPi=1 NpPi=r+1 i = np i−r(vp)i + (vp)1 (a) (vp)1 (b) for every j ≥ 1, (vp)2j (c) for every j ≥ 1, (vp)2j+1 i + (vp)3 i , i = (vp)2j−1 = (vp)2j i i + (vp)2j i , i + (vp)2j+1 i + (vp)2j+3 i , (3) For every 1 ≤ i ≤ r, vp i = 2vp i + rXj=1,j6=i vp j + NpXj=r+1 (4) For every r + 1 ≤ i ≤ Np, np j−rvp j + vp Np+1 , vp i = vp j + rXj=1 NpXj=r+,1j6=i Here, for each r + 1 ≤ i ≤ Np: np j−rvp j + (np i−r + 1)vp i + vp Np+1 + (vp)1 i + Ai + v(i) . • Ai is a nonnegative integral linear combination of the vertices vp 1, . . . , vp Np+1, which • v(i) is a certain vertex in the graph EJ , which will be described below. There is a map i 7→ g(i) from [rp + 1, rp + r +Pn depends on an element g(i) ∈Fn i=1 bGqi, as described below. i=1 Nqi] ∩ Z toFn • establishes a correspondence between [rp + r + 1, rp + r +Pr • establishes a correspondence between [rp + r +Pr andSn j = 1, . . . , Nqi}, and : j = 1, . . . , Nqt}. t=r+1{xqt j i=1 bGqi such that i=1 Nqi] ∩ Z andSr i=1 Nqi + 1, rp + r +Pn • sends the set [rp + 1, rp + r] ∩ Z bijectively to the set of free primes q1, . . . , qr in L(I, p). i=1{gqi : j Of course, we take Ai as the trivial linear combination (with all coefficients being 0) in case i=1 Nqi] ∩ Z Now, we specify the value of the term Ai and the corresponding vertex v(i), which depend on the form of the specific generator g(i). Suppose first that i belongs to the interval [rp + i=1 Nqi]. In this case g(i) is a canonical group generator of a i=1 Nqi. i is larger than rp + r +Pn i=1 Nqi + 1, rp + r +Pn r +Pr 30 PERE ARA AND ENRIQUE PARDO group Gq corresponding to a regular prime q in the lower cover of p. We write for some non-negative integers aji. Then, we define −ϕp,q(g(i)) = ajixp j Np+1Xj=1 Ai = ajivp j , and v(i) = γJ(g(i)). We next consider the case where i belongs to the interval [rp + 1, rp + r], so that g(i) = q for a free prime q in the lower cover of p. Then we write for some non-negative integers aji, and we define −ϕp,q(q) = ajixp j Np+1Xj=1 Ai = ajivp j , and v(i) = vq = γJ (q). Finally, we need to consider the case where g(i) is a canonical semigroup generator of the group Gq for a free prime q in the lower cover of p. This means by definition that either g(i) is of the form ϕq,q′(h), where q′ < q is a regular prime and h is a canonical semigroup generator of Gq′, or that q′ < q is a free prime and g(i) = ϕq,q′(q′). In the former case, we set Np+1Xj=1 Np+1Xj=1 Np+1Xj=1 Np+1Xj=1 for some non-negative integers aji, and we define −ϕp,q′(h) = ajixp j Np+1Xj=1 Ai = ajivp j , and v(i) = γJ(h). In the latter case, we compute the non-negative integers aji using −ϕp,q′(q′), and we define Ai = ajivp j , and v(i) = vq′ = γJ (q′). Note that, in this situation, the same arguments as in Proposition 3.3 give that the subgraph Ep is strongly connected, and that there is a group homomorphism γp : Gp → M(EJ ′)[f ] from the group Gp to the archimedean component M(EJ ′)[f ] of the graph monoid M(EJ ′) corresponding to the vertices in E0 p, where f denotes the neutral element of that component, which sends the canonical semigroup generators xp Np+1 of Gp to the canonical set vp 1, . . . , vp Define a map Np+1 of elements of the group M(EJ ′)[f ]. 1, . . . , xp γJ ′ : M(J ′) → M(EJ ′) REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 31 as follows. The map γJ ′ agrees with the isomorphism γJ when restricted to the order-ideal M(J) of M(J ′). The map γJ ′ restricted to GM (J ′)[p] = GM [p] = Gp is just the map γp above, which sends the canonical semigroup generators of Gp to the elements vp Np+1 In order to prove that γJ ′ gives a well-defined monoid homomorphism, it of GM (EJ ′ )[f ]. suffices by [12, Corollary 1.6] to show that if q < p and x ∈ GM [q] then γJ ′(x) + γJ ′(p) = γJ ′(ϕp,q(x)) + γJ ′(p), that is, γJ ′(x) + f = γJ ′(ϕp,q(x)). By the same argument used in the proof of Theorem 3.6, it is enough to consider the case where q belongs to the lower cover of p. We will consider only the case where q is a free prime. (The case where q is a regular prime is easier and is left to the reader.) Assume that q is a free prime in the lower cover of p. Consider first the case where x = q. There is some i such that g(i) = q and thus v(i) = vq. Observe that, after using the relations R, relation R(4) can be expressed as: 1, . . . , vp i = vp vp i + f + Ai + v(i) = vp i + f − γJ ′(ϕp,q(q)) + vq. So, we obtain γJ ′(ϕp,q(q)) = f + vq, that is, f + γJ ′(x) = γJ ′(ϕp,q(x)), as desired. Now suppose t < q, and ht is either a canonical t is a free prime). that x ∈ Mq. Write x = mq +Pt ϕq,q′ t(ht), where m ∈ N, q′ t is a regular prime) or q′ semigroup generator of Gq′ Assume that we have proven that t (in case q′ t (in case q′ for each t. Then, using (5.2), the induction hypothesis and the fact that γJ and γp are semigroup homomorphisms, we get t(ht)) = f + γJ ′(ht) (5.2) ϕq,q′ γJ ′(bϕp,qϕq,q′ γJ ′(ϕp,q(x)) = γJ ′(ϕp,q(mq +Xt = mγJ ′(ϕp,q(q)) +Xt = f + mγJ (q) + f +Xt = f + γJ(mq +Xt = f + γJ ′(x). t(ht))) γJ(ht) t(ht)) γJ ′(bϕp,qϕq,q′ ht) = f + γJ(mq +Xt ϕq,q′ t(ht)) Thus, it remains to prove (5.2). Assume that q′ index such that g(i) = ϕq,q′ t is a free prime, so that ht = q′ t). Then R(4) gives again t. Let i be the t(q′ t) and v(i) = vq′ vi = vi + f − γJ ′(ϕp,q′ t = γJ ′(q′ t(q′ t)) + γJ ′(q′ t). Since ϕp,q′ t(q′ t(q′ t), we get t) =dϕp,qϕq,q′ t(q′ t)) = f + γJ ′(q′ t), γJ ′(dϕp,qϕq,q′ as desired. The case where q′ Gq′ t is a regular prime and ht is a canonical semigroup generator of t is treated in the same way. This shows that the defining relations of M(J ′) are preserved. So, the map γJ ′ is a well- defined homomorphism from M(J ′) to M(EJ ′). To build the inverse δJ ′ of γJ ′, we follow the idea in the proof of Proposition 3.3. The image by δJ ′ of the vertices in EJ is determined by the inverse of γJ . We define δJ ′(vp i ) = p. It is easily i for 1 ≤ i ≤ Np + 1, and δJ ′((vp)j i ) = xp 32 PERE ARA AND ENRIQUE PARDO checked that δJ ′ preserves the defining relations of M(EJ ′), and so it gives a well-defined homomorphism from M(EJ ′) to M(J ′), which is clearly the inverse of γJ ′. (ii) p is a free prime: isomorphism by hypothesis. Hence, G(ϕp) is surjective and its kernel is generated by a strictly positive element x. Note that Jp is a lower subset of J. In this case, the canonical map G(ϕp) : eGJp → Gp is an almost By Lemma 5.3, we can write x = rXi=1 χqi(ni, gi) + χqi(gi) nXi=r+1 for some ni ∈ N, i = 1, . . . , r, and gi ∈ Gqi, i = 1, . . . , n. By Lemma 5.11, for i ∈ {1, . . . , r} we can write gi = Xq′<qi,q′∈Ifree aq′ i ϕqi,q′(q′) + Xq′′<qi,q′′∈Ireg Nq′′ +1Xj=1 ji ϕqi,q′′(xq′′ bq′′ j ) for some non-negative integers aq′ i and bq′′ ji , and we can write, for i ∈ {r + 1, . . . , n}, gi = ajixqi j , Nqi +1Xj=1 for some non-negative integers aji. Define the graph EJ ′ with E0 E1 formula: J ⊔ {vp}, and with J and a set of edges starting at vp, which are determined by the following J ′ the union of E1 J ′ = E0 where and Ai = Xq′<qi,q′∈Ifree aq′ i vq′ nivqi + vp = vp + rXi=1 rXi=1 , Bi = Xq′′<qi,q′′∈Ireg Nqi +1Xj=1 i=1(Ai + Bi) +Pn i=1 nivqi +Pr ajivqi j , Ci = (Ai + Bi) + Ci, (5.3) nXi=r+1 Nq′′ +1Xj=1 ji vq′′ bq′′ j , (i = 1, . . . , r) (5.4) (i = r + 1, . . . , n). (5.5) If we definebx :=Pr The element vp of M(EJ ′) is a free prime, and so M(EJ ′)[vp] = N × G′ vp for some abelian group G′ vp. It follows easily from the induction hypothesis and the form of the relation (5.3) that the map γJ J extends to an order-isomorphism from J ′ = J ⊔{p} to the set P of primes of M(EJ ′), by sending p to vp. Since M(EJ ′) is a finitely generated conical refinement monoid, the map φp : M(EJ ′)γJ (Jp) → G′ vp induced by the various semigroup homomorphisms i=r+1 Ci, then we have that γJ (x) =bx. q : M(EJ ′)γJ (q) → G′ φp vp y 7→ (vp + y) − vp for q < p is surjective. So, we obtain a surjective group homomorphism G(φp) : G(M(EJ ′)γJ (Jp)) → G′ vp. REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 33 In order to somewhat simplify the notation, we will write M(EJ ′)Jp instead of M(EJ ′)γJ (Jp). J is a hereditary and saturated subset of E0 Since E0 J ) of M(EJ ′) gener- ated by E0 J coincides with the monoid M(EJ ) (by Lemma 2.18), and the component M(EJ ′)Jp coincides with the component M(EJ )Jp. The monoid isomorphism γJ : M(J) → M(EJ ) re- stricts to a semigroup isomorphism MJp → M(EJ )Jp, which induces a group isomorphism J ′, the order ideal I(E0 eγJp : G(MJp) → G(M(EJ )Jp) of the Grothendieck groups. Set K := ker(G(φp)), and notice that the relation (5.3) implies Hence, there is a commutative diagram with exact rows thateγJp(x) =bx ∈ K. 0 −−−→ hxi −−−→ G(MJp) G(ϕp) −−−→ Gp −−−→ 0 y yeγJp yγp 0 −−−→ K −−−→ G(M(EJ )Jp) G(φp) −−−→ G′ vp −−−→ 0 , (5.6) where γp : Gp → G′ the cokernel of the inclusion K ֒→ G(M(EJ )Jp). Notice that γp is an onto map. vp is the map induced from the cokernel of the inclusion hxi ֒→ G(MJ ) to Now, we define a map γJ ′ : M(J ′) → M(EJ ′) extending the monoid isomorphism γJ : M(J) → M(EJ ), and defining γJ ′ on the component Mp ∼= N × Gp of M(J ′) by the formula γJ ′(mp + g) = mvp + γp(g), for m ∈ N and g ∈ Gp. By [12, Corollary 1.6], to show that γJ ′ is a well-defined monoid homomorphism, it suffices to show that if q < p and y ∈ GM [q] = Mq then γJ ′(y) + γJ ′(p) = γJ ′(ϕp,q(y) + p), that is, γJ (y) + vp = γp(ϕp,q(y)) + vp. As y ∈ GM [q] = Mq and we are identifying G(M(Jp)) ∼= G(MJp) = eGJp (Lemma 2.7), there exists a composition map τq : Mq → M(Jp) → G(M(Jp)) ∼= G(MJp) such that ϕp,q = G(ϕp) ◦ τq. Analogously, we have a map τγJ (q) : M(EJ ′)γJ (q) → G(M(EJ ′)Jp) = G(M(EJ )Jp) such that φvp Using this fact, and the commutativity of (5.6), we have that γp(ϕp,q(y)) + vp = γp(G(ϕp)(τq(y))) + vp γJ (q) = G(φp) ◦ τγJ (q), and clearlyeγJp ◦ τq = τγJ (q) ◦ γJ Mq. = G(φp)(eγJp(τq(y))) + vp γJ (q)(γJ (y)) + vp = G(φp)(τγJ (q)(γJ(y))) + vp = φvp = ((vp + γJ(y)) − vp) + vp = vp + γJ(y) , as desired. 34 PERE ARA AND ENRIQUE PARDO This shows that there is a well-defined monoid homomorphism γJ ′ : M(J ′) → M(EJ ′) sending the canonical semigroup generators of M(J ′) to the corresponding canonical sets of vertices seen in M(EJ ′). In particular, γJ ′ is an onto map. In order to prove the injectivity of γJ ′, we can build an inverse map δJ ′ : M(EJ ′) → M(J ′), J , while δJ ′(vp) := p. Notice that the only as follows. On M(EJ ) we define δJ ′ to be γ−1 relation on M(EJ ′) not occurring already in M(EJ ) is vp = vp +bx, where γJ (x) =bx. Thus, δJ ′(bx) = x. But x generates the kernel of the map so that (p + x) − p equals 0 in Gp. Hence, the relation p = p + x holds in M(J ′). Thus, δJ ′ is a well-defined monoid homomorphism, and it is the inverse of γJ ′. This completes the proof of the inductive step, and so the result holds, as desired. (cid:3) G(ϕp) : eGJp → Gp ֒→ bGp = Z × Gp, As an illustration of Theorem 5.6, we obtain the characterization of antisymmetric finitely generated graph monoids [13]. Corollary 5.14. [13, Theorem 5.1] Let M be a finitely generated antisymmetric refinement monoid. Then M is a graph monoid if and only if for each free prime p in M the lower cover of p contains at most one free prime. Proof. Note that in the antisymmetric case we have that the groups Gq are trivial for all primes q. Consequently the maps ϕp,q, for q < p are all the zero map. Let p be a free prime of M. If q1, . . . , qr are the free primes in the lower cover of p, then the component MJp is isomorphic to Nr and thus G(MJp) = Zr. The canonical map G(ϕp) : G(MJp) −→ Gp reduces to the trivial map Zr −→ 0. If r > 1 the kernel of this map cannot be cyclic. Conversely if r ≤ 1 the kernel of this map is generated by a strictly positive element coming (cid:3) fromPq∈L(I,p) q. 6. Concluding remarks We finish the paper with a short overview of the current status of the realization problem for von Neumann regular rings, with special emphasis on the impact of the present paper in this program. After the work done in this paper, we obtain the realization of a large amount of finitely generated conical refinement monoids as graph monoids, and thus, using the results in [5], we can realize those monoids as V-monoids of suitable von Neumann regular rings (see Corollary 5.7). Moreover, we understand the reason why some finitely generated conical refinement monoids are not graph monoids. In order to obtain a general realization result for all finitely generated conical refinement monoids, it would be helpful to find a more comprehensive class of algebras, extending the class of Leavitt path algebras, which should include in particular the class of algebras considered in [4] associated to finite posets. A natural candidate for such a class could be the class of Kumjian-Pask algebras associated to higher rank graphs [14], or some variant of it. In [8], the first-named author and Ken Goodearl have introduced a fundamental distinction between refinement monoids. A refinement monoid M is said to be tame in case it is the direct limit of a directed system of finitely generated refinement monoids, and M is said to be wild REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 35 otherwise. Tame refinement monoids inherit from the finitely generated refinement monoids which approximate them good cancellation and decomposition properties, and include large classes of refinement monoids (see [8, Section 3]). Every graph monoid M(E) of an arbitrary graph E is tame ([8, Theorem 4.1]). Also, every primely generated conical refinement monoid is tame ([12, Theorem 0.1]), but not every graph monoid of a row-finite graph is primely generated (see [11, p. 171]). This raises the question of determining those tame conical refinement monoids which are graph monoids. Even assuming that the monoids are countable, the techniques introduced in the present paper are not enough for this purpose, a first difficulty (but not the only one) is that we cannot assume that the monoids are primely generated. For the monoid M := hp, a, b : p = p + a, p = p + bi considered in Example 5.8, it was proven in [13, Theorem 4.1] that M is not even a direct limit of graph monoids. However, M has been realized as the V-monoid of a von Neumann regular K-algebra for any field K, see [4, Example 3.3]. We present here some examples of conical refinement monoids which can be shown to be graph monoids by using our results. Proposition 6.1. Let M be a conical finite refinement monoid. Then M is a graph monoid. Proof. Every prime of M is necessarily regular, so M is regular and the result follows from Theorem 3.6. (cid:3) Example 6.2. We consider a family of I-systems (where I is a fixed poset) closely related to the ones introduced in Example 2.2. Let I = {p, q1, q2, . . . , qr}, where p > qi for all i, and all qi are pairwise incomparable. We set I = Ifree. Since qi are minimal free primes we must have Gqi = {eqi}. If r = 1, then Gp must be a finite cyclic group (Example 2.2(1)), and the criteria in Theorem 5.6 are then satisfied, so M(J ) is a graph monoid. If r > 1, then we see from Theorem 5.6 that M(J ) is a graph monoid if and only if Gp is of the form Zr/h(n1, n2, . . . , nr)i, where ni are strictly positive integers for all i, and the maps ϕp,qi are the obvious maps N → Zr/h(n1, n2, . . . , nr)i sending the generator of N to the image of the canonical basis vector ei. In terms of the examples given in Example 2.2(2), we see that they give rise to a graph monoid if and only if s = r − 1. Finally we want to add some words about the situation regarding the realization problem for wild refinement monoids. The abelianized Leavitt path algebras Lab(E, C) associated to sepa- rated graphs (E, C) (see [6], and also [7]) have the property that their monoids V(Lab(E, C)) are refinement monoids which may lack any possible cancellation or order-cancellation prop- erty which can fail in a conical abelian monoid. In particular, the monoids V(Lab(E, C)) are often wild refinement monoids. Some particular examples of this sort have been examined in [9], showing that they can be realized by exchange K-algebras ([9, Theorem 4.10]) for an arbitrary field K, and by von Neumann regular K-algebras exactly when the field is count- able ([9, Theorem 5.5]). This shows that there are evident differences with the case of tame refinement monoids, and suggests that different techniques have to be developed to deal with the case of wild refinement monoids. 36 PERE ARA AND ENRIQUE PARDO Acknowledgments Part of this work was done during visits of the second author to the Departament de Matem`atiques de la Universitat Aut`onoma de Barcelona (Spain). The second author thanks the center for its kind hospitality. Both authors thank Kevin O'Meara and the referee for various suggestions which helped to improve the original version of this paper. References [1] G. Abrams, P. Ara, M. Siles Molina, Leavitt path algebras, to appear in Springer's LNM, prelimi- nary version available at http://www.uccs.edu/∼gabrams/documents/FirstThreeChaptersJune2016.pdf. [2] G. Abrams, G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293 (2005), 319 -- 334. [3] P. Ara, The realization problem for von Neumann regular rings, in Ring Theory 2007. Proceedings of the Fifth China-Japan-Korea Conference, (H. Marubayashi, K. Masaike, K. Oshiro, M. Sato, Eds.), Hackensack, NJ (2009) World Scientific, pp. 21 -- 37. [4] P. Ara, The regular algebra of a poset, Trans. Amer. Math. Soc. 362 (2010), no. 3, 1505 -- 1546. [5] P. Ara, M. Brustenga, The regular algebra of a quiver, J. Algebra 309 (2007), 207 -- 235. [6] P. Ara, R. Exel, Dynamical systems associated to separated graphs, graph algebras, and paradoxical decompositions, Adv. Math. 252 (2014), 748 -- 804. [7] P. Ara, K. R. Goodearl, Leavitt path algebras of separated graphs, J. Reine Angew. Math. 669 (2012), 165 -- 224. [8] P. Ara, K. R. Goodearl, Tame and wild refinement monoids, Semigroup Forum 91 (2015), 1 -- 27. [9] P. Ara, K. R. Goodearl, The realization problem for some wild monoids and the Atiyah problem, Trans. Amer. Math. Soc., http://dx.doi.org/10.1090/tran/6889, arXiv:1410.8838v2 [math.RA]. [10] P. Ara, K. R. Goodearl, K.C. O'Meara, E. Pardo, Separative cancellation for projective modules over exchange rings, Israel J. Math. 105 (1998), 105 -- 137. [11] P. Ara, M.A. Moreno, E. Pardo, Nonstable K-Theory for graph algebras, Algebra Rep. Th. 10 (2007), 157-178. [12] P. Ara, E. Pardo, Primely generated refinement monoids, Israel J. Math. 214 (2016), 379 -- 419. [13] P Ara, F. Perera, F. Wehrung, Finitely generated antisymmetric graph monoids, J. Algebra 320 (2008), 1963 -- 1982. [14] G. Aranda Pino, J. Clark, A. an Huef, I. Raeburn, Kumjian-Pask algebras of higher-rank graphs, Trans. Amer. Math. Soc. 365 (2013), 3613 -- 3641. [15] G. Aranda Pino, E. Pardo, M. Siles Molina, Exchange Leavitt path algebras and stable rank, J. Algebra 305 (2006), 912 -- 936. [16] G. Brookfield, Cancellation in primely generated refinement monoids, Algebra Universalis 46 (2001), 343 -- 371. [17] H. Dobbertin, Primely generated regular refinement monoids, J. Algebra 91 (1984), 166 -- 175. [18] K. R. Goodearl, Von Neumann regular rings and direct sum decomposition problems, in "Abelian groups and modules" (Padova, 1994), Math. and its Applics. 343, pp. 249 -- 255, Kluwer Acad. Publ., Dordrecht, 1995. [19] K. R. Goodearl, E. Pardo, F. Wehrung, Semilattices of groups and inductive limits of Cuntz algebras, J. Reine Angew. Math. 588 (2005), 1 -- 25. [20] J. Ketonen, The structure of countable Boolean algebras, Annals of Math. 108 (1978), 41 -- 89. [21] A. Kumjian, D. Pask, I. Raeburn, Cuntz-Krieger algebras of directed graphs, Pacific J. Math. 184 (1998), 161 -- 174. [22] A. Kumjian, D. Pask, I. Raeburn, J. Renault, Graphs, groupoids and Cuntz-Krieger algebras, J. Funct. Anal. 144 (1997), 505 -- 541. [23] P. Menal, J. Moncasi, Lifting units in self-injective rings and an index theory for Rickart C*-algebras, Pacific J. Math. 126 (1987), 295 -- 329. [24] E. Pardo, F. Wehrung, Semilattices of groups and nonstable K-theory of extended Cuntz limits, K-Theory 37 (2006), 1 -- 23. REPRESENTING FINITELY GENERATED REFINEMENT MONOIDS 37 [25] R. S. Pierce, Countable Boolean algebras, in Handbook of Boolean Algebras, Vol. 3 (J. D. Monk and R. Bonnet, Eds.), Amsterdam (1989) North-Holland, pp. 775 -- 876. [26] J. Rosenberg, "Algebraic K-theory and its applications" Graduate Texts in Mathematics, 147. Springer- Verlag, New York, 1994. [27] C.A. Weibel, "The K-book. An Introduction to Algebraic K-theory", Graduate Studies in Mathematics, vol. 145, Amer. Math. Soc., Providence, R.I., 2013. [28] F. Wehrung, Embedding simple commutative monoids into simple refinement monoids, Semigroup Forum 56 (1998), 104 -- 129. [29] F. Wehrung, Non-measurability properties of interpolation vector spaces, Israel J. Math. 103 (1998), 177 -- 206. Departament de Matem`atiques, Universitat Aut`onoma de Barcelona, 08193 Bellaterra (Barcelona), Spain. E-mail address: [email protected] Departamento de Matem´aticas, Facultad de Ciencias, Universidad de C´adiz, Campus de Puerto Real, 11510 Puerto Real (C´adiz), Spain. E-mail address: [email protected] URL: https://sites.google.com/a/gm.uca.es/enrique-pardo-s-home-page/
1906.09979	1	1906	2019-06-21T13:14:38	Manin triples of 3-Lie algebras induced by involutive derivations	[ "math.RA", "math-ph", "math-ph" ]	For any $n$-dimensional 3-Lie algebra $A$ over a field of characteristic zero with an involutive derivation $D$, we investigate the structure of the 3-Lie algebra $B_1=A\ltimes_{ad^} A^ $ associated with the coadjoint representation $(A^, ad^)$. We then discuss the structure of the dual 3-Lie algebra $B_2$ of the local cocycle 3-Lie bialgebra $(A\ltimes_{ad^} A^, \Delta)$. By means of the involutive derivation $D$, we construct the $4n$-dimensional Manin triple $(B_1\oplus B_2,$ $ [ \cdot, \cdot, \cdot]_1,$ $ [ \cdot, \cdot, \cdot]_2,$ $ B_1, B_2)$ of 3-Lie algebras, and provide concrete multiplication in a special basis $\Pi_1\cup\Pi_2$. We also construct a sixteen dimensional Manin triple $(B, [ \cdot, \cdot, \cdot])$ with $\dim B^1=12$ using an involutive derivation on a four dimensional 3-Lie algebra $A$ with $\dim A^1=2$.	math.RA	math	MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS SHUAI HOU AND RUIPU BAI Abstract. For any n-dimensional 3-Lie algebra A over a field of characteristic zero with an in- volutive derivation D, we investigate the structure of the 3-Lie algebra B1 = A ⋉ad∗ A∗ associated with the coadjoint representation (A∗, ad∗). We then discuss the structure of the dual 3-Lie alge- bra B2 of the local cocycle 3-Lie bialgebra (A ⋉ad∗ A∗, ∆). By means of the involutive derivation D, we construct the 4n-dimensional Manin triple (B1 ⊕ B2, [·, ·, ·]1, [·, ·, ·]2, B1, B2) of 3-Lie alge- bras, and provide concrete multiplication in a special basis Π1 ∪ Π2. We also construct a sixteen dimensional Manin triple (B, [·, ·, ·]) with dim B1 = 12 using an involutive derivation on a four dimensional 3-Lie algebra A with dim A1 = 2. 1. Introduction The notion of n-Lie algebras was introduced by Filippov in [12], which are closely related to the fields of mathematics and physics, and the algebraic structure of n-Lie algebras corresponds to Nambu mechanics [14, 8, 21, 1]. In particular, as a special case of n-Lie algebras, 3-Lie algebras are extensively studied because they play a significant role in string theory and M- theory [7, 13, 2, 20, 15]. For example, the basic model of Bagger-Lambert-Gustavsson theory is based on the structure of metric 3-Lie algebras, and the Jacobi equation of 3-Lie algebras is the foundation for defining the N = 8 supersymmetry action. Lie bialgebras have widespread applications in geometry and physics. The structure of Lie bialgebras is very important since it contains coboundary theory, which makes the structure of Lie bialgebras relate to the classical Yang-Baxter equation [16]. In general, a Lie bialgebra is actually a vector space endowed with a Lie algebra structure (A, [·, ·]) and a Lie coalgebra struc- ture (A, ∆) (where ∆ : A → ∧2A is the comultiplication) satisfying the compatibility condition which is proposed based on the Hamitonian dynamics and Poisson Lie groups [10, 11, 19, 17]. In [3, 4], the authors studied 3-Lie bialgebra structures and Manin-triple for 3-Lie algebras. In this paper, we mainly study the structure of 3-Lie algebras with involutive derivations. By means of an involutive derivation of an n-dimensional 3-Lie algebra A, we construct a 4n- dimensional Manin triple (B1 ⊕ B2, [·, ·, ·]1, [·, ·, ·]2, B1, B2) of 3-Lie algebras, and study its structure. We also construct a 16-dimensional Manin triple (B, [·, ·, ·]) of 3-Lie algebras by means of an involutive derivation D on a four dimensional 3-Lie algebra A with dim A1 = 2, which is a 16-dimensional 3-Lie algebra with dim B1 = 12. The paper is organized as follows. In section 2, we recall some elementary facts on 3-Lie algebras, and give the description of the semi-direct product 3-Lie algebras associated with the coadjoint representations of 3-Lie algebras which have involutive derivations. In section 3, 0 Corresponding author: Ruipu Bai, E-mail: [email protected]. 2010 Mathematics Subject Classification. 17B05, 17D99. Key words and phrases. 3-Lie algebras, involutive derivations, semi-direct product 3-Lie algebra, Manin triples. 1 2 SHUAI HOU AND RUIPU BAI by means of involutive derivations, we construct a class of local cocycle 3-Lie bialgebras. In section 4, based on the coadjoint representations of semi-direct product 3-Lie algebras and the dual structures of the 3-Lie coalgebras, we construct a class of Manin triples and Matched pairs of 3-Lie algebras. In the paper, we suppose that all algebras and vector spaces are over a field F of characteristic zero, and for a subset S of a vector space V, we use hS i to denote the subspace of V spanned by S . 2. The semi-direct product 3-Lie algebra A ⋉ad∗ A∗ First we recall the notion of 3-Lie algebras with involutive derivations. A 3-Lie algebra is a vector space A with a linear multiplication (or 3-Lie bracket) [·, ·, ·] : ∧3A → A satisfying (1) [x1, x2, [x3, x4, x5]] = [[x1, x2, x3], x4, x5] + [x3, [x1, x2, x4], x5] + [x3, x4, [x1, x2, x5]], for ∀xi ∈ A, 1 ≤ i ≤ 5. A derivation D of a 3-Lie algebra A is a linear map D : A → A satisfying, (2) D([x1, x2, x3]) = [D(x1), x2, x3] + [x1, D(x2), x3] + [x1, x2, D(x3)], ∀x1, x2, x3 ∈ A. In addition, if D2 = Id, then D is called an involutive derivation on A. Thanks to (1), for ∀x1, x2 ∈ A, the left multiplication defined by (3) satisfies adx1 x2 : ∧2A → gl(A) adx1 x2 x = [x1, x2, x], ∀x ∈ A, (4) adx1 x2 [x3, x4, x5] = [adx1 x2 x3, x4, x5] + [x3, adx1 x2 x4, x5] + [x3, x4, adx1 x2 x5], which is called an inner derivation. Let A be an n-dimensional 3-Lie algebra with an involutive derivation D. Then A has a decomposition (5) A = A1 + A−1, where A1 = {x ∈ A Dx = x} and A−1 = {x ∈ A Dx = −x}, and there is a basis {x1, · · · , xn} of A such that x1, · · · , xs ∈ A1, and xs+1, · · · , xn ∈ A−1. Lemma 2.1. Let A be a finite dimensional 3-Lie algebra with an involutive derivation D. Then (6) (7) [A1, A1, A1] = [A−1, A−1, A−1] = 0, [A1, A1, A−1] ⊆ A1, [A1, A−1, A−1] ⊆ A−1. Proof. Apply Theorem 4 in [5]. (cid:3) A representation (or an A-module)[18] of a 3-Lie algebra A over a field F is a pair (V, ρ), where V is a vector space over F, and ρ is an F-linear map ρ : ∧2A → gl(V) satisfying, for ∀x1, x2, x3, x4 ∈ A, (8) [ρ(x1, x2), ρ(x3, x4)] = ρ([x1, x2, x3], x4) + ρ(x3, [x1, x2, x4]), MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 3 (9) ρ([x1, x2, x3], x4) = ρ(x1, x2)ρ(x3, x4) + ρ(x2, x3)ρ(x1, x4) + ρ(x3, x1)ρ(x2, x4). There is an equivalent description for an A-module, that is, (V, ρ) is an A-module if and only if (A ⊕ V, µ) is a 3-Lie algebra, where µ : (A ⊕ V)3 → A ⊕ V, for all x1, x2, x3 ∈ A, v ∈ V, µ(x1, x2, x3) = [x1, x2, x3], µ(x1, x2, v) = ρ(x1, x2)v, [A, V, V] = [V, V, V] = 0. The 3-Lie algebra (A ⊕ V, µ) is called the semi-direct product 3-Lie algebra of A associated with (V, ρ), which is denoted by A ⋉ρ V. [9] Let (A, [·, ·, ·]) be a 3-Lie algebra. Thanks to (3) and (4), (A, ad) is a representation, where ad : ∧2A → gl(A), for all x1, x2 ∈ A, ad(x1 ∧ x2) = adx1 x2 , which is called the adjoint representation of A. The dual representation (A∗, ad∗) of (A, ad) is called the coadjoint representation, where ad∗ : ∧2A → gl(A∗) is defined by (10) had∗ x1 x2 x∗ c, xti = −hx∗ c, adx1 x2 xti, ∀x1, x2, xt ∈ A, x∗ c ∈ A∗. The semi-direct product 3-Lie algebra (A ⋉ad∗ A∗, µ) associated with (A∗, ad∗) is denoted by B1. Then for all xi ∈ A, x∗ i ∈ A∗, 1 ≤ i ≤ 3, (11) µ(x1 + x∗ 1, x2 + x∗ 2, x3 + x∗ 3) = [x1, x2, x3] + ad∗ x1 x2 x∗ 3 + ad∗ x3 x1 x∗ 2 + ad∗ x2 x3 x∗ 1. For describing the structure of the semi-direct product 3-Lie algebra (A ⋉ad∗ A∗, µ), we need the structural constant in a basis of the 3-Lie algebra A. Let A be a 3-Lie algebra with a basis {x1, · · · , xn}, and let {x∗ dual space A∗, that is, (12) hxi, x∗ ji = δi j, 1 ≤ i, j ≤ n. Suppose the multiplication of A in the basis {x1, · · · , xn} is 1, · · · , x∗ n} be the dual basis of the (13) [xa, xb, xc] = n Xk=1 Γk abc xk, Γk abc ∈ F, 1 ≤ a, b, c, k ≤ n. By the above notations, we have the following result. Theorem 2.2. Let A be a 3-Lie algebra with an involutive derivation D, and let the multiplica- tion of A in the basis {x1, · · · , xn} be (13), where A1 = hx1, · · · , xsi, A−1 = hxs+1, · · · , xni. Let furtherd {x∗ the semi-direct product 3-Lie algebra B1 = A ⋉ad∗ A∗ satisfies 1, · · · , x∗ n} be the dual basis of the dual space A∗. Then the mulitiplication µ of (14) µ(xa, xb, xc) = Γk abc xk,1 ≤ a, b ≤ s < c ≤ n, Γk abc xk,1 ≤ a ≤ s < b, c ≤ n, 0, 1 ≤ a, b, c ≤ s or s + 1 ≤ a, b, c ≤ n. s Xk=1 Xk=s+1 n  4 SHUAI HOU AND RUIPU BAI − − − − n s Xk=s+1 Xk=1 Xk=1 Xk=s+1 n s  (15) µ(xa, xb, x∗ c) = and Γc abk x∗ k, 1 ≤ a, b, c ≤ s, Γc abk x∗ k, s + 1 ≤ a, b, c ≤ n, Γc abk x∗ k, 1 ≤ a, c ≤ s < b ≤ n, Γc abk x∗ k, 1 ≤ a ≤ s < b, c ≤ n, 0, 1 ≤ a, b ≤ s < c ≤ n, 0, 1 ≤ c ≤ s < a, b ≤ n, µ(xa, x∗ b, x∗ c) = µ(x∗ a, x∗ b, x∗ c) = 0, ∀1 ≤ a, b, c ≤ n. Proof. From (11), we can suppose (16) µ(xa, xb, x∗ c) = ad∗ xa xb x∗ c = n Xk=1 λk abc x∗ k, λk abc ∈ F, 1 ≤ a, b, c, k ≤ n. Thanks to Lemma 2.1 and Eq (13), a direct computation yields (14). By Eqs (10) and (16), for ∀xa, xb, xt ∈ A, x∗ c ∈ A∗, n hµ(xa, xb, x∗ λk abc x∗ k, xti = λt abc, c), xti =h Xk=1 c), xti = − hx∗ hµ(xa, xb, x∗ c, adxa,xb xti = −hx∗ c, n Xk=1 Γk abt xki = −Γc abt. Therefore, By (16), λt abc = −Γc abt, ∀1 ≤ a, b, c, t ≤ n. µ(xa, xb, x∗ c) = n Xk=1 λk abc x∗ k = − n Xk=1 Γc abk x∗ k, ∀1 ≤ a, b, c, t ≤ n. Follows from (14), we obtain (15). The proof is complete. (cid:3) 3. The local cocycle 3-Lie bialgebras induced by involutive derivations For a 3-Lie algebra A with an involutive derivation D, we will construct a 3-Lie algebra B2 on the dual space (A ⊕ A∗)∗. But before that, we need to construct a local cocycle 3-Lie bialgebra structure on the semi-direct product 3-Lie algebra A ⋉ad∗ A∗. First we recall some definitions. Let A be a 3-Lie algebra, A∗ be the dual space of A, and ∆ : A → A ∧ A ∧ A be a linear mapping. Then the dual mapping ∆∗ of ∆ is a linear mapping ∆∗ : A∗ ∧ A∗ ∧ A∗ → A∗ satisfying, (17) h∆∗(α, β, γ), xi = hα ⊗ β ⊗ γ, ∆(x)i, ∀α, β, γ ∈ A∗, x ∈ A. MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 5 A local cocycle 3-Lie bialgebra [4] is a pair (A, ∆), where A is a 3-Lie algebra, and ∆ = ∆1 + ∆2 + ∆3 : A → A ∧ A ∧ A is a linear mapping satisfying that • (A∗, ∆∗) is a 3-Lie algebra, • ∆1 is a 1-cocycle associated to the A-module (A ⊗ A ⊗ A, ad ⊗ 1 ⊗ 1), • ∆2 is a 1-cocycle associated to the A-module (A ⊗ A ⊗ A, 1 ⊗ ad ⊗ 1), • ∆3 is a 1-cocycle associated to the A-module (A ⊗ A ⊗ A, 1 ⊗ 1 ⊗ ad). For r = Pi xi ⊗ yi ∈ A ⊗ A, denotes [[r, r, r]] :≡Xi, j,k (cid:0)[xi, x j, xk] ⊗ yi ⊗ y j ⊗ yk + xi ⊗ [yi, x j, xk] ⊗ y j ⊗ yk + xi ⊗ x j ⊗ [yi, y j, xk] ⊗ yk + xi ⊗ x j ⊗ xk ⊗ [yi, y j, yk](cid:1). (18) The equation (19) [[r, r, r]] = 0 is called the 3-Lie classical Yang-Baxter equation in the 3-Lie algebra A, and simply denoted by CYBE. Let A be a 3-Lie algebra with a basis {x1, · · · , xn}, {x∗ 1, · · · , x∗ n} the dual basis of A∗, and D an involutive derivation of A. Define the tensor D ∈ A∗ ⊗ A by, (20) D(x, ξ) = hξ, Dxi, ∀x ∈ A, ξ ∈ A∗. Thanks to Theorem 3.3 in [6] (21) r = D − σ12D is a skew-symmetric solution of CYBE in the semi-direct product 3-Lie algebra A ⋉ad∗ A∗, and (22) D = n Xi=1 x∗ i ⊗ Dxi, r = n Xi=1 x∗ i ⊗ Dxi − n Xi=1 Dxi ⊗ x∗ i ∈ A∗ ⊗ A, where σ12 is the exchanging mapping. And r in (22) induces a local cocycle 3-Lie bialgebra (A ⋉ad∗ A∗, ∆) on the semi-direct product 3-Lie algebra (A ⋉ad∗ A∗, µ), where ∀x ∈ A ⋉ad∗ A∗, [x, −Dx∗ i , x∗ j] ⊗ Dx∗ j ⊗ x∗ i n Pi, j=1 ∆1(x) = [x, x∗ i , −Dx j] ⊗ x∗ j ⊗ Dxi +P j ⊗ x∗ i , [x, Dx∗ i , Dx j] ⊗ x∗ n n Pi, j=1 Pi, j=1 + ∆2(x) = σ13σ12∆1(x), ∆3(x) = σ12σ13∆1(x), ∆(x) = ∆1(x) + ∆2(x) + ∆3(x). (23)  6 SHUAI HOU AND RUIPU BAI If we suppose that x1, · · · , xs ∈ A1 and xs+1, · · · , xn ∈ A−1. By a direct computation we have s s s n (24) (25) ∆1(x) = + + + + + s s s n Xj=1 Xi=1 Xj=1 Xi=s+1 Xi=1 Xj=1 Xi=s+1 Xj=1 Xj=1 Xi=1 Xj=1 Xi=s+1 n n s s s s s n n Xi=1 Xj=s+1 Xi=s+1 Xj=s+1 Xi=1 Xj=s+1 Xi=s+1 Xj=s+1 Xi=1 Xi=s+1 + n n n n n s Xj=s+1 [x, x∗ i , −x j] ⊗ x∗ j ⊗ xi + [x, x∗ i , −x j] ⊗ x∗ j ⊗ (−xi) + [x, x∗ i , x j] ⊗ x∗ j ⊗ xi n Xj=s+1 [x, x∗ i , x j] ⊗ x∗ j ⊗ (−xi) [x, −xi, x∗ j] ⊗ x j ⊗ x∗ i + [x, −xi, x∗ j] ⊗ (−x j)∗ ⊗ x∗ i [x, xi, x∗ j] ⊗ x j ⊗ x∗ i + [x, xi, x∗ j] ⊗ (−x j) ⊗ x∗ i [x, xi, x j] ⊗ x∗ j ⊗ x∗ i + [x, xi, −x j] ⊗ x∗ j ⊗ x∗ i [x, −xi, x j] ⊗ x∗ j ⊗ x∗ i [x, xi, x j] ⊗ x∗ j ⊗ x∗ i , ∆2(x) = σ13σ12∆1(x) xi ⊗ [x, x∗ i , −x j] ⊗ x∗ j + i , x j] ⊗ x∗ j xi ⊗ [x, x∗ i , −x j] ⊗ x∗ j − i , x j] ⊗ x∗ j = − − + + − s s s s s n Xj=1 Xi=1 Xj=1 Xi=s+1 Xj=1 Xi=1 Xj=1 Xi=s+1 Xj=1 Xi=1 Xj=1 Xi=s+1 n n s s s s n xi ⊗ [x, x∗ Xj=s+1 Xj=s+1 i ⊗ [x, xi, x∗ x∗ xi ⊗ [x, x∗ s n s n n Xi=1 Xi=s+1 Xj=s+1 Xi=1 Xj=s+1 Xi=s+1 Xj=s+1 Xi=1 Xj=s+1 Xi=s+1 n n n n n s i ⊗ [x, xi, x∗ x∗ j] ⊗ x j + j] ⊗ x∗ j x∗ i ⊗ [x, xi, x∗ j] ⊗ x j − x∗ i ⊗ [x, xi, x∗ j] ⊗ x j x∗ i ⊗ [x, xi, x j] ⊗ x∗ j − x∗ i ⊗ [x, xi, x j] ⊗ x∗ j x∗ i ⊗ [x, xi, x j] ⊗ x∗ j + i ⊗ [x, xi, x j] ⊗ x∗ x∗ j, MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 7 ∆3(x) = σ12σ13∆1(x) s s s n x∗ j ⊗ xi ⊗ [x, x∗ i , −x j] + x∗ j ⊗ xi ⊗ [x, x∗ i , x j] − x∗ j ⊗ xi ⊗ [x, x∗ i , x j] x∗ j ⊗ xi ⊗ [x, x∗ i , x j] x j ⊗ x∗ i ⊗ [x, xi, x∗ j] + j ⊗ x∗ x∗ i ⊗ [x, xi, x∗ j] x j ⊗ x∗ i ⊗ [x, xi, x∗ j] − x j ⊗ x∗ i ⊗ [x, xi, x∗ j] x∗ j ⊗ x∗ i ⊗ [x, xi, x j] + x∗ j ⊗ x∗ i ⊗ [x, xi, −x j] x∗ j ⊗ x∗ i ⊗ [x, xi, x j] + x∗ j ⊗ x∗ i ⊗ [x, xi, x j]. (26) = + − + + − s s s n Xj=1 Xi=1 Xj=1 Xi=s+1 Xj=1 Xi=1 Xi=s+1 Xj=1 Xi=1 Xj=1 Xj=1 Xi=s+1 n n s s s s s n n n Xi=1 Xi=s+1 Xi=1 Xi=s+1 Xi=1 Xi=s+1 Xj=s+1 Xj=s+1 Xj=s+1 Xj=s+1 Xj=s+1 Xj=s+1 n n n n n s For convenience, in the following, the semi-direct product 3-Lie algebra (A ⋉ad∗ A∗, µ) is denoted by B1, and the 3-Lie algebra ((A ⊕ A∗)∗, ∆∗) is denoted by B2, where ∆∗ : (A ⊕ A∗)∗ ∧ (A ⊕ A∗)∗ ∧ (A ⊕ A∗)∗ → (A ⊕ A∗)∗ is the dual mapping of ∆ = ∆1 + ∆2 + ∆3 defined as (17). Suppose that (27) is a basis of B1, and (28) Π1 = {x1, · · · , xs, xs+1, · · · , xn, x∗ 1, · · · , x∗ s, x∗ s+1, · · · , x∗ n} Π2 = {y1, · · · , ys, ys+1, · · · , yn, y∗ 1, · · · , y∗ s, y∗ s+1, · · · , y∗ n} is a basis of B2 such that xi ∈ A, x∗ s + 1 ≤ l ≤ n, satisfying i ∈ A∗ for 1 ≤ i ≤ n and xk ∈ A1, xl ∈ A−1, 1 ≤ k ≤ s, (29) hxi, y∗ ji = hx∗ i , y ji = hxi, x∗ ji = δi j, hxi, y ji = hx∗ i , y∗ ji = 0, 1 ≤ i, j ≤ n, and the multiplication of the 3-Lie algebra A in the basis x1, · · · xs, xs+1, · · · , xn is (13). 8 SHUAI HOU AND RUIPU BAI Thanks to Theorem 2.2 and Eqs (24), (25), (26) and (13), for ∀1 ≤ t ≤ n, (30) (31) s ∆(xt) =(cid:0) − −(cid:0) +(cid:0) − Xi, j,k=1 Xi=s+1 +(cid:0) + n s s n s n n + + Xk=s+1 Xi, j=1 Xj=s+1 Xi,k=1 Xi, j,k=1 Xi, j,k=s+1(cid:1)Γi t jk x∗ Xi=s+1 Xj,k=1 Xi,k=s+1 Γk j ⊗ x∗ ti j x∗ + n n s i ⊗ xk −(cid:0) Xi,k=s+1 Xj=1 + n s s Xj,k=1 − s Xj=1 + Xi,k=s+1 k ⊗ x∗ n s n s s s s n n + + − − t jk x∗ Xj=1 Xi, j=s+1 Xk=1 (cid:1) Γi Xi, j=1 Xk=s+1 Xi=1 j ⊗ xi +(cid:0) Xi, j,k=1 Xi, j,k=s+1(cid:1) Γ i ⊗ x∗ tik x∗ Xi,k=1 Xj=s+1 Xi=1 Xi, j=s+1 Xk=1 (cid:1)Γk ti j x∗ Xj,k=s+1(cid:1)Γk j ⊗ x∗ i ⊗ xk, ti j x∗ + n n n s s s j k ⊗ x∗ j ⊗ xi n Xj,k=s+1(cid:1)Γ k ⊗ x j j ⊗ x∗ i ⊗ xk j tik x∗ i ⊗ x∗ k ⊗ x j ∆(x∗ t ) = s s n s s s n n n + + Γt Γt k ⊗ x∗ i jk)(x∗ Xi=1 Xj=1 Xk=s+1(cid:0) − Γt Xi=1 Xj=s+1 Xk=1 i jk(cid:0)x∗ Xi=1 Xj=s+1 Xk=s+1 i jk(cid:0)x∗ Xi=s+1 Xj=1 Xk=1 k ⊗ x∗ i jk(cid:0)x∗ Xi=s+1 Xj=1 Xk=s+1 i jk(cid:0)x∗ Xi=s+1 Xj=s+1 Xk=1 (cid:0) − Γt i jk)(x∗ Γt Γt + + n n n n n s s s s k ⊗ x∗ j ⊗ x∗ i + x∗ i ⊗ x∗ k ⊗ x∗ j + x∗ j ⊗ x∗ j ⊗ x∗ i + x∗ i ⊗ x∗ k ⊗ x∗ j + x∗ j ⊗ x∗ k ⊗ x∗ j ⊗ x∗ i + x∗ i ⊗ x∗ k ⊗ x∗ j + x∗ j ⊗ x∗ k ⊗ x∗ j ⊗ x∗ i + x∗ i ⊗ x∗ k ⊗ x∗ j + x∗ j ⊗ x∗ k ⊗ x∗ j ⊗ x∗ i + x∗ i ⊗ x∗ k ⊗ x∗ j + x∗ j ⊗ x∗ i ⊗ x∗ k(cid:1) j ⊗ x∗ i + x∗ i ⊗ x∗ k ⊗ x∗ j + x∗ j ⊗ x∗ i ⊗ x∗ k(cid:1) i ⊗ x∗ k(cid:1) i ⊗ x∗ k(cid:1) i ⊗ x∗ k(cid:1) i ⊗ x∗ k(cid:1). Then we get the following result. Theorem 3.1. Let A be a 3-Lie algebra with an involutive derivation D, and let the multiplica- tion of A in the basis {x1, · · · , xs, xs+1, · · · , xn} be (13), where A1 = hx1, · · · , xsi and A−1 = hxs+1, · · · , xni. Then the multiplication of the 3-Lie algebra (B2, ∆∗) in the basis {y1, · · · , ys, ys+1 · · · yn, y∗ 1, · · · , y∗ n} is as follows s+1, · · · , y∗ s, y∗ MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 9  (32) Proof. Suppose ∆∗(ya, yb, yc) = ∆∗(ya, yb, y∗ c) = − s Xk=1 Xk=s+1 n − Γk abcyk, 1 ≤ a, b ≤ s < c ≤ n, Γk abcyk, 1 ≤ a ≤ s < b, c ≤ n, 0, 1 ≤ a, b, c ≤ s, or s + 1 ≤ a, b, c ≤ n. n s Xk=s+1 Xk=1 Xk=1 Xk=s+1 n s Γc abky∗ k, 1 ≤ a, b, c ≤ s, Γc abky∗ k, s + 1 ≤ a, b, c ≤ n, Γc abky∗ k, 1 ≤ a, c ≤ s < b ≤ n, Γc abky∗ k, 1 ≤ a ≤ s < b, c ≤ n, 0, 1 ≤ a, b ≤ s < c ≤ n, or 1 ≤ c ≤ s < a, b ≤ n,   ∆∗(ya, y∗ b, y∗ c) = ∆∗(y∗ a, y∗ b, y∗ c) = 0, 1 ≤ a, b, c ≤ n. Xk=1 Xk=1 n Xk=1 Xk=1 n (33) ∆∗(ya, yb, yc) = Thanks to (29), n Xk=1 λk abcyk + n Xk=1 µk abcy∗ k, λk abc, µk abc ∈ F, 1 ≤ a, b, c, k ≤ n. n n h∆∗(ya, yb, yc), xti = h λk abcyk + µk abcy∗ k, xti = µt abc, 1 ≤ a, b, c, t ≤ n, h∆∗(ya, yb, yc), x∗ t i = λk abcyk + µk abcy∗ k, x∗ t i = λt abc, 1 ≤ a, b, c, t ≤ n. By Eqs (30) and (31), if 1 ≤ a, b, c ≤ s, 1 ≤ t ≤ n, then h∆∗(ya, yb, yc), xti = hya ⊗ yb ⊗ yc, ∆(xt)i = 0, h∆∗(ya, yb, yc), x∗ t i = hya ⊗ yb ⊗ yc, ∆(x∗ t )i = 0. Therefore, λt abc = µt abc = 0, and ∆∗(ya, yb, yc) = 0. For ∀ 1 ≤ a, b ≤ s < c ≤ n, 1 ≤ t ≤ n, h∆∗(ya, yb, yc), xti = hya ⊗ yb ⊗ yc, ∆(xt)i = 0, h∆∗(ya, yb, yc), x∗ t i = hya ⊗ yb ⊗ yc, ∆(x∗ t )i = ( −Γt abc, 1 ≤ t ≤ s, 0, s + 1 ≤ t ≤ n. 10 SHUAI HOU AND RUIPU BAI Therefore, µt abc = 0, λt abc = −Γt abc, and ∆∗(ya, yb, yc) = − s Xk=1 Γk abcyk. For ∀ 1 ≤ a ≤ s < b, c ≤ n, 1 ≤ t ≤ n, h∆∗(ya, yb, yc), xti = hya ⊗ yb ⊗ yc, ∆(xt)i = 0, h∆∗(ya, yb, yc), x∗ t i = hya ⊗ yb ⊗ yc, ∆(x∗ t )i = ( 0, 1 ≤ t ≤ s, −Γt abc, s + 1 ≤ t ≤ n. we get µt abc = 0, λt abc = −Γt abc, and ∆∗(ya, yb, yc) = − n Xk=s+1 Γk abcyk. Similarly, suppose ∆∗(ya, yb, y∗ c) = n Xk=1 λk abcyk + n Xk=1 Thanks to (29), µk abcy∗ k, λk abc, µk abc ∈ F, 1 ≤ a, b, c, k ≤ n. h∆∗(ya, yb, y∗ h∆∗(ya, yb, y∗ c), xti = hya ⊗ yb ⊗ y∗ c), x∗ t i = hya ⊗ yb ⊗ y∗ c), ∆(xt)i = µt t )i = λt c), ∆(x∗ abc, 1 ≤ a, b, c, t ≤ n, abc, 1 ≤ a, b, c, t ≤ n. For ∀ 1 ≤ a, b, c ≤ s, 1 ≤ t ≤ n, by Eqs (30) and (31), h∆∗(ya, yb, y∗ c), xti = hya ⊗ yb ⊗ y∗ c, ∆(xt)i = ( t i = hya ⊗ yb ⊗ y∗ 0, 1 ≤ t ≤ s, abt, s + 1 ≤ t ≤ n. Γc h∆∗(ya, yb, y∗ c), x∗ c, ∆(x∗ t )i = 0 Therefore, λt abc = 0, µt abc = Γc abt, ∆∗(ya, yb, y∗ c) = n Xk=s+1 Γc abky∗ k. For ∀ 1 ≤ a, c ≤ s < b ≤ n, 1 ≤ t ≤ n, h∆∗(ya, yb, y∗ c), xti = hya ⊗ yb ⊗ y∗ c, ∆(xt)i = ( Γc abt, 1 ≤ t ≤ s, 0, s + 1 ≤ t ≤ n. h∆∗(ya, yb, y∗ c), x∗ t i = hya ⊗ yb ⊗ y∗ c, ∆(x∗ t )i = 0 Therefore, λt abc = 0, µt abc = Γc abt, ∆∗(ya, yb, y∗ c) = s Xk=1 Γc abky∗ k. For ∀ 1 ≤ a ≤ s < b, c ≤ n, 1 ≤ t ≤ n, h∆∗(ya, yb, y∗ c), xti = hya ⊗ yb ⊗ y∗ c, ∆(xt)i = ( t i = hya ⊗ yb ⊗ y∗ 0, 1 ≤ t ≤ s, abt, s + 1 ≤ t ≤ n. Γc h∆∗(ya, yb, y∗ c), x∗ c, ∆(x∗ t )i = 0. MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 11 Therefore, λt abc = 0, µt abc = Γc abt, ∆∗(ya, yb, y∗ c) = n Xk=s+1 Γc abky∗ k. By the similar discussion to the above, we get the result for other cases. We omit the compu- (cid:3) tation process. 4. Manin triples and matched pairs of 3-Lie algebras induced by involutive derivations A metric on a 3-Lie algebra A is a non-degenerate symmetric bilinear form (·, ·) : A ⊗ A → F satisfying (34) ([x1, x2, x3], x4) + ([x1, x2, x4], x3) = 0, ∀x1, x2, x3, x4 ∈ A. The pair (A, (·, ·)) is called a metric 3-Lie algebra, or simply A is a metric 3-Lie algebra. If there are two subalgebras A1 and A2 of (A, (·, ·)) such that A = A1 ⊕ A2 (as vector spaces), (A1, A1) = 0, (A2, A2) = 0, [A1, A1, A2] ⊆ A2 and [A2, A2, A1] ⊆ A1, then the 5-tuple (A, [·, ·, ·], (·, ·), A1, A2) ( or 4-tuple (A, (·, ·), A1, A2)) is called a Manin triple [4]. Let (A, (·, ·)A, A1, A2) be a Manin triple, and (A′, (·, ·)A′) be a metric 3-Lie algebra. If there is a 3-Lie algebra isomorphism f : A → A′ satisfying (x, y)A = ( f (x), f (y))A ′ , ∀x, y ∈ A, then (A′, (·, ·)A′, f (A1), f (A2)) is also a Manin triple. And in this case we say that (A, (·, ·)A, A1, A2) is isomorphic to (A′, (·, ·)A′, A′ ′ ′ 2. Let (A, [·, ·, ·]) and (A 1, A′ f (A2) = A , [·, ·, ·]′) be two 3-Lie algebras. Suppose 2), where f (A1) = A 1, ′ ρ : A ∧ A → Der(A′), and χ : A′ ∧ A′ → Der(A) are linear mappings. For ∀xi ∈ A and ai ∈ A ′ , 1 ≤ i ≤ 3 we give the following identities: (35) (36) (37) (38) adx1 x2 χ(a1, a2)x3 − χ(a1, a2)adx1 x2 x3 =χ(ρ(x1, x2)a1, a2)x3 + χ(a1, ρ(x1, x2)a2)x3, ad′ a1a2 ρ(x1, x2)a3 − ρ(x1, x2)ad′ a1a2 a3 =ρ(χ(a1, a2)x1, x2)a3 + ρ(x1, χ(a1, a2)x2)a3, adx1 x2 χ(a1, a2)x3 = χ(ρ(x1, x2)a1, a2)x3 −χ(a1, ρ(x3, x1)a2)x2 − χ(a1, ρ(x2, x3)a2)x1, ad′ a1a2 ρ(x1, x2)a3 = ρ(χ(a1, a2)x1, x2)a3 −ρ(x1, χ(a3, a1)x2)a2 − ρ(x1, χ(a2, a3)x2)a1, where (A′, ad′) is the adjoint representation of the 3-Lie algebra (A′, [·, ·, ·]′). 12 SHUAI HOU AND RUIPU BAI , [·, ·, ·]′) are 3-Lie algebras, and there are linear Theorem 4.1. Suppose that (A, [·, ·, ·]) and (A mappings ρ : ∧2A → Der(A′) and χ : ∧2A → Der(A) such that (A′, ρ) is an A-module, (A, χ) is an A′-module, and Eqs (35), (36), (37) and (38) hold. Then (A ⊕ A′, [·, ·, ·]A⊕A′) is a 3-Lie algebra, where for ∀xi ∈ A, ai ∈ A , 1 ≤ i ≤ 3, ′ ′ ′ (39) [x1 + a1, x2 + a2, x3 + a3]A⊕A ′ =[x1, x2, x3] + ρ(x1, x2)a3 + ρ(x3, x1)a2 + ρ(x2, x3)a1 +[a1, a2, a3] ′ + χ(a1, a2)x3 + χ(a3, a1)x2 + χ(a2, a3)x1. Proof. Apply Proposition 4.4 in [4]. (cid:3) The 4-tuple (A, A′, ρ, χ) is called a matched pair of 3-Lie algebras. Let (A, [·, ·, ·]) and (A∗, [·, ·, ·]∗) be 3-Lie algebras, where A∗ is the dual space of A. There is a natural non-degenerate symmetric bilinear form (·, ·) : (A ⊕ A∗) ⊗ (A ⊕ A∗) → F given by (40) (x1 + ξ, x2 + η) = hx1, ηi + hξ, x2i, ∀x1, x2 ∈ A, ξ, η ∈ A∗. Then (A, A) = (A∗, A∗) = 0. Define linear multiplication [·, ·, ·]A⊕A∗ : (A⊕A∗)∧3 → A⊕A∗ by, ∀xi ∈ A, yi ∈ A∗ for 1 ≤ i ≤ 3, [x1 + y1, x2 + y2, x3 + y3]A⊕A∗ (41) =[x1, x2, x3] + ad∗ +aφ∗ x3 + aφ∗ y3y1 y1y2 y3 + ad∗ x1 x2 x2 + aφ∗ y2 + ad∗ x3 x1 x1 + [y1, y2, y3]∗, x2 x3 y1 y2y3 where (A∗, ad∗) is the coadjoint representation of the 3-Lie algebra (A, [·, ·, ·], and (A, aφ∗) is the coadjoint representation of the 3-Lie algebra (A∗, [·, ·, ·]∗). Then by (40) and (41), for ∀xi ∈ A, yi ∈ A∗ for 1 ≤ i ≤ 4, ([x1 + y1, x2 + y2, x3 + y3]A⊕A∗, x4 + y4) + (x3 + y3, [x1 + y1, x2 + y2, x4 + y4]A⊕A∗) x3 + aφ∗ y2 + ad∗ y3y1 x2 x3 y1y2 =([x1, x2, x3] + aφ∗ y3 + ad∗ + (ad∗ + (y3, [x1, x2, x4] + aφ∗ + (x3, ad∗ y4 + ad∗ x1,x2 x3 x1 x4 x1 x1 x2 y2y3 x1, y4) x2 + aφ∗ y1 + [y1, y2, y3]∗, x4) x1) y4y1 y1 + [y1, y2, y4]∗) x2 + aφ∗ y2y4 x2 x4 x4 + aφ∗ y1y2 y2 + ad∗ =([x1, x2, x3], y4) − (x3, [y1, y2, y4]∗) − (x2, [y3, y1, y4]∗) − (x1, [y2, y3, y4]∗) + (x4, [y1, y2, y3]∗) − ([x1, x2, x4], y3) − ([x3, x1, x4], y2) − ([x2, x3, x4], y1) + ([x1, x2, x3], y4) − (x4, [y1, y2, y3]∗) − (x2, [y4, y1, y3]∗) − (x1, [y2, y4, y3]∗) + (x3, [y1, y2, y4]∗) − ([x1, x2, x3], y4) − ([x4, x1, x3], y2) − ([x2, x4, x3], y1) = 0, and (A, A) = (A∗, A∗) = 0. We find that the non-degenerate symmetric bilinear form (·, ·) defined by (40) satisfies product invariance, and A, A∗ are isotropic in the 3-algebra (A ⊕ A∗, [·, ·, ·]A⊕A∗). Therefore, if (A ⊕ A∗, [·, ·, ·]A⊕A∗) is a 3-Lie algebra, then (A ⊕ A∗, [·, ·, ·]A⊕A∗, (·, ·)) is a metric 3-Lie algebra with isotropic subalgebras A and A∗, and (A ⊕ A∗, [·, ·, ·]A⊕A∗, (·, ·), A, A∗) is a Manin triple, which is called the standard Manin triple. Thanks to Theorem 4.1 and Eq (41), for arbitrary two 3-Lie algebras (A, [·, ·, ·]) and (A∗, [·, ·, ·]∗), the 5-tuple (A ⊕ A∗, [·, ·, ·]A⊕A∗, (·, ·), A, A∗) is a standard Manin triple if and only if (A, A∗, ad∗, aφ∗) is a matched pair. MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 13 In the following we suppose that A is a 3-Lie algebra with an involutive derivation D, and the multiplication of A in the basis {x1, · · · , xs, xs+1, · · · , xn} is (13), where xi ∈ A1, x j ∈ A−1, 1 ≤ i ≤ s, s + 1 ≤ j ≤ n; B1 = (A ⋉ A∗, µ) is the semi-direct product 3-Lie algebra of A with the multiplication (14) and (15) in the basis Π1 (defined by (27) ), and B2 = ((A ⊕ A∗)∗, ∆∗) is the 3-Lie algebra induced by the local cocycle 3-Lie bialgebra (A ⊕ A∗, ∆) with the multiplication (32) in the basis Π2 (defined by (28) ). Let aδ∗ : B1 ∧ B1 → gl(B2), be the coadjoint representations of the 3-Lie algebra (B1, µ) on B2 = (A ⊕ A∗)∗ = B∗ 1, and aψ∗ : B2 ∧ B2 → gl(B1 = B∗ 2) be the coadjoint representation of the 3-Lie algebra (B2, ∆∗) on B1 = B∗ 2. Then for ∀ya, yb, yc, y∗ a, y∗ c, yt, y∗ r ∈ Π2, and ∀xa, xb, xc, x∗ e, x∗ d, xt, x∗ t ∈ Π1, we have (42) (43) di = −hyt, µ(xa, xb, x∗ yt, xci = −hyt, µ(xa, xb, xc)i, yt, x∗ y∗ r , xci = −hy∗ y∗ r , x∗ di = −hy∗ yt, xci = −hyt, µ(x∗ d)i, r , µ(xa, xb, xc)i, r , µ(xa, xb, x∗ d)i, d, xb, xc)i, d, xb, x∗ ei = −hyt, µ(x∗ yt, x∗ e)i. xt, yci = −hxt, ∆∗(ya, yb, yc)i, ci = −hxt, ∆∗(ya, yb, y∗ xt, y∗ t , yci = −hx∗ x∗ x∗ t , y∗ ci = −hx∗ xt, yci = −hxt, ∆∗(y∗ c)i, t , ∆∗(ya, yb, yc)i, t , ∆∗(ya, yb, y∗ c)i, a, yb, yc)i, a, yb, y∗ c)i. xt, y∗ ci = −hxt, ∆∗(y∗ haδ∗ xa xb haδ∗ xa xb haδ∗ xa xb haδ∗ xa xb haδ∗ x∗ d xb haδ∗ x∗ d xb haψ∗ yayb haψ∗ yayb haψ∗ yayb haψ∗ yayb haψ∗ y∗ ayb haψ∗ y∗ ayb   By the above notations, we have the following result. Lemma 4.2. The multiplication [·, ·, ·]1 : ∧3(B1 ⊕ B2) → B1 ⊕ B2 of the semi-direct product 3-Lie algebra B1 ⋉aδ∗ B2 = (B1 ⊕ B2, [·, ·, ·]1) in the basis Π1 ∪ Π2 = {x1, · · · , xs, xs+1, · · · , xn, x∗ 1, · · · , x∗ n}, is as following, for ∀1 ≤ a, b, c, d, e, g, p, q, r, f , h, t ≤ n, n, y1, · · · , ys, ys+1, · · · , yn, y∗ s+1, · · · , x∗ s+1, · · · , y∗ 1, · · · , y∗ s, x∗ s, y∗ [xa, xb, xc]1 = µ(xa, xb, xc), [xa, xb, x∗ d]1 = µ(xa, xb, x∗ d), are defined as (14) and (15), and [xa, x∗ d, x∗ e]1 = [x∗ d, x∗ e, x∗ g]1 = [xa, yp, yq]1 = [xa, yp, y∗ t ]1 = 0, [xa, y∗ f , y∗ t ]1 = [yp, yq, yr]1 = [yp, y∗ q, y∗ t ]1 = [y∗ f , y∗ h, y∗ t ]1 = 0, 14 SHUAI HOU AND RUIPU BAI (44)  [xa, xb, yt]1 = aδ∗ xa xb yt = [xa, xb, y∗ t ]1 = aδ∗ xa xb y∗ t = [x∗ a, xb, yt]1 = aδ∗ x∗ a xb yt = s s Xk=1 Xk=1 Xk=s+1 Xk=s+1 n n n Γk abtyk, 1 ≤ a, b ≤ s < t ≤ n, Γk abtyk, 1 ≤ a, t ≤ s < b ≤ n, Γk abtyk, 1 ≤ t ≤ s < a, b ≤ n, Γk abtyk, 1 ≤ a ≤ s < b, t ≤ n, 0, 1 ≤ a, b, t ≤ s or s + 1 ≤ a, b, t ≤ n; − − − − s Xk=s+1 Xk=1 Xk=1 Xk=s+1 n s Γt abky∗ k, 1 ≤ a, b, t ≤ s, Γt abky∗ k, s + 1 ≤ a, b, t ≤ n, Γt abky∗ k, 1 ≤ a, t ≤ s < b ≤ n, Γt abky∗ k, 1 ≤ a ≤ s < b, t ≤ n, 0, 1 ≤ a, b ≤ s < t ≤ n, or 1 ≤ t ≤ s < a, b ≤ n; n s Xk=s+1 Xk=1 Xk=1 s Γa bkty∗ k, 1 ≤ a, b, t ≤ s, Γa bkty∗ k, s + 1 ≤ a, b, t ≤ n, Γa bkty∗ k, 1 ≤ a, b ≤ s < t ≤ n, or 1 ≤ a, t ≤ s < b ≤ n, n Xk=s+1 Γa bkty∗ k, 1 ≤ t ≤ s < a, b ≤ n, or 1 ≤ b ≤ s < a, t ≤ n, 0, 1 ≤ a ≤ s < b, t ≤ n, or 1 ≤ b, t ≤ s < a ≤ n.    Proof. For ∀1 ≤ a, b, c, t ≤ n, suppose [xa, xb, yt]1 = n Xk=1 λk abtyk + n Xk=1 By Eq (29), if 1 ≤ a, b, c ≤ n, 1 ≤ t ≤ n, then νk abty∗ k, λk abt, νk abt ∈ F. MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 15 haδ∗ xa xb haδ∗ xa xb yt, xci = h yt, x∗ ci = h n Xk=1 Xk=1 n λk abtyk + λk abtyk + n Xk=1 Xk=1 n νk abty∗ k, xci = νc abt, νk abty∗ k, x∗ ci = λc abt. Thanks to Eqs (14), (15), (29) and (42), for all 1 ≤ c ≤ n, haδ∗ xa xb yt, xci = −hyt, µ(xa, xb, xc)i = 0, 1 ≤ a, b ≤ s, 1 ≤ t ≤ n, haδ∗ xa xb yt, x∗ ci = −hyt, µ(xa, xb, x∗ c)i = ( 0, 1 ≤ a, b ≤ s, 1 ≤ t ≤ s, abt, 1 ≤ a, b ≤ s, , s + 1 ≤ t ≤ n. Γc Therefore, νc abt = 0, λc abt = Γc abt for 1 ≤ a, b ≤ s, s + 1 ≤ t ≤ n, and [xa, xb, yt]1 =  0, 1 ≤ a, b, t ≤ s, Γk abtyk, 1 ≤ a, b ≤ s < t ≤ n. s Xk=1 In the case s + 1 ≤ a, b ≤ n, 1 ≤ t ≤ n, we have that for all 1 ≤ c ≤ n, haδ∗ xa xb yt, xci = −hyt, µ(xa, xb, xc)i = 0, haδ∗ xa xb yt, x∗ ci = −hyt, µ(xa, xb, x∗ c)i = ( Γc abt, 1 ≤ t ≤ s, 0, s + 1 ≤ t ≤ n, and νc abt = 0, λc abt = Γc abt, 1 ≤ t ≤ s. Therefore, [xa, xb, yt]1 =  0, s + 1 ≤ a, b, t ≤ n, Γk abtyk, 1 ≤ t ≤ s < a, b ≤ n. n Xk=s+1 If ∀1 ≤ a ≤ s < b ≤ n, 1 ≤ t ≤ n, then for all 1 ≤ c ≤ n, haδ∗ xa xb yt, xci = −hyt, µ(xa, xb, xc)i = 0, haδ∗ xa xb yt, x∗ ci = −hyt, µ(xa, xb, x∗ c)i = ( Γc Γc abt, 1 ≤ t ≤ s, abt, s + 1 ≤ t ≤ n, and νc abt = 0, λc abt = Γc abt, 1 ≤ t ≤ n. Therefore, [xa, xb, yt]1 = s Xk=1 Xk=s+1 n Γk abtyk, s + 1 ≤ a, t ≤ s < b ≤ n, Γk abtyk, 1 ≤ a ≤ s < b, t ≤ n.  By a completely similar discussion to the above, for others cases, we get the expression of t ]1, respectively. We omit the calculation process. The identities in (44) t ]1 and [x∗ a, xb, y∗ (cid:3) [xa, xb, y∗ follow. 16 SHUAI HOU AND RUIPU BAI Lemma 4.3. The multiplication [·, ·, ·]2 : ∧3(B2 ⊕ B1) → B2 ⊕ B1 of the semi-direct product 3-Lie algebra B2 ⋉aψ∗ B1 = (B1 ⊕ B2, [·, ·, ·]2) in the basis Π1 ∪ Π2 = {x1, · · · , xs, xs+1, · · · , xn, x∗ 1, · · · , x∗ s, x∗ s+1, · · · , x∗ n, y1, · · · , ys, ys+1, · · · , yn, y∗ 1, · · · , y∗ s, y∗ s+1, · · · , y∗ n}, is as follows, for ∀ 1 ≤ a, b, c, d, e, g, p, q, r, f , h, t ≤ n, [ya, yb, yc]2 = ∆∗(ya, yb, yc), [ya, yb, y∗ d]2 = ∆∗(ya, yb, y∗ d) are defined as (32), [ya, y∗ d, y∗ e]2 = [y∗ d, y∗ e, y∗ g]2 = [ya, xp, xq]2 = [ya, xp, x∗ t ]2 = 0, [ya, x∗ f , x∗ t ]2 = [xp, xq, xr]2 = [xp, x∗ q, x∗ t ]2 = [x∗ f , x∗ h, x∗ t ]2 = 0, and (45) [ya, yb, xt]2 = aψ∗ yayb xt = (46) [ya, yb, x∗ t ]2 = aψ∗ yayb x∗ t =   − − s s Xk=1 Xk=1 Xk=s+1 Xk=s+1 n n − − Γk abt xk, 1 ≤ a, b ≤ s < t ≤ n, Γk abt xk, 1 ≤ a, t ≤ s < b ≤ n, Γk abt xk, 1 ≤ t ≤ s < a, b ≤ n, Γk abt xk, 1 ≤ a ≤ s < b, t ≤ n, 0, 1 ≤ a, b, t ≤ s, or s + 1 ≤ a, b, t ≤ n; n s Xk=s+1 Xk=1 Xk=1 Xk=s+1 n s Γt abk x∗ k, 1 ≤ a, b, t ≤ s, Γt abk x∗ k, s + 1 ≤ a, b, t ≤ n, Γt abk x∗ k, 1 ≤ a, t ≤ s < b ≤ n, Γt abk x∗ k, 1 ≤ a ≤ s < b, t ≤ n, 0, 1 ≤ a, b ≤ s < t ≤ n, or 1 ≤ t ≤ s < a, b ≤ n; MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 17 − n s Xk=s+1 Xk=1 Xk=1 s − − − n Xk=s+1  (47) [y∗ a, yb, xt]2 = aψ∗ y∗ ayb xt = Proof. For ∀1 ≤ a, b, t ≤ n, suppose Γa bkt x∗ k, 1 ≤ a, b, t ≤ s, Γa bkt x∗ k, s + 1 ≤ a, b, t ≤ n, Γa bkt x∗ k, 1 ≤ a, b ≤ s < t ≤ n, or 1 ≤ a, t ≤ s < b ≤ n, Γa bkt x∗ k, 1 ≤ t ≤ s < a, b ≤ n, or 1 ≤ b ≤ s < a, t ≤ n, 0, 1 ≤ a ≤ s < b, t ≤ n, or 1 ≤ b, t ≤ s < a ≤ n. [ya, yb, xt]2 = n Xk=1 λk abt xk + n Xk=1 νk abt x∗ k, λk abt, νk abt ∈ F. By Eqs (30) and (31), for 1 ≤ a, b, c ≤ n, 1 ≤ t ≤ n, we have haψ∗ yayb haψ∗ yayb xt, yci = h xt, y∗ ci = h n Xk=1 Xk=1 n λk abt xk + λk abt xk + n Xk=1 Xk=1 n νk abt x∗ k, yci = νc abt, νk abt x∗ k, y∗ ci = λc abt. Thanks to Eqs (29), (32) and (43), for ∀ya, yb, yc ∈ Π2, xt ∈ Π1, if 1 ≤ a, b ≤ s, 1 ≤ c, t ≤ n, then haψ∗ yayb haψ∗ yayb xt, yci = −hxt, ∆∗(ya, yb, yc)i = 0, xt, y∗ ci = −hxt, ∆∗(ya, yb, y∗ c)i = ( 0, 1 ≤ t ≤ s, abt, s + 1 ≤ t ≤ n. −Γc Therefore, νc abt = 0, λc abt = −Γc abt, and If 1 ≤ a ≤ s < b ≤ n, 1 ≤ c, t ≤ n, then νc [ya, yb, xt]2 =   [ya, yb, xt]2 = − s Xk=1 Xk=s+1 n − − s Xk=1 Γk abt xk, 1 ≤ a, b ≤ s < t ≤ n, 0, 1 ≤ a, b, t ≤ s. = 0, λc abt = −Γc abt, and abt Γk abt xk, 1 ≤ a, t ≤ s < b ≤ n, Γk abt xk, 1 ≤ a ≤ s < b, t ≤ n. 18 SHUAI HOU AND RUIPU BAI We get (45). By Eqs (30), (31), (29), (32) and (43). A similar discussion to the above, we get (46) and (47). We omit the calculation process. The proof is complete. (cid:3) Lemma 4.4. Let A be a 3-Lie algebra with an involutive derivation D, and the multiplication of A in the basis {x1, · · · , xs, xs+1, · · · , xn} be (13), where xi ∈ A1, x j ∈ A−1, 1 ≤ i ≤ s, s +1 ≤ j ≤ n. Then we have Γt kbc + Γk i jb Γt akc + Γk i jc Γt i ja Γk i jt + Γk i ja Γt kbt + Γk i jb Γc abk abk(cid:1) = 0, akt(cid:1) = 0, Γc jkt + Γc jak Γk bit + Γc jbk Γk iat) = 0, n Γc k jt + Γk ab j Γc ikt) − Xk=s+1 Γt aki − Γk abi Γc abk Γk i jt) = 0, n Γt kbi + Γk c jb c ja Γt c jk) + Γk c ji Γt abk) = 0, Γk c jt + Γk c ja Γb kit − Γb c jk Γk ait) + s Γk c ji Γb akt) = 0, (Γk ai j Γb ckt + Γb cak Γk i jt) + (Γb cik Γk jat + Γb c jk Γk ait)) = 0, Xk=s+1 Xk=s+1 s Xk=1 Xk=s+1 n Γt ak j + Γk bc j Γt aik) − Γk ai j Γt bck) = 0, where 1 ≤ a, b, c ≤ s, s + 1 ≤ i, j ≤ n, n s abi s s s s s n ( abi (Γk (Γk Xt=1 (cid:0)Γk Xk=s+1 Xt=1 (cid:0)Γc Xk=s+1 Xt=1 Xk=1 Xt=1 Xk=1 Xt=1 Xk=1 Xt=1 Xk=1 Xt=1 Xk=s+1 Xt=1 Xk=1 (Γk (Γk bci (Γb aik ( ( ( ( n s s s s s s s  n n s s n Xt=s+1(cid:0) Xk=s+1 Xt=s+1(cid:0)Γi Xk=1 Xt=s+1 Xk=1 Xt=s+1(cid:0) Xk=1 (Γb n n s (48) (49)  Γc abk Γk i jt + s Xk=1 (cid:0)Γk i ja Γc kbt + Γk i jb Γc akt + Γc i jk Γk abt(cid:1)(cid:1) = 0, Γk bct + Γi jbk Γk cat jak + Γi Γk ict + Γk ic j Γb akt) + a jk Γk a jt = 0, Γk s n jck Γb abt(cid:1) = 0, Xk=s+1 Xt=1 Xk=s+1(cid:0)Γi cak ick n Γk (Γk ab j Γi ckt + Γi c jk Γk abt) + b jt + Γi cbk Γk jat(cid:1)(cid:1) = 0, where 1 ≤ a, b, c, i ≤ s, s + 1 ≤ j ≤ n, MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 19 Γt kbc + Γk i jb Γt akc + Γk i jc Γt i ja Γk i jt + Γk i ja Γc kbt + Γk i jb Γc abk abk(cid:1) = 0, akt(cid:1) = 0, (Γk ibc Γa jkt + Γa jbk Γk cit + Γa jck Γk ibt) = 0, Γt kac + n Xk=s+1 (Γk jba Γt ikc + Γk jbc iak − Γk Γt iac Γt jbk)) = 0, jbi (Γc iak Γk jbt + Γk jba Γc ikt − Γc jbk Γk iat) − s s Xk=1 Γk jbi Γc kat) = 0, Γt ikc) − Γk i jt) = 0, abk (Γk abi (Γk abi Γt k jc + Γk ab j Γt k jc + Γk ab j Γt ikc) − Γt abk) = 0, s Γc Γk Xk=1 Xk=1 Xk=s+1(cid:0)Γa i jc n s n s s n n n n n ( Γk Xk=1 Xk=1 Xk=s+1 Xt=s+1 Xt=s+1 Xt=s+1 Xt=s+1 Xt=s+1(cid:0) Xt=s+1(cid:0)Γk Xt=s+1(cid:0)Γc Xt=s+1 Xk=1 Xk=s+1 Xk=s+1 Xk=s+1 Xk=1 ( ( ( n n n n n n s  (50) (51) (Γk i jc Γa bkt + Γa bck Γk i jt) + bik Γk jct + Γa b jk Γk cit(cid:1)(cid:1) = 0, where s + 1 ≤ a, b, c ≤ n, 1 ≤ i, j ≤ s, Γc abk Γk jit + n Xk=s+1 Γk jia Γc kbt + n Xk=s+1 Γk jib Γc akt(cid:1) + n n Xk=s+1 Xt=s+1 Γc jik Γk abt = 0, (Γi jak Γk bct + Γi jbk Γk cat + Γi jck Γk abt) = 0, s (Γk jbc Γi akt + Γi a jk Γk bct) + (Γc jbk Γk iat + Γk ia j Γc kbt) + Xk=1 Xk=1 s Γc iak Γk jbt) = 0, (Γi abk Γk c jt + Γi ack Γk jbt)) = 0, s s s n Xt=1 (cid:0) Xk=s+1 Xt=1 Xt=1 Xk=1 Xt=1 Xk=s+1 Xk=s+1 ( ( n n s s  where 1 ≤ j ≤ s, s + 1 ≤ a, b, c, i ≤ n. Proof. Thanks to (1), (14) and (15), for all 1 ≤ a, b, c ≤ s, s + 1 ≤ i, j ≤ n, 0 = µ(xi, x j, µ(xa, xb, xc)) = µ(µ(xi, x j, xa), xb, xc) + µ(xa, µ(xi, x j, xb), xc) + µ(xa, xb, µ(xi, x j, xc)) Γk i ja = ( n Pk=s+1 Therefore, Γt kbc + s Pt=1 Γk i jb n Pk=s+1 s Pt=1 Γt akc + Γk i jc n Pk=s+1 s Pt=1 Γt abk)xt. n Xk=s+1 Γk i ja s Xt=1 Γt kbc + n Xk=s+1 Γk i jb s Xt=1 Γt akc + n Xk=s+1 Γk i jc s Xt=1 Γt abk = 0, SHUAI HOU AND RUIPU BAI 20 and µ(xi, x j, µ(xa, xb, x∗ c)) = − Γc abk Γk i jt x∗ t , n Pk=s+1 s Pt=1 µ(µ(xi, x j, xa), xb, x∗ c) + µ(xa, µ(xi, x j, xb), x∗ c) + µ(xa, xb, µ(xi, x j, x∗ c)), n = ( Pk=s+1 Γk i ja s Pt=1 Γt kbt + Γc akt)x∗ t . It follows n i jb Γk Pk=s+1 Xk=s+1 n s s Pt=1 Xt=1 (cid:0)Γc abk Γk i jt + Γk i ja Γt kbt + Γk i jb Γc akt(cid:1) = 0. By a similar discussion, for others cases of 1 ≤ a, b, c, i, j ≤ n, we get (49)-(51). (cid:3) Lemma 4.5. Let aδ∗ and aψ∗ be defined as (42) and (43), respectively. Then aδ∗ and aψ∗ satisfy aδ∗(B1 ∧ B1) ⊆ DerB2 and aψ∗(B2 ∧ B2) ⊆ DerB1. Proof. First we prove aδ∗(B1 ∧ B1) ⊆ DerB2. Thanks to (32), and (44)-(47), we have the following identities ya, yb, yc) + ∆∗(ya, aδ∗ ∆∗(aδ∗ aδ∗ where1 ≤ a, b, c ≤ s, 1 ≤ i, j ≤ s, or 1 ≤ i ≤ s < j ≤ n; xi x j ∆∗(ya, yb, yc) = 0, yb, yc) + ∆∗(ya, yb, aδ∗ xi x j xi x j xi x j yc) = 0, or 1 ≤ a, b ≤ s < c ≤ n, 1 ≤ i, j ≤ s; ∆∗(aδ∗ xi x j n ya, yb, yc) + ∆∗(ya, aδ∗ xi x j n s = ( Xk=s+1 Γk i ja Xt=1 Γt kbc + Γk i jb Xk=s+1 Xt=1 xi x j ∆∗(ya, yb, yc) = 0, aδ∗ where s + 1 ≤ i, j ≤ n, 1 ≤ a, b, c ≤ s; yb, yc) + ∆∗(ya, yb, aδ∗ xi x j s s s yc) Γt akc + Γk i jc Xk=s+1 Xt=1 Γt abk)yt, ∆∗(aδ∗ xi x j n = ( Xk=s+1 ya, yb, yc) + ∆∗(ya, aδ∗ xi x j yb, yc) + ∆∗(ya, yb, aδ∗ xi x j yc) Γk i ja n Xt=s+1 Γt kbc + n Xk=s+1 Γk i jb n Xt=s+1 Γt akc)yt, aδ∗ xi x j ∆∗(ya, yb, yc) = s Xk=1 Γk abc n Xt=s+1 Γt i jkyt, where s + 1 ≤ i, j ≤ n, 1 ≤ a, b ≤ s < c ≤ n; ∆∗(aδ∗ ya, yb, yc) + ∆∗(ya, aδ∗ xi x j xi x j s = (− Γk i ja Xk=1 s Xt=1 yb, yc) + ∆∗(ya, yb, aδ∗ s s s xi x j yc) Γt akc − Γk i jc Xk=s+1 Xt=1 Γt abk)yt, Γt kbc − Γk i jb s Xk=1 Xk=1 s Xt=1 Xt=1 s aδ∗ xi x j ∆∗(ya, yb, yc) = − Γk abc Γt i jkyt, where 1 ≤ a, b ≤ s < c ≤ n, 1 ≤ i ≤ s < j ≤ n, or 1 ≤ j ≤ s < i ≤ n.     MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 21 By the above discussion and Eqs (48)-(51), we get aδ∗ xi x j ∆∗(ya, yb, yc) = ∆∗(aδ∗ xi x j ya, yb, yc) + ∆∗(ya, aδ∗ xi x j yb, yc) + ∆∗(ya, yb, aδ∗ xi x j yc), that is, aδ∗ xi x j ∈ Der(B2), for all 1 ≤ i, j ≤ n. Apply Lemma 4.3, Lemma 4.4 and a similar discussion to the above, we get aψ∗ for all 1 ≤ i, j ≤ n. The proof is complete. xi x j ∈ Der(B1), (cid:3) Theorem 4.6. Let (B1 ⊕ B2, [·, ·, ·]1) and (B1 ⊕ B2, [·, ·, ·]2) be 3-Lie algebras in Lemma 4.2 and Lemma 4.3, respectively. Then the 5-tuple (B1 ⊕ B2, [·, ·, ·]B1⊕B2, (·, ·), B1, B2) is a 4n-dimensional standard Manin triple of 3-Lie algebras, where for ∀u, v, w ∈ B1, α, β, ξ ∈ B2, (52) [u + α, v + β, w + ξ]B1⊕B2 = [u + α, v + β, w + ξ]1 + [u + α, v + β, w + ξ]2, and for ∀x ∈ B1, θ ∈ B2, (x, θ) = hx, θi. Proof. By Lamma 4.5, aδ∗ and aψ∗ satisfy aδ∗(B1 ∧ B1) ⊆ DerB2 and aψ∗(B2 ∧ B2) ⊆ DerB1, respectively. Next we only need to prove that identities (53), (54), (55) and (56) below hold, since (53) is equivalent to (35), (54) is equivalent to (36), (55) is equivalent to (37), and (56) is equivalent to (38) in the 3-algebra (B1 ⊕ B2, [·, ·, ·]B1⊕B2), respectively, where, ∀xa, xb, xc, yi, y j, ya, yb, yc, xi, x j, x∗ j, y∗ a, y∗ b, y∗ c, x∗ i , x∗ a, x∗ b, x∗ c, y∗ i , y∗ j ∈ Π1 ∪ Π2, yiy j yiy j µ(xa, xb, aψ∗ µ(xa, xb, aψ∗ µ(xa, xb, aψ∗ y∗ i y j a, xb, aψ∗ µ(x∗ µ(x∗ a, xb, aψ∗ µ(x∗ a, xb, aψ∗ y∗ i y j yiy j yiy j xc) = [aδ∗ x∗ c) = [aδ∗ xc) = [aδ∗ xc) = [aδ∗ x∗ c) = [aδ∗ xc) = [aδ∗ xa xb xa xb xa xb xa xb xa xb xa xb xa xb yi, y j, xc]2 + [yi, aδ∗ yi, y j, x∗ c]2 + [yi, aδ∗ xa xb i , aδ∗ i , y j, xc]2 + [y∗ y∗ yi, y j, xc]2 + [yi, aδ∗ yi, y j, x∗ c]2 + [yi, aδ∗ xa xb i , aδ∗ i , y j, xc]2 + [y∗ y∗ xa xb xa xb xa xb y j, xc]2 + aψ∗ µ(xa, xb, xc), yiy j y j, x∗ c]2 + aψ∗ µ(xa, xb, x∗ c), yiy j y j, xc]2 + aψ∗ µ(xa, xb, xc), y∗ i y j y j, xc]2 + aψ∗ µ(xa, xb, xc), yiy j y j, x∗ c]2 + aψ∗ µ(xa, xb, x∗ c), yiy j y j, xc]2 + aψ∗ µ(xa, xb, xc), y∗ i y j xi x j xi x j ∆∗(ya, yb, aδ∗ ∆∗(ya, yb, aδ∗ ∆∗(ya, yb, aδ∗ x∗ i x j a, yb, aδ∗ ∆∗(y∗ ∆∗(y∗ a, yb, aδ∗ a, yb, aδ∗ ∆∗(y∗ x∗ i x j xi x j xi x j yc) = [aψ∗ y∗ c) = [aψ∗ yc) = [aψ∗ yc) = [aψ∗ y∗ c) = [aψ∗ yc) = [aψ∗ yayb yayb yayb yayb yayb yayb yayb xi, x j, yc]1 + [xi, aψ∗ xi, x j, y∗ c]1 + [xi, aψ∗ yayb x∗ i , x j, yc]1 + [x∗ i , aψ∗ xi, x j, yc]1 + [xi, aψ∗ xi, x j, y∗ c]1 + [xi, aψ∗ yayb x∗ i , x j, yc]1 + [x∗ i , aψ∗ yayb yayb yayb xi x j x j, yc]1 + aδ∗ ∆∗(ya, yb, yc), x j, y∗ c]1 + aδ∗ ∆∗(ya, yb, y∗ c), xi x j x j, yc]1 + aδ∗ ∆∗(ya, yb, yc), x∗ i x j ∆∗(ya, yb, yc), x j, yc]1 + aδ∗ x j, y∗ ∆∗(yayb, y∗ c]1 + aδ∗ c), xi x j x j, yc]1 + aδ∗ ∆∗(ya, yb, yc), x∗ i x j xi x j yiy j yiy j µ(xa, xb, aψ∗ µ(xa, xb, aψ∗ µ(xa, xb, aψ∗ y∗ i y j a, xb, aψ∗ µ(x∗ a, xb, aψ∗ µ(x∗ a, xb, aψ∗ µ(x∗ y∗ i y j yiy j yiy j xa xb xa xb xc) = [aδ∗ x∗ c) = [aδ∗ xc) = [aδ∗ xc) = [aδ∗ x∗ a xb x∗ c) = [aδ∗ x∗ a xb xc) = [aδ∗ x∗ a xb xa xb xc xa yi, y j, xc]2 − [yi, aδ∗ c]2 − [yi, aδ∗ yi, y j, x∗ x∗ c xa y∗ i , y j, xc]2 − [y∗ i , aδ∗ yi, y j, xc]2 − [yi, aδ∗ c]2 − [yi, aδ∗ yi, y j, x∗ c x∗ x∗ a y∗ i , y j, xc]2 − [y∗ i , aδ∗ xc x∗ a xb xc y j, xb]2 − [yi, aδ∗ y j, xb]2 − [yi, aδ∗ xb x∗ c y j, xb]2 − [y∗ i , aδ∗ y j, xb]2 − [yi, aδ∗ y j, xb]2 − [yi, aδ∗ xb x∗ c y j, xb]2 − [y∗ i , aδ∗ xb xc xc xa y j, xa]2, y j, xa]2, y j, xa]2, y j, x∗ a]2, y j, x∗ a]2, y j, x∗ a]2, xb xc xb xc xc x∗ a (53) (54) (55)    22 SHUAI HOU AND RUIPU BAI xi x j xi x j ∆∗(ya, yb, aδ∗ ∆∗(ya, yb, aδ∗ ∆∗(ya, yb, aδ∗ x∗ i x j a, yb, aδ∗ ∆∗(y∗ a, yb, aδ∗ ∆∗(y∗ ∆∗(y∗ a, yb, aδ∗ x∗ i x j xi x j xi x j yayb yayb yc) = [aψ∗ y∗ c) = [aψ∗ yc) = [aψ∗ yc) = [aψ∗ y∗ a,yb y∗ c) = [aψ∗ y∗ ayb yc) = [aψ∗ y∗ ayb yayb ycya xi, x j, yc]1 − [xi, aψ∗ c]1 − [xi, aψ∗ xi, x j, y∗ y∗ cya i , aψ∗ i , x j, yc]1 − [x∗ x∗ xi, x j, yc]1 − [xi, aψ∗ c]1 − [xi, aψ∗ xi, x j, y∗ cy∗ y∗ a x∗ i , x j, yc]1 − [x∗ i , aψ∗ ycy∗ a ybyc x j, yb]1 − [xi, aψ∗ x j, yb]1 − [xi, aψ∗ yby∗ c i , aψ∗ x j, yb]1 − [x∗ x j, yb]1 − [xi, aψ∗ x j, yb]1 − [xi, aψ∗ yby∗ c x j, yb]1 − [x∗ i , aψ∗ ybyc ycya x j, ya]1, x j, ya]1, x j, ya]1, x j, y∗ a]1, x j, y∗ a]1, x j, y∗ a]1. ybyc ybyc ycy∗ a (56)  First we discuss Eq (53). Thanks to (44) in Lemma 4.2, (45)-(47) in Lemma 4.3, and (14), we obtain [aδ∗ xa xb µ(xa, xb, aψ∗ yi, y j, xc]2 + [yi, aδ∗ xc) = 0, where 1 ≤ a, b, c ≤ s, 1 ≤ i, j ≤ s, or 1 ≤ i ≤ s < j ≤ n. µ(xa, xb, xc) = 0, and y j, xc]2 + aψ∗ xa xb yiy j yiy j [aδ∗ yi, y j, xc]2 + [yi, aδ∗ y j, xc]2 + aψ∗ µ(xa, xb, xc) = 0, and xa xb xc) = 0, where s + 1 ≤ a, b, c ≤ n, s + 1 ≤ i, j ≤ n, or 1 ≤ i ≤ s < j ≤ n. xa xb yiy j µ(xa, xb, aψ∗ yiy j [aδ∗ xa xb yi, y j, xc]2 + [yi, aδ∗ xa xb y j, xc]2 + aψ∗ yiy j µ(xa, xb, xc) n n (Γk abi Γt k jc + Γk ab j Γt ikc)xt, and Xt=s+1 = − Xk=s+1 Xt=s+1 Xk=1 n s µ(xa, xb, aψ∗ yiy j xc) = − Γk i jc Γt abk xt, where 1 ≤ a, b, c ≤ s, s + 1 ≤ i, j ≤ n. [aδ∗ xa xb yi, y j, xc]2 + [yi, aδ∗ xa xb y j, xc]2 + aψ∗ yiy j µ(xa, xb, xc) s Xt=1 (Γk abi Γt k jc + Γk ab j Γt ikc)xt, and = − s Xk=1 Xt=1 s µ(xa, xb, aψ∗ yiy j xc) = − n Xk=s+1 Γk i jc Γt abk xt, where s + 1 ≤ a, b, c ≤ n, 1 ≤ i, j ≤ s + 1. Thanks to Eqs (48) and (50), we get s n Xk=1 Xt=s+1 Γk i jc Γt abk xt = n n Xk=s+1 Xt=s+1 (Γk abi Γt k jc + Γk ab j Γt ikc)xt, for 1 ≤ a, b, c ≤ s, s + 1 ≤ i, j ≤ n; and n s Xk=s+1 Xt=1 Γk i jc Γt abk xt = s s Xk=1 Xt=1 (Γk ab j Γt ikc + Γk abc Γt i jk)xt, for s + 1 ≤ a, b, c ≤ n, 1 ≤ i, j ≤ s. Therefore, for 1 ≤ a, b, c ≤ s, or s + 1 ≤ a, b, c ≤ n, µ(xa, xb, aψ∗ yi, y j, xc]2 + [yi, aδ∗ xc) = [aδ∗ yiy j xa xb y j, xc]2 + aψ∗ yiy j xa xb µ(xa, xb, xc), 1 ≤ i, j ≤ n. By a similar discussion to the above, we have [aδ∗ xa xb yi, y j, xc]2 + [yi, aδ∗ xa xb y j, xc]2 + aψ∗ yiy j µ(xa, xb, xc) = 0, and MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 23 where 1 ≤ a, b ≤ s < c ≤ n, 1 ≤ i, j ≤ s or 1 ≤ c ≤ s < a, b ≤ n, s + 1 ≤ i, j ≤ n. µ(xa, xb, aψ∗ yiy j xc) = 0, [aδ∗ xa xb yi, y j, xc]2 + [yi, aδ∗ xa xb y j, xc]2 + aψ∗ yiy j µ(xa, xb, xc) = − s n Xk=1 Xt=s+1 (Γk abi Γt k jc + Γk ab j Γt ikc + Γk abc Γt i jk)xt, and µ(xa, xb, aψ∗ yiy j xc) = 0, where 1 ≤ a, b ≤ s < c ≤ n, s + 1 ≤ i, j ≤ n. [aδ∗ xa xb yi, y j, xc]2 + [yi, aδ∗ xa xb y j, xc]2 + aψ∗ yiy j µ(xa, xb, xc) = − s s Xk=1 Xt=1 (Γk ab j Γt ikc + Γk abc Γt i jk)xt, and µ(xa, xb, aψ∗ yiy j xc) = − n s Xk=s+1 Xt=1 Γk i jc Γt abk xt, where 1 ≤ a, b ≤ s < c ≤ n, 1 ≤ i ≤ s < j ≤ n. [aδ∗ xa xb yi, y j, xc]2 + [yi, aδ∗ xa xb y j, xc]2 + aψ∗ yiy j µ(xa, xb, xc) = − s s Xk=1 Xt=1 (Γk abi Γt k jc + Γk abc Γt i jk)xt, µ(xa, xb, aψ∗ yiy j xc) = − n s Xk=s+1 Xt=1 Γk i jc Γt abk xt, where 1 ≤ a, b ≤ s < c ≤ n, 1 ≤ j ≤ s < i ≤ n. [aδ∗ xa xb yi, y j, xc]2 + [yi, aδ∗ xa xb y j, xc]2 + aψ∗ yiy j µ(xa, xb, xc) = − n s Xk=s+1 Xt=1 (Γk abi Γt k jc + Γk ab j Γt ikc + Γk abc Γt i jk)xt, and µ(xa, xb, aψ∗ yiy j xc) = 0, where 1 ≤ c ≤ s < a, b ≤ n, 1 ≤ i, j ≤ s. [aδ∗ xa xb yi, y j, xc]2 + [yi, aδ∗ xa xb y j, xc]2 + aψ∗ yiy j µ(xa, xb, xc) n Xt=s+1 (Γk abi Γt k jc + Γk abc Γt i jk)xt, and = − n Xk=s+1 Xt=s+1 n µ(xa, xb, aψ∗ yiy j xc) = − s Xk=1 Γk i jc Γt abk xt, where 1 ≤ c ≤ s < a, b ≤ n, 1 ≤ i ≤ s < j ≤ n. [aδ∗ xa xb yi, y j, xc]2 + [yi, aδ∗ xa xb y j, xc]2 + aψ∗ yiy j µ(xa, xb, xc) = − n n Xk=s+1 Xt=s+1 (Γk ab j Γt ikc + Γk abc Γt i jk)xt, and µ(xa, xb, aψ∗ yiy j xc) = − s n Xk=1 Xt=s+1 Γk i jc Γt abk xt, where 1 ≤ c ≤ s < a, b ≤ n, 1 ≤ j ≤ s < i ≤ n. 24 SHUAI HOU AND RUIPU BAI Thanks to (48) (in Lemma 4.4), for all 1 ≤ a, b ≤ s < c ≤ n, we have s Xt=1 (Γk abi Γt k jc + Γk ab j Γt ikc + Γk abc Γt i jk)xt = 0, s + 1 ≤ i, j ≤ n; s s Γk i jc Γt abk xt = (Γk ab j Γt ikc + Γk abc Γt i jk)xt, 1 ≤ i ≤ s < j ≤ n; Γk i jc Γt abk xt = (Γk abi Γt k jc + Γk abc Γt i jk)xt, , 1 ≤ j ≤ s < i ≤ n. Xk=1 Xk=1 Xt=1 Xt=1 s s n s Xk=s+1 Xt=1 Xt=1 s n Xk=s+1 Xk=s+1 n s n Xk=1 Xt=s+1 Xt=s+1 n Γk i jc Γk i jc s Xk=1 Xk=1 s By (50) (in Lemma 4.4), for 1 ≤ c ≤ s < a, b ≤ n, we have n Xt=s+1 (Γk abi Γt k jc + Γk ab j Γt ikc + Γk abc Γt i jk)xt = 0, 1 ≤ i, j ≤ s; Γt abk xt = Γt abk xt = n Xk=s+1 Xk=s+1 n n Xt=s+1 Xt=s+1 n (Γk abi Γt k jc + Γk abc Γt i jk)xt, 1 ≤ i ≤ s < j ≤ n; (Γk ab j Γt ikc + Γk abc Γt i jk)xt, 1 ≤ j ≤ s < i ≤ n. Therefore, for 1 ≤ i, j ≤ n, 1 ≤ a, b ≤ s < c ≤ n, or 1 ≤ c ≤ s < a, b ≤ n, we have µ(xa, xb, aψ∗ yiy j xc) = [aδ∗ xa xb yi, y j, xc]2 + [yi, aδ∗ xa xb y j, xc]2 + aψ∗ yiy j µ(xa, xb, xc). Summarizing the above discussion, we get that for 1 ≤ a, b, c ≤ n, 1 ≤ i, j ≤ n, µ(xa, xb, aψ∗ yiy j xc) = [aδ∗ xa xb yi, y j, xc]2 + [yi, aδ∗ xa xb y j, xc]2 + aψ∗ yiy j µ(xa, xb, xc), this is, the first identity in (53) holds. By a similar discussion to the above, we get (54)-(56). Thanks to Theorem 4.1, (B1 ⊕ B2, [·, ·, ·]B1⊕B2, (·, ·), B1, B2) is a standard Manin triple of 3-Lie algebras. The proof is complete. (cid:3) Corollary 4.7. Let (B1 ⊕ B2, [·, ·, ·]1) and (B1 ⊕ B2, [·, ·, ·]2) be 3-Lie algebras in Lemma 4.2 and Lemma 4.3, respectively. Then ((B1 ⊕ B2, [·, ·, ·]1), (B1 ⊕ B2, [·, ·, ·]2), aδ∗, aψ∗) is an 4n- dimensional matched pair. Proof. Apply Proposition 4.7 in [4] and Theorem 4.9. (cid:3) At last of the paper, we construct a sixteen dimensional Manin triple of 3-Lie algebras by an involutive derivation. Example 4.8. Let A be a 4-dimensional 3-Lie algebra with dim A1 = 2, and the multiplication of A in the basis {x1, x2, x3, x4} be as follows [x1, x3, x4] = x2, [x1, x3, x4] = x1. Then the linear mapping D : A → A defined by D(xi) = xi for 1 ≤ i ≤ 3 and D(x4) = −x4 is an involutive derivation of A, and satisfies x1, x2, x3 ∈ A1 and x4 ∈ A−1. By Theorem 4.9, (B1 ⊕ B2, [·, ·, ·]B1⊕B2, (·, ·), B1, B2) is a sixteen dimensional Manin triple of 3-Lie algebras in the basis {x1, · · · , x16}, where B1 = hx1, · · · , x8i, B2 = hx9, · · · , x16i, and the multiplication [·, ·, ·]B1⊕B2 is defined as (52). MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 25 For convenience, let B = B1 ⊕ B2, and [·, ·, ·]B the multiplication [·, ·, ·]B1⊕B2. Then the multi- plication of the Manin triple of 3-Lie algebras in the basis {x1, · · · , x16} is as follows: [x2, x3, x4]B = x1, [x1, x3, x4]B = x2, [x1, x4, x6]B = x7, [x2, x5, x3]B = x8, [x2, x4, x5]B = x7, [x3, x5, x4]B = x6, [x3, x6, x4]B = x5, [x3, x4, x9]B = x10, [x3, x1, x14]B = x16, [x3, x2, x13]B = x16, [x6, x3, x9]B = x16, [x3, x6, x12]B = x13, [x4, x6, x9]B = x15, [x6, x4, x11]B = x13, [x2, x3, x12]B = x10, [x4, x1, x11]B = x10, [x2, x3, x12]B = x9, [x4, x2, x11]B = x9, [x1, x14, x4]B = x15, [x2, x4, x13]B = x15, [x4, x3, x13]B = x14, [x4, x3, x14]B = x13, [x1, x5, x11]B = x16, [x5, x2, x12]B = x15, [x5, x3, x10]B = x16, [x3, x5, x12]B = x14, [x4, x5, x10]B = x15, [x5, x4, x11]B = x14, [x1, x6, x11]B = x16, [x6, x1, x12]B = x15, [x9, x4, x11]B = x2, [x9, x12, x3]B = x2, [x10, x4, x11]B = x1, [x10, x12, x3]B = x1, [x11, x1, x12]B = x2, [x11, x10, x12]B = x1, [x10, x11, x5]B = x8, [x10, x5, x12]B = x7, [x11, x12, x5]B = x6, [x9, x11, x6]B = x8, [x9, x6, x12]B = x7, [x11, x12, x6]B = x5, [x13, x10, x3]B = x8, [x13, x4, x10]B = x7, [x13, x2, x11]B = x8, [x13, x11, x4]B = x6, [x13, x12, x2]B = x7, [x13, x3, x12]B = x6, [x14, x9, x3]B = x8, [x14, x4, x9]B = x7, [x14, x1, x11]B = x8, [x14, x11, x4]B = x5, [x14, x12, x1]B = x7, [x14, x3, x12]B = x5, [x10, x12, x11]B = x9, [x9, x12, x11]B = x10, [x9, x11, x14]B = x16, [x9, x14, x12]B = x15, [x10, x11, x13]B = x16, [x10, x13, x12]B = x15, [x1, x6, x3]B = x8, [x3, x4, x10]B = x9, [x11, x12, x13]B = x14, [x11, x14, x12]B = x13. Theorem 4.9. The 16-dimensional 3-Lie algebra B in Example 4.8 is 2-solvable but non- nilpotent, and B has the smallest ideal I = hx1, x2, x7, x8, x9, x10, x15, x16i, which I is an abelian ideal, and dim B1 = 12. Proof. From the multiplication, I = hx1, x2, x7, x8, x9, x10, x15, x16i is the smallest ideal of B, and satisfies [I, I, B]B = 0. Therefore, I is an abelian ideal. It is clear that B is a 16-dimensional 3-Lie algebra with dim B1 = 12, and B1 = [B, B, B]B = hx1, x2, x5, x6, x7, x8, x9, x10, x13, x14, x15, x16i. Since for any positive integer r, Br = B1, B is non-nilpotent. From B(2) = [B1, B1, B]B = I, B(3) = [B(2), B(2), B]B = [I, I, B]B = 0, B is a 2-solvable 3-Lie algebra. (cid:3) Acknowledgements. Ruipu Bai was supported by the Natural Science Foundation of Hebei Province (A2018201126). 26 SHUAI HOU AND RUIPU BAI References [1] H. Awata, M. Li, D. Minic, et al, On the quantization of Nambu brackets, Journal of High Energy Physics, 2001, 2001(2): 69-82. [2] I. A. Bandos, On multiple M2-brane models and its N=8 superspace formulations, Acta Polytechnica, 2010, 50(3):1-10. [3] R. Bai, W. Guo, L. Lin, n-Lie Bialgebras, Linear and Multilinear Algebra, 2018, 66(2): 382-397. [4] C. Bai, L. Guo, Y. Sheng, Bialgebras, the classical Yang-Baxter equation and Manin triples for 3-Lie algebras, arXiv:1604.05996. [5] R. Bai, S. Hou, Y. Gao, Structure of n-Lie Algebras with Involutive Derivations, Interna- tional Journal of Mathematics and Mathematical Sciences, 2018, (2018): 1-9. [6] R. Bai, S. Hou S, 3-Lie bialgebras and 3-pre-Lie algebras induced by involutive deriva- tions, arXiv:1906.06771. [7] J. Bagger, N. Lambert, Gauge symmetry and supersymmetry of multiple M2-branes, Phys- ical Review D, 2008, 77(6): 065008. [8] P. Dirac, Generalized hamiltonian dynamics, Proceedings of the Royal Society A Mathe- matical Physical and Engineering Sciences, 1958, 246(1246): 2405-2412. [9] A. S. Dzhumadil' daev, Representation of Vector Porduct n-Lie Algebras, Communication in Algebra, 2014, 32: 3315-3326. [10] V. G. Drinfeld, Hamiltonian structures of lie groups, lie bialgebras and the geometric meaning of the classical Yang-Baxter equations, Soviet Math Doklady, 1983, 27(2): 222- 225. [11] P. Etingof, D. Kazhdan, Quantization of Lie bialgebras, II, Selecta Mathematica, 1998, 4(2): 213. [12] V. T. Filippov, n-Lie algebras, Siberian Mathematical Journal, 1985, 26 (6): 879-891. [13] A. Gustavsson, Algebraic structures on parallel M2-branes, Nuclear Physics B, 2009, 811(1): 66-76. [14] P. Gautheron, Some remarks concerning Nambu mechanics, Letters in Mathematical Physics, 1996, 37(1): 103-116. [15] J. P. Gauntlett, J. B. Gutowski, Constraining maximally supersymmetric membrane ac- tions, Journal of High Energy Physics, 2008, 2008(6): 53. [16] M. Gerstenhaber, S. D. Schack, Bialgebra cohomology, deformations, and quantum groups, Proceedings of the National Academy of Sciences of the United States of America, 1990, 87(1):478-481. [17] S. Joni, C. Rota, Coalgebras and bialgebras in combinatorics, Studies in Applied Mathe- matics, 1979, 61(2): 93-139. [18] S. M. Kasymov, Theory of n-lie algebras, Algebra and Logic, 1987, 26(3): 155-166. MANIN TRIPLES OF 3-LIE ALGEBRAS INDUCED BY INVOLUTIVE DERIVATIONS 27 [19] J. Loday, Generalized bialgebras and triples of operads, Mathematics, 2008, 320(320): 1-103. [20] M. M. Sheikh-Jabbari, A new three-algebra representation for the Superconformal Chern- Simons theory, Journal of High Energy Physics, 2008, 6(12): 111-111. [21] L. Takhtajan, On foundation of the generalized Nambu mechanics, Communications in Mathematical Physics, 1994, 160(2): 295-315. College of Mathematics and Information Science, Hebei University, Baoding 071002, China E-mail address: [email protected] College of Mathematics and Information Science, Hebei University, Key Laboratory of Machine Learning and Computational, Intelligence of Hebei Province, Baoding 071002, P.R. China E-mail address: [email protected]
1009.5543	1	1009	2010-09-28T11:59:33	On maximal distances in a commuting graph	[ "math.RA" ]	We study maximal distances in the commuting graphs of matrix algebras defined over algebraically closed fields. In particular, we show that the maximal distance can be attained only between two nonderogatory matrices. We also describe rank-one and semisimple matrices using the distances in the commuting graph.	math.RA	math	ON MAXIMAL DISTANCES IN A COMMUTING GRAPH GREGOR DOLINAR, BOJAN KUZMA, AND POLONA OBLAK Abstract. We study maximal distances in the commuting graphs of matrix algebras defined over algebraically closed fields. In particular, we show that the maximal distance can be attained only between two nonderogatory matrices. We also describe rank-one and semisimple ma- trices using the distances in the commuting graph. 1. Introduction and preliminaries One of the options how to study properties in certain non-commutative algebraic domains is a commutator. For example, in algebras the additive commutator, i.e., Lie product [A, B]a = AB−BA is usually used and with its help some beautiful results were obtained. Let us only mention the famous Kleinecke-Shirokov Theorem [15, 21]. In groups the multiplicative commu- tator [A, B]m = A−1B−1AB is used, and it is a central tool in studying solvability of groups and hence in Galois theory of solvability of equations by radicals [11]. Additional information about non-commuting elements is obtained by studying the properties of a commuting graph. For example, if the com- muting graphs over two finite semisimple rings are isomorphic, then their noncommutative parts are also isomorphic [3]. Let us remark that commut- ing graph can also be used in algebraic domains where commutator is not available, e.g., in semigroups or in semirings. Up until now, one of the prime concerns when studying commuting graphs was calculating its diameter [1, 4, 6, 9, 10, 17, 19]. It turned out that we obtain essentially different results if the matrix algebra Mn(F) is defined over an algebraically closed field F than if it is defined over non-closed one. While in the former case the diameter is always equal to four, provided n ≥ 3, in the later case the graph may be disconnected, and if it is connected the diameter is known to be at most six. The hypothesis is that if the commuting graph is connected, its diameter is at most five [4, Conjecture 18]. Note that for n = 2 the commuting graph over any field is disconnected [5, Remark 8]. 2000 Mathematics Subject Classification. 15A27, 05C50, 05C12. Key words and phrases. Algebraically closed field, Matrix algebra, Centralizer, Com- muting graph, Distance, Path, Minimal matrix, Maximal Matrix. The work is partially supported by a grant from the Ministry of Higher Education, Science and Technology, Slovenia. 1 2 DOLINAR, KUZMA, AND OBLAK In the present paper we are interested in commuting graphs of matrix algebras Mn(F) over algebraically closed fields F with n ≥ 3. In particular we study the maximal distances between its vertices. It was already proved in [4, Proof of Theorem 3] that an elementary Jordan matrix is always at the maximal distance (i.e., four) from its transpose. In the present paper we show that the maximal distance cannot be achieved when one of the matrices is derogatory. However, if both A and B are non-derogatory we construct an invertible matrix S so that A and S−1BS are at the distance four. We also show that there exist an infinite collection of matrices, pairwise at the maximal distance. Next, we describe rank-one matrices as the ones which are not at the maximal distance from any derogatory matrix. A similar result classifies semisimple (i.e., diagonalizable) matrices. Our paper concludes with a specific example of matrix algebra over algebraically non-closed field, such that the diameter of its commuting graph is greater than four. Let us briefly recall some standard definitions and notations. Unless ex- plicitly stated otherwise, F is an algebraically closed field of an arbitrary characteristics. Further, Mm,n(F) is the space of m × n matrices over F with a standard basis Eij, and Mn(F) = Mn,n(F) is the matrix alge- bra with identity I. Let e1, . . . , en be the standard basis of column vec- tors in Fn (i.e., of n × 1 matrices). Given an integer k ≥ 2 denote by i=1 Ei(i+1) ∈ Mk(F) the upper-triangular elementary Jor- dan cell with µ on its main diagonal, and let J1(µ) = µ ∈ F. We write shortly Jk = Jk(0). Matrix B is a conjugated matrix of A if B = S−1AS for some invertible matrix S. As usual, Atr is a transpose of A ∈ Mn(F) and rk A its rank. Jk(µ) = µIk +Pk−1 For a matrix algebra Mn(F) over a field F its commuting graph Γ(Mn(F)) is a simple graph (i.e., undirected and loopless), with the vertex set consist- ing of all non-scalar matrices. Two vertices X, Y form an edge X Y if the corresponding matrices are different and commute, i.e., if X 6= Y and XY = Y X. The sequence of successive connected vertices X0 X1, X1 X2, ..., Xk−1 Xk is a path of length k and is denoted by X0 X1 . . . Xk. The distance d(A, B) between vertices A and B is the length of the shortest path between them. The diameter of the graph is the maximal distance between any two vertices of the graph. Given a subset Ω ⊆ Mn(F), let C(Ω) = {X ∈ Mn(F); AX = XA for every A ∈ Ω} If Ω = {A} then we write shortly C(A) = C({A}). be its centralizer. In graph terminology, the set of all non-scalar matrices from the centralizer of A is equal to the set of all vertices X such that d(A, X) ≤ 1. Note that FI ∈ C(A) for any matrix A and that, by a double centralizer theorem, C(C(A)) = F[A] (see [22, Theorem 2, pp. 106] or [16]). We remark that in different articles a centralizer is also called a commutant and is denoted by A′ = C(A). ON MAXIMAL DISTANCES IN A COMMUTING GRAPH 3 A centralizer induces two natural relations on Mn(F). One is the equiv- alence relation, defined by A ∼ B if C(A) = C(B). We call any such two matrices equivalent. The other relation is a preorder given by A ≺ B if C(A) ⊆ C(B). It was already observed that minimal and maximal matrices in this poset are of special importance, see for example [7, 20, 8]. Recall that a matrix A is minimal if C(X) ⊆ C(A) implies C(X) = C(A). It was shown in [20, Lemma 3.2] that the matrix A is minimal if and only if it is nonderogatory, which means that each of its eigenvalue has geometric multi- plicity one, which is further equivalent to the fact that its Jordan canonical form is equal to J = Jn1(λ1) ⊕ · · · ⊕ Jnk (λk), with λi 6= λj for i 6= j. In this case, C(J) = F[Jn1(λ1)] ⊕ · · · ⊕ F[Jnk (λk)] = F[J], where F[X] is an F-algebra generated by X, see [22, Theorem 1, pp. 105] or [12, Theorem 3.2.4.2]. Recall also that a non-scalar matrix A is maximal if C(A) ⊆ C(X) implies C(A) = C(X) or X is a scalar matrix. It is known (see [7, Lemma 4] and also [20, Lemma 3.1]) that a matrix is maximal if and only if it is equal to αI + βP or αI + βN , where P 2 = P is a non-scalar idempotent, N 6= 0 is square-zero (i.e., N 2 = 0), and a scalar β is nonzero. It should be noted that the proof of this fact was done only for the field of complex numbers, but can be repeated almost unchanged in an arbitrary algebraically closed field. 2. Results Throughout this section, with an exception of the last example, F is an algebraically closed field and n ≥ 3. We start with three technical lemmas which will be needed in the sequel. First we observe that every matrix commutes with a rank-one matrix. Lemma 2.1. For every matrix A ∈ Mn(F) there exists a rank-one matrix R ∈ Mn(F) with d(A, R) ≤ 1. Proof. Given any A ∈ Mn(F), it suffices to show that A commutes with at least one matrix of rank one. Since F = F, the matrix A has at least one eigenvalue λ. So, we may assume without loss of generality that A is singular, otherwise we would consider A − λI. Now, let x and y be nonzero vectors in the kernels of A and Atr , respectively. Then, R = xytr is a rank-one matrix with AR = (Ax)ytr = 0 = x(Atr y)tr = RA. (cid:3) Using Lemma 2.1 we can give an alternative proof of the already known fact about the diameter of a commuting graph [4, Corollary 7]. Corollary 2.2. The distance between any two matrices in the commuting graph is at most four. Proof. Let A and B be arbitrary matrices. By Lemma 2.1 there exist rank- one matrices R1 = xf tr ∈ C(A) and R3 = ygtr ∈ C(B). Since n ≥ 3 we 4 DOLINAR, KUZMA, AND OBLAK can find a nonzero z ∈ Fn with f tr z = 0 = gtr z and a nonzero h ∈ Fn with htr x = 0 = htr y. Then for a rank-one matrix R2 = xhtr we obtain A = R0 R1 R2 R3 R4 = B. (cid:3) Lemma 2.3. Let A = Jk1 ⊕ Jk2 ∈ Mk1+k2(F) be a nilpotent matrix with two Jordan cells of sizes k1, k2 ≥ 1. Then d(A, R) ≤ 2 for an arbitrary rank-one R ∈ Mk1+k2(F). Proof. If k1 = k2 = 1 then A is a zero matrix and the conclusion is then imminent. Otherwise, k1 ≥ 2 or k2 ≥ 2. Let k = k1 + k2. It is elementary that the matrix Z = x1E1k1 + x2E1k + x3E(k1+1)k1 + x4E(k1+1)k commutes with A for any choice of x1, x2, x3, x4 ∈ F. Actually, ZA = AZ = 0. Moreover, the matrix Z is non-scalar, except when x1 = x2 = x3 = x4 = 0. Therefore, it suffices to show that for an arbitrary rank-one matrix R there exist x1, . . . , x4 ∈ F, such that at least one of them is nonzero and ZR = RZ = 0. To this end, write R = abtr for some column vectors a = (a1, . . . , ak)tr and b = (b1, . . . , bk)tr . Then ZR = RZ = 0 is equivalent to Za = Z tr b = 0, hence we must solve a homogeneous system of four linear equations (1) x1ak1 + x2ak = 0, x3ak1 + x4ak = 0, x1b1 + x3bk1+1 = 0, x2b1 + x4bk1+1 = 0, x1, x2, x3, x4 unknown. The corresponding matrix of coefficients is equal to and it is easy to check that it is always singular. Therefore the system (1) has a nontrivial solution. This solution defines a non-scalar matrix Z, which commutes with A and R, so d(A, R) ≤ 2 in Γ(Mk1+k2(F)). (cid:3) Lemma 2.4. Suppose A is not minimal. Then d(A, R) ≤ 2 for an arbitrary rank-one matrix R ∈ Mn(F). Proof. Using conjugation we might assume A is already in its Jordan form. Since it is not minimal, hence it is derogatory, at least two Jordan cells contain the same eigenvalue. Let k1, k2 ≥ 1 be their sizes. Define also k = k1 + k2. Moreover, C(A) = C(A − λI) so we may also assume that these two Jordan cells are nilpotent and that A = Jk1 ⊕ Jk2 ⊕ A. It is elementary that C(cid:0)Jk1 ⊕ Jk2(cid:1) ⊕ (FIn−k) ⊆ C(A). Now, let R = xytr be an arbitrary rank-one matrix. Decompose x = x1 ⊕ x2 ∈ Fk ⊕ Fn−k and y = y1 ⊕ y2 ∈ Fk ⊕ Fn−k. We claim that there exists a non-scalar matrix bZ ∈ C(Jk1 ⊕Jk2) satisfying simultaneously bZx1 = λx1 as ak1 ak 0 0 0 b1 0 b1 0 ak1 bk1+1 0 0 ak 0 bk1+1   ON MAXIMAL DISTANCES IN A COMMUTING GRAPH 5 well as bZ tr y1 = λy1 for some λ ∈ F. In fact, this is trivial when x1 = y1 = 0. Otherwise we let x1ytr 1 ; x1, y1 6= 0 e1ytr 1 ; x1 = 0 x1etr 1 ; y1 = 0 ∈ Mk(F), bR = where e1 ∈ Fk is the first vector of the standard basis. By Lemma 2.3 there exists at least one non-scalar matrix bZ ∈ Mk(F) which commutes with bR as well as with Jk1 ⊕ Jk2 ∈ Mk(F). Therefore, if x1, y1 6= 0, then 1 = x1(bZ tr y1)tr and we obtain bZx1 = λx1, and bZ tr y1 = λy1 for some bZx1ytr λ ∈ F. If x1 = 0, then similarly as above bZe1 = λe1, and bZ tr y1 = λy1. Obviously bZx1 = λx1. Likewise we argue if y1 = 0. With the help of bZ we define Z = bZ ⊕λIk ∈ C(Jk1 ⊕Jk2)⊕(FIn−k) ⊆ C(A). Clearly, Zx = bZx1 ⊕ λx2 = λx, and similarly, Z tr y = λy, so Z commutes with R = xytr and with A. Akbari, Mohammadian, Radjavi, and Raja proved in [4, Lemma 2] that, for matrices of size n ≥ 3, the diameter of the commuting graph is at most four (see also Corollary 2.2 above) and that d(J, J tr ) = 4, thus showing that the diameter of the commuting graph of matrix algebra over algebraically closed fields is equal to four. It is well-known [12, p. 134] that the transpose of a matrix is conjugate to the original, so [4, Lemma 2] implies that the maximal distance from J to some of its conjugates is equal to four. Our next lemma will strengthen their result by considering maximal distances between an arbitrary minimal matrix A ∈ Mn(F) and matrices from conjugation orbit {S−1BS; S invertible} of another minimal matrix B ∈ Mn(F). Recall that (cid:3) kLi=1 a minimal matrix is conjugate to Jni(λi), where λi 6= λj for i 6= j, and where (n1, n2, . . . , nk) is a partition of n. We will show below that for any two given partitions of n, we can find two minimal matrices with their Jordan forms corresponding to these two partitions, at distance four. One of the matrices is already in its Jordan canonical form, while the other is a matrix, conjugated to its Jordan canonical form by an invertible matrix with all of its minors nonzero. Such invertible matrix is for example a Cauchy matrix (cid:2) 1 xi−yj(cid:3)ij (see [18]). Theorem 2.5. Let S be any matrix with all of its minors nonzero. For any two minimal matrices A = Mn(F), we have d(A, S−1BS) = 4. kLi=1 Jni(λi) ∈ Mn(F) and B = Jmi(µi) ∈ lLi=1 Proof. Assume erroneously that A and B, as defined in Lemma, are not at the maximal distance, i.e., d(A, S−1BS) ≤ 3. Since C(A) = C(αA) for all nonzero α ∈ F, we can lengthen every path by adding vertices which correspond to scalar multiples of matrices. So, there exists a path 6 DOLINAR, KUZMA, AND OBLAK S−1BS of length 3 in Γ(Mn(F)). We can assume without loss A X Y of generality that X and Y are maximal matrices. Namely, if X is not max- imal, then there exists a maximal X ′ ≻ X, and since A, Y ∈ C(X) ⊆ C(X ′), S−1BS of length 3. Likewise for Y . we could consider a path A X ′ We will show that no two maximal matrices X ∈ C(A) and Y ∈ C(S−1BS) commute and thus obtain a contradiction to the assumption d(A, S−1BS) ≤ 3. Since all maximal matrices are equivalent either to a square-zero matrix or to an idempotent matrix we will consider three cases. Y First, let us assume that both X and Y are square-zero but nonzero. i=1 Jmi(µi) are minimal, so λi 6= λj and Since A =Lk i=1 Jni(λi) and B =Ll µi 6= µj for i 6= j, we have that X = T1 ⊕ T2 ⊕ . . . ⊕ Tk and Y = S−1(T ′ 1 ⊕ T ′ 2 ⊕ . . . ⊕ T ′ l )S, where all Ti and T ′ j are upper triangular Toeplitz matrices. Clearly then Im X = Lin{eσ(1), eσ(2), . . . , eσ(r)} for some permutation σ of length n and integer r, 1 ≤ r ≤ n 2 . Moreover, by the block-Toeplitz structure of X 6= 0 there exist indices t and s such that Xet = αes 6= 0. For the sake of simplicity let us denote T = T ′ l . Now, if Y X = XY , we would have that S−1T SXet ∈ Im X. This would imply, 2 ⊕ . . . ⊕ T ′ 1 ⊕ T ′ αT Ses ∈ S(Im X) = Lin{Seσ(1), Seσ(2), . . . , Seσ(r)} (cid:2) 1 α Seσ(2), . . . , 1 α Seσ(r)(cid:3) is the same as the rank of the augmented which is clearly possible if and only if the rank of the n × r matrix M = α Seσ(1), 1 matrix [M T Ses]. However, we will show that this is not the case. Since all minors of S are nonzero, its s-es column Ses has no zero entries and as such cannot be annihilated by a nonzero block-Toeplitz matrix T . Note that T is also square-zero and so it has at least n 2 zero rows. Recall that r ≤ n 2 , consequently there exists an (r + 1) × (r + 1) submatrix of the augumented matrix, having in the last column exactly r zeros and one nonzero element. By expanding this (r +1)×(r +1) minor by the last column, we observe that it is equal to a multiple of an r × r minor of matrix M which is equal to ( 1 α )r times an r × r minor of S. By the assumption, every minor of S is nonzero and so r + 1 = rk [M T Ses] > rk M = r. This implies T Ses /∈ S(Im X), a contradiction. 2 , otherwise take I − X instead of X. So, X = Pr Second, suppose a non-scalar idempotent X ∈ C(A) commutes with a non-scalar square-zero Y ∈ C(S−1BS). Without loss of generality, r = rk X ≤ n i=1 Eσ(i)σ(i) and Y = S−1T S, where σ and T = T ′ l are as above. Define t = σ(1). Similarly as before, if Y X = XY we would have that S−1T SXet ∈ Im X = Lin{eσ(1), eσ(2), . . . , eσ(r)}, or, equivalently, T Set ∈ Lin{Seσ(1), Seσ(2), . . . , Seσ(r)}. We proceed as in the first case to obtain a contradiction. 2 ⊕ . . . ⊕ T ′ 1 ⊕ T ′ By the symmetry the only case remaining is the case when X and Y are i=1 Eσ(i)σ(i) and Y = S−1P S 2 , since otherwise both non-scalar idempotents. Write X = Pr for P =Ps i=1 Eτ (i)τ (i). Without loss of generality, r, s ≤ n ON MAXIMAL DISTANCES IN A COMMUTING GRAPH 7 we would substitute X by I − X or Y by I − Y . Again, take t = σ(1). If Y X = XY then S−1P SXet ∈ Im X = Lin{eσ(1), eσ(2), . . . , eσ(r)}, or, equivalently, P Set ∈ Lin{Seσ(1), Seσ(2), . . . , Seσ(r)}. Since rk P ≤ n 2 , it follows that the vector P Set has at least n 2 ≥ r and Set is the t-th column of S, so it has no zero entries. This gives P Set 6= 0, a contradiction as in the first case. 2 zero entries. Note that n This shows d(A, S−1BS) ≥ 4. But the diameter of commuting graph is (cid:3) equal to four (see [4, Lemma 2]), hence d(A, S−1BS) = 4. Remark 2.6. The matrix S−1BS from Theorem 2.5 can be rather compli- cated. In a special case, when A is nilpotent we can take A = Jn to achieve that d(Jn, B) = 4 for any companion matrix B in the lower-triangular form. This can be seen by a slight adaptation of the proof of [4, Lemma 2]. For convenience we sketch the main points of the proof. First, it suffices to prove that each maximal D ∈ C(Jn) = F[Jn] satisfies C(B) ∩ C(D) = FI. We may 2 ≤ r ≤ n − 1, is square-zero and hence write it as a 3 × 3 block matrix with block at position (1, 3) being invertible upper-triangular Toeplitz of size (n − r) × (n − r), while all the rest blocks are zero. If Z ∈ C(B) ∩ C(D) then in particular it commutes with D. By direct computation using block-matrix structure we see that the (n, 1) entry of Z is zero. However, Z ∈ C(B) and since companion matrices are non- i=0 λiBi. By considering the images of basis further assume D = Pn−1 i=r diJ i n, n derogatory we have Z = Pn−1 vectors we see that Bi = (cid:20)0i,(n−i) ⋆i,i In−i ⋆(n−i),i(cid:21). Since (n, 1) entry of Z is 0 we see that λn−1 = 0. Proceeding inductively we see that λi = 0 for every i = (n − 1), . . . , 1, whence Z is scalar. By Theorem 2.5 there exist different types of matrices which are at the maximal distance. Next we show that we can find infinitely many matrices which are in the commuting graph pairwise at the maximal distance. Actu- ally, we find an induced graph which is a tree with an internal vertex and all of its leaves at distance two from the internal vertex. Theorem 2.7. There exist an infinite family of matrices (Xα)α ∈ Mn(F) and a rank-one matrix Z such that d(Xα, Xβ) = 4 for α 6= β and d(Xα, Z) = 2 for all α. Proof. We consider three cases separately. Case n = 3. Choose Z = E11 and let the infinite family consist of rank one nilpotent matrices Rα = (0, 1, α)tr (0, α, −1), α ∈ F. It is easy to see that each member commutes with E11 and that the elements of the family are pairwise at distance two. For each index α ∈ F choose a nilpotent Xα such that X 2 α = Rα. Since n = 3 all non-scalar matrices, which commute with Xα are equivalent to Xα or to X 2 α = Rα. Therefore, as d(X 2 β) = 2 for α 6= β, we see that d(Xα, Xβ) = 4 for α 6= β. α, X 2 8 DOLINAR, KUZMA, AND OBLAK Case n = 4. Choose λ ∈ F \ {0, 1}. For nonzero α ∈ F consider rank-one nilpotent matrix Nα = (0, λ, λα, λ)tr (0, −α, 1, 0) and rank-one idempotent Pα = (0, 1, α, 0)tr (0, 1, 0, −1). It is a straightforward calculation that all these matrices are pairwise non-commutative but they all commute with E11, hence (2) d(Nα, Nβ) = d(Pα, Pβ) = d(Nα, Pβ) = 2 for every α 6= β ∈ F \ {0}. Moreover, there exists a conjugation such that α NαSα = E13 and S−1 S−1 α PαSα = E44, for example, take Sα = 1 0 0 0 λ 0 0 −1 αλ 0 1 −α 0 λ 0 0  . Then, for each α we can find a minimal matrix Xα = Sα(J3 ⊕ 1)S−1 Xα ≺ Pα and Xα ≺ Nα. α with We claim that d(Xα, Xβ) = 4. In fact, if a maximal matrix M satisfies M ≻ Xα, then, up to equivalence, either M = Nα, or M = Pα. Hence, if Yα,β Zα,β Xβ would be a path of length three, connecting Xα and Xα Xβ for α 6= β, then we may assume without loss of generality that Yα,β and Zα,β are maximal matrices (see the proof of Theorem 2.5). Hence Yα,β is equivalent either to Nα or Pα and Zα,β is equivalent either to Nβ or Pβ. This contradicts equation (2), so d(Xα, Xβ) ≥ 4 for α 6= β. Observe that one of the paths from Xα to Xβ is Xα Nα E11 Nβ Xβ. Case n ≥ 5. Let A = diag(λ1, . . . , λn) where λi are pairwise distinct. Consider an infinite family of rank one nilpotent matrices Rα indexed by scalars α ∈ F: Rα = R + α R; R = xf tr , R = xgtr , x =" 1 ... 1# , f = 2−n 1 ... 1 0  , g = 1−n 1 ... 1 1  , Note that Sα = I + Rα is invertible with S−1 α Sβ = (I − Rα)(I + Rβ) = I + (β − α) R. Let us define for every α ∈ F the matrix Xα = SαAS−1 α . We will prove that d(Xα, Xβ) = 4 for α 6= β. α = I − Rα, and that S−1 Note first that the distance is invariant for simultaneous conjugation. So, α XβSα) = (A, SAS−1), where we may replace (Xα, Xβ) with (S−1 α Sβ = I + (β − α) R. Now, to prove d(A, SAS−1) = 4 it suffices S = S−1 to show that, given any non-scalar matrices D1 ∈ C(A) and SD2S−1 ∈ S C(A)S−1, they do not commute. α XαSα, S−1 By the choice of minimal A, C(A) consists of diagonal matrices only, hence D1 and D2 are diagonal. Assume erroneously that D1 and SD2S−1 do commute, i.e., that D1(SD2S−1) = (SD2S−1)D1, or equivalently, D1(I + xgtr )D2(I − xgtr ) = (I + xgtr )D2(I − xgtr )D1, ON MAXIMAL DISTANCES IN A COMMUTING GRAPH 9 where x = (β − α)x. Since diagonal matrices commute, we get after expan- sion and simplification (3) (D1 x)(D2g)tr − (D1D2 x)gtr − (gtr D2 x) · (D1 x)gtr = x(D1D2g − (gtr D2 x) · D1g)tr − (D2 x)(D1g)tr . Notice that an eigenvector of a non-scalar diagonal matrix has at least one nonzero entry. Hence, x = (β − α)(1, . . . , 1)tr and g = (1 − n, 1, . . . , 1, 1)tr can not be eigenvectors of a non-scalar diagonal matrix. In particular, g and D2g are linearly independent and so there exists a vector y such that gtr y = 0 and (D2g)tr y = 1. Post-multiplying both sides of equation (3) with y, we now have D1 x = µx + νD2 x, µ = (D1D2g − (gtr D2 x)D1g)tr y and ν = −(D1g)tr y. We infer that (D1 − νD2)x = µx, hence (D1 − νD2) is a scalar matrix because x has all its entries nonzero. Thus, D1 = λI + νD2 for some λ. This simplifies the starting equation D1(SD2S−1) = (SD2S−1)D1 into D(SDS−1) = (SDS−1)D; D = D2, wherefrom also the derived equation (3) simplifies into (4) (Dx)(Dg)tr − (D2 x)gtr − (gtr Dx) · (Dx)gtr = x(D2g − (gtr Dx) · Dg)tr − (Dx)(Dg)tr . By the similar arguments as above we find a vector z such that gtr z = 0 and (Dg)tr z = 1, and continuing along the lines we see that Dx = ((D2g)tr z − (gtr Dx)) · x − Dx. If char F 6= 2 then the above equation implies that x is an eigenvector of a diagonal matrix D which is possible only when D2 = D is scalar, a contra- diction. However, if char F = 2 then we choose a vector, still named z, such that gtr z = 1 and (Dg)tr z = 0. Similarly as above, this simplifies equation (4) into D2 x + (gtr Dx) · Dx = µx; µ = (D2g)tr z. Arguing as above, D2+(gtr Dx)D−µI = 0. Thus, D2 = D, being non-scalar diagonal, has exactly two distinct eigenvalues: d1 and d2 (with multiplicities k and n−k, respectively), because it is annihilated by a quadratic polynomial p(λ) = λ2 + (gtr Dx)λ − µ = (λ − d1)(λ − d2). With no loss of generality we assume that d1 = D1,1. Then, comparing the coefficients in characteristics 2, gives d1 + d2 = (gtr Dx) = (β − α)(cid:0)(1 − n)d1 + (k − 1)d1 + (n − k)d2(cid:1) = (β −α)(n−k)(d1 +d2). Since char F = 2 and D is not a scalar matrix, we can divide by d1 + d2 to obtain (β − α)(n − k) = 1. Observe that in characteristic two, (β − α)(n − k) is either equal to 0 or β − α and since (β − α)(n − k) = 1, we have that β −α = (β −α)(n−k) = 1. and thus β = α+1. Clearly, we can choose an infinite subset of indices A = {0, α1, α1 +α2, α1 +α2 +α3, . . . } ⊂ F such that α − β 6= 1 for α, β ∈ A. For this subset, d(Xα, Xβ) = 4. 10 DOLINAR, KUZMA, AND OBLAK To prove the rest, observe that the rank-one matrix (I + x(f + αg)tr )e1etr 1 (I − x(f + αg)tr ) = SαE11S−1 α commutes with Xα = SαAS−1 α . Now, since n ≥ 5 there exists a nonzero vector w with wtr e1 = 0 = wtr x = wtr f = wtr g. Then, a rank-one matrix Z = wwtr commutes with SαE11S−1 α which gives the path Xα SαE11S−1 α Z. Hence, d(Xα, Z) ≤ 2 for every α. Actually, no shorter path exists, because otherwise, we could join the shorter path for some α with the above path for some other index β to obtain that d(Xα, Xβ) ≤ 3, a contradiction. (cid:3) We next proceed with the classification of matrices which are equivalent to rank-one matrices. In this classification we will need the following lemma. Lemma 2.8. Let n ≥ 4. Suppose A ∈ Mn(F) is (i) either a maximal matrix with 2 ≤ rk A ≤ n − 2, or (ii) a nilpotent matrix with A3 = 0 and rk(A2) = 1. Then there exists a nonminimal matrix X, such that d(A, X) ≥ 3. Proof. (i) As already observed in Preliminaries, a maximal matrix A is either a square-zero matrix or an idempotent, up to equivalence. Let k = rk A. 2 . We define sℓ = (1, 1, . . . , 1)tr ∈ Fℓ 2 ∈ M2ℓ(F). If A is square-zero, then 2 ≤ k ≤ n Note that N 2 i=1 J tr 2ℓ = 0 and rk N2ℓ = ℓ. It is easy to see that a matrix and z2ℓ = (0, 1, 0, 1, . . . , 0, 1)tr ∈ F2ℓ. Also, let N2ℓ =Lℓ  z2k−2 sn−2k+1 0n−2k+1,n−2k+1  N2k−2 01,1 0 0 (5) 0 0 is a square-zero of rank k, hence conjugate to A. So, we can assume without loss of generality that A is already in the form (5). Next, let us define a matrix X = J2 ⊕01 ⊕D, where D is a diagonal matrix with n − 3 distinct nonzero diagonal entries. Clearly, X is nonminimal. We will prove that d(A, X) ≥ 3, i.e., any matrix that commutes with A and X is a scalar matrix. First, let us assume k = 2. It is easy to see that every matrix B ∈ C(X) can be decomposed in the following way (6) B = S1 D′ T S2 01,2 01,n−3 02,1 0n−3,1 λ  where T = (cid:20)a b 0 a(cid:21) ∈ M2(F), D′ = diag(d3, d4, . . . , dn−1) ∈ Mn−3(F), S1 ∈ M2,n−3(F) has the only nonzero entry in its upper left corner, S2 ∈ Mn−3,2(F) has the only nonzero entry in its upper right corner, and λ ∈ F. Note that the blocks in the decomposition of B correspond to the blocks in the decomposition of A. Suppose B ∈ C(X) also commutes with A = ON MAXIMAL DISTANCES IN A COMMUTING GRAPH 11 N2 0 0  0 0n−3,n−3 0 z2 sn−3 01,1 . Then, N2S1 = 0 and S2N2 = 0 imply that S1 = 0 and S2 = 0. Moreover, from D′sn−3 = λsn−3 and T z2 = λz2 we easily see that D′ = λIn−3 and T = λI2. Thus, B = λI, so d(A, X) ≥ 3. Now, let us consider the case k ≥ 3. Again we decompose every B ∈ C(X) to the blocks that correspond to the block decomposition of A: B1 0n−2k+1,2k−2 02k−2,n−2k+1 D′ 01,2k−2 01,n−2k+1 02k−2,1 0n−2k+1,1 λ  B = where B1 =  b a a c1 0 0 0 c2 d3  ⊕ diag(d4, d5, . . . , d2k−2) ∈ M2k−2(F) and D′ ∈ Mn−2k+1(F) is a diagonal matrix. Note that B is the same as in (6) but decomposed in a different way. Suppose B also commutes with A as defined in (5). Similarly as before, we have D′sn−2k+1 = λsn−2k+1 and thus D′ = λIn−2k+1. Moreover, it is straightforward that from B1z2k−2 = λz2k−2 and N2k−2B1 = B1N2k−2 we obtain B1 = λI2k−2. This completes the proof that d(A, X) ≥ 3. If A is an idempotent, then rk(I − A) = n − rk A. Since A and I − A are equivalent, we can thus assume without loss of generality that A is of rank k with n 2 ≤ k ≤ n − 2. Let W be a k × (n − k) matrix with the only nonzero elements being W1,n−k = W2,n−k−1 = W3,n−k−2 = . . . = Wn−k,1 = Wk,1 = Wk,n−k = 1 . Note that, if k = n conjugation we can additionally assume that 2 , the rows k and n − k coincide. Using an appropriate A =(cid:20) Ik 0n−k,k 0n−k,n−k(cid:21) . W Let us define a nonminimal matrix X = Jk ⊕ 01 ⊕ In−k−1. We will prove that d(A, X) ≥ 3, i.e., any matrix B ∈ C(A) ∩ C(X) is a scalar matrix. It is a straightforward calculation that C(A) =(cid:26)(cid:20) M M W − W N 0n−k,k N (cid:21) ; M ∈ Mk(F), N ∈ Mn−k(F)(cid:27) and that C(X) consists of all matrices of the form U vtr s λ 0n−k−1,k 0n−k−1,1  0k,n−k−1 01,n−k−1 Y  , where U is an upper triangular Toeplitz k × k matrix, s = (s1, 0, . . . , 0)tr ∈ Fk, v = (0, . . . , 0, vk)tr ∈ Fk, Y =(cid:2)yij(cid:3)2≤i,j≤n−k ∈ Mn−k−1(F), and λ ∈ F. 12 DOLINAR, KUZMA, AND OBLAK 0 N (cid:21) ∈ C(A) ∩ C(X). It follows that M = , N = λ ⊕ Y and (M W − W N )ij = 0 except possibly for Suppose B = (cid:20)M M W − W N Pk i=1 miJ i−1 i = j = 1. k By equations 0 = (M W − W N )k,1 = m1 − λ, 0 = (M W − W N )k,n−k = m1 − yn−k,n−k, 0 = (M W − W N )i,n−k = mk−i+1 it follows that m1 = yn−k,n−k = λ and m2 = m3 = . . . = m2k−n+1 = 0. Moreover, by 0 = (M W − W N )i,1 = mk−i+1 for i = 2, 3, . . . , n − k − 1, it follows that m2k−n+2 = . . . = mk−1 = 0. Now, equation 0 = (M W − W N )1,n−k = mk completes the proof that M = λIk. for all i with (k − 1) ≥ i ≥ (n − k) We proceed by 0 = (M W − W N )i,n−k−i+1 = m1 − yn−k−i+1,n−k−i+1 for i = 2, . . . , n − k − 1 and 0 = (M W − W N )i,j = −yn−k−i+1,j for i = 1, 2, . . . , n − k − 1 and j = 2, 3, . . . , n − k, such that i + j 6= n − k + 1. It follows that N = λIn−k and (M W − W N ) = 0. Thus, B = λI and d(A, X) ≥ 3. (ii) Let A be a nilpotent matrix such that A3 = 0 and rk(A2) = 1. We may assume A is already in its Jordan canonical form, i.e., A = J3 ⊕ J2 ⊕ 0n−3−2k. kMi=1 (cid:20) T S1 S2 V(cid:21), where T = t0I3 + t1J3 + t2J 2 The centralizer of A is contained in the set of matrices of the form B = 3 ∈ M3(F), V ∈ Mn−3(F), and where the first column of S2 ∈ Mn−3,3(F) as well as the last row of S1 ∈ M3,n−3(F) contain only zero entries. Now, let us define the nonminimal matrix X = 1 ⊕ 0 ⊕ Jn−2 and take any B ∈ C(A) ∩ C(X). Since B ∈ C(X), its off-diagonal entries on the first row and the first column are all zero. Comparing with the above form for B we deduce that T = t0I. Moreover, B ∈ C(X) also implies that the bottom-right (n − 2) × (n − 2) block of B is upper triangular Toeplitz matrix, which is moreover equal to t′ 0 ∈ F by the fact that the third row of S1 vanishes. Actually, t0 = t′ 0 because a 3 × 3 block T overlaps with (n − 2) × (n − 2) bottom right block. Further, B ∈ C(X) implies that the only possible off-diagonal nonzero entries in the second row and column lie at positions (2, n), and (3, 2). Actually, B32 = T32 = 0, while from B ∈ C(A) we deduce that if B2n 6= 0 then also B1(n−1) 6= 0, which would contradict the fact that the first row of B has zero off-diagonal entries. Hence, B2n = B32 = 0 and so B = t0I is a scalar and therefore d(A, X) ≥ 3. (cid:3) 0In−2 for some t′ Theorem 2.9. The following statements are equivalent for a non-scalar matrix R. ON MAXIMAL DISTANCES IN A COMMUTING GRAPH 13 (i) R is equivalent to a matrix of rank one. (ii) d(R, X) ≤ 2 for every nonminimal matrix X. Proof. If n = 3, then every nonminimal matrix is equivalent to a rank-one matrix, so we may assume that n ≥ 4. To prove that (i) =⇒ (ii), we can assume without loss of generality that rk R = 1. Let X be an arbitrary nonminimal matrix. Then d(R, X) ≤ 2 by Lemma 2.4. ¬(i) =⇒ ¬(ii). Suppose R is not equivalent to a rank-one matrix. Note that there exists at least one maximal matrix M ≻ R. In fact, M = p(R) for some polynomial p. Moreover, we can assume that every maximal M ≻ R is either a nonzero square-zero matrix or a non-scalar idempotent. Hence 1 ≤ rk M ≤ n − 1. Note that rk M = n − 1 implies M is an idempotent and therefore it is equivalent to a maximal matrix of rank one. So we can assume that 1 ≤ rk M ≤ n − 2. If for a maximal M ≻ R we have 2 ≤ rk M ≤ n − 2, then by Lemma 2.8 there exists a nonminimal matrix X with d(M, X) ≥ 3. Hence also d(R, X) ≥ 3 because C(R) ⊆ C(M ). Otherwise, every maximal matrix M ≻ R is equivalent to a rank-one matrix. This implies that (i) R is either equivalent to a nilpotent matrix with exactly one Jordan block of dimension 3 and all other cells of dimension at most 2, or (ii) R is equivalent to a matrix whose Jordan structure is equal to 1 ⊕ J2 ⊕ 0n−3, or (iii) R is equivalent to a matrix whose Jordan structure i=1 J2 ⊕ 0n−3−2k. In the first case, Lemma 2.8 assures that there exists a nonminimal X with d(R, X) ≥ 3. In the case (ii) we have, modulo conjugation, R = 1 ⊕ J2 ⊕ 0n−3. It is easy to see that X = J2 ⊕ Jn−2 is nonminimal and d(R, X) ≥ 3. In case (iii) we have, modulo conjugation, i=1 J2 ⊕0n−3−2k. Again, Lemma 2.8 gives a nonminimal (cid:3) is equal to 1 ⊕ J3 ⊕Lk R ≺ R′ = 0⊕J3 ⊕Lk matrix X with d(R′, X) ≥ 3, so also d(R, X) ≥ 3. In the previous theorem rank-one matrices are classified with the help of matrices which are not minimal. We next classify minimal matrices as the ones which maximize the distance in a commuting graph. Theorem 2.10. The following are equivalent for a matrix A ∈ Mn(F). (i) A is minimal. (ii) There exists a matrix X such that d(A, X) = 4. Proof. (i) =⇒ (ii). This follows from Theorem 2.5. ¬(i) =⇒ ¬(ii). Let A be a non-minimal matrix, and let X be any matrix. By Lemma 2.1, there exists a rank-one matrix R with d(X, R) ≤ 1. By Theorem 2.9 we have d(A, R) ≤ 2, so triangle inequality gives d(A, X) ≤ 3. (cid:3) Remark 2.11. Combining the previous two theorems yields that R is equiv- alent to rank-one matrix if and only if d(R, X) ≤ 2 for every matrix X such that d(X, Z) ≤ 3, for all Z ∈ Mn(F). 14 DOLINAR, KUZMA, AND OBLAK Semisimple matrices can also be classified using the distance in the com- muting graph. Before doing that we need two lemmas. Lemma 2.12. Suppose a minimal matrix B ∈ Mn(F) is semisimple. Then for any Y X B there exists a minimal matrix M with Y M X. Proof. Assume with no loss of generality that B is diagonal. Then, every X ∈ C(B) is also diagonal. Using simultaneous conjugation on (B, X) we may further assume that X = λ1In1 ⊕ · · · ⊕ λkInk , with λ1, . . . , λk pairwise distinct and n1, . . . , nk ≥ 1. Now, since Y commutes with X we have that Y ∈ C(X) = Mn1(F) ⊕ · · · ⊕ Mnk (F). Consequently, Y = Y1 ⊕ · · · ⊕ Yk is block-diagonal and we may find an invertible block-diagonal matrix S = i YiSi is in i=1 Jmi (µi), with mi ≥ 1, s ≥ k. Then we can choose distinct ν1, . . . , νs ∈ F, such that the matrix i=1 Jmi(νi)S−1 is neither equal to X nor Y . Also, since ν1, . . . , νs are distinct, M is nonderogatory, hence minimal, and it commutes with X and with Y . (cid:3) S1 ⊕ · · · ⊕ Sk such that S−1XS = X and S−1Y S = Lk Jordan upper-triangular form; say S−1Y S = Ls M = SLs i=1 S−1 Lemma 2.13. Suppose a minimal B ∈ Mn(F) is not semisimple. Then there exist matrices X, Y with Y X B, but such that no minimal matrix commutes with both X and Y . Proof. With no loss of generality assume B is already in its upper-triangular Jordan form, B = Jn1(λ1) ⊕ · · · ⊕ Jnk (λk) with λ1, . . . , λnk distinct and n1 ≥ 2. Define X = E1n1 = J n1−1 n1 ⊕ 0n−n1 ∈ C(B) and for an arbitrary k ∈ {1, . . . , n} \ {1, n1} define Y = E1k. Clearly, X commutes with Y . Let us show that no minimal A commutes with both X and Y . Assume A = j=1 Jnj (µj) is already in its Jordan canonical form. Since X ∈ C(A) is of nj2 −1 nj2 rank one, it follows that X ∈ FJ for some j2. However, rk(X + Y ) = 1 and so j1 = j2, which gives X and Y must be linearly dependent, a contradiction. (cid:3) Ls nj1 −1 nj1 for some j1, and likewise Y ∈ FJ Theorem 2.14. Let A ∈ Mn(F) be a non-scalar matrix. Then the following are equivalent. (i) A is semisimple. (ii) There exists a minimal B ∈ C(A) such that for any Y X B there exists a minimal matrix M with Y M X. Proof. (i) =⇒ (ii). Assume without loss of generality that A is already diagonal. Choose distinct scalars µ1, . . . , µn to form a minimal matrix B = diag(µ1, . . . , µn) which clearly commutes with A. Then, (ii) follows from Lemma 2.12. ¬(i) =⇒ ¬(ii). Choose any minimal B which commutes with non- i=1 Jni(λi)S−1, i=1 Jni(λi)S−1 commute). Since C(B) = F[B] it follows that A ∈ F[B] which implies that semisimple A (at least one does exist, for example, if A = SLk then for distinct scalars λ1, . . . , λk matrices A and B = SLk ON MAXIMAL DISTANCES IN A COMMUTING GRAPH 15 B itself is not semisimple. It now follows from Lemma 2.13 that there exist X, Y with Y X B, but no minimal matrix commutes with both of them. So (ii) does not hold. (cid:3) Let us conclude with an example of a connected commuting graph over algebraically non-closed field with the diameter strictly larger than 4. Example 2.15. The commuting graph for M9(Z2) is connected with diam- eter at least 5. Note that Z2 permits only one field extension of degree n = 9, and this is the Galois field GF (29) which contains GF (23) as the only proper inter- mediate subfield. So, by [2, Theorem 6] the commuting graph of M9(Z2) is connected. To see that its diameter is at least 5, consider an irreducible the field GF (23) contains no idempotents other than 0 and 1 we see that powers of the same irreducible polynomials. Likewise, the field contains of field extensions for finite fields. In the sequel we will identify the two. Since the field extension Z2 ⊂ GF (29) contains only GF (23) as a proper well known Frobenius result on dimension of centralizer (see for example [2, Corollary 1]), and this is a field extension of Z2 [14, Theorem 4.14, pp. 472] polynomial m(λ) = λ9 +λ8 +λ4 +λ2 +1 ∈ Z2[λ] and let bA = C(m) ∈ M9(Z2) be its companion matrix. Since bA has a cyclic vector, C(bA) = Z2[bA] by a of index n = 9. Actually, C(bA) is isomorphic to GF (29) by the uniqueness intermediate subfield, we see that each X ∈ C(bA) \ GF (23) satisfies Z2[X] = Z2[bA] = C(bA) and in particular X and bA are polynomials in each other so they are equivalent. Moreover, each non-scalar bY ∈ GF (23) satisfies Z2[bY ] = GF (23), because no proper intermediate subfields exist between Z2 ⊂ GF (23), and in particular, C(bY1) = C(bY2) for any two non-scalar bY1,bY2 ∈ GF (23) ⊂ GF (29) = C(bA). There exists a polynomial p so that bY = p(bA) ∈ GF (23) \ {0, 1}. As the rational canonical form of bY consists only of cells which correspond to no nonzero nilpotents, so each cell of bY corresponds to the same irreducible Z2, so Z2[bY ] = GF (23) and hence the minimal polynomial of bY ∈ GF (23) derivative, so in a splitting field, bY has three distinct eigenvalues. It easily follows that bY is conjugate to a matrix C ⊕ C ⊕ C, with C being a 3 × 3 invertible matrix such that bY = S−1 Clearly then p(A) = S1bY S−1 C(p(A)) = A = S1bAS−1 Z2[C] Z2[C] Z2[C] . has degree [GF (23) : Z2] = 3. This polynomial is relatively prime to its polynomial, raised to power 1. Moreover, GF (23) has no subfields other that 1 (C ⊕ C ⊕ C)S1 and define 1 . 1 = C ⊕ C ⊕ C and it follows that Z2[C] Z2[C] Z2[C] Z2[C] Z2[C] Z2[C] companion matrix of some irreducible polynomial of degree 3. Let S1 be an (7) 16 DOLINAR, KUZMA, AND OBLAK Since Z2[bY ] = GF (23) we obtain Z2[C] = GF (23). Consider a 3 × 3 block matrix N = 0 0 E13 E13 0 0 E32 0 0  , E13, E13, E32 ∈ M3(Z2). It is immediate that N 3 = 0, so I + N is invertible. Define B = (I + N )A(I + N )−1. We will show that d(A, B) ≥ 5. Suppose there exists a path A V Z W B of length 4. Note that V ∈ GF (23) ⊂ C(A). Otherwise, if V ∈ C(A) \ GF (23) then C(V ) = C(A) and such V has exactly the same neighbours as A. Since B = (I + N )A(I + N )−1, it follows W = (I + N )U (I + N )−1 for some U ∈ GF (23) ⊂ C(A) = (I + N )−1 C(B)(I + N ). Recall that any two non-scalar elements in GF (23) have the same centralizer. So in particular we might take U = V = p(A) = C ⊕ C ⊕ C where polynomial p was defined before. For any Z ∈ C(V ) ∩ C((I + N )V (I + N )−1) we have and hence, by postmultiplying with (I + N ) and rearranging, Z = (I + N )bZ(I + N )−1, Z,bZ ∈ C(V ) (8) Z − bZ = NbZ − ZN. of them is either zero or invertible. Then (8) implies Let us write Z =(cid:2)Zij(cid:3)1≤i,j≤3 and bZ =(cid:2)bZij(cid:3)1≤i,j≤3 as 3 × 3 block matrices and by (7) we have that Zij,bZij ∈ Z2[C] = GF (23) ⊆ M3(Z2), hence each (cid:2)Zij −bZij(cid:3)ij =  . −Z11E13 − Z13E32 + E13bZ11 E13bZ12 E13bZ13 − Z12E13 −Z21E13 − Z23E32 + E13bZ31 E13bZ32 E13bZ33 − Z22E13 −Z31E13 − Z33E32 + E32bZ11 E32bZ12 E32bZ13 − Z32E13 Observe that each block on the left side belongs to Z2[C] = GF (23) ⊆ M3(Z2), and so is either zero or invertible. On the other hand, on the right side, each block in the last two columns has rank at most two. We deduce that the last two columns on both sides are zero. In particular, comparing the second columns we see that bZ12 = Z12 = 0 and bZ32 = 0, so Z22 = bZ22, and Z32 = bZ32 = 0. Putting this in the above equation and simplifying, the last column then gives bZ13 = 0, so Z13 = bZ13 = 0, Z23 = bZ23, and Z33 = bZ33. Also, comparing the (2, 3) positions gives 0 = Z23 − bZ23 = E13bZ33 − bZ22E13 = e1(bZ tr Moreover, bZ tr 33e3 = λe3 and bZ22e1 = λe1, λ ∈ Z2. Since bZ33,bZ22 ∈ Z2[C] and every vector is cyclic for C we see that bZ33 = bZ22 = λI3. The matrix 33e3)tr − bZ22e1etr 3 . ON MAXIMAL DISTANCES IN A COMMUTING GRAPH 17 equation therefore simplifies to  Z11 − bZ11 0 0 Z21 − bZ21 0 0 Z31 − bZ31 0 0  = 0 0 −Z11E13 + E13bZ11 −Z21E13 − bZ23E32 + E13bZ31 0 0 −Z31E13 − λE32 + E32bZ11 0 0  . 0 = (µ − λ)E32 − Z31E13 = (µ − λ)e3etr Z11 = µI3. Inserting this into the equation we see after rearrangement that the rank of the block at position (3, 1) is equal to rk((µ−λ)E32−Z31E13) ≤ 2, Comparing the position (1, 1) gives by similar arguments as above that bZ11 = which forces the two blocks at position (3, 1) to be zero, i.e., Z31 − bZ31 = Z31 = bZ31 = 0 = (µ − λ). Therefore, Z11 = Z22 = Z33 = λI3. Finally, and arguing as above, Z21 = bZ21 = 0. Hence, Z is scalar. So, C(V ) ∩ C(W ) Z21 − bZ21 = −Z21E13 − bZ23E32, contains only scalars, which gives that d(A, B) ≥ 5. 2 − Z31e1etr 3 . We immediately get comparing the (2, 1) positions gives References [1] A. Abdollahi, Commuting graphs of full matrix rings over finite fields, Linear Algebra Appl. 428 (2008), 2947 -- 2954. [2] S. Akbari, H. Bidkhori, A. Mohammadian, Commuting graphs of matrix algebras, Commun. Algebra 36 (2008), 4020 -- 4031. [3] S. Akbari, M. Ghandehari, M. Hadian, A. Mohammadian, On commuting graphs of semisimple rings, Linear Algebra Appl. 390 (2004), 345 -- 355. [4] S. Akbari, A. Mohammadian, H. Radjavi, P. Raja, On the diameters of commuting graphs, Linear Algebra Appl. 418 (2006), 161 -- 176. [5] S. Akbari, P. Raja, Commuting graphs of some subsets in simple rings, Linear Algebra Appl. 416 (2006), 1038 -- 1047. [6] J. Araujo, M. Kinyon, J. Konieczny, Minimal paths in the commuting graphs of semi- groups, European J. of Combinatorics, to appear. [7] G. Dolinar, P. Semrl, Maps on matrix algebras preserving commutativity, Linear and Multilinear Algebra, 52 (2004), 69 -- 78. [8] G. Dolinar, B. Kuzma, General preservers of quasi-commutativity, Can. J. Math. 62 (2010), 758 -- 786. [9] D. Dolzan, P. Oblak, Commuting graphs of matrices over semirings, Linear Algebra Appl., to appear. [10] M. Giudici, A. Pope, The diameters of commuting graphs of linear groups and matrix rings over the integers modulo m, Australas. J. Combin., to appear. [11] L. C. Grove, Algebra. AP, New-York, 1980. [12] R. A. Horn, C. R. Johnson, Matrix Analysis. Cambridge UP, 1985. [13] A. Iranmanesh, A. Jafarzadeh, On commuting graph associated with the symmetric and alternating groups, J. Alg. Appl. 7 (2008), 129 -- 146. [14] K. D. Joshi, Foundations of discrete mathematics. New age international publishers, Bangalore, 1989. [15] D. C. Kleinecke, On operator commutators, Proc. Amer. Math. Soc. 8 (1957) 535 -- 536. [16] P. Lagerstrom, A proof of a theorem on commutative matrices, Bull. Amer. Math. Soc. 51 (1945), 535 -- 536. [17] A. Mohammadian, On commuting graphs of finite matrix rings, Commun. Algebra 38 (2010), 988 -- 994. 18 DOLINAR, KUZMA, AND OBLAK [18] S. Schechter, On the inversion of certain matrices, Math. Tables Aids Comput. 13 (1959), 73 -- 77. [19] Y. Segev, The commuting graph of minimal nonsolvable groups, Geom. Dedicata. 88 (2001), 55 -- 66. [20] P. Semrl, Non-linear commutativity preserving maps, Acta Sci. Math. (Szeged) 71 (2005), 781 -- 819. [21] F. V. Shirokov, Proof of a conjecture by Kaplansky, Uspekhi Mat. Nauk 11 (1956), 167 -- 168 (in Russian). [22] J. H. M. Wedderburn, Lectures on Matrices, American Mathematical Society Collo- quium Publications, Volume XVII, 1934. (Gregor Dolinar) Faculty of Electrical Engineering, University of Ljubl- jana, Trzaska cesta 25, SI-1000 Ljubljana, Slovenia. E-mail address, Gregor Dolinar: [email protected] (Bojan Kuzma) 1University of Primorska, Glagoljaska 8, SI-6000 Koper, Slovenia, and 2IMFM, Jadranska 19, SI-1000 Ljubljana, Slovenia. E-mail address, Bojan Kuzma: [email protected] (Polona Oblak) Faculty of Computer and Information Science, University of Ljubljana, Trzaska cesta 25, SI-1000 Ljubljana, Slovenia. E-mail address, Polona Oblak: [email protected]
1608.08600	1	1608	2016-08-30T19:05:49	A Note On Noeherian Rings	[ "math.RA" ]	In this paper we introduce the definition of a noetherian disjoint ring and that of a noetherian non-disjoint ring . For a noetherian ring R , with nilradical N if P and Q represent the semiprime ideals of R called as the right and the left krull-homogenous parts of N as defined in [8] , then we prove the main theorem of this paper for the ring R whose statement is given below. Main Theorem :- Let R be a Noetherian ring with nilradical N . Let P and Q represent the right and the left krull-homogenous parts of N . Then the following hold true for the ring R ; (a) If R is a disjoint ring , then the nilradical N of R is a right and a left weakly ideal invariant ideal of R . Hence N is a right and a left localizable semiprime ideal of R . (b) If R is a non-disjoint ring then the following are equivalent conditions on R ; (i) N is a right and a left weakly ideal invariant ideal of R . (ii) P = Q is a right and a left localizable semiprime ideal of R .	math.RA	math	A NOTE ON NOETHERIAN RINGS . C.L.Wangneo Jammu,J&K,India,180002 (E-mail:[email protected] ) 1 Abstract:- In this paper we introduce the definition of a noetherian disjoint ring and that of a noetherian non-disjoint ring . For a noetherian ring R , with nilradical N if P and Q represent the semiprime ideals of R called as the right and the left krull- homogenous parts of N as defined in [8] , then we prove the main theorem of this paper for the ring R whose statement is given below. Main Theorem :- Let R be a Noetherian ring with nilradical N . Let P and Q represent the right and the left krull-homogenous parts of N . Then the following hold true for the ring R ; (a) If R is a disjoint ring , then the nilradical N of R is a right and a left weakly ideal invariant ideal of R . Hence N is a right and a left localizable semiprime ideal of R . (b) If R is a non-disjoint ring then the following are equivalent conditions on R ; (i) N is a right and a left weakly ideal invariant ideal of R . (ii) P = Q is a right and a left localizable semiprime ideal of R . Introduction :- In this paper we introduce the definition of a noetherian disjoint ring and that of a noetherian non-disjoint ring . For a noetherian ring R , with nilradical N if P and Q represent the semiprime ideals of R called as the right and the left krull- homogenous parts of N as defined in [8] , then we prove the main 2 theorem of this paper , namely theorem (3.8) for the ring R which states the following . Theorem (3.8) :- Let R be a Noetherian ring with nilradical N . Let P and Q represent the right and the left krull-homogenous parts of N . Then the following hold true for the ring R ; (a) If R is a disjoint ring , then the nilradical N of R is a right and a left weakly ideal invariant ideal of R . Hence N is a right and a left localizable semiprime ideal of R . (b) If R is a non-disjoint ring then the following are equivalent conditions on R ; (i) N is a right and a left weakly ideal invariant ideal of R . (ii) P = Q is a right and a left localizable semiprime ideal of R . The paper is divided into three sections . In section (1) we introduce the preliminaries for this paper . We thus recall the following definitions of ideals A(R) and B(R) of a noetherian ring R , where A(R) is the sum of all right Ideal I of R such that Ir < Rr , and B(R) is the sum of all left ideal K of R with Kl < Rl . We then state some some basic results regarding the ideals A(R) and B(R) . Next we recall for a noetherian ring R certain subsets Λ(R) and Λ'(R) of min. spec.R as follows ; Λ(R)= { Pi in spec. (R) / R/Pir = Rr } and Λ'(R)= { Qi in spec. (R) / R/Q il = Rl } . These definitions allow us to consider the semiprime ideals P and Q of R , where P = ∩ pi ; pi Ɛ Λ and Q = ∩ qi ; qi Ɛ Λ' . We call P and Q as the right and the left krull-homogenous parts of N as defined in [8] . We come up with two obvious cases for the noetherian ring R , namely , Case (i) :- When Λ ∩ Λ' = , and Case (ii) :- When Λ ∩ Λ' ≠ . 3 We then characterize in section(1) the above two cases by conditions equivalent to them . Finally we end section (1) with the following definitions . We call a noetherian ring R a disjoint ring if Λ ∩ Λ' = and we call a noetherian ring R a non-disjoint ring if Λ ∩ Λ' ≠ . In section (2) we study a noetherian disjoint ring and in section(3) we study a noetherian non-disjoint ring . We prove our key theorems of these sections , namely, theorems (2.9) and (3.7) , respectively . Finally in section (3) we combine the key theorems of sections (2) and (3) as theorem (3.8) which we call as our main theorem . Notation and Terminology:- Throughout in this paper a ring is meant to be an associative ring with identity which is not necessarily a commutative ring. We will throughout adhere to the same notation and terminology as in [3] and [8] . Thus by a noetherian ring R we mean that R is both a left as well as a right noetherian ring . By a module M over a ring R we mean that M is a right R-module unless stated otherwise . For the basic definitions regarding noetherian modules and noetherian rings and all those regarding Krull dimension , we refer the reader to [3] . Moreover we will use the following terminology throughout . If R is a ring then we denote by Spec.R , the set of prime ideals of R and by min.Spec.R , the set of minimal prime ideals of R . Again if R is a ring and M is a right R module , then we denote by r(T) the right annihilator of a subset T of M and by l(T) the left annihilator of a subset T of W in case W is a left R module . Also Mr denotes the right Krull dimension of the 4 right R-module M if it exists and Wl denotes the left Krull dimension of the left R module W if it exists . For two subsets A and B of a given we denote by A ≤ B that B contains A and A < B denotes A ≤ B but A≠B . Also for two sets A and B , AȻB denotes that the subset B does not contain the subset A . For an ideal A of R , c(A) denotes the set of elements of R that are regular modulo A . Section (1) (Preliminaries) :- In this section we first recall the following definitions from [ 8 ] . Definition and Notation(1.1) :- Let R be a noetherian ring. Denote by A(R) the sum of all right Ideal I of R such that Ir < Rr . Similarly denote by B(R) the sum of all left ideal K of R with Kl < Rl . If there is no confusion regarding the underlying ring R we may write A and B for A(R) and B(R) respectively. We now state the following basic results regarding the above definitions from [ 8 ] ; Proposition (1.2) :- Let R be a Noetherian ring. Let A and B be as in definition(1.1) above. Then the following hold true ; (i) A is an ideal of R and is the unique largest right ideal of R with Ar < Rr and R/Ar = Rr . (ii) Moreover R/A is a right k-homogenous ring and if A ≠ 0 , then r(A) ≠ 0 and R/r(A)r < Rr . Proof :- For the proof of the above result we ask the reader to see the proof of the proposition (1.3 ) of [ 8 ] . We now state proposition (1.3) below which is similar to proposition (1.2) above and may be considered as its left analogue ; Proposition(1.3) :- Let R be a Noetherian ring. Let A and B be as in definition(1.1) above. Then the following hold true ; (i) B is an ideal of R and is the unique largest left ideal of R with Bl < Rl 5 and R/Bl = Rl . (ii) Moreover R/B is a left k-homogenous ring and if B ≠ 0 , then l(B) ≠ 0 and R/l(B)l < Rl . Next we recall the following definition and notation from [ 8 ] . Definition and Notation (1.4) :- (i) Let R be a Noetherian ring with nilradical R . Define subsets of min. spec. R denoted by Λ(R) and Λ'(R) as follows ; Λ(R)= { Pi in spec. (R) / R/Pir = Rr } and Λ'(R)= { Qi in spec. (R) / R/Q il = Rl } . We define another subset namely X(R) of min.spec.(R) as follows ; X(R) = { l ε min.Spec.(R) / R/lr < Rr , and R/ll < Rl } If there is no confusion about the underlying ring R we set Λ = Λ(R) , Λ' =Λ'(R) and X = X (R) . (ii) It is not difficult to see that if , Pi ε Λ(R), then Pi ε min. spec.(R) . We denote by P(R) the ideal P(R) = ∩ Pi , Pi Ɛ Λ and by Q(R) the ideal Q(R) = ∩ Qi , Qi Ɛ Λ' . Then the ideals P(R) and Q(R) are called the right and the left krull homogenous parts respectively of the nilradical N(R) of the ring R . If there is no confusion regarding the underlying ring R we may write P for P(R) and Q for Q(R) . We may note that in this case the factor rings R/P and R/Q are semiprime right and left krull homogenous rings respectively . We call X the non krull-homogenous part of N . It is also clear that N ≤ P and N ≤ Q . 6 (iii) For an ideal I of R we have ; Λ(R/I)= { Pi/I in spec. (R/I) / I≤ Pi and R/Pir = R/Ir } and Λ'(R/I)= { Qi/I in spec. (R/I) / I≤ Qi and R/Qil = R/Il } . We now recall definition (2.3) of [8] of a right ( or a left ) weakly krull symmetric ring , a weakly krull symmetric and a weakly ideal krull symmetric noetherian ring R and call it as our definition (1.5) given below . Definition (1.5) :- Let R be a Noetherian ring with nilradical N and let Λ , Λ' , A , B be as defined in definitions (1.4 ) and (1.1) above . Let P = ∩ pi ; pi Ɛ Λ and let Q = ∩ qi ; qi Ɛ Λ' . Then we have the following definitions , (i) R is called a right ( or a left) weakly krull symmetric ring if Λ'(R) ≤ Λ (R) ( or Λ(R) ≤ Λ' (R)) . (ii) R is called a weakly krull symmetric ring if Λ(R) = Λ' (R) . (iii) R is called a weakly ideal krull symmetric ring if A = B . Definition (1.6) ( Two cases ) :- We now state and define two obvious cases for a noetherian ring R , namely , Case (i) ( the disjoint case ) :- When Λ ∩ Λ' = , and Case (ii) ( the non-disjoint case ) :- When Λ ∩ Λ' ≠ . We characterise the above two cases by conditions equivalent to them in the theorems stated below ; Theorem (1.7) :- Let R be a Noetherian ring with nilradical N and let Λ , Λ' , A , B be as defined in definitions (1.4) and (1.1) above . Then the following are equivalent ; (a) Λ ∩ Λ' = . (b) R/Al < Rl . (c) R/Br < Rr . 7 Proof :- We first prove (a) => (b) . For this asume (a) , then we show that (b) is true . Suppose not and let R/Al = Rl . Then there exists a prime ideal p ,such that A ≤ p and R/pl = Rl . Thus p ε Λ' ( R ) and so p ε Λ' (R/A ) . Since R/A is right krull homogenous , so by the main theorem of [8] , namely theorem (2.10) of [ 8] , we have that Λ' (R/A) ≤ Λ ( R/A) . Thus we must have that p/A ε Λ ( R/A ) . Hence R/pr = Rr . Thus p ε Λ ( R ) also . Hence Λ ∩ Λ' ≠ , which is a contradiction to (a) . Hence we must have that R/Al < Rl . (b) => (a) We now prove (b) => (a) . If R/Al < Rl , then we show that Λ ∩ Λ' = . Suppose not , then Λ ∩ Λ' ≠ . Let p be a prime ideal such that p ε Λ ∩ Λ' , then using propositions (1.2) and (1.3) above we get that A ≤ p , and B ≤ p . Since R/pl = Rl , this must imply that R/Al = Rl which contradicts (b) . Hence we must have that Λ ∩ Λ' = . This proves (a) < => (b) . (a) < => (c) The proof of (a) < => (c) is similar to the proof of (a) < => (b) given above . A reformulation of the above theorem (1.7) is theorem (1.8) stated below ; Theorem (1.8) :- Let R be a Noetherian ring with nilradical N and let Λ , Λ' , A , B be as defined in definitions (1.4) and (1.1) above . Then the following are equivalent ; (a) Λ ∩ Λ' ≠ (b) R/Al = Rl (it is always true that R/Ar = Rr ) . (c) R/Br = Rr (it is always true that R/Bl = Rl ) . 8 Deinition and Notation (1.9) :- If Λ ∩ Λ' ≠ , then set V(R) = Λ(R) ∩ Λ' (R) and denote by Y(R) the ideal Y(R) = ∩ x / { x ε Λ(R) ∩ Λ' (R)} . Then we have the following theorem regarding the case (2) mentioned above . Theorem (1.10) :- Let R be a Noetherian ring with nilradical N and let Λ , Λ' , A , B be as defined in definitions (1.4) and (1.1) above . If Λ ∩ Λ' ≠ , then the following hold true ; (a) Y is a semiprime ideal of R such that A ≤ Y and B ≤ Y . Hence A+B ≤ Y . (b) Moreover V ( R/A+B) = Λ (R/A+B) = Λ' (R/A+B ) . Proof :- (a) Using propositions (1.2) and (1.3) respectively it is not difficult to see that Y is a semiprime ideal of R such that A ≤ Y and B ≤ Y . Hence A+B ≤ Y . (b) (b) follows easily from the definition of the ideals A , B and subsets Λ ( R/A+B) and Λ' ( R/A+B) of min.spec.(R/A+B) . We finally end section (1) with the following definition . Definition (1.11) :- Let R be a noetherian ring . If A , B and Λ , Λ' are as defined in definitions (1.1) and (1.4) respectively , then we have the following definitions ; (a) We call R a disjoint ring if Λ ∩ Λ' = . (b) We call R a non-disjoint ring if Λ ∩ Λ' ≠ . Section (2) (Noetherian disjoint rings ) :- In this section we study a noetherian disjoint ring and we prove our key theorem of this section , namely theorem (2.9) . To do so we must recall the definitions of the notion of the weak ideal invariance (w.i.i for short ) and that of the localizability of an ideal of a noetherian ring R as stated for example in [2] . This we do below ; 9 Definition (2.1) (Weak ideal invariance) :- Let R be a noetherian ring . If I is any ideal of R we call I a right weakly ideal invariant ideal ( right w.i.i for short ) if for any right ideal J of R with R/Jr < R/Ir , we have that I/JIr < R/Ir . Similarly we define the notion of the left w.i.i of an ideal I of a noetherian ring R . An ideal is said to be w.i.i if it is left w.i.i as well as right w.i.i . Definition (2.2) ( localisability ) :- Let I be an ideal of a ring R . Let C(I) ={ c Ɛ R/ c+I is regular in R/I } . The ideal I is said to be right localisable if the elements of C(I) satisfy the right ore condition ; that is , given c Ɛ C(I) and x Ɛ R , there exists d Ɛ C(I) and y Ɛ R such that xd = cy . An ideal is said to be localisable if it is left and right localisable . Define T(I) ={ x Ɛ R / x c=0 , for some c Ɛ C(I) } . Note that , if I is right localisable , then T(I) is actually an ideal . Remark :- We now state theorem (2.3) below which states that the nilradical N of a noetherian ring R is a right and a left localizable semiprime ideal of R if we assume that N is a right and a left w.i.i ideal of R . Theorem (2.3) :- Let R be a Noetherian ring with nilradical N . If N is right and left weakly ideal invariant ideal of R , then N is a right and a left localizable semiprime ideal of R . Proof :- We assume that N is a right and a left w.i.i semiprime ideal of R . We then show that N is a right and a left localizable semiprime ideal of R . To see this we make the following claim first ; Claim (1) :- For any d ε c(N) , R/dR+Nr < Rr if and only if R/dRr < Rr . 10 Proof :- To prove the above claim we first observe that N is a nilpotent ideal of R . Now given that N is a right w.i.i ideal of R , so apply lemma (3) of [5 ] to conclude that R/dR+Nr < Rr implies that R/dRr < Rr . The converse is clearly easy . Now use the fact that N is a right w.i.i ideal of R together with the proof outlined in the beginning of theorem (8) of [5 ] to conclude that N is a right localizable semiprime ideal of R . That N is a left localizable semiprime ideal follows similarly . Remark:- To the best of the authors knowledge the converse of the above theorem is an open question . We now recall briefly from [3] the following definitions ; Definition (2.4) :- (a) For a finitely generated right R- module M over a noetherian ring R we denote by Ass.M the set of prime ideals of R associated to M on the right . We use a similar notation for a left R-module M causing no confusion . (b) Again from [3] we call a right (or a left) R-module M to be right ( or left ) primary if it has a unique right ( or left) associated prime ideal . We now recall further the following definition from [1] ; Definition (2.5) :- A noetherian ring R is said to have the right large condition if for any essential right ideal J of R we have that R/Jr < Rr With these definitions we now state the following proposition ; 11 Proposition (2.6) :- Let R be a Noetherian ring . Let p be a prime ideal of R such that R/pr = Rr . Let M be a finitely generated critical right R-module M with Mr = Rr and Ass.(M) = p . If r(M) = I , then we have the following ; (i) I ≤ p . (ii) R/I is a right krull homogenous ring with R/Ir = Rr . (iii) R/I is a p/I primary ring such that p/I is a non-essential right ideal of R/I and moreover R/I satisfies the right large condition . (iv) Either I=p or if I ≠ p , then there exists a nonzero ideal J ≠ p , I < J , with J p ≤ I and R/Jr < Rr . Proof :- Use [4] , lemma (1.9) and [1 ] corollary (2.9) for the proof of (i) ,(ii) and (iii) respectively . We will now prove (iv) . To prove (iv) observe that since R/I is a p/I primary ring and since p/I is non- essential as a right ideal of R/I thus there must exist an ideal J of R with I < J and Jp ≤ I . Now two cases arise ; case (i) Either J=R in which case I=p . case (ii) In this case J/I is a proper large right ideal of R/I and hence we must have that R/Jr < Rr . Remark:- The above proposition immediately yields the important lemma (2.7) below which is what we will need throughout this section ; Lemma (2.7) :- Let R be a Noetherian ring . Let p be a prime ideal of R such that R/pr = Rr and R/pl < Rl . Then for any finitely generated critical right R-module M with Mr = Rr and Ass.(M) = p , we have that r( M) = p . Proof :- Let r(M) = I . Clearly I ≤ p . From proposition (2.6) above , we have the following , 12 If I ≠ p ,then there exists a nonzero ideal J ≠ p , I < J , and J p ≤ I , and R/Jr < Rr . Now since R/I is a right krull homogenous ring , thus from theorem (2.10) of [8] we get that Λ' (R/I) ≤ Λ( R/I) . Hence we must have that R/Jl < Rl . Since Jp ≤ I , this yields that pl < Rl , and thus R/pl = Rl , a contradiction to the given hypothesis . Hence , we must have that I = p . We now state Theorem (2.8) below which gives conditions equivalent to the w.i.i of the nilradical of a noetherian ring R . Theorem(2.8) :- Let R be a noetherian ring with nilradical N . Then the following conditions are equivalent ; (i) N is right w.i.i ideal . (ii) For any prime ideal q of R with R/qr = Rr , q is right w.i.i . (iii) For any ideal I of R with R/Ir = Rr , I is right w.i.i . (iv) If M is any finitely generated critical right R-module with Mr = Rr and Ass.(M) =q, then r(M) = q . Proof :- This is proved as theorem (2.5 ) of [ 2 ] . We now prove the key theorem of this section below . Theorem (2.9) :- Let R be a Noetherian ring with nilradical N . If R is a disjoint ring then N is a right and a left weakly ideal invariant ideal of R . Hence N is a right and a left localizable semiprime ideal of R . Proof :- Let Λ , Λ' be as defined in definition (1.4) of section (1) above . If R is a disjoint ring , then we have that Λ ∩ Λ' = . We will first show that N is a right w.i.i ideal of R . Let M be a finitely generated critical right R-module with Mr = Rr . If Ass.(M) = p , then clearly R/pr = Rr . Since Λ ∩ Λ' = , thus we must have that R/pl < Rl . Thus from the above lemma (2.7) , we get that r(M) 13 = p . Hence using theorem (2.8 ) stated above , we get that N is right w.i.i ideal of R . Similarly we can show that N is a left w.i.i ideal of R . That N is a right and a left localizable semiprime ideal of R follows from theorem (2.3 ) stated above . Corollary (2.10) :- Let R be a Noetherian ring with nilradical N (R) . Let R/I be a factor ring of R with R/It = Rr . Then we have either (i) R/Il = Rl or (ii) R/Il < Rl , in which case we must have that N (R/I) is right weakly ideal invariant ideal of R/I . Proof :- If (i) is true ,then there is nothing to prove . So assume (ii) ,namely, that R/Il < Rl . We show in this case that N(R/I) is a right w.i.i ideal of R/I . To see this let M be a finitely generated , critical right R/I- module with Mr = R/Ir = Rr . Let Ass.(M) = p/I , for some prime ideal p of R such that I ≤ p . Clearly R/pr = Rr and R/pl < Rl . Now obviously, we can consider M as a finitely generated ,critical, right R-module , with Ass.(M) = p and Mr = Rr . From lemma (2.7) stated above we get that r(M) =p . Now this is true for any finitely generated, critical right R/I module with Mr = R/Ir . Hence from theorem (2.8 ) above , we must have that N(R/I) is a right w.i.i ideal of R/I . Section (3) (Noetherian non-disjoint rings ) :- In this section we study a noetherian non-disjoint ring . We will prove our key theorem of this section , namely, theorem (3.7) , whose statement is given below ; Theorem (3.7) :- Let R be a noetherian ring with nilradical N . If R is a non-disjoint ring then the following are equivalent conditions on R ; (a) N is right and left w.i.i. ideal of R . 14 (b) P = Q is both a right and left localizable ideal of R . To prove this theorem we recall the following notation from section (1) above . Notation (3.1) :- (i) For a noetherian ring R with nilradical N recall from section (1) above that Λ and Λ' are the subsets of min.spec.R as defined in definition (1.4) . (ii) Next for the noetherian ring R we recall again from section (1) above the definitions of the ideals P and Q of R which we call as the right and the left krull homogenous parts respectively of the nilradical N of the ring R . Remark:- We now state theorem (3.2) below which is the main theorem of [9], namely , theorem (2.10) of [9] . Theorem (3.2) :- Let R be a Noetherian ring with nilradical N and let Λ , Λ' , A , B be as defined in definitions (1.4) and (1.1) of section (1) above . Let P = ∩ pi ; pi Ɛ Λ and Q = ∩ qi ; qi Ɛ Λ' . Then the following hold true ; (a) N is a right w.i.i ideal of R if and only if P is a right localizable ideal of R . (b) N is a left w.i.i ideal of R if and only if Q is a left localizable ideal of R . Remark :- To continue with the proof of our main theorem we recall that we have already stated two obvious cases above for a noetherian ring R , namely , Case (i) :- When Λ ∩ Λ' = , and Case (ii) :- When Λ ∩ Λ' ≠ . We now discuss case (2) ; 15 Definition and notation (3.3) :- Let R be a noetherian ring with nilradical N . Let Λ , Λ' be as in definition (1.4 ) of section (1) above . If R is a non-disjoint ring , that is if Λ ∩ Λ' ≠ , then set V = Λ ∩ Λ' and let Y = { ∩ p / p ε V } .With these notation we have the following theorem ; Theorem(3.4) :- Let R be a Noetherian ring with nilradical N and let Λ , Λ' , A , B be as defined in definitions (1.4) and (1.1) respectively . Let P = ∩ pi ; pi Ɛ Λ and let Q = ∩ qi ; qi Ɛ Λ' . Then the following are true ; (i) If R is right krull homogenous , then Λ' ≤ Λ , and P ≤ Q . (ii) If R is right krull homogenous and N is right w.i.i ideal of R , then N = P . Proof :- (i) Since R is right krull homogenous , so apply the main theorem of [8] to conclude that Λ' ≤ Λ . Clearly then P ≤ Q . (ii) If R is right krull homogenous and N is right w.i.i ideal of R then we prove that N = P . To prove this we first make the following claim ; Claim(1) :- Let q , p be minimal prime ideals of R , with R/pr = R r, and R/qr < R r . Then there exists an ideal B with R/Br < R r , such that pB ≤ qp . Proof :- This is true because N is right w.i.i ideal of R . Proof of (ii) :- We now prove (ii) . To prove (ii) assume A1 A2A3….Am….An = 0 , where Ai are all the minimal prime ideals of R .Then applying claim (1) we get that p1p2…pk D =0 , where each pi Ɛ Λ and D is an ideal such that R/Dr < R r . Hence since R is right krull homogenous we must have that P is a nilpotent ideal of R . Thus N= P . 16 Theorem(3.5) :- Let R be a Noetherian ring with nilradical N and let Λ , Λ' , A , B be as defined in definitions (1.4) and (1.1) respectively . Let P = ∩ pi ; pi Ɛ Λ and let Q = ∩ qi ; qi Ɛ Λ' . If R is right krull homogenous and N is right w.i.i ideal of R , then R is also a left krull homogenous ring . Proof :- To prove the theorem we first observe from theorem (3.4)(i) above that since R is a right krull homogenous ring , hence Λ' ≤ Λ and thus P ≤ Q . Again since R is right krull homogenous and N is right w.i.i ideal of R so using theorem (3.4) (ii) above we get that N = P . We prove now that R is also a left krull homogenous ring . Suppose this is not true . So assume B ≠ 0 . Then by proposition (1.3 ) of section (1) above we have that l(B) ≠ 0 , and R/l(B)l < Rl . Note that by theorem (3.4) (i) above we have that Λ' ≤ Λ and P ≤ Q . Since R/Q is a left krull homogenous semiprime ring with R/Q)l = Rl, so we must have that l(B) Ȼ Q . Since P ≤ Q , so we must have that l(B) Ȼ P as well . Since R/P is a right krull homogenous semiprime ring with R/P)r = Rr , so (l(B) + P)/P ∩ C(R/P) ≠ . It is not difficult to see that this implies that l(B) ∩ c(P) ≠ . Let d ε l(B) ∩ c(P) . Note that by theorem (8) of [5] R has an artinian quotient ring . Hence , since N = P and R has an artinian quotient ring , so using Small's theorem (see [6] ) we get that d ε l(B) ∩ c(0) . Therefore d B = 0 , gives that B = 0 , a contradiction to our original assumption that B ≠ 0 . Thus we must have that B = 0 and hence R is also a left krull homogenous ring . Theorem (3.6) :- Let R be a noetherian ring with nilradical N . Let Λ , Λ' be as in definition (1.4) above . Let Λ ∩ Λ' ≠ . If V = Λ ∩ Λ' 17 let Y = { ∩ p / p ε V } as in definition (3.3) above , then the following hold true ; (a) If N is right w.i.i ideal , then V = Λ , and Y = P is a right localizable ideal of R . (b) If N is left w.i.i ideal , then V = Λ' , and Y = Q is a left localizable ideal of R . Proof :- (a) Consider the factor ring R/A . Then R/A is a right krull homogenous ring with R/A r = R r . Since V ≠ , thus by theorem (1.8) of section (1) above we must have that R/A l = R l . Since N is a right w.i.i ideal , so by theorem (3.5 ) proved above R/A is also a left krull homogenous ring . Hence by the main theorem of [7] , namely, theorem (4) of [7] , we must have that Λ' ( R/A ) = Λ( R/A ) . Moreover , since by proposition (1.2 ) above , Λ (R/A) = Λ (R) , so we get that Λ(R) = Λ' (R/A) . By theorem (1.5 ) of section (1) , since R/A l = R l , so we must have that Λ' (R/A) ≤ Λ'(R) . Thus from the foregoing we finally get that Λ (R) ≤ Λ'(R) . Hence we must have that V = Λ(R) and hence Y= P , where P = { ∩ p / p ε Λ } , which is the right krull homogenous part of N . Thus by the main theorem of [9] , namely theorem (2.10 ) of [9] ( see theorem (3.2) of this paper ) we get that P is right localizable . So Y is a right localizable ideal of R . (b) The proof for (b) is the left analogue of (a) . In this case V = Λ' ( R ) and Y= Q , and hence Y = Q is a left localizable ideal of R . From the above theorem we get the main theorem of section (3) for the case Λ ∩ Λ' ≠ , which we state below ; Theorem (3.7) :- Let R be a noetherian ring with nilradical N . Let R be a non-disjoint ring . Then the following are equivalent 18 conditions on R ; (a) N is right and left w.i.i. ideal of R . (b) P = Q is both a right and left localizable ideal of R . Proof :- (a) => (b) :- Let Λ , Λ' be as in definition (1.4 ) above . If R is a non-disjoint ring , that is if Λ ∩ Λ' ≠ set V = Λ ∩ Λ' and let Y = { ∩ p / p ε V } . Then since N is a right and a left w.i.i of R , so from theorem (3.6) above we have that V = Λ = Λ' , and Y = P = Q is a right and a left localizable ideal of R . (b) => (a) follows from the main theorem of [9] ( see theorem (3.2) of this paper stated above ) . Combining theorem (2.9) and theorem (3.7) of this paper we get theorem (3.8) below which is also the main theorem of this paper ; Theorem (3.8) (Main theorem ) :- Let R be a Noetherian ring with nilradical N . Let P and Q be the semiprime ideals that are the right and the left krull -homogenous parts of N . Then the following hold true ; (a) If R is a disjoint ring , then the nilradical N of R is a right and a left weakly ideal invariant ideal of R . Hence N is a right and a left localizable semiprime ideal of R . (b) If R is a non-disjoint ring then the following are equivalent conditions on R ; (a) N is right and left w.i.i. ideal of R . (b) P = Q is a right and left localizable ideal of R . 19 An interesting feature of a non-disjoint ring is the statement of the following corollary (3.9) given below ; Corollary (3.9) :- Let R be a noetherian ring with nilradical N . Let Λ , Λ' , A and B be as in definitions (1.4 ) and (1.1 ) above . Let Λ ∩ Λ' ≠ . If V = Λ ∩ Λ' and let Y = { ∩ p / p ε V } as in definition (3.3) above , then the following hold true ; (a) N is a right and a left w.i.i. ideal of R implies that R is a weakly ideal krull symmetric ring ( that is A = B ) . (b) A=B implies that Λ = Λ' . Equivalently , this means that a weakly ideal krull symmetric ring is a weakly krull symmetric ring . Proof :- (a) Suppose first that N is a right w.i.i ideal of R . Then theorem (2.6) stated above guarantees that N(R/A) is a right w.i.i and thus theorem (3.5) stated above guarantees that R/A is a left krull –homogenous ring . From theorem (1.9) above we have that R/A l = R l , hence we must have that B ≤ A . Similarly using the left w.i.i of N , we get that A ≤ B . Thus A = B . (b) (b) is the same as the statement of theorem (2.9) of [8] . Remark :- In connection with corollary (3.9) above we mention that converse questions of the statements in (a) and (b) of corollary (3.9) are open questions . In general in the absence of the w.i.i of the nilradical of the ring R of the above theorem (3.7) we have the following result ; Theorem (3.10) :- Let R be a noetherian ring with nilradical N . Let Λ , Λ' be as in definition (1.4 ) above . If Λ ∩ Λ' ≠ , then set V = Λ ∩ Λ' and Y = { ∩ p / p ε V } , then the following hold true ; (i) A+B ≤ Y and V( R/(A+B) = Λ(R/A+B) = Λ'(R/A+B) . 20 (ii) Y is a right localizable semiprime ideal of R if and only if N(R/A+B) , the nilradical of the ring R/A+B is a right w.i.i ideal of the ring R . (iii) Y is a left localizable semiprime ideal of R if and only if N(R/A+B) , the nilradical of the ring R/A+B is a left w.i.i ideal of the ring R . References:- (1) A.K. Boyle, E.H. Feller; Semi critical modules and K-primitive rings, "L.M.N"; No. 700, Springer Verlag (2) K.A. Brown, T.H. Lenagan and J.T. Stafford, "Weak Ideal Invariance and localisation"; J. Lond. math.society (2), 21 , 1980 , 53-61 . (3) K.R. Goodearl and R.B. Warfield ,"An Introduction to Non commutative Noetherain Rings" , L.M.S., student texts ,16. (4) R. Gordon , "some Aspects of Non Commutative Noetherian Rings" , L.N.M.; vol.545; Spg. Vlg. , 1975, 105-127 (5) G.Krause , T.H. Lenagan and J.T. Stafford, " Ideal Invariance and Artinian Quotient Rings "; Journal of Algebra 55, 145-154 (1978) . (6) L.W.Small; Orders In Artinian Rings , J.Algebra 4 (1966) . 13-41. MR 34, #199 . (7) C.L. Wangneo ; On A Certain Krull Symmetry Of a Noetherian Ring. arXiv:1110.3175, v2 [math.RA] , Dec.2011. (8) C.L. Wangneo ; On the weak Krull Symmetry Of a Noetherian Ring. arXiv: 1511.02678v2 [math.RA] , April 2,2016 . (9) C.L. Wangneo ; On some conditions on a noetherian ring arXiv: :1503.00462v5 [math.RA] , April 2 , 2016 21
1307.8247	1	1307	2013-07-31T08:16:30	Distributive lattices and the poset of pre-projective tilting modules	[ "math.RA" ]	D.Happel and L.Unger defined a partial order on the set of basic tilting modules. We study the poset of basic pre-projective tilting modules over path algebra of infinite type. We give an equivalent condition for that this poset is a distributive lattice. We also give an equivalent condition for that a distributive lattice is isomorphic to the poset of basic pre-projective tilting modules over a path algebra of representation infinite type.	math.RA	math	DISTRIBUTIVE LATTICES AND THE POSET OF PRE-PROJECTIVE TILTING MODULES RYOICHI KASE Abstract. D.Happel and L.Unger defined a partial order on the set of basic tilting modules. We study the poset of basic pre-projective tilting modules over path algebra of infinite type. We give an equivalent condition for that this poset is a distributive lattice. We also give an equivalent condition for that a distributive lattice is isomorphic to the poset of basic pre-projective tilting modules over path algebra of infinite type. Introduction Tilting theory first appeared in the article by Brenner and Butler [4]. In this article the notion of a tilting module for finite dimensional algebra was introduced. Tilting theory now appear in many areas of mathematics, for example algebraic geometry, theory of algebraic groups and algebraic topol- ogy. Let T be a tilting module for finite dimensional algebra A and let B = EndA(T ). Then Happel showed that the two bounded derived cate- gories Db(A) and Db(B) are equivalent as triangulated category. Therefore classifying tilting modules is an important problem. Theory of tilting-mutation introduced by Riedtmann and Schofield is one of the approach to this problem. Riedtmann and Schofield defined the tilting quiver related with tilting-mutation. Happel and Unger defined the partial order on the set of basic tilting modules and showed that tilting quiver is coincided with Hasse quiver of this poset. These combinatorial structure are now studied by many authors. notations. Let Q be a finite connected quiver without loops or oriented cycles. We denote by Q0 (resp. Q1) the set of vertices (resp. arrows) of Q. For any arrow α ∈ Q1 we denote by s(α) its starting point and denote by t(α) its target point (i.e. α is an arrow from s(α) to t(α)). Let kQ be the path algebra of Q over an algebraically closed field k. Denote by mod-kQ the category of finite dimensional right kQ modules and by ind-kQ the category of indecomposable modules in mod-kQ. For any module M ∈ mod-kQ we denote by M the number of pairwise non isomorphic indecomposable 2000 Mathematics Subject Classification. Primary 16G20; Secondary 16D80. Key words and phrases. Tilting modules, representations of quivers, distributive lattices. 1 2 RYOICHI KASE direct summands of M . For any paths w : a0 w β2→ · · · β1→ b1 : b0 ′ α1→ a1 α2→ · · · αr→ ar and βs→ bs, α1→ a1 ′ w · w :=( a0 0 α2→ · · · αr→ ar = b0 β1→ b1 β2→ · · · βs→ bs if ar = b0 if ar 6= b0, in kQ. Let P (i) be an indecomposable projective module in mod-kQ asso- ciated with vertex i ∈ Q0. In this paper we will consider the set Tp(Q) of basic pre-projective tilting modules and study its combinatorial structure. In [13] we showed following: Theorem 0.1. If Q satisfies the following condition (C), (C) δ(a) := #{α ∈ Q1 s(α) = a or t(α) = a} ≥ 2 ∀a ∈ Q0, then for any T ∈ Tp there exists (ri)i∈Q0 ∈ ZQ0 ≥0 such that T ≃ ⊕i∈Q0τ −ri Q P (i). Moreover ⊕i∈Q0τ −riP (i) 7→ (ri)i∈Q0 induces a poset inclusion, (Tp(Q), ≤) → (ZQ0, ≤op), where (ri) ≤op (si) def⇔ ri ≥ si for any i ∈ Q0. One of the result of this paper is that Q satisfies the condition (C) if and only if (Tp(Q), ≤) is a distributive lattice. We note that under the condition (C) the poset (Tp(Q), ≤) has inner poset inclusion τ −1 Q . Question 0.2. Let L be a distributive lattice equipped with inner poset in- clusion τ −1. When (L, τ −1) ≃ (Tp(Q), τ −1 Q ) for some Q? As the goal of this paper we will give an answer of this question. Moreover we will construct a quiver Q satisfying (L, τ −1) ≃ (Tp(Q), τ −1 Q ). We now give an outline of this paper. In Section 1 we recall definitions of tilting modules, tilting quivers, lattices and distributive lattices. In Section 2 we define the pre-projective part of tilting quiver and recall results of [13]. In Section 3 we first show that Q satisfies the condition (C) if and only if Tp(Q) is an infinite distributive lattice. Next we give an answer of Ques- tion 0.2. Acknowledgement The author would like to express his gratitude to Professor Susumu Ariki for his mathematical supports and warm encouragements. 1. Preliminary 1.1. Tilting modules. In this sub-section we will recall the definition of tilting modules and basic results for combinatorics of the set of tilting mod- ules. DISTRIBUTIVE LATTICES AND THE POSET OF PRE-PROJECTIVE TILTING MODULES3 Definition 1.1. A module T ∈ mod-kQ is tilting module if, (1) Ext1 (2) T = #Q0. kQ(T, T ) = 0, Remark 1.2. In general, a module T over a finite dimensional algebra A is called a tilting module if (1) its projective dimension is finite, (2) Exti 0 for any i > 0 and (3) there is a exact sequence, A(T, T ) = 0 → AA → T0 → T1 → · · · → Tr → 0, with Ti ∈ add T . If A is hereditary, then it is well-known that this definition is equivalent to our definition. We denote by T (Q) the set of (isomorphism classes of) basic tilting mod- ules in mod-kQ. Definition-Proposition 1.3. [10, Lemma 2.1] Let T, T the following relation ≤ define a partial order on T (Q), ′ ∈ T (Q). Then T ≥ T ′ kQ(T, T ′ def⇔ Ext1 −→ T (Q) is defined as follows: ) = 0. −→ T (Q)0 := T (Q), Definition 1.4. The tilting quiver (1) (2) T → T M ∈ mod-kQ and there is a non split exact sequence, −→ T (Q) if T ≃ M ⊕ X , T in ′ ′ ≃ M ⊕ Y for some X, Y ∈ ind-kQ, 0 → X → M ′ → Y → 0, with M ′ ∈ add M . Theorem 1.5. [9, Theorem 2.1] The tilting quiver the Hasse-quiver of (T (Q), ≤). −→ T (Q) is coincided with Remark 1.6. In this paper we define the Hasse-quiver poset (P, ≤) as follows: −→ (1) P 0 := P , (2) x → y in −→ P if x > y and there is no z ∈ P such that x > z > y. −→ P of (finite or infinite) 1.2. Lattices and distributive lattices. In this subsection we will recall definition of a lattice and a distributive lattice. Definition 1.7. A poset (L, ≤) is a lattice if for any x, y ∈ L there is the minimum element of {z ∈ L z ≥ x, y} and there is the maximum element of {z ∈ L z ≤ x, y}. In this case we denote by x ∨ y the minimum element of {z ∈ L z ≥ x, y} and denote by x ∧ y the maximum element of {z ∈ L z ≤ x, y}. Definition 1.8. A lattice L is a distributive lattice if (x ∨ y) ∧ z = (x ∧ z) ∨ (y ∧ z) holds for any x, y, z ∈ L. Remark 1.9. It is well-known that L is a distributive lattice if and only if (x ∧ y) ∨ z = (x ∨ z) ∧ (y ∨ z) holds for any x, y, z ∈ L. 4 RYOICHI KASE In this paper we use the following notation. Definition 1.10. Let (L1, ≤1) and (L2, ≤2) are posets and φ : L1 → L2 be an order preserving map. (1) We call φ a poset inclusion if φ(x) ≤2 φ(y) implies x ≤1 y. (2) Assume that L1 and L2 are lattices. We call φ lattice inclusion if φ is a poset inclusion and φ(x ∨ y) = φ(x) ∨ φ(y), φ(x ∧ y) = φ(x) ∧ φ(y) holds for any x, y ∈ L1. Definition 1.11. Let L be a lattice. We call an element x ∈ L join- irreducible if x = y ∨ z implies either y = x or z = x. Definition 1.12. Let P be a poset and I ⊂ P . We call I poset-ideal of P if x ≤ y ∈ I implies x ∈ I. Then we denote by I(P ) the poset ({I : poset-ideal of P }, ⊂) and call it the ideal-poset of P . Theorem 1.13. (Birkhoff 's representation theorem, c.f. [3], [7]) Let L be a finite distributive lattice and J ⊂ L be the poset of join-irreducible elements of L. Then L is isomorphic to I(J). 2. Pre-projective tilting modules In this section we will review [13]. Denote by τ = τQ the Auslander-Reiten translation of kQ. First we collect basic properties of the Auslander-Reiten translation. Proposition 2.1. (cf.[1], [2], [6]) Let A = kQ be a path algebra and M, N ∈ ind-A. Then the following assertions hold. (1) If M and N are non-injective modules, then HomA(M, N ) ≃ HomA(τ −1M, τ −1N ). (2) (Auslander-Reiten duality) There is a functorial isomorphism, D HomA(M, N ) ≃ Ext1 A(N, τ M ), where D := Homk(−, k). (3) For any indecomposable non-projective module X and almost split se- quence 0 → τ X → E → X → 0, we get dim Hom(M, τ X) − dim Hom(M, E) + dim Hom(M, X) =(cid:26) 1 X ≃ M 0 otherwise. Definition 2.2. Let Tp(Q). −→ Tp(Q) be a full sub-quiver of −→ T (Q) with −→ Tp(Q)0 = Lemma 2.3. −→ Tp(Q) is coincided with the Hasse-quiver of (Tp(Q), ≤). DISTRIBUTIVE LATTICES AND THE POSET OF PRE-PROJECTIVE TILTING MODULES5 Now we consider the condition, (C) δ(x) := #{α ∈ Q1 s(α) = x or t(α) = x} ≥ 2 for any x ∈ Q0. Let d(X, Y ) := dim Ext1 If Q satisfies the condition (C), then kQ is representation infinite and pre-projective part of its Auslander-Reiten quiver is the translation quiver Z≤0Q (cf.[2]). Let Y = τ −sP (y). Then Proposition 2.1 implies the following. kQ(X, Y ). if (r, x) 6(cid:23) (s, y) if (r, x) = (s, y) if (r, x) ≻ (s, y), d(τ −rP (x), Y ) = Pα:s(α)=x d(τ −rP (t(α)), Y ) +Pα:t(α)=x d(τ −r+1P (s(α)), Y ) −d(τ −r+1P (x), Y ) 0 1   where (r, x) (cid:23) (s, y) means either (i) r > s or (ii) r = s and there is a path from x to y hold. Lemma 2.4. Assume Q satisfies the condition (C). If there is an arrow γ : x → y in Q, then dim Ext1 for any r ≥ 0 and M ∈ mod-kQ. kQ(τ −rP (x), M ) ≤ dim Ext1 kQ(τ −rP (y), M ) ≤ dim Ext1 kQ(τ −r−1P (y), M ). We define a map lQ : Q0 × Q0 → Z≥0 as follows: Let Q be a quiver with Q0 := Q0 and Q1 := Q1` −Q1 where for any arrow α : x → y in α2→ · · · αr→ xr in Q we Q we set −α : y → x. For any path w : x0 put c+(w) := #{i αi ∈ Q1}. Then we set lQ(x, y) := min{c+(w) w : path from x to y in Q}. α1→ x1 Proposition 2.5. If Q satisfies the condition (C), then Ext1 kQ(τ −rP (i), τ −sP (j)) = 0 ⇔ r ≤ s + lQ(j, i) α1→ x1 → · · · αt→ xt = i be a path such that Proof. (⇒): Let w : j = x0 l(j, i) := lQ(j, i) = c+(w) and {k1 < k2 < · · · < kl(j,i)} = {k αk ∈ Q1}. If there exists r > l(j, i) such that Ext1(τ −rP (i), P (j)) = 0, then, by Lemma 2.4, we obtain 0 = d(τ −rP (xt), P (j)) ≥ d(τ −rP (xkl(j,i)), P (j)) ≥ d(τ −r+1P (xkl(j,i)−1), P (j)) ≥ · · · ≥ d(τ −r+l(j,i)−1P (xk1), P (j)) ≥ d(τ −r+l(j,i)P (xk1−1), P (j)) ≥ d(τ −r+l(j,i)P (j), P (j)) ≥ · · · ≥ d(τ −1P (j), P (j)) > 0. Therefore we get a contradiction. In particular if Ext1 0 with r > s + l(j, i), then by Proposition 2.1, we obtain a contradiction. kQ(τ −rP (i), P (j)) 6= 0}. If A(j) 6= ∅, then we take r := min{r (i, r) ∈ A(j) for some i}. Let i ∈ Q0 such that (i, r) ∈ A(j) and (i (⇐) Let A(j) := {(i, r) r ≤ l(j, i), Ext1 kQ(τ −rP (i), τ −sP (j)) = , r) /∈ A(j) for any i ′ ← i in Q. Since ′ 0 < d(τ −rP (i), P (j)) ≤ Xα:s(α)=i d(τ −rP (t(α)), P (j))+ Xβ:t(β)=i d(τ −r+1P (s(β)), P (j)), 6 RYOICHI KASE we obtain (1) dj(t(α)+rn) 6= 0 for some α ∈ s(i) or (2) dj(s(β)+(r −1)n) 6= 0 for some β ∈ t(i). Note that r ≤ l(j, i) ≤ l(j, t(α)) for any α ∈ Q1 with s(α) = i and r − 1 ≤ l(j, i) − 1 ≤ l(j, s(β)) for any β ∈ Q1 with t(β) = i. By the definition of (i, r), we obtain d(τ −rP (t(α)), P (j)) = 0 = d(τ −r+1P (s(β)), P (j)) for any α ∈ Q1 with s(α) = i and β ∈ Q1 with t(β) = i. Therefore we get a contradiction. In particular we obtain A(j) = ∅. Suppose there exists (i, r, s) ∈ Q0 × Z≥0 × Z≥0 such that r ≤ s + l(j, i) and Ext1 kQ(τ −rP (i), τ −sP (j)) 6= 0. If r < s, then Proposition 2.1 shows Ext1 kQ(τ −rP (i), τ −sP (j)) = 0. If r ≥ s, then Proposition 2.1 implies (i, r − s) ∈ A(j). Therefore we obtain a contradiction. (cid:3) We note that in the proof of (⇐) of the above Proposition we did not use the condition (C). In particular we obtain the following Corollary. Corollary 2.6. Let i, j ∈ Q0 and r, s ∈ Z≥0. If r ≤ s + lQ(i, j), then Ext1 kQ(τ −rP (i), τ −sP (j)) = 0. Theorem 2.7. Assume that Q satisfies the condition (C). Then we get followings: (1) Let T ∈ Tp(Q). Then there exists (rx)x∈Q0 ∈ ZQ0 ≥0 such that T = τ −rxP (x). Lx∈Q0 (2) Lx∈Q0 τ −rxP (x) → (rx)x∈Q0 induces both a poset inclusion, Tp(Q) → (ZQ0 ≥0, ≤op) and a quiver inclusion, −→ Tp(Q) → −−−−−−−→ (ZQ0 ≥0, ≤op) . In this case we set Tx := rx for any T ≃Lx∈Q0 Remark 2.8. Assume that Q satisfies the condition (C). We define τ −rxP (x). T (a) := Mx∈Q0 τ −lQ(a,x)P (x) ′ for any a ∈ Q0. Then τ −rT (a) is the minimum element of {Tp(Q) ∋ T ≃ τ −rxP (x) ra ≤ r}. Lx∈Q0 Tp(Q) P (a) ∈ add T }. Let T ≃ Lx∈Q0 Proof. Proposition 2.5 shows that T (a) is a minimum element of {T ∈ τ −rxP (s) ∈ Tp(Q) such that x := max{rx, r + lQ(a, x)} for any x ≥ r for any x ∈ Q0, . ≥ τ −rT (a). (cid:3) ra ≤ r and T x ∈ Q0. It is easy to check that T we have τ rT In particular we obtain τ rT is a basic pre-projective tilting module with P (a) ∈ add T =Lx∈Q0 τ −r ≥ T (a). Therefore we have T ∈ Tp(Q). Since r xP (x) where r ′ ′ ′ ′ ′ ′ ′ ′ DISTRIBUTIVE LATTICES AND THE POSET OF PRE-PROJECTIVE TILTING MODULES7 3. Main results In this section we give our main results. Denote by Q the set of finite con- nected quivers without loops or oriented cycles. First we show the following Theorem. Theorem 3.1. Let Q ∈ Q. Then Tp(Q) is an infinite distributive lattice if and only if Q satisfies the condition (C). Proof. First we assume that Q doesn't satisfy the condition (C). Then one of the following holds, (a) there is a source s in Q such that δ(s) = 1, (b) there is a sink s in Q such that δ(s) = 1. In the case (a), let x be the unique direct successor of s. We denote by I the set of successors of x. Let C := (⊕i∈I τ −2P (i))⊕(⊕i6∈I τ −1P (i)). Then we consider following five modules T := P (s) ⊕ τ −1P (x) ⊕ C, T := τ −2P (s) ⊕ τ −2P (x) ⊕ C, X1 := τ −1P (s) ⊕ τ −1P (x) ⊕ C, X2 := τ −1P (s) ⊕ τ −2P (x) ⊕ C and Y := P (s) ⊕ τ −2P (s) ⊕ C. We note that Corollary 2.6 implies T , T , X1 and X2 are in Tp. We also note that Ext1 kQ(τ −2P (s), P (s)) = 0. Indeed dim Ext1 kQ(τ −2P (s), P (s)) = dim HomkQ(P (s), τ −1P (s)) ′ ′ = dim HomkQ(P (s), τ −1P (x)) − dim HomkQ(P (s), P (s)) −dim HomkQ(P (s), P (x)) = dim HomkQ(P (s),Ly→x P (y)) = dim HomkQ(P (s),Ly→x P (y)) +dim HomkQ(P (s),Lz←x τ −1P (z)) kQ(Lz←x τ −2P (z), P (s)) +dim Ext1 −dim HomkQ(P (s), P (x)) = 0. Therefore we obtain Y ∈ Tp(Q). Since there is the following diagram in −→ Tp(Q), Y Tp is not a distributive lattice. T ′ T X1 X2 plies that Tp is an infinite distributive lattice. Similarly, in the case (b), we obtain that Tp(Q) is not a distributive lattice. Next we assume Q satisfies the condition (C). Then Theorem 2.7 im- Indeed it is easy to check ≃ x}P (x) that for any basic pre-projective modules T ≃ Lx∈Q0 Lx∈Q0 x}P (x) andLx∈Q0 xP (x), bothLx∈Q0 τ −rxP (x) and T τ −max{rx,r τ −min{rx,r ′ τ −r ′ ′ ′ 8 RYOICHI KASE are also basic pre-projective tilting modules (Remark. Let a := (rx)x∈Q0, b = x})x∈Q0 and (r a ∧ b = (max{rx, r ≥0. Then it is obvious that a ∨ b = (min{rx, r x})x∈Q0 in the distributive lattice (ZQ0, ≤op).). ′ ′ x)x∈Q0 ∈ ZQ0 ′ Example 3.2. We give three examples of (1) Let Q be the quiver: −→ Tp(Q). (cid:3) Then −→ Tp(Q) is given by the following: (2) Let Q be the quiver: DISTRIBUTIVE LATTICES AND THE POSET OF PRE-PROJECTIVE TILTING MODULES9 Then −→ Tp(Q) is given by the following: (3) Let Q be the quiver: Then −→ Tp(Q) is given by the following: 10 RYOICHI KASE Lemma 3.3. Assume that Q satisfies the condition (C). Then the set of join-irreducible elements of Tp(Q) is {τ −rT (a) a ∈ Q0, r ∈ ZQ0 ≥0}. Proof. Theorem 2.7 implies T ∈ Tp(Q) is join-irreducible if and only if there −→ Tp(Q). Let T be a join irreducible is the unique direct successor of T in element and T a = Ta +1. Without loss of generality, we can assume that Tx = 0 for some x ∈ Q0. Then it is suffice to show that T = T (a). be the its direct successor. Let a ∈ Q0 such that T ′ ′ We now define a partial order ≤Q on Q0 as follows: x ≤Q y def⇔ there exists a path from x to y. Let b ∈ Q0 be a minimal element of {y ∈ Q0 Ty = 0}. Then T Lx6=b τ −TxP (x) ⊕ τ −1P (b) ∈ Tp(Q). Therefore we obtain Ta = 0. In partic- ular T ≥ T (a). Suppose that T > T (a). Then there is a path ′′ := T → T 1 → T 2 → · · · → T r = T (a). Since Ta = 0 = T (a)a, we have T 1 6= T ′ . We now get a contradiction. (cid:3) Definition 3.4. We define a poset J = J(Q) as follows: • J = Z≥0 × Q0 as a set. • (r, a) ≤ (s, b) def⇔ lQ(a, x) + r ≥ lQ(b, x) + s for any x ∈ Q0. We set T (j) := τ −rT (x) for any j = (r, x) ∈ J. Note that j1 ≤ j2 ⇔ T (j1) ≤ T (j2). Corollary 3.5. Assume that Q satisfies the condition (C). Then a map ρ : I(Q) \ {∅} ∋ I 7→Wi∈I T (i) ∈ Tp(Q) induces a poset isomorphism I(Q) \ {∅} ≃ Tp(Q), where I(Q) be a ideal-poset of J(Q). ′ Proof. Let I ∈ I(Q) \ {∅}. Then it is easy to check that there is a finite subset {i1, · · · im} of I such that I = {j ∈ J j ≤ it for some t}. Then ∈ I(Q) \ {∅} with ρ(I) ≤ ρ(I Tp(Q) well-defined. I, I Wi∈I T (i) =Wm Wi∈I T (i) ≤ Wi∈I t=1 T (it). In particular a map ρ : I(Q)\{∅} ∋ I 7→Wi∈I T (i) ∈ It is obvious that ρ is an order-preserving map. Let ) and (r, x) ∈ I. Then T ((r, x)) ≤ . Since T ((r, x)) is the minimum element of {T ∈ Tp(Q) Tx ≤ r}, we obtain T ((r, x)) ≤ T (i . In particular we obtain (r, x) ∈ I ). Therefore we have (r, x) ≤ i , x) ≤ r for some i ′ T (i) implies r + lQ(x := (r ) ∈ I , x ′ . ′ ′ ′ ′ ′ ′ ′ ′ ′ If ρ(I) = ρ(I We show that ρ is bijection. ′ ⊂ I. Therefore ρ is injection. Let T ∈ Tp(Q). Then it is easy to check that ′ x ≤ Tx}. Therefore we obtain T ≥ τ −TxT (x) for any x ∈ Q0. In particular T τ −TxT (x))a ≤ Ta for any τ −TxT (x). Indeed τ −TxT (x) is the minimum element of {T ), then I ⊂ I and I ′ ′ ′ T = Wx∈Q0 we have T ≥ Wx∈Q0 τ −TxT (x). Since (Wx∈Q0 DISTRIBUTIVE LATTICES AND THE POSET OF PRE-PROJECTIVE TILTING MODULES11 a ∈ Q0, we obtain T ≤ Wx∈Q0 I := {j ∈ J j ≤ (Tx, x) for some x ∈ Q0}. In particular ρ is bijection. τ −TxT (x). Therefore we obtain ρ(I) = T for (cid:3) Lemma 3.6. For j = (r, a) ∈ Z≥0 × Q0, set P (j) := τ −rP (a). Then j1 ≤ j2 if and only if there is a path from P (j2) to P (j1) in Auslander-Reiten quiver of kQ. Proof. Let j1 = (r, a) and j2 = (s, b). First we assume that there exists an arrow P (j2) → P (j1) in Γ(kQ). Then we have (1) a → b in Q and r = s or (2) b → a and r = s + 1. In both of two cases, we have lQ(a, x) + r ≥ lQ(b, x) + s for any x ∈ Q0. Next we assume that j1 ≤ j2 and let t := lQ(b, a). Then we have r ≥ s + t. By definition of lQ, we can take a sub-quiver b ← · · · ← b1 → a1 ← · · · ← b2 → a2 · · · bt → at ← · · · ← a of Q. In particular we obtain a path P (j2) → · · · → τ −sP (b1) → τ −s−1P (a1) → · · · → τ −s−tP (at) → · · · → τ −s−tP (a) in Γ(kQ). Now r ≥ s + t implies that there is a path from τ −s−tP (a) to τ −rP (a) = P (j1) in Γ(kQ). (cid:3) Definition 3.7. For any acyclic quiver Γ, we define a poset P(Γ) as follows: • P(Γ) = Γ0 as a set. • x ≤ y if there is a path from y to x in Γ. Corollary 3.8. Let Γp(Q) be the pre-projective component of Auslander- Reiten quiver of kQ. Then the poset Tp(Q) is isomorphic to I(P(Γp(Q))) \ {∅}. Example 3.9. Let Q be a quiver: Q : 0 1 2 12 RYOICHI KASE Then I(P(Γp(kQ))) \ {∅} is given by the following: Let L be an infinite distributive lattice with the maximum element o and −→ L . τ −1 be a lattice inclusion L → L which induces a quiver inclusion −→ L → Definition 3.10. Let ∼ be an equivalence relation on following: (a)α ∼ τ −1α. (b)α ∼ β if there is a full sub-quiver S(α, β) of −→ L . −→ L 1 generated by the α s(α) t(α) S(α, β) Then we put Λ = Λ(L, τ −1) := s(β) t(β) −→ L 1/∼. β DISTRIBUTIVE LATTICES AND THE POSET OF PRE-PROJECTIVE TILTING MODULES13 Let P := L \ τ −1L. Proposition 3.11. Assume that (L, τ −1) satisfies the following conditions, (c0) P 6= ∅ is finite. (c1) L =`r≥0 τ −rP . (c2) x 6< τ −ry for any x, y ∈ P and r > 0. (c3) x → y implies y ≥ τ −1x. (c4) For any x ∈ P0 there exists a path w : x that {αi}1≤i≤r is a minimal representable of Λ. Then the following assertions hold. α1→ x1 α2→ · · · αr→ τ −1x in −→ L such −→ L . In particular −→ L ′ , λ). α1→ x1 If s(w) = s(w α2→ · · · αr→ xr in (1) If x > y in L, then there is a path from x to y in is a connected quiver. (2) For any path w : x0 φ(w, λ) := #{i αi/∼ = λ}. φ(w, λ) = φ(w (3) For any x ∈ L we put φ(x) := (φ(w, λ))λ∈Λ, where w is a path from o −→ L to the Hasse-quiver of the to x. Then φ induces a quiver inclusion from poset (ZΛ (4) Let x, y ∈ L. Then x < y if and only if φ(x) ≤op φ(y) (5) Let x, y ∈ L and λ ∈ Λ. Then φ(x ∨ y)λ = min{φ(x)λ, φ(y)λ} and φ(x ∧ y)λ = max{φ(x)λ, φ(y)λ}. −→ L and λ ∈ Λ we put ), then ) and t(w) = t(w ≥0, ≤op). ′ ′ −→ L (x) is a full sub-quiver of Proof. Let L(x) := {y ∈ L0 y ≥ x}. We note that L(x) is a distributive −→ lattice and its Hasse-quiver L . We claim that Indeed the condition (c1) implies there exists r ≥ 0 such L(x) is finite. i=0 τ −iP . Therefore the condition (c0) implies L(x) is finite. In particular L(x) is a finite distributive lattice for any x ∈ L. (1) Let x, y ∈ L with x > y. Since x ∈ L(y), there is a path from x to y in that x ∈ τ −rP and then the condition (c2) implies L(x) ⊂ `r −→ L (y) is a finite full sub-quiver of −→ L and β1→ x (2) Let w : x prove with the using of an induction on l(w) = l(w α1→ x1 : x ′ ′ ′ 1 ) = r. β2→ · · · βr→ x ′ r = y. We −→ L . α2→ · · · αr→ xr = y and w (r = 1) In this case the assertion is obvious. (r > 1) Without loss of generality, we can assume x1 6= x ′ 1. Put s = min{i xi+1 ≤ x ′ 1} then xs ∧ x ′ 1 = xs+1. We put if i ≤ s if i ≥ s + 1 1 ′ x ′′ i :=(cid:26) xi−1 ∧ x xi Then we get the following diagram, α1 x γ1 x ′ 1 = x ′′ 1 γ2 x1 x ′′ 2 α2 γ3 x2 x ′′ 3 xs−1 αs xs αs+1 x ′′ s+1 x ′′ s γs+1 x ′′ s+2 x ′′ r = y 14 RYOICHI KASE −→ L . We consider a path w in pothesis of induction we get φ(w it is sufficient to show φ(w, λ) = φ(w : x ′′ ′ γ1→ x ′ 1 = x ′′ , λ) = φ(w ′′ γ2→ · · · γr→ x ′′ r = y. By hy- 1 , λ) for any λ ∈ Λ. Therefore ′′ , λ). By the definition of ∼, we get γi+1 γ1 γi if i ≤ s if i = s + 1 if i ≥ s + 2. αi ∼  Therefore we obtain φ(w, λ) = φ(w (3) First we show that φ is injective. Let w : x ′′ , λ) for any λ ∈ Λ. α1→ x1 α2→ · · · αr→ xr, w −→ L . We assume (φ(w, λ))λ∈Λ = (φ(w ′ ′ : x β1→ , λ))λ∈Λ. ′ β2→ · · · ′ βr→ x 1 x Then we show xr = x r are paths in ′ r with the using of an induction on r. ′ ′ (r = 1) We note that the condition (c3) implies x1 ∧ x 1 ≥ τ −1x. Indeed, 1) is a finite distributive lattice and there are arrows x → −→ L (x1 ∧ x 1 or there are 1 = x −→ 1, x L ). Therefore if 1, then φ(p, α1/∼) ≥ 2 for any path p from x to τ −1x. This contradict 1), we obtain either x1 = x1 ∧ x 1 → x1 ∧ x 1) (i.e. in since L(x1 ∧ x x1, x → x 1 in arrows x1 → x1 ∧ x x1 6= x to the condition (c4). −→ L (x1 ∧ x 1 in ′ ′ ′ ′ ′ ′ ′ ′ ′ (r > 1) Without loss of generality, we can assume α1 6∼ β1. Let s := min{i αs ∼ β1}. Then there is an arrow xi note that α Now we take a path s−1 ∼ · · · ∼ α s ∼ α ′ ′ ′ 2 ∼ β1 ∼ αs. Therefore we get xs−1 ∧x 1 for any i < s. We 1 = xs. ′ ′ α i+1→ xi ∧ x ′ ′′ w : x β1→ x ′ 1 γ2→ x1 ∧ x ′ 1 γ3→ x2 ∧ x ′ 1 γ4→ · · · γs→ xs αs+1→ xs+1 αs+2→ · · · αr→ xr, with γi ∼ αi−1 (2 ≤ i < s). Since (φ(w (φ(w , λ))λ∈Λ, we obtain xr = x ′ ′ r (we use a hypothesis of an induction). ′′ , λ))λ∈Λ = (φ(w, λ))λ∈Λ = By applying (2) of this Proposition, we obtain that φ is injective. Now the assertion follows from the definition of φ. (4) We only show that φ(x) >op φ(y) implies x > y. Let x, y ∈ L with α2→ φ(x) >op φ(y). Suppose x 6> y then there are two paths x ∨ y βt→ yt = y. Then φ(x) >op φ(y) implies · · · αs→ xs = x and x ∨ y I := {i βi ∼ α1} 6= ∅. We put i := min I. Then we obtain following diagram, α1→ x1 β2→ · · · β1→ y1 β1 x ∨ y y1 β2 y2 α1 yi−1 γ y ′ l := yl−1 ∧ x1 (l = 1, 2 · · · i) x1 = y ′ 1 y ′ 2 y ′ 3 y ′ i −→ L . Since γ ∼ α we get x1 ∧ yi−1 = yi. Therefore we obtain x ∨ y > x1 ≥ in yi ≥ y and x1 ≥ x. We now get a contradiction. DISTRIBUTIVE LATTICES AND THE POSET OF PRE-PROJECTIVE TILTING MODULES15 (5) Since x ∨ y ≥ x, y there are paths, x ∨ y α1→ x1 α2→ · · · αs→ xs = x and x ∨ y β1→ y1 β2→ · · · βt→ yt = y. Suppose I := {i αi ∼ βk for some k} 6= ∅. Let i := min I, j := min{k αi ∼ βk} and λ := αi/∼. We claim xi ∧ yj−1 = xi−1 ∧ yj. Since xi−1 ∧ yj ≤ xi−1, xi−1 ∧ yj ≤ yj, we obtain φ(xi−1 ∧ yj)λ′ ≥(cid:26) φ(xi−1)λ + 1 if λ φ(xi−1)λ if λ ′ ′ = λ ′ 6= λ. Above inequalities imply φ(xi−1 ∧ yj) ≤op φ(xi). Therefore we obtain xi−1 ∧ yj ≤ xi. In particular we get xi−1 ∧ yj ≤ xi ∧ yj−1. Similarly we obtain xi ∧ yj−1 ≤ xi−1 ∧ yj. Since yj−1 = (x ∨ y) ∧ yj−1 = (xi ∨ yj) ∧ yj−1 = (xi ∧ yj−1) ∨ (yj ∧ yj−1) = (xi−1 ∧ yj) ∨ yj = yj, we get a contradiction. In particular I = ∅. Since φ(x ∨ y) ≥op (min{φ(x)λ, φ(y)λ})λ, it is sufficient to show that φ(x ∨ y) 6>op (min{φ(x)λ, φ(y)λ})λ. If φ(x ∨ y) >op (min{φ(x)λ, φ(y)λ})λ, then there exists λ ∈ Λ such that φ(x ∨ y)λ < min{φ(x)λ, φ(y)λ}. This implies there exists (i, j) such that αi/∼ = λ = βj/∼. In particular I 6= ∅. We obtain a contradiction. (cid:3) Corollary 3.12. Assume that (L, τ −1) satisfies the conditions (c0) ∼ (c4). Then a map φ defined in Proposition 3.11 induces a lattice inclusion L → (ZΛ, ≤op). Lemma 3.13. Assume that Q satisfies the condition (C). Then (Tp(Q), τ −1 Q ) satisfies the conditions (c0) ∼ (c4). Moreover we get Λ = Q0 and φ(T ) = (rx)x∈Q0 for a basic pre-projective tilting module T ≃ ⊕x∈Q0τ −rxP (x). Proof. For any T ≃ ⊕x∈Q0τ −rxP (x), we set Tx := rx. For any arrow α : T → T −→ Tp(Q) we set v(α) ∈ Q0 such that T v(α) = Tv(α) + 1. in ′ ′ First we will show that α ∼ β if and only if v(α) = v(β). Let α : T → T Q α or there is . We note that v(α) = v(β) if either β = τ −1 ′′ → T ′′′ ′ and β : T the diagram, S(α, β) α ′ T T ′′ T ′′′ T β RYOICHI KASE 16 in −→ Tp(Q). Therefore α ∼ β implies v(α) = v(β). We assume v(α) = v(β) = x. We show α ∼ β. At first we assume Tx = T x −→ Tp(Q). : T = Y 0 → (1 ≤ i ≤ r). Now we and T ≥ T Since v(X i−1 → X i) 6= x for any i > 0, we obtain a path w Y 1 → Y 2 → · · · → Y r = T note that there is a diagram, . Let w : T = X 0 → X 1 → · · · → T r = T , where Y i := X i−1 ∧ T be a path in ′′′ ′′ ′′ ′′ ′ ′ αi−1 X i−1 Y i−1 (∀i > 0) αi In particular we obtain α ∼ β. Xi Y i Next we show α ∼ β in arbitrary case. Since α ∼ τ −max{0,T Q ′′ x −Tx} α and ′′ x } Q β ∼ τ −max{0,Tx−T ′′ → T γ : T ∧ T ′′ ≤ T T ≥ T ∧ T ′′ β, we can assume that Tx = T ′′′ ′ ∧ T , we obtain α ∼ γ ∼ β. with v(γ) = x. Since (T ∧ T ′′ x . Then there is an arrow x and )x = Tx = T ′′ ′ Therefore v induces Q0 ≃ Λ. In particular, we obtain φ(T )x = Tx. Now, Q ) satisfies the by applying Theorem 2.7, we can easily check that (Tp(Q), τ −1 conditions (c0) ∼ (c4). (cid:3) From now on we assume that (L, τ −1) satisfies the conditions (c0) ∼ (c4) in Proposition 3.11. Put Λ := Λ(L, τ −1). Then we can identify L with its −→ L . Indeed Proposition 3.11 (1) shows that x > y in L if and Hasse-quiver −→ only if there exists a path from x to y in L . Moreover, by Proposition 3.11 and Corollary 3.12, we can regard L as a sub-lattice of (ZΛ, ≤op). Then for x ∈ L and λ ∈ Λ we denote by xλ the λ-th entry of x. We note that (τ −1x)λ = xλ + 1 for any x ∈ L and λ ∈ Λ. Now we define τ x := (xλ − 1)λ∈Λ ∈ L for any x ∈ τ −1L. Lemma 3.14. Let λ ∈ Λ. Then there is the minimum element x(λ) of the set L(λ) := {x ∈ L0 xλ = 0}. Moreover, λ 7→ x(λ) induces an inclusion Λ → P0. Proof. Since xλ ≥ 1 for any x ∈ L0 \ P0, we obtain #L(λ) < ∞. Therefore we can take x(λ) := ∧x∈L(λ)x. Now we assume x(λ1) = x(λ2). Let α be an arrow x(λ1) α→ y in L. Since yλ1 6= 0 and yλ2 6= 0, we obtain λ1 = α/∼ = λ2. Remark 3.15. τ −rx(λ) is the minimum element of {x ∈ L xλ ≤ r}. Indeed for any x ∈ L with xλ = s ≤ r, we have x λ = 0. := τ −so∧x ∈ τ −sP and τ sx (cid:3) ′ ′ DISTRIBUTIVE LATTICES AND THE POSET OF PRE-PROJECTIVE TILTING MODULES17 Therefore τ sx ′ ≥ x(λ). In particular we have ′ x ≥ x ≥ τ −sx(λ) ≥ τ −rx(λ). For any λ ∈ Λ we denote by y(λ) the unique direct successor of x(λ). Definition-Lemma 3.16. We can define a partial order ≤ on Λ as follows: λ1 ≤ λ2 def⇔ xλ1 ≥ xλ2 ∀x ∈ L. Proof. It is obvious that (1) λ ≤ λ for any λ ∈ Λ and (2) λ1 ≤ λ2 ≤ λ3 implies λ1 ≤ λ3. Therefore it is sufficient to show that (3) λ1 ≤ λ2 ≤ λ1 implies λ1 = λ2. Since λ1 ≤ λ2, we get x(λ1)λ2 ≤ x(λ1)λ1 = 0. Therefore we obtain x(λ1) ≥ x(λ2). Similarly λ2 ≤ λ1 implies x(λ2) ≥ x(λ1). Then Lemma 3.14 shows λ1 = λ2. (cid:3) Lemma 3.17. Let λ1, λ2 ∈ Λ. Then λ1 ≤ λ2 if and only if x(λ1)λ2 = 0. Proof. First we assume that λ1 ≤ λ2. Then we get x(λ1)λ2 ≤ x(λ1)λ1 = 0. Next we assume that there exists x ∈ L such that xλ1 < xλ2. We consider an element c := τ xλ1 (τ −xλ1 o ∧ x) ∈ P (note that τ −xλ1 o ∧ x ∈ τ −xλ1 P ). Since cλ1 = 0 and cλ2 = xλ2 − xλ1 > 0, we obtain 0 < cλ2 ≤ x(λ1)λ2. (cid:3) Definition 3.18. Let (L, τ −1) be a pair satisfying the conditions (c0) ∼ (c4). We define a quiver Q = Q(L, τ −1) having Λ as the set of vertices as follows: We draw an arrow λ → λ ′ in Q if x(λ)λ′ = 1 and x(λ ′ )λ = 0. that there is an edge λ − λ in G(Λ, τ −1) if and only if x(λ)λ Now we denote by G(λ, τ −1) the underlying graph of Q(λ, τ −1). Note )λ = 1. Definition 3.19. Let (L, τ −1) be a pair satisfying the conditions (c0) ∼ (c4). (L, τ −1) having Λ as a set of vertices as follows: We define a graph G in G in the underlying graph of the Hasse quiver of = G We draw an edge λ − λ (1) there is an edge λ − λ (Λ, ≤). (2) there is an arrow α ∈ L1 such that s(α) = y(λ) and α/∼ = λ . (3) there is an arrow β ∈ L1 such that s(β) = y(λ ) and β/∼ = λ. if one of the following hold. ′ ′ + x(λ ′ ′ ′ ′ ′ ′ ′ ′ For any quiver Q ∈ Q we define a quiver Q satisfying the condition (C) as follows: (1) Q0 = Q0. (2) For any pair (x, α) ∈ Q0 × Q1 with δ(x) = 1 and α being an edge satisfying either s(α) = x or t(α) = x, draw new edge αc : s(α) → t(α) in Q. For example, if we consider the following quiver Q: then Q is given by the following: 18 RYOICHI KASE Now we give an necessary and sufficient condition for (L, τ −1) being iso- Q ) (i.e. there exists poset isomorphism ρ : L ≃ Tp(Q) Q ρ(x) holds for any x ∈ L) for some quiver Q ∈ Q. morphic to (Tp(Q), τ −1 such that ρ(τ −1x) = τ −1 Theorem 3.20. Let L be an infinite distributive lattice with the maximum element o and τ −1 be an inner lattice inclusion of L which induces an inner quiver inclusion of (a) (L, τ −1) ≃ (Tp(Q), τ −1 Q ) for some quiver Q ∈ Q. (b) (L, τ −1) satisfies the conditions (c0) ∼ (c4) and, −→ L . Then the following are equivalent. ′ (c5) : G (L, τ −1)1 ⊂ G(L, τ −1)1. In this case we can take Q = Q(L, τ −1). Proof. ((a) ⇒ (b)) Let Q ∈ Q such that (L, τ −1) ≃ (Tp(Q)), τ −1 Q . Then The- orem 3.1 implies that Q satisfies the conditions (C). Therefore Lemma 3.13 implies that (Tp(Q), τ −1 Q ) satisfies the conditions (c0) ∼ (c4). We show that (Tp(Q), τ −1 Q ) satisfies the condition (c5). In this case we note that Λ = Q0, x(a)b = lQ(a, b) and a ≤ b ⇔ lQ(a, b) = 0. Let a, b ∈ Q0 satisfying one of the following (see Definition 3.19): (1)There is an edge a − b in the underlying graph of the Hasse-quiver of (Q0, ≤). (2) There is an arrow α ∈ (3) There is an arrow β ∈ −→ Tp(Q)1 such that s(α) = y(a) and v(α) = b. −→ Tp(Q)1 such that s(β) = y(b) and β/∼ = a. It is sufficient to show that lQ(a, b) + lQ(b, a) = 1. First we assume (a, b) satisfies (1). Then it is obvious that lQ(a, b) + lQ(b, a) = 1. Next we assume (a, b) satisfies (2). Let lQ(a, b) = l. If l = 0, then there exists a path a ← a1 ← · · · ← ar = b in Q. If a1 6= b, then t(α)b = lQ(a, b) + 1 > 0 = lQ(a, a1) + lQ(a1, b) = t(α)a1 + lQ(a1, b). Therefore we get a contradiction. In particular b = a1. We assume l > 0. Then there exists (a1, a2 · · · , al) ∈ Ql 0 such that a ≤ a1, ai → bi ≤ ai+1 (i = 1, · · · l) and bl ≤ b in Q. If a 6= a1, then we obtain 0 and (b1, · · · bl) ∈ Ql t(α)b = lQ(a, b) + 1 > lQ(a, b) = lQ(a, a1) + lQ(a1, b) = t(α)a1 + lQ(a1, b). Therefore we get a contradiction. If b 6= b1, then we obtain t(α)b = lQ(a, b) + 1 > lQ(a, b) = lQ(a, b1) + lQ(b1, b) = t(α)b1 + lQ(b1, b). We get a contradiction. Therefore a = a1 and b = b1. In particular we get lQ(a, b) + lQ(b, a) = 1. Similarly we obtain lQ(a, b) + lQ(b, a) = 1 in the case of (3). ((b) ⇒ (a)) Let Q = Q(L, τ −1). It is sufficient to show that x ∈ ZΛ ≥0 is in L if and only if (∗) · · · xλ ≤ xλ′ + lQ(λ ′ , λ), ∀λ, λ ′ ∈ Λ. DISTRIBUTIVE LATTICES AND THE POSET OF PRE-PROJECTIVE TILTING MODULES19 Let L∗ := {z ∈ ZΛ fλ′ (x) := τ x ′ (τ −x λ λ ≥0 z satisfies (∗)}(≃ Tp(Q)). For any x ∈ L we consider ′ o ∧ x). It is easy to check that, Therefore we obtain fλ we claim, fλ ′ (x) ≥ x(λ ′ (x)λ = max{0, xλ − xλ ) and xλ − xλ ′ ′ }. ′ ≤ fλ ′ (x)λ ≤ x(λ ′ )λ. Then (∗∗) · · · x(λ)λ ′ ≤ lQ(λ, λ ′ ′ ) (∀λ, λ ∈ Λ). We note that (∗∗) implies L ⊂ L∗. Let l = lQ(λ, λ ). Without loss of generality, we can assume 0 < l < ∞. Then there exists (λ1, λ2 · · · , λl) ∈ Λl i ≤ λi+1 (i = 1, · · · l) in Q and and (λ l ≤ λ λ 1, · · · λ . In particular we obtain l) ∈ Λl such that λ ≤ λ1, λi → λ ′ ′ ′ ′ ′ ′ x(λ)λ′ ≤ x(λ1)λ′ ≤ x(λ ′ 1)λ′ +1 ≤ x(λ2)λ′ +1 ≤ · · · ≤ x(λl)λ′ +l−1 ≤ x(λ ′ l)λ′ +l = l. Therefore we obtain (∗∗). In particular L ⊂ L∗. We show that L∗ \ L = ∅. Let z ∈ L∗ \ {o} and λ ′′ ∈ Λ}. We define z of Λ(z) := {λ ∈ Λ zλ ≥ zλ ′′ ∀λ ′ be a maximal element ′ ∈ ZΛ ≥0 as follows: zλ :=(cid:26) zλ′ − 1 if λ = λ if λ 6= λ zλ ′ ′ . We can easily check that z ′ ∈ L∗. Indeed it is sufficient to check zλ ≤ zλ ′ − 1 + x(λ ′ )λ, ∀λ ∈ Λ(z) \ {λ ′ }). Now above inequalities followed from maximality of λ is a path ′ . In particular there w : o = z0 → z1 → · · · → zr = z, ≥0 such that zi ∈ L∗. This implies that if L∗ \ L 6= ∅, then ∆∗ := {z ∈ in ZΛ L∗ \ L ∃x ∈ L such that x → z in ZΛ ≥0} 6= ∅. Suppose L∗ \ L 6= ∅ and let z ∈ ∆∗. Then, by the definition of ∆∗, There ≥0 and zλ = xλ + 1. We consider exists (x, λ) ∈ L × Λ such that x → z in ZΛ a path, x = x0 → x1 → · · · → xr = τ −1x, in L. Put s := min{i xi λ = xλ + 1 = zλ}. Then there is a path, z = z0 → z1 → · · · zs−1 = xs ∈ L, in L∗ where zi := xi ∧ z. In particular there exists t < s such that zt−1 ∈ L∗ \ L and zt ∈ L. Let λ ′ + 1. Then we note that λ ′ ∈ Λ such that xt λ + 1 = zt−1 xt−1 xt−1 λ′ + 1 = xt λ ≤ zt−1 λ′ + lQ(λ λ + lQ(λ, λ λ′ ≤ xt ′ λ′ = xt−1 , λ) = xt−1 = xt−1 ) ′ ′ λ′ + lQ(λ λ + lQ(λ, λ , λ), ), ′ In particular, we obtain 2 ≤ lQ(λ, λ ′ ) + lQ(λ ′ , λ). Therefore it is sufficient to prove the following claim (†): (†) Let w : x α→ y β → z be a path in −→ L with α/∼ = λ and β/∼ = λ ′ . Let 20 RYOICHI KASE p = (pλ)λ∈Λ ∈ ZΛ ′ lQ(λ, λ ) + lQ(λ ′ , λ) = 1. ≥0 such that pλ′′ := (cid:26) xλ′ + 1 if λ xλ′′ if λ ′′ ′′ ′ ′ = λ 6= λ . If p 6∈ L, then We prove (†) with the using of an induction on r := max{zλ, zλ′ }. (r = 1) We note that x ≥ x(λ). If x(λ)λ′ > 0, then (x(λ) ∨ z)λ 0 1 xλ′′ ′′ ′′ = λ if λ if λ = λ otherwise. ′ ′′ :=  ′ ′ ′′ ′ = 0, we obtain λ < λ , then zλ In particular we get p = x(λ) ∨ z ∈ L. Therefore we get a contradiction. ′′ ∈ Λ such that Since x(λ)λ ′ . This is a contradiction. λ < λ < λ ′ → λ in Q. In Therefore the condition (c5) implies that there is an arrow λ particular we get lQ(λ, λ (r > 1) We consider the following three cases: (1) xλ > 0 and xλ ′ > 0. (2) xλ = 0 and xλ ′ > 0. (3) xλ > 0 and xλ ′′ ≤ xλ = 0 < 1 = zλ If there exists λ ) + lQ(λ ′′ = xλ , λ) = 1. . ′ ′ In the case of (1) let x ′ := τ (τ −1o ∧ x), y ′ τ (τ −1 ∧ z). Then there exists a path x ′ ′ α → y ′ = 0. := τ (τ −1o ∧ y), and z ′ β ′ ′ ′ ′ → z . Let p ∈ ZΛ ≥0 with := ′ ′ ′ ′ ′′ λ p x ′′ := ( x λ′ + 1 if λ if λ λ′′ ′ ∈ L, then p = (τ −1p p 6∈ L. Since z obtain p from the hypothesis of induction. = λ ′′ 6= λ ′ ∨ z) ∧ x ∈ L. This is a contradiction. Therefore we λ = zλ − 1 and z λ′ = zλ′ − 1, the assertion follows /∼ = λ and β We note that α /∼ = λ . If . ′ ′ ′ ′ ′ ′ ′ In the case of (2), we first show that xλ ′ , then we get p = x(λ) ∨ z ∈ L. This is a contradiction. Therefore, since x ≥ x(λ), we obtain xλ := y(λ) ∧ z. Since ′ . If x(λ)λ ′ . Now let z ′ = x(λ)λ ′ = x(λ)λ ′ > xλ ′ ′′ 1 if λ x(λ)λ′ + 1 if λ x(λ)λ′′ = λ = λ otherwise, ′′ ′ z ′ λ′′ =  we obtain a path x(λ) → y(λ) follows from the condition (c5). ′ β → z ′ ′ with β ′ /∼ = λ . Therefore the assertion Finally we consider the case of (3). Let p′ := (τ −1o ∨ z) ∧ x. Then it is ′′ ∈ Λ. In particular we get p ∈ L. ′′ = p ′ ′′ for any λ λ easy to check that pλ This is a contradiction. Corollary 3.21. An infinite distributive lattice L is isomorphic to Tp(Q) for some Q ∈ Q if and only if there is a poset inclusion τ −1 : L → L which −→ L and satisfies the conditions (c0) ∼ (c5). induces a quiver inclusion −→ L → (cid:3) DISTRIBUTIVE LATTICES AND THE POSET OF PRE-PROJECTIVE TILTING MODULES21 References [1] M. Auslander, I. Reiten and S. Smalø, Representation theory of artin algebras, Cam- bridge University Press, 1995. [2] I. Assem, D. Simson and A. Skowro´nski, Elements of the representation theory of as- sociative algebras Vol. 1, London Mathematical Society Student Texts 65, Cambridge University Press, 2006. [3] G. Birkhoff, Lattice Theory, 3rd ed. Providence, RI: Amer. Math. Soc., 1967. [4] S. Brenner, M.C.R Butler, Generalizations of the Bernstein-Gelfand-Ponomarev re- flection functors. Representation theory, II (Proc. Second Internat. Conf., Carleton Univ., Ottawa, Ont., 1979), pp.103-169, Lecture Notes in Math., 832, Springer, Berlin-New York, 1980. [5] F. Coelho, D. Happel and L. Unger, Complements to partial tilting modules, J. Al- gebra 170 (1994), no.3, 184-205. [6] P. Gabriel, Auslander-Reiten sequences and representation-finite algebras, Repre- sentation theory, I(Proc. Workshop, Carleton Univ., Ottawa, Ont., 1979), pp.1-71, Lecture Notes in Math., 831, Springer, Berlin, 1980. [7] G. Gratzer, Lattice Theory: First Concepts and Distributive Lattices, San Francisco, CA: W. H. Freeman, 1971. [8] D. Happel and C. M. Ringel, Tilted algebras, Trans. Amer. Math. Soc. 274 (1982), no.2, 399-443. [9] D. Happel and L. Unger, On a partial order of tilting modules, Algebr. Represent. Theory 8 (2005), no.2, 147-156. [10] D. Happel and L. Unger, On the quiver of tilting modules, J. Algebra 284 (2005), no.2, 857-868. [11] D. Happel and L. Unger, Reconstruction of path algebras from their posets of tilting modules, Trans. Amer . Math. Soc 361 (2009), no.7, 3633-3660. [12] D. Happel and L. Unger, Links of faithful partial tilting modules, Algebr. Represent. Theory 13 (2010), no.6, 637-652. [13] R. Kase, A pre-projective part of tilting quivers of certain path algebras, arXiv: 1212.0359 [14] I. Reiten, Tilting theory and homologically finite subcategories, Handbook of tilting theory, L.Angeleri Hugel, D.Happel, H.Krause, eds., London Mathematical Society Lecture Note Series 332, Cambridge University Press, 2007. [15] C. Riedtmann and A. Schofield, On a simplicial complex associated with tilting mod- ules, Comment. Math. Helv 66 (1991), no.1, 70-78. [16] L. Unger, Combinatorial aspects of the set of tilting modules, Handbook of tilting theory, L.Angeleri Hugel, D.Happel, H.Krause, eds., London Mathematical Society Lecture Note Series 332, Cambridge University Press, 2007. Department of Pure and Applied Mathematics Graduate School of Infor- mation Science and Technology ,Osaka University, Toyonaka, Osaka 560-0043, Japan E-mail address: [email protected]
1302.6873	2	1302	2013-03-13T14:33:20	Quasipolar Subrings of $3\times 3$ Matrix Rings	[ "math.RA" ]	An element $a$ of a ring $R$ is called \emph{quasipolar} provided that there exists an idempotent $p\in R$ such that $p\in comm^2(a)$, $a+p\in U(R)$ and $ap\in R^{qnil}$. A ring $R$ is \emph{quasipolar} in case every element in $R$ is quasipolar. In this paper, we determine conditions under which subrings of $3\times 3$ matrix rings over local rings are quasipolar. Namely, if $R$ is a bleached local ring, then we prove that $\mathcal{T}_3(R)$ is quasipolar if and only if $R$ is uniquely bleached. Furthermore, it is shown that $T_n(R)$ is quasipolar if and only if $T_n\big(R[[x]]\big)$ is quasipolar for any positive integer $n$.	math.RA	math	QUASIPOLAR SUBRINGS OF 3 × 3 MATRIX RINGS ORHAN GURGUN, SAIT HALICIOGLU, AND ABDULLAH HARMANCI Abstract. An element a of a ring R is called quasipolar provided that there exists an idempotent p ∈ R such that p ∈ comm2(a), a + p ∈ U (R) and ap ∈ Rqnil. A ring R is quasipolar in case every element in R is quasipolar. In this paper, we determine conditions under which subrings of 3 × 3 matrix rings over local rings are quasipolar. Namely, if R is a bleached local ring, then we prove that T3(R) is quasipolar if and only if R is uniquely bleached. Furthermore, it is shown that Tn(R) is quasipolar if and only if Tn(cid:0)R[[x]](cid:1) is quasipolar for any positive integer n. Keywords: Quasipolar ring, local ring, 3 × 3 matrix ring. 2010 Mathematics Subject Classification: 16S50, 16S70, 16U99 1. Introduction Throughout this paper all rings are associative with identity unless other- wise stated. Following Koliha and Patricio [11], the commutant and double commutant of an element a ∈ R are defined by comm(a) = {x ∈ R xa = ax}, comm2(a) = {x ∈ R xy = yx for all y ∈ comm(a)}, respectively. If Rqnil = {a ∈ R 1 + ax ∈ U (R) for every x ∈ comm(a)} and a ∈ Rqnil, then a is said to be quasinilpotent [10]. An element a ∈ R is called quasipolar provided that there exists an idempotent p ∈ R such that p ∈ comm2(a), a + p ∈ U (R) and ap ∈ Rqnil. A ring R is quasipolar in case every element in R is quasipolar. Properties of quasipolar rings were studied in [6, 7, 14]. For a ring R, let T3(R) =    a11 0 0 a21 a22 a23 0 0 a33   a11, a21, a22, a23, a33 ∈ R  . Then T3(R) is a ring under the usual addition and multiplication, and so T3(R) is a subring of M3(R). Motivated by results in [3] and [5], we study quasipolar subrings of 3 × 3 matrix rings over local rings. We prove that 1 2 ORHAN GURGUN, SAIT HALICIOGLU, AND ABDULLAH HARMANCI   Z(2) Z(2) Z(2) Z(2) Z(2) 0 0 quasipolar. 0 0 is quasipolar but the full matrix ring M3(Z(2)) is not   In this paper, Mn(R) and Tn(R) denote the ring of all n × n matrices and the ring of all n × n upper triangular matrices over R, respectively. We write R[[x]], U (R) and J(R) for the power series ring over a ring R, the set of all invertible elements and the Jacobson radical of R, respectively. For A ∈ Mn(R), χ(A) stands for the characteristic polynomial det(tIn − A). 2. Quasipolar Elements In [12], Nicholson gives several equivalent characterizations of strongly clean rings through the endomorphism ring of a module. Analogously, we present similar results for quasipolar rings. For convenience, we use left modules and write endomorphisms on the right. For a module RM , we write E = EndR(M ) for the ring of endomorphisms of RM . Lemma 2.1. [12, Lemma 2] Let β, π2 = π ∈ EndR(M ). Then both M π and M (1 − π) are β-invariant if and only if πβ = βπ. Similar to [4, Theorem 2.1] we have the following results for quasipolar endomorphisms of a module. Theorem 2.2. Let α ∈ E = EndR(M ). The following are equivalent. (1) α is quasipolar in E. (2) There exists π2 = π ∈ E such that π ∈ comm2 E(α), απ is a unit in πEπ and α(1 − π) is a quasinilpotent in (1 − π)E(1 − π). (3) M = P ⊕Q, where P and Q are β-invariant for every β ∈ commE(α), αP is a unit in End(P ) and αQ is a quasinilpotent in End(Q). (4) M = P1 ⊕ P2 ⊕ · · · ⊕ Pn for some n ≥ 1, where Pi is β-invariant for every β ∈ commE(α), αPi is quasipolar in End(Pi) for each i. Proof. (1) ⇒ (2) Since α is quasipolar in E, there exists an idempotent τ ∈ R such that τ ∈ comm2 E(α), α + τ = η ∈ U (E) and ατ ∈ Eqnil. Let π = 1 − τ . Clearly, π2 = π ∈ comm2 E(α). Note that α, π, η and τ all commute. Now, QUASIPOLAR SUBRINGS OF 3 × 3 MATRIX RINGS 3 multiplying α + τ = η by π yields απ = ηπ = πη ∈ πEπ. Since η−1π ∈ πEπ this gives (απ)(η−1π) = (πη)(η−1π) = π. Similarly, (η−1π)(απ) = π so απ is a unit in πEπ. Let (1 − π)γ(1 − π) ∈ comm(1−π)E(1−π)(α(1 − π)). Then (1 − π)γ(1 − π) ∈ commE(α(1 − π)). The remaining proof is to show that (1− π)+ α(1− π)γ(1− π) is a unit in (1− π)E(1− π). Since α(1− π) ∈ Eqnil, 1 + α(1 − π)γ(1 − π) is a unit in E and so (1 − π) + α(1 − π)γ(1 − π) is a unit (1 − π)E(1 − π). (2) ⇒ (3) Given π as in (2), let P = M π and Q = M (1 − π). Then M = P ⊕ Q. For any β ∈ commE(α), the hypothesis π ∈ comm2 E(α) implies that πβ = βπ. By Lemma 2.1, both P and Q are β-invariant. As in the proof [12, Theorem 3], απ = αP is a unit in End(P ). Let γ ∈ commEnd(Q)(αQ). We show that 1Q + αQγ is a unit in End(Q). Clearly, γ ∈ comm(1−π)E(1−π)(α(1 − π)). Since α(1 − π) is a quasinilpotent in (1 − π)E(1 − π), (1 − π) + α(1 − π)γ is a unit in (1 − π)E(1 − π). Let [(1 − π) + α(1 − π)γ]−1 = (1 − π)τ (1 − π) = τ0 ∈End(Q) and let q ∈ Q. Then (q)[1Q + αQγ]τ0 = (q + q(1 − π)αγ)τ0 = (q(1 − π) + qα(1 − π)γ)τ0 = q[(1 − π) + α(1 − π)γ]τ0 = (q)1Q. Hence (1Q + αQγ)τ0 = 1Q. Similarly, τ0(1Q + αQγ) = 1Q. Thus αQ is a quasinilpotent in End(Q). (3) ⇒ (4) Suppose M = P ⊕Q as in (3). Since αP is a unit in End(P ), αP is a quasipolar in End(P ) by [6, Example 2.1]. As αQ is a quasinilpotent in End(Q), 1Q + αQ is a unit in End(Q). Further, 12 comm2 End(Q)(αQ) so αQ is quasipolar in End(Q). Q = 1Q and 1Q ∈ (4) ⇒ (1) Let λi ∈End(Pi). Given the situation in (4), extend maps λi n in End(Pi) to λi in End(M ) by defining ( pj)λi = (pi)λi for any pj ∈ Pj. Then λi λj = 0 if i 6= j while λi µi = λiµi and λi + µi = λi + µi for all µi ∈End(Pi). By hypothesis, there exists π2 End(Pj )(αPj ), If σj ∈ U(cid:0)End(Pj)(cid:1) such that αPj + πj = σj and αPj πj ∈End(Pj)qnil. 2 = π ∈End(M ) and σ is a j = πj ∈ comm2 σj then π2 = πj and σ = π = πj Pj=1 n Pj=1 n Pj=1 n Pj=1 n Pj=1 unit in E because σ−1 = σj −1. Since α = n n αPj = (−πj + σj) = −π + σ, we show that π ∈ comm2 E(α) and απ ∈ Eqnil. Since for each β ∈ commE(α), P and Q are β-invariant. Hence, πβ = βπ by Lemma 2.1 Pj=1 Pj=1 4 ORHAN GURGUN, SAIT HALICIOGLU, AND ABDULLAH HARMANCI and so π ∈ comm2 E(α). For any β ∈ commE(απ), we only need to show that 1E + βαπ is an isomorphism in E. Note that βPj ∈ commEnd(Pj)(αPj ) and 1Pj + βPj αPj = (π + βα)Pj . Since αPj πj ∈ End(Pj)qnil, 1Pj + βPj αPj πj = (π + βπα)Pj is a unit in End(Pj). Let γj ∈End(Pj) be such that (1Pj + βPj αPj πj)γj = 1Pj = γj(1Pj + βPj αPj πj) and let m = n n n n pj)(1E + βαπ)γ = (cid:0) n pj + ( Pj=1 Pj=1 pj)βαπ(cid:1)γ (pj)[1Pj + βPj αPj πj](cid:1)γ = Pj=1 n (pj)[1Pj ]=m where γ = γj. Similarly, (pj)[βPj αPj πj](cid:1)γ = (cid:0) Pj=1 n n n Pj=1 (pj)1Pj + ( pj with pj ∈ Pj. So ( Pj=1 = (cid:0) Pj=1 (pj)[1Pj + βPj αPj πj]γj(cid:1) = (cid:0) Pj=1 n n we have ( pj)γ(1E + βαπ) = Pj=1 proof is completed. n Pj=1 Pj=1 (pj)[1Pj ] = m. Therefore απ ∈ Eqnil, the (cid:3) Pj=1 The following result is a direct consequence of Theorem 2.2. Corollary 2.3. Let R be a ring. The following are equivalent for a ∈ R. (1) a ∈ R is quasipolar. (2) There exists e2 = e ∈ R such that e ∈ comm2 R(a), ae ∈ U (eRe) and a(1 − e) ∈ (1 − e)R(1 − e)qnil. 3. The Rings T3(R) For a ring R, let a ∈ R, la : R → R and ra : R → R denote, respec- tively, the abelian group endomorphisms given by la(r) = ar and ra(r) = ra for all r ∈ R. Thus, for a, b ∈ R, la, rb is an abelian group endomor- phism such that (la − rb)(r) = ar − rb for any r ∈ R. A local ring R is called bleached [1] if, for any a ∈ J(R) and any b ∈ U (R), the abelian group endomorphisms lb − ra and la − rb of R are both surjective. A local ring R is called uniquely bleached if, for any a ∈ J(R) and any b ∈ U (R), the abelian group endomorphisms lb − ra and la − rb of R are isomor- phic. According to [8, Example 2.1.11], commutative local rings, division rings, local rings with nil Jacobson radicals, local rings for which some power of each element of their Jacobson radicals is central are uniquely bleached. Clearly uniquely bleached local rings are bleached. But so far it is unknown whether a bleach local ring is uniquely bleached. Obviously, QUASIPOLAR SUBRINGS OF 3 × 3 MATRIX RINGS 5 a11 0 0 0 0 a33 a21 a22 a23   a11 ∈ U(cid:0)T3(R)(cid:1) if and only if a11, a22, a33 ∈ U (R). Further,   J(cid:0)T3(R)(cid:1) =    A − E ∈ U(cid:0)T3(R)(cid:1) and EA ∈ J(cid:0)T3(R)(cid:1) ⊆ T3(R)qnil, then −A is quasipolar a11, a22, a33 ∈ J(R), a21, a23 ∈ R  that if, for every A ∈ T3(R), there exists E2 = E ∈ comm2(A) such that and so T3(R) is quasipolar. We use this fact in the proof of Theorem 3.1 a21 a22 a23 0 a33   0 0 0 . Note without mention. By [13, Example 1] and [9, Remark 3.2.11], M3(R) is not quasipolar in general. Our next aim is to determine to find conditions under which T3(R) is quasipolar. In this direction we can give the following theorem. Theorem 3.1. Let R be a bleached local ring. The following are equivalent. (1) R is uniquely bleached. (2) T3(R) is quasipolar. (3) T2(R) is quasipolar. Proof. (1) ⇒ (2) Let A =  ing cases. a11 0 0 a21 a22 a23 0 0 a33   ∈ T3(R). Consider the follow- Case 1. a11, a22, a33 ∈ J(R). Then A + I3 ∈ U (T3(R)) and AI3 = A ∈ J(cid:0)T3(R)(cid:1) ⊆ T3(R)qnil. So A is quasipolar. Case 2. a11, a22, a33 ∈ U (R). Then A + 0 ∈ U (T3(R)) and A0 = 0 ∈ T3(R)qnil. So A is quasipolar. Case 3. a11 ∈ U (R), a22, a33 ∈ J(R). There exists a unique element e21 ∈ R such that a22e21 − e21a11 = a21. Let E =   0 0 0 e21 1 0 0 0 1 E2 = E, A − E ∈ U(cid:0)T3(R)(cid:1) and AE ∈ J(cid:0)T3(R)(cid:1) ⊆ T3(R)qnil. We show that . Then   6 ORHAN GURGUN, SAIT HALICIOGLU, AND ABDULLAH HARMANCI E ∈ comm2(A). Let X =  and so x11 0 0 x21 x22 x23 0 0 x33   ∈ comm(A). Then XA = AX a11x11 = x11a11, a22x22 = x22a22, a33x33 = x33a33 x21a11 + x22a21 = a21x11 + a22x21 (i) (ii) x22a23 + x23a33 = a22x23 + a23x33 (iii) Since a22e21 − e21a11 = a21, a22[x22e21 − e21x11 − x21] − [x22e21 − e21x11 − x21]a11 = 0 by (i) and (ii). By (1), la22 − ra11 is injective and so x22e21 − e21x11 = x21. That is, XE = EX. Hence E ∈ comm2(A). Case 4. a11 ∈ J(R), a22 ∈ U (R), a33 ∈ J(R). There exist unique ele- ments e21, e23 ∈ R such that a22e21 − e21a11 = −a21 and a22e23 − e23a11 = 0 1 0 0 0 1 e21 0 e23  . Then E2 = E, A − E ∈ U(cid:0)T3(R)(cid:1)  and AE ∈ J(cid:0)T3(R)(cid:1) ⊆ T3(R)qnil. We prove E ∈ comm2(A). Let X =   −a23. Let E =     −a21, a22[−x22e21 + e21x11 − x21]− [−x22e21 + e21x11 − x21]a11 = 0 by (i) and ∈ comm(A). Then XA = AX. Since a22e21 − e21a11 = x21 x22 x23 x11 x33 0 0 0 0 (ii). By (1), la22 −ra11 is injective and so x22e21+x21 = e21x11. Since a22e23− e23a11 = −a23, a22[−x22e23 + e23x33 − x23] − [−x22e23 + e23x33 − x23]a11 = 0 by (i) and (iii). By (1), la22 − ra11 is injective and so x22e23 + x23 = e23x33. That is, XE = EX. Hence E ∈ comm2(A). Case 5. a11, a22 ∈ J(R), a33 ∈ U (R). There exists a unique element e23 ∈ R such that a22e23 − e23a33 = a23. Let E =     E2 = E, A − E ∈ U(cid:0)T3(R)(cid:1) and AE ∈ J(cid:0)T3(R)(cid:1) ⊆ T3(R)qnil. We show that E ∈ comm2(A). Let X =   1 0 0 0 1 e23 0 0 0 ∈ comm(A). Then x21 x22 x23   . Then x33 x11 0 0 0 0 QUASIPOLAR SUBRINGS OF 3 × 3 MATRIX RINGS 7 XA = AX. Since a22e23 − e23a33 = a23, a22[x22e23 − e23x33 − x23] − [x22e23 − e23x33 − x23]a33 = 0 by (i) and (iii). By (1), la22 − ra33 is injective and so x22e23 − e23x33 = x23. That is, XE = EX. Hence E ∈ comm2(A). Case 6. a11 ∈ J(R), a22, a33 ∈ U (R). There exists a unique element e21 ∈ R such that a22e21 − e21a11 = −a21. Let E =     E2 = E, A − E ∈ U(cid:0)T3(R)(cid:1) and AE ∈ J(cid:0)T3(R)(cid:1) ⊆ T3(R)qnil. We prove that E ∈ comm2(A). Let X =   XA = AX. Since a22e21 − e21a11 = −a21, a22[−x22e21 + e21x11 − x21] − 1 0 0 e21 0 0 0 0 0 ∈ comm(A). Then x21 x22 x23   . Then x11 x33 0 0 0 0 [−x22e21 + e21x11 − x21]a11 = 0 by (i) and (ii). By (1), la22 − ra11 is injective and so x22e21 + x21 = e21x11. That is, XE = EX. Hence E ∈ comm2(A). Case 7. a11 ∈ U (R), a22 ∈ J(R), a33 ∈ U (R). There exist unique ele- ments e21, e23 ∈ R such that a22e21 − e21a11 = a21 and a22e23 − e23a33 = 0 0   0 0 0 e21 1 e23 a23. Let E =     . Then E2 = E, A − E ∈ U(cid:0)T3(R)(cid:1) and AE ∈ J(cid:0)T3(R)(cid:1) ⊆ T3(R)qnil. To show E ∈ comm2(A) let X =   a21, a22[x22e21 − e21x11 − x21] − [x22e21 − e21x11 − x21]a11 = 0 by (i) and ∈ comm(A). Then XA = AX. Since a22e21 − e21a11 = x21 x22 x23 x11 x33 0 0 0 0 0 (ii). By (1), la22 − ra11 is injective and so x22e21 − e21x11 = x21. Since a22e23 −e23a33 = a23, a22[x22e23−e23x33−x23]−[x22e23−e23x33−x23]a33 = 0 by (i) and (iii). By (1), la22 − ra33 is injective and so x22e23 − e23x33 = x23. That is, XE = EX. Hence E ∈ comm2(A). Case 8. a11, a22 ∈ U (R), a33 ∈ J(R). There exists a unique element e23 ∈ . Then E2 = E, R such that a22e23 − e23a33 = −a23. Let E =  0 0 0 0 0 e23 0 0 1   A − E ∈ U(cid:0)T3(R)(cid:1) and AE ∈ J(cid:0)T3(R)(cid:1) ⊆ T3(R)qnil. The remaining proof 8 ORHAN GURGUN, SAIT HALICIOGLU, AND ABDULLAH HARMANCI is to show that E ∈ comm2(A). Let X =   x11 0 0 x21 x22 x23 0 0 x33   Then XA = AX. Since a22e23 −e23a33 = −a23, a22[−x22e23 +e23x33 −x23]− ∈ comm(A). [−x22e23 + e23x33 − x23]a33 = 0 by (i) and (iii). By (1), la22 − ra33 is injective and so x22e23 + x23 = e23x33. That is, XE = EX. Hence E ∈ comm2(A). (2) ⇒ (3) Assume that T3(R) is quasipolar. Let E =   T3(R). Then T2(R) ∼= ET3(R)E. Thus T2(R) is quasipolar by [14, Proposi- tion 3.6]. 1 0 0 0 1 0 0 0 0   ∈ (3) ⇒ (1) It follows from [7, Proposition 2.9]. (cid:3) An element a ∈ R is strongly rad clean provided that there exists an idempotent e ∈ R such that ae = ea and a − e ∈ U (R) and ea ∈ J(eRe). A ring R is strongly rad clean in case every element in R is strongly rad clean (cf. [8]). Due to the proof of Theorem 3.1, we have the following. Corollary 3.2. Let R be a local ring. The following are equivalent. (1) T3(R) is strongly rad clean. (2) R is bleached. For a ring R, let L3(R) =     a11 0 0 a22 0 0 a31 0 a33   a11, a31, a22, a33 ∈ R  . Then L3(R) is a ring under the usual addition and multiplication, and so L3(R) is a subring of M3(R). Our next endeavor is to find conditions under which L3(R) is quasipolar. Proposition 3.3. Let R be a bleached local ring. The following are equiv- alent. (1) R is uniquely bleached. (2) L3(R) is quasipolar. QUASIPOLAR SUBRINGS OF 3 × 3 MATRIX RINGS 9 Proof. Let ϕ : L3(R) → T2(R) ⊕ R given by a11 0 0 a22 0 0 a31 0 a33     7→(cid:18)" a33 a31 a11 # , a22(cid:19). 0 Then ϕ is an isomorphism (see [3, Proposition 2.2]). Since R is local, it is quasipolar. Hence L3(R) is quasipolar if and only if T2(R) is quasipolar. Therefore it follows from Theorem 3.1. (cid:3) Corollary 3.4. Let R be a bleached local ring. The following are equivalent. (1) R is uniquely bleached. lar. (2) The ring  (3) The ring  lar. a11 0 0 a22 0 0 a31 a32 a33 a11 0 a13 0 0 a22 a23 0 a33         a11, a31, a32, a22, a33 ∈ R  a11, a13, a23, a22, a33 ∈ R  is quasipo- is quasipo- Proof. (1) ⇔ (2) Let   a11, a31, a32, a22, a33 ∈ R  0 a11 0 ϕ : T3(R) →    a11 0 0 a22 0 0 a31 a32 a33 7→   0 0 0 0 a33 a11 a21 a22 a23 given by A =     T3(R) is quasipolar if and only if  is quasipolar, as asserted. (1) ⇔ (3) is symmetric. Then ϕ is an isomorphism (see [2, Corollary 3.4]). In view of Theorem 3.1, 0 a33 0 a21 a23 a22 a11 0 0 a22 0 0 a31 a32 a33     for any A ∈ T3(R).   a11, a31, a32, a22, a33 ∈ R  (cid:3) Corollary 3.5. Let R be a bleached local ring. The following are equivalent. 10 ORHAN GURGUN, SAIT HALICIOGLU, AND ABDULLAH HARMANCI (1) R is uniquely bleached. lar. (2) The ring S1 =    (3) The ring S2 =    lar. a11 0 a13 0 0 a11 0 0 a22 0 0 0 a22 a33 0 0 a32 a33     a11, a13, a22, a33 ∈ R  a11, a31, a22, a33 ∈ R  is quasipo- is quasipo- Proof. (1) ⇔ (2) As in the proof of Proposition 3.3, R is uniquely bleached if and only if S1 is quasipolar, as asserted. (2) ⇔ (3) Let ϕ : S1 → S2 given by A =  a11 0 a13 0 0 a22 0 0 a33   7→  a22 0 0 0 a33 a13 a11 0 0   for any A ∈ S1. Then ϕ is an isomorphism. Hence S1 is quasipolar if and only if S2 is quasipolar. (cid:3) Let R be a commutative local ring. By Theorem 3.1, Proposition 3.3, Corollary 3.4 and Corollary 3.5, the rings R 0 0 0 R 0 R 0 R     ,  R 0 0 R R R 0 0 R R 0 R 0 R 0 0 0 R   ,  R 0 0 0 R 0 R R R   ,    are all quasipolar. Remark 3.6. Let Z(2) = { m n m, n ∈ Z, 2 ∤ n}. By [13, Example 1] and [9, Remark 3.2.11], M3(Z(2)) is not strongly clean and so it is not quasipolar. However, by Theorem 3.1, Proposition 3.3, Corollary 3.4 and Corollary 3.5, QUASIPOLAR SUBRINGS OF 3 × 3 MATRIX RINGS Z(2) 0 Z(2) 0 Z(2) 0 0 0 Z(2) 0 0 Z(2) Z(2) Z(2) Z(2) Z(2) 0 0  ,   Z(2) 0 0 0 Z(2) 0 Z(2) 0 Z(2)   ,  11 ,   the rings   0 Z(2) 0   0 0   Z(2) are all quasipolar. Z(2) Z(2) Z(2) 4. Matrices Over Power Series Rings ∞ Pi=0 ∞ Pi=0 ∞ Pi=0 In this section, we characterize quasipolar matrices over the power series ring of a local ring. In order to prove Theorem 4.2, we need the following lemma. Lemma 4.1. Let R be a commutative local ring and A(x) ∈ M2(cid:0)R[[x]](cid:1). The following are equivalent. (1) χ(cid:0)A(0)(cid:1) has a root in J(R) and a root in U (R). (2) χ(cid:0)A(x)(cid:1) has a root in J(cid:0)R[[x]](cid:1) and a root in U(cid:0)R[[x]](cid:1). Proof. (1) ⇒ (2) Assume that χ(cid:0)A(0)(cid:1) = y2−µy−λ has a root α ∈ J(R) and a root β ∈ U (R). Let y = bixi. Then y2 = cixi where ci = bkbi−k. ∞ i Pi=0 Pk=0 Let µ(x) = µixi, λ(x) = λixi ∈ R[[x]] where µ0 = µ and λ0 = λ. Then, y2 − µ(x)y − λ(x) = 0 holds in R[[x]] if the following equations are satisfied: b2 0 − b0µ0 − λ0 = 0; (b0b1 + b1b0) − (b0µ1 + b1µ0) − λ1 = 0; (b0b2 + b2 1 + b2b0) − (b0µ2 + b1µ1 + b2µ0) − λ2 = 0; ... Obviously, µ0 = trA(0) = α + β ∈ U (R). Let b0 = α. Since R is commuta- tive local, there exists some b1 ∈ R such that b0b1 + b1(b0 − µ0) = λ1 + b0µ1. Further, there exists some b2 ∈ R such that b0b2 + b2(b0 − µ0) = λ2 − b2 1 + b0µ2 + b1µ1. 12 ORHAN GURGUN, SAIT HALICIOGLU, AND ABDULLAH HARMANCI By iteration of this process, we get b3, b4, · · · . Then y2 − µ(x)y − λ(x) = 0 If b0 = β, analogously, we show that y2 − has a root α(x) ∈ J(cid:0)R[[x]](cid:1). µ(x)y − λ(x) = 0 has a root β(x) ∈ U(cid:0)R[[x]](cid:1). (2) ⇒ (1) Suppose that χ(cid:0)A(x)(cid:1) = y2 − µ(x)y − λ(x) has a root α(x) ∈ J(cid:0)R[[x]](cid:1) and a root β(x) ∈ U(cid:0)R[[x]](cid:1). Then µ(x) = trA(x) and −λ(x) = detA(x). Hence µ(0) = trA(0) and −λ(0) = detA(0). Thus, χ(cid:0)A(0)(cid:1) = y2 − µ(0)y − λ(0). Since α(x)2 − µ(x)α(x)− λ(x) = 0 and β(x)2 − µ(x)β(x)− λ(x) = 0, α(0)2 − µ(0)α(0) − λ(0) = 0 and β(0)2 − µ(0)β(0) − λ(0) = 0. Then χ(cid:0)A(0)(cid:1) = y2 − µ(0)y − λ(0) has a root α(0) ∈ J(R) and a root β(0) ∈ U (R). (cid:3) Theorem 4.2. Let R be a commutative local ring. The following are equiv- alent. (1) A(0) ∈ M2(R) is quasipolar. (2) A(x) ∈ M2(cid:0)R[[x]](cid:1) is quasipolar. Proof. (1) ⇒ (2) It is known that R[[x]] is local. To complete the proof we consider the following cases: (i) A(0) ∈ GL2(R), (ii) detA(0), trA(0) ∈ J(R), is quasipolar by [6, Example 2.1]. If detA(0), trA(0) ∈ J(R), then trA(x), (iii) detA(0) ∈ J(R), trA(0) ∈ U (R) and χ(cid:0)A(0)(cid:1) is solvable in R. If A(0) ∈ GL2(R), then A(x) ∈ GL2(cid:0)R[[x]](cid:1) and so A(x) ∈ M2(cid:0)R[[x]](cid:1) detA(x) ∈ J(cid:0)R[[x]](cid:1) and so A(x) is quasipolar by [6, Theorem 2.6]. Now suppose that detA(0) ∈ J(R), trA(0) ∈ U (R) and χ(cid:0)A(0)(cid:1) has two roots α, β ∈ R. Then detA(x) ∈ J(cid:0)R[[x]](cid:1) and trA(x) ∈ U(cid:0)R[[x]](cid:1). Since detA(0) ∈ J(R) and trA(0) ∈ U (R), either α ∈ J(R) or β ∈ J(R). Without loss of generality, we assume that α ∈ J(R) and β ∈ U (R). According to Lemma 4.1, χ(cid:0)A(x)(cid:1) has a root in J(cid:0)R[[x]](cid:1) and a root in U(cid:0)R[[x]](cid:1). Hence A(x) is quasipolar in M2(cid:0)R[[x]](cid:1) by [6, Proposition 2.8]. (2) ⇒ (1) is similar to the proof of (1) ⇒ (2). (cid:3) QUASIPOLAR SUBRINGS OF 3 × 3 MATRIX RINGS 13 Example 4.3. Let R = Z4[[x]], and let 0 − ∞ ∞ 1 3 − (1 + 3n)xn Pn=1 Pn=1 A(x) =    1 3 #, Since A(0) = " 0 0 χ(cid:0)A(0)(cid:1) = t2 − trA(0)t + detA(0) = t2 − 3t = t(t − 3) is solvable in Z4. By [6, Proposition 2.8], A(0) ∈ M2(Z4) is quasipolar. In view of Theorem 4.2, Obviously, Z4 is a commutative local ring. (1 + 3n)xn ∈ M2(R). A(x) ∈ M2(R) is quasipolar. Theorem 4.4. Let R be a commutative local ring and for m ≥ 1 A(x) ∈ M2(cid:0)R[[x]]/(xm)(cid:1). The following are equivalent. (1) A(0) ∈ M2(R) is quasipolar. (2) A(x) ∈ M2(cid:0)R[[x]]/(xm)(cid:1) is quasipolar. Proof. The proof is similar to that of Theorem 4.2. Example 4.5. Let R = Z4[[x]]/(x2), and let (cid:3) 3 + (x2) 2 + 2x + (x2) A(x) =" 2 + x + (x2) 2 + 3x + (x2) # ∈ M2(R). Obviously, Z4 is a commutative local ring. Since A(0) =" 3 2 2 2 #, χ(cid:0)A(0)(cid:1) = t2 − trA(0)t + detA(0) = t2 − t + 2 = (t − 3)(t + 2) is solvable in Z4. By [6, Proposition 2.8], A(0) ∈ M2(Z4) is quasipolar. In view of Theorem 4.4, A(x) ∈ M2(R) is quasipolar. Lemma 4.6. Let R be a local ring. Then R is uniquely bleached if and only if R[[x]] is uniquely bleached. Proof. Assume that R is uniquely bleached. Then lu − rj is an isomorphism aixi ∈ R[[x]]. Since for any j ∈ J(R) and u ∈ U (R) and let f (x) = ∞ R is bleached, by [8, Example 2.1.11(6)], R[[x]] is bleached. If, for j(x) = ∞ Pi=1 then jixi ∈ J(cid:0)R[[x]](cid:1) and u(x) = ∞ Pi=1 uixi ∈ U(cid:0)R[[x]](cid:1), (lj(x)−ru(x))(f (x)) = 0, Pi=1 14 ORHAN GURGUN, SAIT HALICIOGLU, AND ABDULLAH HARMANCI j0a0 = a0u0 j0a1 + j1a0 = a0u1 + a1u0 j0a2 + j1a1 + j2a0 = a0u2 + a1u1 + a2u0 ... (i1) (i2) (i3) ... By assumption, lj0 − ru0 is an isomorphism and so a0 = 0 by (i1). As a0 = 0, by (i2), j0a1 = a1u0 and so a1 = 0 by assumption. Since a0 = 0 = a1, by (i3), j0a2 = a2u0 and so a2 = 0 by assumption. By iteration of this process, we deduce that f (x) = 0. Hence lj(x) − ru(x) is an isomorphism and so R[[x]] is uniquely bleached. Conversely, suppose that R[[x]] is uniquely bleached. ∞ and u(x) = Then lj(x) − ru(x) is an isomorphism for any j(x) = jixi ∈ J(cid:0)R[[x]](cid:1) uixi ∈ U(cid:0)R[[x]](cid:1) and let r ∈ R. Let (lj − ru)(r) = 0 with j ∈ J(R) and u ∈ U (R). Since j ∈ J(cid:0)R[[x]](cid:1) and u ∈ U(cid:0)R[[x]](cid:1), by assumption, r = 0 and so lj − ru is injective. The remaining proof is to show Pi=1 ∞ Pi=1 that lj − ru is surjective. Since lj(x) − ru(x) is an isomorphism where j(0) = j aixi ∈ R[[x]] such and u(0) = u, for any r ∈ R, we can find some f (x) = ∞ Pi=1 that j(x)f (x) − f (x)u(x) = r. Hence ja0 − a0u = r with a0 ∈ R and so lj − ru is surjective. Thus R is uniquely bleached. (cid:3) Proposition 4.7. Let R be a bleached local ring. The following are equiv- alent. (1) T3(R) is quasipolar. (2) T3(cid:0)R[[x]](cid:1) is quasipolar. Proof. (1) ⇒ (2) Assume that T3(R) is quasipolar. By Theorem 3.1, R is uniquely bleached. Note that if R is local, then so is R[[x]] because R/J(R) ∼= R[[x]]/J(cid:0)R[[x]](cid:1). According to Lemma 4.6, R[[x]] is uniquely bleached. Hence T3(cid:0)R[[x]](cid:1) is quasipolar by Theorem 3.1. (2) ⇒ (1) Suppose that T3(cid:0)R[[x]](cid:1) is quasipolar. Then R[[x]] is uniquely bleached by Theorem 3.1. In view of Lemma 4.6, R is uniquely bleached. Hence T3(R) is quasipolar by Theorem 3.1. (cid:3) QUASIPOLAR SUBRINGS OF 3 × 3 MATRIX RINGS 15 Corollary 4.8. Let R be a bleached local ring. For any positive integer n, the following are equivalent. (1) Tn(R) is quasipolar. (2) Tn(cid:0)R[[x]](cid:1) is quasipolar. Proof. By [7, Proposition 2.9] and Lemma 4.6, the proof is completed. (cid:3) References [1] H. Chen, Rings Related Stable Range Conditions, Series in Algebra 11, World Scien- tific, Hackensack, NJ, 2011. [2] H. Chen, Some strongly nil clean matrices over local rings, Bull. Korean Math. Soc. 48(4)(2011), 759-767. [3] H. Chen, A note on strongly clean matrices, Int. Electron. J. Algebra 10(2011), 192- 204. [4] H. Chen, Strongly J-clean matrices over local rings, Comm. Algebra 40(4)(2012), 1352-1362. [5] J, Cui and J. Chen, Strongly clean 3 × 3 matrices over a class of local rings, Nanjing Daxue Xuebao Shuxue Bannian Kan 27(1)(2010), 31-40. [6] J, Cui and J. Chen, When is a 2 × 2 matrix ring over a commutative local ring quasipolar, Comm. Algebra 39(2011), 3212-3221. [7] J, Cui and J. Chen, Quasipolar triangular matrix rings over local rings, Comm. Al- gebra 40(2012), 784-794. [8] A. J. Diesl, Classes of Strongly Clean Rings, Ph.D. Thesis, University of California, Berkeley, 2006. [9] T. J. Dorsey, Cleanness and Strong Cleanness of Rings of Matrices, Ph.D. Thesis, University of California, Berkeley, 2006. [10] R. E. Harte, On quasinilpotents in rings, Panam. Math. J. 1(1991), 10-16. [11] J. J. Koliha and P. Patricio, Elements of rings with equal spectral idempotents, J. Austral. Math. Soc. 72(2002), 137-152. [12] W. K. Nicholson, Strongly clean rings and Fittings lemma, Comm. Algebra 27(1999), 3583-3592. [13] Z. Wang and J. L. Chen, On two open problems about strongly clean rings, Bull. Austral. Math. Soc. 70(2004), 279-282. [14] Z. Ying and J. Chen, On quasipolar rings, Algebra Colloq. 4(2012), 683-692. 16 ORHAN GURGUN, SAIT HALICIOGLU, AND ABDULLAH HARMANCI Orhan Gurgun, Department of Mathematics, Ankara University, Turkey E-mail address: [email protected] Sait Halıcıoglu, Department of Mathematics, Ankara University, Turkey E-mail address: [email protected] Abdullah Harmanci, Department of Maths, Hacettepe University, Turkey E-mail address: [email protected]
1607.07117	1	1607	2016-07-25T00:47:53	Higher Order and Secondary Hochschild Cohomology	[ "math.RA" ]	In this note we give a generalization for the higher order Hochschild cohomology and show that the secondary Hochschild cohomology is a particular case of this new construction.	math.RA	math	Higher Order and Secondary Hochschild Cohomology Bruce R. Corrigan-Salter and Mihai D. Staic Abstract. In this note we give a generalization for the higher order Hochschild cohomology and show that the secondary Hochschild cohomology is a particular case of this new construction. 5 2 ] . A R h t a m [ 1 v 7 1 1 7 0 . 7 0 6 1 : v i X r a Introduction Hochschild cohomology is a useful tool for studying deformation theory and it was studied exten- sively over the years (for example see [3], [4], [5], [9], [10] and [14]). Higher order Hochschild (co)homology was introduced by Pirashvili in [11] (see also [1] and [6]). It associated to a simplicial set X•, a commutative k-algebra A and an A-bimodule M , the higher Hochschild cohomology groups H n (A, M ). When X• is the standard simplicial set associated to the X• sphere S1, one recovers the usual Hochschild cohomology. One important feature of these cohomology groups is that they depend only on the homotopy type of the geometric realization of the simplicial set X•. For more recent results about higher Hochschild cohomology see [2] and [7]. Secondary Hochschild cohomology was introduced in [12] where it was used to study B-algebra structures on the algebra A[[t]]. It associates to a triple (A, B, ε) (where ε gives the B-algebra structure on A), and an A-bimodule M that is B-symmetric, the secondary Hochschild cohomology groups H n((A, B, ε), M ). The main difference from the usual Hochschild cohomology is that the secondary Hochschild cohomology gives information about both, the product and the B-algebra structure on A[[t]]. Many of the results that are known for Hochschild cohomology can be extended to secondary Hochschild cohomology. One can study B-extensions of the algebra A, show that H •((A, B, ε), A) is a multiplicative operad and admits a Hodge type decomposition ([13]). There is also a cyclic version of the secondary cohomology that can be described using a generalization of the bar resolution ([8]). Our main goal in this paper is to show that secondary Hochschild cohomology is a certain version of higher order Hochschild cohomology. More precisely, we consider a simplicial pair (X•, Y•) (where Y• is a simplicial set and X• is simplicial subset of Y•), a triple (A, B, ε) (where A and B are commutative k-algebras, and ε : B → A is a morphism of k-algebras) and M a symmetric A-bimodule. To this setting we associate the groups H n (X•,Y•)((A, B, ε), M ). When X• = Y• we recover the higher order (A, M ). When (X•, Y•) = (S1, D2) with the natural simplicial structure, Hochschild cohomology H n X• we recover the secondary Hochschild cohomology H n((A, B, ε), M ). In this paper we fix a field k and denote ⊗k by ⊗. We assume that the reader is familiar with Hochschild cohomology, but provide some details for the discussion of higher order and secondary Hochschild cohomology. We also assume familiarity with simplicial sets. 1. Preliminaries 2010 Mathematics Subject Classification. Primary 16E40, Secondary 18G30. Key words and phrases. Hochschild Cohomology. 1 2 BRUCE R. CORRIGAN-SALTER AND MIHAI D. STAIC 1.1. Higher order Hochschild cohomology. We follow the description in [6] (see also [11]). Assume that A is a commutative k-algebra and M is a symmetric A-bimodule. Let V be a finite pointed set such that V = v + 1 (we identify it with v+ = {0, 1, ..., v} with 0 the fixed element) and define H(A, M )(V ) = H(A, M )(v+) = Homk(A⊗v, M ). For φ : V = v+ → W = w+ we define H(A, M )(φ) : H(A, M )(w+) → H(A, M )(v+) determined as follows: if f ∈ H(A, M )(w+) then H(A, M )(φ)(f )(a1 ⊗ ... ⊗ av) = b0f (b1 ⊗ ... ⊗ bw) where bi = Y {j∈V j6=0,φ(j)=i} aj. Take X• to be a pointed simplicial set. Suppose that Xn = sn + 1, we identify the set Xn with (sn)+ = {0, 1, ..., sn} then define C n X• = H(A, M )(Xn) = Homk(A⊗sn , M ). For each di : Xn+1 → Xn we define d∗ defined as ∂n = Pn+1 i=0 (−1)i(di)∗. The homology of this complex is denoted by H n X• (A, M ) and is called the higher order Hochschild cohomology group. One interesting fact is that these groups depend only on the homotopy type of the geometric realization of the simplicial set X•. i = H(A, M )(di) : C n X• → C n+1 X• and take ∂n : C n X• → C n+1 X• 1.2. Secondary Hochschild cohomology. We recall the construction from [12]. Let A be a k-algebra, B a commutative k-algebra, ε : B → A a morphism of k-algebras such that ε(B) ⊂ Z(A) and M an A-bimodule that is B-symmetric. We define C n((A, B, ε); M ) := Homk(A⊗n ⊗ B⊗ n(n−1) 2 , M ) and n−1 : C n−1((A, B, ε); M ) → C n((A, B, ε); M ) δε such that for f ∈ C n−1((A, B, ε); M ), we have: (1.1) δε n−1(f )   ⊗   a1 α1,2 1 a2 . . 1 1 1 1 ... α1,n−1 ... α2,n−1 ... ... ... an−1 . 1 α1,n α2,n . αn−1,n an     = HIGHER ORDER AND SECONDARY HOCHSCHILD COHOMOLOGY 3 a1ε(α1,2α1,3...α1,n)f  ⊗    a2 α2,3 1 a3 . . 1 1 1 1 ... α2,n−1 ... α3,n−1 ... ... ... an−1 1 . α2,n α3,n . αn−1,n an     + i=1 (−1)if Pn−1   (−1)nf   ⊗ ⊗   a1 α1,2 1 a2 . . 1 1 . . 1 1 1 1  a1 α1,2 1 a2 . . 1 1 1 1  . α1,iα1,i+1 α2,iα2,i+1 ... ... ... ... aiai+1ε(αi,i+1) ... ... ... . . . α1,n α2,n α1,n−1 α2,n−1 ... ... ... ... αi,n−1αi+1,n−1 αi,nαi+1,n ... ... ... αn−1,n an−1 . . an . . 1     + ... α1,n−2 ... α2,n−2 ... ... ... an−2 1 . α1,n−1 α2,n−1 . αn−2,n−1 an−1     anε(α1,nα2,n...αn−1,n). where ai ∈ A and αi,j ∈ B. It was proved in [12] that (C n((A, B, ε); M ), δε n) is a complex. The homol- ogy of this complex is called the secondary Hochschild cohomology and is denoted by H n((A, B, ε); M ). The homology and the cyclic version of this theory were discussed in [8]. 2. Main Construction In this section we introduce a new cohomology associated to a simplicial pair (X•, Y•) and a triple (A, B, ε) (where A and B are commutative k-algebra and ε : B → A is a morphism of k-algebras). We start with a few notations. Definition 2.1. We consider Γ2 to be the category whose objects are pairs (U, V ), where V is a finite pointed set with base point ∗, and U is a pointed subset of V . A morphism f ∈ HomΓ2((U1, V1), (U2, V2)) is a map of pointed sets f : V1 → V2 such that f (U1) ⊂ U2. Remark 2.2. The category of finite pointed sets Γ can be see as a full subcategory of Γ2 in two different ways. First we can take the inclusion given by V → (V, V ), second we can take the inclusion V → ({∗}, V ). Definition 2.3. A Γ2-module is a functor from Γop 2 to k-modules. Example 2.4. Let A and B be two commutative k-algebras, ε : B → A a morphism of k-algebras and M a symmetric A-bimodule. We construct L((A, B, ε); M ) : Γop 2 → k − mod to be the Γ2-module determined as follows. For (U, V ) ∈ Γ2 such that U = 1 + m and V = 1 + m+ n define L((A, B, ε); M )((U, V )) = Homk(A⊗m ⊗ B⊗n, M ). If f : (U1, V1) → (U2, V2) is a morphism in Γ2, we define L((A, B, ε); M )(f ) : Homk(A⊗m2 ⊗ B⊗n2 , M ) → Homk(A⊗m1 ⊗ B⊗n1 , M ), for ψ ∈ Homk(A⊗m2 ⊗ B⊗n2 , M ) then L((A, B, ε); M )(f )(ψ)(a1 ⊗ ... ⊗ am1 ⊗ α1 ⊗ ... ⊗ αn1 ) = b0 · ψ(b1 ⊗ ... ⊗ bm2 ⊗ β1 ⊗ ... ⊗ βn2 ) where for i ∈ U2 we have (2.1) bi = Y {j∈U1j6=∗, f (j)=i} aj Y {k∈V1\U1k6=∗, f (k)=i} ε(αk) ∈ A, 4 BRUCE R. CORRIGAN-SALTER AND MIHAI D. STAIC and for p ∈ V2 \ U2 we have βp = Y {q∈V1\U1q6=∗, f (q)=p} αq ∈ B. With the convention that if the product is taken over the empty set then we put bi = 1 ∈ A and βp = 1 ∈ B. We say that a pair (X•, Y•) is a simplicial pair if Y• is a simplicial set and X• a simplicial subset of Y•. In other words we have a functor (X•, Y•) : ∆op → Γop 2 . For a simplicial pair (X•, Y•) we define the higher order Hochschild cohomology associated to the triple (A, B, ε) and a symmetric A-bimodule M , to be the homology of the complex defined as follows. For every q ∈ N we consider (Xq, Yq) ∈ Γop (X•,Y•) = L((A, B, ε); M )((Xq, Yq)). We construct a complex by taking the differential induced by the simplicial structure on (X•, Y•). More precisely if di : Yq+1 → Yq then we define 2 and take C q δi = L((A, B, ε); M )(di) : C q (X• ,Y•) → C q+1 (X• ,Y•) and take ∂(X•,Y•) : C q (X• ,Y•) → C q+1 (X• ,Y•), (2.2) ∂(X•,Y•) = q+1 X i=0 (−1)iδi. Definition 2.5. The homology of the above complex is called the higher order Hochschild coho- mology associated to the simplicial pair (X•, Y•), of the triple (A, B, ε) with coefficients in M and is denoted by H q (X•,Y•)((A, B, ε); M ). Remark 2.6. In the event that X• = Y• this definition agrees with the definition of higher order Hochschild cohomology H q X• (A, M ). 3. Secondary Cohomology as a Higher Order Cohomology In this section we show that when A is commutative and M is a symmetric A-bimodule, then the secondary Hochschild cohomology H n((A, B, ε); M ) is a particular case of the construction from the previous section. Consider the simplicial pair (X•, Y•) = (S1, D2), where the sphere S1 is a obtained from the interval I = [01] by identifying the ends of the interval, and the disk D2 is obtained from the 2- simplex ∆ = [012] by collapsing the edges [01] and [12] (i.e. the boundary of D2 is the edge [02]). More precisely, we take X• to be the simplicial set where the only nondegenerate 1-simplex is b the simplex in dimension 0 is 0 ) = I 0 0 I = [02]. We denote by ∗n the base point in dimension n, and by I a n = a + b + 1, where we iterate the [0] vertex a times, and the [2] vertex b times. For example, I 0 the interval I with d0(I 0 and d2(I 1 0 is a 2-simplex [002] such that d0(I 1 0 ) = ∗0, and I 1 0 ) = d1(I 1 0 ) = d1(I 0 0 ) = ∗1. 0 and take d0(0∆0 For Y•, besides the above simplices, we also have a nondegenerate 2-simplex ∆ = [012]. Denote it by 0∆0 c the a+b+c+2- dimensional simplex obtained by iterating the [0] vertex a times, the [1] vertex b times, and the [2] vertex c times. For example 1∆0 0) = I 1 0 , and d3(1∆0 0 is a 3-simplex [0012] with d0(1∆0 0 . More generally, take a∆b 0) = ∗1 and d1(0∆0 0) = d1(1∆0 0) = d2(0∆0 0) = I 0 0) = 0∆0 0, d2(1∆0 0) = ∗2. In general we have X n = {∗n} ∪ {I a b a, b ∈ N, a + b = n − 1} and Y n = X n ∪ {a∆b c a, b, c ∈ N, a + b + c = n − 2}. The di : Y n → Y n−1 are defined as follows: (3.1) di(∗n) = ∗n−1, HIGHER ORDER AND SECONDARY HOCHSCHILD COHOMOLOGY 5 (3.2) di(I a b ) = ∗a+b I a−1 b I a b−1 ∗a+b   if a = 0 and i = 0 if a 6= 0 and i ≤ a if b 6= 0 and i > a if b = 0 and i = n = a + 1, (3.3) di( a∆b c) =   ∗a+b+c+1 a−1∆b c I a c a∆b−1 ∗a+b+c+1 a∆b c−1 c if a = 0 and i = 0 if a 6= 0 and i ≤ a if b = 0 and i = a + 1 if b 6= 0 and a < i ≤ a + b + 1 if c = 0 and i = n = a + b + 2 if c 6= 0 and i ≥ a + b + 2. Notice that I a b is degenerate if a + b > 0, and a∆b c is degenerate if a + b + c > 0. Also we have that X n = 1 + n and Y n = 1 + n + n(n−1) . In particular we get that 2 L((A, B, ε); M )((X n, Y n)) = Homk(A⊗n ⊗ B⊗ n(n−1) 2 , M ). Next we need to make the identification with the notation from [12]. First recall that an element in A⊗n ⊗ B⊗ n(n−1) was represented by a tensor matrix 2 a1,1 α1,2 a2,2 T = ⊗   1 . 1 1 . 1 1 α1,n α2,n α1,n−1 α2,n−1 ... ... ... ... an−1,n−1 αn−1,n ... an,n . . 1   , where ai,i ∈ A and αi,j ∈ B. For a, b ∈ N with a + b + 1 = n the element I a b ∈ Y n corresponds to the position (a + 1, a + 1) c ∈ Y n corresponds to in the tensor matrix. For a, b, c ∈ N with a + b + c + 2 = n the element a∆b the position (a + 1, a + b + 2) = (a + 1, n − c) in the tensor matrix. We also add the symbol (0, 0) to correspond to ∗n. With the above identifications the formulas (3.2) and (3.3) become: di((a + 1, a + 1)) = (0, 0) (a, a) (a + 1, a + 1) (0, 0) if a = 0 and i = 0 if a 6= 0 and i ≤ a if b 6= 0 and i > a if b = 0 and i = n = a + 1,   di(a + 1, a + b + 2) = (0, 0) (a, a + b + 1) (a + 1, a + 1) (a + 1, a + b + 1) (0, 0) (a + 1, a + b + 2) if a = 0 and i = 0 if a 6= 0 and i ≤ a if b = 0 and i = a + 1 if b 6= 0 and a < i ≤ a + b + 1 if c = 0 and i = n = a + b + 2 if c 6= 0 and i ≥ a + b + 2.   If we use the above identification the formula for ∂(S1,D2) from equation (2.2) is the same as the formula for differential δε n−1 from equation (1.1). To summarize we have the following result. Theorem 3.1. Let A and B be commutative k-algebras, ε : B → A a morphism of k-algebras and M a symmetric A-bimodule, then we have H q((A, B, ε); M ) ≃ H q (S1,D2)((A, B, ε); M ). 6 Bibliography 4. Some Remarks One can see that H q (X•,Y•)((A, B, ε); M ) is functorial with respect to all of its entries. More precisely, let (A1, B1, ε1) and (A2, B2, ε2) be two triples, M a symmetric A2-bimodule, and f : A1 → A2 a morphism of k-algebras such that f (B1) ⊆ B2 and f ε1(b) = ε2(f (b)), then we have the natural morphism f ∗ : H q (X•,Y•)((A2, B2, ε2); M ) → H q (X•,Y•)((A1, B1, ε1); M ), where the A1-bimodule structure on M is induced by f . Also, if g : M → N is a morphism of symmetric A-bimodules then g∗ : H q (X•,Y•)((A, B, ε); M ) → H q (X• ,Y•)((A, B, ε); N ). Moreover if h : (X•, Y•) → (Z•, T•) is a morphism of simplicial sets then h∗ : H q (Z•,T•)((A, B, ε); M ) → H q (X•,Y•)((A, B, ε); M ). If we take the natural inclusion of simplicial pairs i : (S1, S1) → (S1, D2) with the simplicial structure discussed in the previous section, then in : H n (S1,D2)((A, B, ε); M ) → H n (S1,S1)((A, B, ε); M ) is nothing else but the morphism Φn : H n((A, B, ε); M ) → H n(A, M ) discussed in [13]. One natural question is whether the construction in this paper depends only of the homotopy type of the geometric realization of the simplicial pair (X•, Y•) (or maybe invariant under a certain equivalence relation among simplicial pairs). We explored this problem but we were not able to prove any interesting result. The main issue is finding an equivalence relation among simplicial pairs that is manageable at the algebraic level. The authors would like to thank Andrew Salch for conversations and suggestions about this research. Bruce would also like to thank his wife Kendall for her continued support. Acknowledgment References [1] D. W. Anderson. Chain functors and homology theories. In Symposium on Algebraic Topology (Battelle Seattle Res. Center, Seattle, Wash., 1971), pages 1 -- 12. Lecture Notes in Math., Vol. 249. Springer, Berlin, 1971. [2] Bruce R. Corrigan-Salter. Coefficients for higher order Hochschild cohomology. Homology Homotopy Appl., 17(1):111 -- 120, 2015. [3] Murray Gerstenhaber. The cohomology structure of an associative ring. Ann. of Math. (2), 78:267 -- 288, 1963. [4] Murray Gerstenhaber. On the deformation of rings and algebras. Ann. of Math. (2), 79:59 -- 103, 1964. [5] Murray Gerstenhaber and Samuel D. Schack. Algebraic cohomology and deformation theory. In Deformation theory of algebras and structures and applications (Il Ciocco, 1986), volume 247 of NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., pages 11 -- 264. Kluwer Acad. Publ., Dordrecht, 1988. [6] Gr´egory Ginot. Higher order Hochschild cohomology. C. R. Math. Acad. Sci. Paris, 346(1-2):5 -- 10, 2008. [7] Gr´egory Ginot, Thomas Tradler, and Mahmoud Zeinalian. Higher Hochschild homology, topological chiral homology and factorization algebras. Comm. Math. Phys., 326(3):635 -- 686, 2014. [8] Jacob Laubacher, Mihai D. Staic, and Alin Stancu. Bar Simplicial Modules and Secondary Cyclic (Co)homology. ArXiv e-prints, May 2016. [9] Olav Arnfinn Laudal. Formal moduli of algebraic structures, volume 754 of Lecture Notes in Mathematics. Springer, Berlin, 1979. [10] Jean-Louis Loday. Cyclic homology, volume 301 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, 1992. Appendix E by Mar´ıa O. Ronco. [11] Teimuraz Pirashvili. Hodge decomposition for higher order Hochschild homology. Ann. Sci. ´Ecole Norm. Sup. (4), 33(2):151 -- 179, 2000. [12] Mihai D. Staic. Secondary Hochschild cohomology. Algebr. Represent. Theory, 19(1):47 -- 56, 2016. [13] Mihai D. Staic and Alin Stancu. Operations on the secondary Hochschild cohomology. Homology Homotopy Appl., 17(1):129 -- 146, 2015. [14] Donald Yau. Deformation theory of modules. Comm. Algebra, 33(7):2351 -- 2359, 2005. Bibliography 7 Department of Mathematics, Wayne State University, Detroit, MI 48202, USA E-mail address: [email protected] Department of Mathematics and Statistics, Bowling Green State University, Bowling Green, OH 43403 E-mail address: [email protected]
1009.2061	3	1009	2012-04-06T09:05:59	Moduli spaces for point modules on naive blowups	[ "math.RA", "math.AG" ]	The naive blow-up algebras developed by Keeler-Rogalski-Stafford, after examples of Rogalski, are the first known class of connected graded algebras that are noetherian but not strongly noetherian. This failure of the strong noetherian property is intimately related to the failure of the point modules over such algebras to behave well in families: puzzlingly, there is no fine moduli scheme for such modules, although point modules correspond bijectively with the points of a projective variety X. We give a geometric structure to this bijection and prove that the variety X is a coarse moduli space for point modules. We also describe the natural moduli stack \tilde{X} for embedded point modules---an analog of a "Hilbert scheme of one point"---as an infinite blow-up of X and establish good properties of \tilde{X}. The natural map \tilde{X} -> X is thus a kind of "Hilbert-Chow morphism of one point" for the naive blow-up algebra.	math.RA	math	MODULI SPACES FOR POINT MODULES ON NAIVE BLOWUPS THOMAS A. NEVINS AND SUSAN J. SIERRA Abstract. The naıve blow-up algebras developed by Keeler-Rogalski-Stafford [KRS05], after examples of Rogalski [Rog04], are the first known class of connected graded algebras that are noetherian but not strongly noetherian. This failure of the strong noetherian property is intimately related to the failure of the point modules over such algebras to behave well in families: puzzlingly, there is no fine moduli scheme for such modules, although point modules correspond bijectively with the points of a projective variety X. We give a geometric structure to this bijection and prove that the variety X is a coarse moduli space for point modules. We also describe the natural moduli stack X∞ for embedded point modules -- an analog of a "Hilbert scheme of one point" -- as an infinite blow-up of X and establish good properties of X∞. The natural map X∞ → X is thus a kind of "Hilbert-Chow morphism of one point" for the naıve blow-up algebra. 1. Introduction One of the important achievements of noncommutative projective geometry is the classification of noncom- mutative projective planes, such as the 3-dimensional Sklyanin algebra Skl3, by Artin-Tate-Van den Bergh [ATV90]. More formally, these are Artin-Schelter regular algebras of dimension 3: noncommutative graded rings that are close analogs of a commutative polynomial ring in 3 variables; see [SV01] for a discussion. The key method of Artin-Tate-Van den Bergh is to study point modules; that is, cyclic graded modules with the Hilbert series of a point in projective space. Given a noncommutative projective plane R, Artin-Tate-Van den Bergh describe a moduli scheme for its point modules. This allows them to construct a homomorphism from R to a well-understood ring, providing a first step in describing the structure of the noncommutative plane itself. The techniques described above work in a more general context. Let k be an algebraically closed field; we assume k is uncountable, although for some of the results quoted this hypothesis is unnecessary. A k-algebra R is said to be strongly noetherian if for any commutative noetherian k-algebra C, the tensor product R ⊗k C is again noetherian. By a general result of Artin-Zhang [AZ01, Theorem E4.3], if R is a strongly noetherian N-graded k-algebra, then its point modules are parameterized by a projective scheme. Rogalski-Zhang [RZ08] used this result to extend the method of [ATV90] to strongly noetherian connected graded k-algebras that are generated in degree 1. (An N-graded k-algebra R is connected graded if R0 = k.) Their method constructs a map from the algebra to a twisted homogeneous coordinate ring (see Section 2 for definitions) on the scheme X parameterizing point modules. For example, Sklyanin algebras are strongly noetherian, and here X is an elliptic curve. The homomorphism here gives the well-known embedding of an elliptic curve in a noncommutative P2. Although it was believed for a time that all connected graded noetherian algebras would be strongly noetherian, Rogalski [Rog04] showed this was not the case. His example was generalized in joint work with Keeler and Stafford [KRS05, RS07] to give a geometric construction of a beautiful class of noncommutative graded algebras, known as naıve blow-ups, that are noetherian but not strongly noetherian. Along the way, they showed that point modules for naıve blow-ups -- viewed as objects of noncommutative projective geometry, in a way we make precise below -- cannot behave well in families: there is no fine moduli scheme of finite type for such modules. In the present paper, we systematically develop the moduli theory of point modules for the naıve blow-ups S of [KRS05, RS07]. Roughly speaking, we show that there is an analog of a "Hilbert scheme of one point on Proj(S)" that is an infinite blow-up of a projective variety. This infinite blow-up is quasi-compact and noetherian as an fpqc-algebraic stack (a notion we make precise in Section 4). Furthermore, we show there is a coarse "moduli space for one point on Proj(S)" -- it is, in fact, the projective variety from which the naıve 1 blowup was constructed. These are the first descriptions in the literature of moduli structures for point modules on a naıve blow-up. More precisely, let X be a projective k-variety of dimension at least 2, let σ be an automorphism of X, and let L be a σ-ample (see Section 2) invertible sheaf on X. We follow the standard convention that Lσ := σ∗L. Let P ∈ X (in the body of the paper we let P be any 0-dimensional subscheme of X), and assume that the σ-orbit of P is critically dense: that is, it is infinite and every infinite subset is Zariski dense. For n ≥ 0, let In := IP I σ P · · · I σn−1 P and Ln := L ⊗ Lσ ⊗ · · · ⊗ Lσn−1 . Define Sn := In ⊗ Ln and let S := S(X, L, σ, P ) :=Mn≥0 H 0(X, Sn). The algebra S is the naıve blow-up associated to the data (X, L, σ, P ). If L is sufficiently ample, then S is generated in degree one; alternatively, a sufficiently large Veronese of S is always generated in degree one. We will assume throughout that S is generated in degree one. A point module is a graded cyclic S-module M with Hilbert series 1+t+t2 +· · · . We say M is an embedded point module if we are given, in addition, a surjection S → M of graded modules. Two embedded point modules M and M ′ are isomorphic if there is an S-module isomorphism from M to M ′ that intertwines the maps from S. We begin by constructing a moduli stack for embedded point modules. Recall that X∞ is a fine moduli space (or stack) for embedded point modules if there is an S-module quotient S ⊗k OX∞ → M that is a universal family for point modules: that is, M is an X∞-flat family of embedded S-point modules, with the property that, if S ⊗k C → M ′ is any C-flat family of embedded point modules for a commutative k-algebra −→ X∞ and an isomorphism f ∗M ∼= M ′ of families of embedded S- C, then there is a morphism Spec(C) point modules. Let Xn be the blowup of X at In; there is an inverse system · · · → Xn → Xn−1 → · · · → X of schemes. Let X∞ := lim Xn. This inverse limit exists as a stack. More precisely, in Definition 4.1, we ←− introduce the notion of an fpqc-algebraic stack. We then have: f Theorem 1.1. The inverse limit X∞ is a noetherian fpqc-algebraic stack. The morphism X∞ → X is quasicompact. Moreover, X∞ is a fine moduli space for embedded S-point modules. We have been told that similar results were known long ago to M. Artin; however, they seem not to have been very widely known even among experts, nor do they seem to have appeared in the literature. Note that the stack X∞ is discrete: its points have no stabilizers. Thus, X∞ is actually a k-space in the terminology of [LMB00]; in particular, this justifies our use of the phrase "fine moduli space" in the statement of the theorem. However, X∞ does not seem to have an ´etale cover by a scheme, and hence does not have the right to be called an algebraic space. We recall that, by definition, the noncommutative projective scheme associated to S is the quotient category Qgr- S = Gr- S/ Tors- S of graded right S-modules by the full subcategory of locally bounded modules. A point object in Qgr- S is the image of (a shift of) a point module. If S is a commutative graded algebra generated in degree 1, Qgr- S is equivalent to the category of quasicoherent sheaves on Proj(S); this justifies thinking of Qgr- S as the noncommutative analog of a projective scheme. If R is strongly noetherian and generated in degree 1, then a result of Artin-Stafford [KRS05, The- orem 10.2] shows that point objects of Qgr-R are parameterized by the same projective scheme X that parameterizes embedded point modules. On the other hand, for naıve blowups S = S(X, L, σ, P ) as above, Keeler-Rogalski-Stafford show: Theorem 1.2 ([KRS05], Theorem 1.1). The algebra S is noetherian but not strongly noetherian. Moreover, there is no fine moduli scheme of finite type over k parameterizing point objects of Qgr-S. By contrast, [KRS05] gives a simple classification (that fails in families), namely that point objects are in bijective correspondence with points of X: to a point x ∈ X we associate the S-moduleL H 0(X, kx ⊗ Ln). In the present paper, we explain how these two facts about point objects of Qgr-S naturally fit together. Assume that L is sufficiently ample (in the body of the paper we work with any σ-ample L by considering shifts of point modules). Let F be the moduli functor of embedded point modules over S. Define an equivalence relation ∼ on F (C) by saying that M ∼ N if (their images) are isomorphic in Qgr- S ⊗k C. We 2 obtain a functor G : Affine schemes −→ Sets by sheafifying (in the fpqc topology) the presheaf Gpre of sets defined by Spec C 7→ F (C)/ ∼. A scheme Y is a coarse moduli scheme for point objects if it corepresents the functor G: that is, there is a natural transformation G → Homk(−, Y ) that is universal for natural transformations from G to schemes. Our main result is: Theorem 1.3. The variety X is a coarse moduli scheme for point objects in Qgr- S. This gives a geometric structure to the bijection discovered by Keeler-Rogalski-Stafford. Corollary 1.4. There is a fine moduli space X∞ for embedded S-point modules but only a coarse moduli scheme X for point objects of Qgr- S. It may be helpful to compare the phenomenon described by Corollary 1.4 to a related, though quite different, commutative phenomenon. Namely, let Y be a smooth projective (commutative) surface. Fix n ≥ 1. Let R = C[Y ] denote a homogeneous coordinate ring of Y (associated to a sufficiently ample invertible sheaf on Y ), and consider graded quotient modules R → M such that dim Mℓ = n for ℓ ≫ 0. By a general theorem of Serre, the moduli space for such quotients is the Hilbert scheme of n points on Y , denoted Hilbn(Y ). This is a smooth projective variety of dimension 2n. Alternatively, remembering only the corresponding objects [M ] of Qgr- R ≃ Qcoh(Y ), and imposing the further S-equivalence relation (see Example 4.3.6 of [HL97]), we get the moduli space Symn(Y ) for semistable length n sheaves on Y , which equals the nth symmetric product of Y . The latter moduli space is only a coarse moduli space for semistable sheaves. One has the Hilbert-Chow morphism Hilbn(Y ) → Symn(Y ) which is defined by taking a quotient R → M to the equivalence class of M . It is perhaps helpful to view the moduli spaces and map X∞ → X associated to the algebra S in light of the theorems stated above: that is, as a kind of "noncommutative Hilbert-Chow morphism of one point" for a naıve blow-up algebra S(X, L, σ, P ). In work in preparation, we generalize the results in [RZ08] by proving a converse, of sorts, to Theorem 1.3. Namely, suppose R is a connected graded noetherian algebra generated in degree one; that R has a fine moduli space X∞ for embedded point modules; that R has a projective coarse moduli scheme X for point objects of Qgr- R; and that the spaces X∞ and X and the morphism X∞ → X between them have geometric properties similar to those of the spaces we encounter in the theorems above. Then, we show, there exist an automorphism σ of X, a zero-dimensional subscheme P ⊂ X supported on points with critically dense orbits, an ample and σ-ample invertible sheaf L on X, and a homomorphism φ : R → S(X, L, σ, P ) from R to the naıve blow-up associated to this data; furthermore, φ is surjective in large degree. This construction gives a new tool for analyzing the structure of rings that are noetherian but not strongly noetherian. Details will appear in [NS]. Acknowledgements. The authors are grateful to B. Conrad, J.T. Stafford, and M. Van den Bergh for helpful conversations; to S. Kleiman for help with references; and to the referee for several helpful questions and comments. The first author was supported by NSF grant DMS-0757987. The second author was supported by an NSF Postdoctoral Research Fellowship, grant DMS-0802935. 2. Background In this section, we give needed definitions and background. We begin by discussing bimodule algebras: this is the correct way to think of the sheaves Sn defined above. Most of the material in this section was developed in [Van96] and [AV90], and we refer the reader there for references. Our presentation follows that in [KRS05] and [Sie11]. Convention 2.1. Throughout the paper, by variety (over k) we mean an integral separated scheme of finite type over k. Throughout this section, let k be an algebraically closed field and let A denote an affine noetherian k-scheme, which we think of as a base scheme. Definition 2.2. Let X be a scheme of finite type over A. An OX -bimodule is a quasicoherent OX×X - module F , such that for every coherent submodule F ′ ⊆ F , the projection maps p1, p2 : Supp F ′ → X are both finite morphisms. The left and right OX -module structures associated to an OX -bimodule F are 3 defined respectively as (p1)∗F and (p2)∗F . We make the notational convention that when we refer to an OX -bimodule simply as an OX -module, we are using the left-handed structure (for example, when we refer to the global sections or higher cohomology of an OX -bimodule). All OX -bimodules are assumed to be OA-symmetric. There is a tensor product operation on the category of bimodules that has the expected properties; see [Van96, Section 2]. All the bimodules that we consider will be constructed from bimodules of the following form: Definition 2.3. Let X be a projective scheme over A and let σ, τ ∈ AutA(X). Let (σ, τ ) denote the map X → X ×A X defined by x 7→ (σ(x), τ (x)). If F is a quasicoherent sheaf on X, we define the OX -bimodule σFτ to be σFτ = (σ, τ )∗F . If σ = 1 is the identity, we will often omit it; thus we write Fτ for 1Fτ and F for the OX -bimodule 1F1 = ∆∗F , where ∆ : X → X ×A X is the diagonal. Definition 2.4. Let X be a projective scheme over A. An OX -bimodule algebra, or simply a bimodule algebra, B is an algebra object in the category of bimodules. That is, there are a unit map 1 : OX → B and a product map µ : B ⊗ B → B that have the usual properties. We follow [KRS05] and define associativity conditions. Definition 2.5. Let X be a projective scheme over A and let σ ∈ AutA(X). A bimodule algebra B is a graded (OX , σ)-bimodule algebra if: (2) B0 = OX ; (3) the multiplication map µ is given by OX -module maps Bn ⊗ Bσn (1) There are coherent sheaves Bn on X such that B =Ln∈Z 1(Bn)σn ; Definition 2.6. Let X be a projective scheme over A and let σ ∈ AutA(X). Let R = Ln∈Z(Rn)σn be is a direct sum decomposition M = Mn∈Z a graded (OX , σ)-bimodule algebra. A right R-module M is a quasicoherent OX -module M together with a right OX -module map µ : M ⊗ R → M satisfying the usual axioms. We say that M is graded if there (Mn)σn with multiplication giving a family of OX -module maps m → Bn+m, satisfying the obvious Mn ⊗ Rσn m → Mn+m, obeying the appropriate axioms. We say that M is coherent if there are a coherent OX -module M′ and a surjective map M′ ⊗ R → M of ungraded R-modules. We make similar definitions for left R-modules. The bimodule algebra R is right (left) noetherian if every right (left) ideal of R is coherent. A graded (OX , σ)-bimodule algebra is right (left) noetherian if and only if every graded right (left) ideal is coherent. We recall here some standard notation for module categories over rings and bimodule algebras. Let C be a commutative ring and let R be an N-graded C-algebra. We define Gr-R to be the category of Z-graded right R-modules; morphisms in Gr-R preserve degree. Let Tors-R be the full subcategory of modules that are direct limits of right bounded modules. This is a Serre subcategory of Gr-R, so we may form the quotient category Qgr-R := Gr-R/ Tors-R. (We refer the reader to [Gab62] as a reference for the category theory used here.) There is a canonical quotient functor from Gr-R to Qgr-R. We make similar definitions on the left. Further, throughout this paper, we adopt the convention that if Xyz is a category, then xyz is the full subcategory of noetherian objects. Thus we have gr-R and qgr-R, R-qgr, etc. If X is a scheme, we will denote the category of quasicoherent (respectively coherent) sheaves on X by OX -Mod (respectively OX -mod). Given a module M ∈ gr-R, we define M [n] :=Li∈Z M [n]i, where M [n]i = Mn+i. For a graded (OX , σ)-bimodule algebra R, we likewise define Gr-R and gr-R. The full subcategory Tors-R of Gr-R consists of direct limits of modules that are coherent as OX -modules, and we similarly define Qgr-R := Gr-R/ Tors-R. We define qgr-R in the obvious way. 4 H 0(X, Rn) ⊗A H 0(X, Rm) 1⊗σn µ / H 0(X, Rn+m). If R is an OX -bimodule algebra, its global sections H 0(X, R) inherit an OA-algebra structure. We call H 0(X, R) the section algebra of R. multiplication on H 0(X, R) is induced from the maps If R = L(Rn)σn is a graded (OX , σ)-bimodule algebra, then If M is a graded right R-module, then H 0(X, M) =Ln∈Z H 0(X, Mn) is a right H 0(X, R)-module in If R = H 0(X, R) and M is a graded right R-module, define M ⊗R R to be the sheaf associated to the ⊗R R : Gr-R → Gr-R is a presheaf V 7→ M ⊗R R(V ). This is a graded right R-module, and the functor right adjoint to H 0(X, ) is a functor from Gr-R to Gr-H 0(X, R). / H 0(X, Rn) ⊗A H 0(X, Rσn m ) the obvious way; thus H 0(X, ). The following is a relative version of a standard definition. Definition 2.7. Let A be an affine k-scheme and let q : X → A be a projective morphism. Let σ ∈ AutA(X), and let {Rn}n∈N be a sequence of coherent sheaves on X. The sequence of bimodules {(Rn)σn}n∈N is right ample if for any coherent OX -module F , the following properties hold: (1) F ⊗ Rn is globally generated for n ≫ 0 -- that is, the natural map q∗q∗(F ⊗ Rn) → F ⊗ Rn is surjective for n ≫ 0; (2) Riq∗(F ⊗ Rn) = 0 for n ≫ 0 and i ≥ 1. The sequence {(Rn)σn }n∈N is left ample if for any coherent OX -module F , the following properties hold: (1) the natural map q∗q∗(Rn ⊗ F σn (2) Riq∗(Rn ⊗ F σn If A = k, we say that an invertible sheaf L on X is σ-ample if the OX -bimodules ) = 0 for n ≫ 0 and i ≥ 1. is surjective for n ≫ 0; ) → Rn ⊗ F σn {(Ln)σn}n∈N = {L⊗n σ }n∈N form a right ample sequence. By [Kee00, Theorem 1.2], this is true if and only if the OX -bimodules {(Ln)σn }n∈N form a left ample sequence. The following result is a special case of a result due to Van den Bergh [Van96, Theorem 5.2], although we follow the presentation of [KRS05, Theorem 2.12]: Theorem 2.8 (Van den Bergh). Let X be a projective k-scheme and let σ be an automorphism of X. Let R = L(Rn)σn be a right noetherian graded (OX , σ)-bimodule algebra, such that the bimodules {(Rn)σn} form a right ample sequence. Then R = H 0(X, R) is also right noetherian, and the functors H 0(X, ⊗R R induce an equivalence of categories qgr-R ≃ qgr-R. ) and Castelnuovo-Mumford regularity is a useful tool for measuring ampleness and studying ample sequences. We will need to use relative Castelnuovo-Mumford regularity; we review the relevant background here. In the next three results, let X be a projective k-scheme, and let A be a noetherian k-scheme. Let XA := X × A and let p : XA → X and q : XA → A be the projection maps. Fix a very ample invertible sheaf OX (1) on X. Let OXA (1) := p∗OX (1); note OXA (1) is relatively ample for q : XA → A. If F is a coherent sheaf on XA and n ∈ Z, let F (n) := F ⊗XA OXA (1)⊗n. We say F is m-regular with respect to OXA (1), or just m-regular, if Riq∗F (m − i) = 0 for all i > 0. Since OXA (1) is relatively ample, F is m-regular for some m. The regularity of F is the minimal m for which F is m-regular; we write it reg(F ). Castelnuovo-Mumford regularity is usually defined only for k-schemes, so we will spend a bit of space on the technicalities of working over a more general base. First note that we have: Lemma 2.9. Let F be a coherent sheaf on X. Then reg(F ) = reg(p∗F ). (cid:3) The fundamental result on Castelnuovo-Mumford regularity is due to Mumford: Theorem 2.10. ([Laz04, Example 1.8.24]) Let F be an m-regular coherent sheaf on XA. Then for every n ≥ 0: (1) F is (m + n)-regular; (2) F (m + n) is generated by its global sections: that is, the natural map q∗q∗F (m + n) → F (m + n) is surjective; 5 / / (3) the natural map q∗F (m) ⊗A q∗OXA (n) → q∗F (m + n) is surjective. We also have Lemma 2.11. For any 0-regular invertible sheaf H on XA and any A-point y of X, the natural map is surjective. q∗H α→ q∗(Oy ⊗XA H) Proof. This is standard, but we check the details. Since cohomology commutes with flat base change, it suffices to consider the case that A = Spec C where C is a local ring. Then for any n ∈ Z, we may consider Oy ⊗XA H(n) as an invertible sheaf on A. Since C is local, as a C-module this is isomorphic to C. We thus have q∗(Oy ⊗XA H) ∼= C. Let I := Im(α); this is an ideal of C. Let n ≥ 0 and consider the natural maps (2.12) q∗H ⊗C q∗OXA (n) µ q∗H(n) -❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ α⊗1 f / q∗(Oy ⊗XA H) ⊗C q∗OXA (n) I ⊗C q∗OXA (n) / q∗(Oy ⊗XA H(n)) ∼= C. This diagram clearly commutes, and by construction α ⊗ 1 factors through I ⊗C q∗OXA (n). Thus Im f ⊆ I for all n. On the other hand, by Theorem 2.10(3), µ is surjective. As OXA (1) is relatively ample, for n ≫ 0 the (cid:3) right hand vertical map is surjective. Thus f is surjective for n ≫ 0, and so I = C. Let Z be a closed subscheme of XA. We say that Z has relative dimension ≤ d if for all x ∈ A, the fiber q−1(x) has dimension ≤ d as a k(x)-scheme. The following is a relative version of a result of Keeler. Proposition 2.13. (cf. [Kee06, Proposition 2.7]) Let X be a projective k-scheme. There is a constant D, depending only on X and on OX (1), so that the following holds: for any noetherian k-scheme A, and for any coherent sheaves F , G on XA so that the closed subscheme of XA where F and G both fail to be locally free has relative dimension ≤ 2, we have reg(F ⊗XA G) ≤ reg(F ) + reg(G) + D. Proof. The statement is local on the base, so we may assume without loss of generality that A = Spec C is affine. Since standard results such as Theorem 2.10 and Lemma 2.11 hold in this relative context, we may repeat the proof of [Kee06, Proposition 2.7]. The relative dimension assumption ensures the vanishing of Rq∗ that is needed in the proof. (cid:3) To end the introduction, we define naıve blowups: these are the algebras and bimodule algebras that we will work with throughout the paper. Let X be a projective k-variety. Let σ ∈ Autk(X) and let L be a σ-ample invertible sheaf on X. Let P be a 0-dimensional subscheme of X. We define ideal sheaves for n ≥ 0. Then we define a bimodule algebra S(X, L, σ, P ) := Ln≥0(Sn)σn , where Sn := InLn. Define S(X, L, σ, P ) := H 0(X, S(X, L, σ, P )). We recall the main results of [RS07]: In := IP I σ P · · · I σn−1 P Theorem 2.14 ([RS07], Theorems 1.2 and 3.1). Let X be a projective k-variety with dim X ≥ 2. Let σ ∈ Autk(X) and let L be a σ-ample invertible sheaf on X. Let P be a 0-dimensional subscheme of X. Let S := S(X, L, σ, P ) and let S := S(X, L, σ, P ). If all points in P have critically dense σ-orbits, then: (1) The sequence of bimodules {(Sn)σn } is a left and right ample sequence. (2) S and S are left and right noetherian, and the categories qgr-S and qgr-S are equivalent via the global sections functor. Likewise, S-qgr and S-qgr are equivalent. 6 / / - / / (3) The isomorphism classes of simple objects in qgr-S ≃ qgr-S are in 1-1 correspondence with the closed points of X, where x ∈ X corresponds to the S-module L kx ⊗ Ln. However, the simple objects in qgr-S are not parameterized by any scheme of finite type over k. For technical reasons, we will want to assume that our naıve blowup algebras S is generated in degree one. By [RS07, Propositions 3.18, 3.19], this will always be true if we either replace S by a sufficiently large Veronese or replace L by a sufficiently ample line bundle (for example, if L is ample, by a sufficiently high tensor power of L). If S is generated in degree one, then by [RS07, Corollary 4.11] the simple objects in qgr-S are the images of shifts of point modules. 3. Blowing up arbitrary 0-dimensional schemes For the rest of the paper, let k be an uncountable algebraically closed field. Let X be a projective variety over k, let σ ∈ Autk(X), and let L be a σ-ample invertible sheaf on X. Let P be a 0-dimensional subscheme of X, supported at points with dense (later, critically dense) orbits. Let S := S(X, L, σ, P ) and let S := S(X, L, σ, P ). In this paper, we compare three objects: the scheme parameterizing length n truncated point modules over S, the scheme parameterizing length n truncated point modules over S, and the blowup of X at the ideal sheaf In = IP · · · I σn−1 . In this section, we focus on the blowup of X. We first give some general lemmas on blowing up the defining ideals of 0-dimensional schemes. These are elementary, but we give proofs for completeness. P Suppose that X is a variety and that f : Y → X is a surjective, projective morphism of schemes. Let η be the generic point of X. We define Y o := f −1(η), and refer to Y o, by abuse of terminology, as the relevant component of Y . In our situation, f will always be generically one-to-one and Y o will be irreducible, with f Y o birational onto its image. Lemma 3.1. Let A be a variety of dimension ≥ 2. Let I be the ideal sheaf of a 0-dimensional subscheme of A, and let π : X → A be the blowup of A at I. Let W be the scheme parameterizing colength 1 ideals inside I. Let φ : W → A be the canonical morphism that sends an ideal J to the support of I/J . Then there is a closed immersion c : X → W that gives an isomorphism between X and W o. Further, the diagram X ❅❅❅❅❅❅❅❅ π c A W ~⑥⑥⑥⑥⑥⑥⑥⑥ φ commutes. Proof. Without loss of generality A = Spec C is affine; let I := I(A). We may identify W with Proj SymC(I) (see Proposition 2.2 of [Kle90]); under this identification φ : W → A is induced by the inclusion C ֒→ SymC (I). There is a canonical surjective map of graded C-algebras SymC (I) → C ⊕Ln≥1 I n, which is the identity on C. This induces a closed immersion c : X → W with φc = π as claimed. Further, both π : X → A and φ : W → A are isomorphisms away from Cosupp I. Thus c gives a birational closed immersion (and therefore an isomorphism) onto W o. (cid:3) Lemma 3.2. Let A be a variety of dimension ≥ 2, and let I and J be ideal sheaves on A. Let K := IJ . Define i : X → A to be the blowup of A at I, j : Y → A to be the blowup of A at J , and k : Z → A to be the blowup of A at K. (a) There are morphisms ξ : Z → X and ω : Z → Y so that the diagram Z ω Y commutes. (b) We have Z ∼= (X ×A Y )o. ❅❅❅❅❅❅❅❅ k X i / A ξ j 7 / / ~ / / / (c) Let W be the moduli scheme of subsheaves of K of colength 1, and let V be the moduli scheme of subsheaves of I of colength 1. Let c : Z → W and d : X → V be the maps from Lemma 3.1, and let Z ′ := c(Z) and X ′ := d(X). Then the map ξ′ : Z ′ → X ′ induced from ξ sends K′ ⊂ K to (K′ : J ) ∩ I. Proof. (a). Since ξ−1(K)OZ = ξ−1(I)ξ−1(J )OZ is invertible, the inverse images of both I and J on Z are invertible. By the universal property of blowing up ([Har77, Proposition 7.14]), the morphisms ξ : Z → X and ω : Z → Y exist and commute as claimed. For (b), (c) we may without loss of generality assume that A = Spec C is affine. (b). Let U := (X ×A Y )o. Let A′ := A r Cosupp K. Then U is the closure of A′ in Pn A ×A Pm A for appropriate n, m. Let φ : Pn A ×A Pm Z ⊆ W ⊆ Pnm+m+n A closure of A′ in Pnm+n+m A , we have φ′(U ) = Z. A → Σn,m ⊂ Pnm+m+n be the Segre embedding. Note that the canonical embeddings actually have W ⊆ Σn,m. Since φ′ := φU is the identity over A′ and Z ⊆ Σn,m is the A Let p : X ×A Y → X and q : X ×A Y → Y be the projection maps. From the commutative diagram in (a) we obtain a morphism r : Z → X ×A Y with qr = ω and pr = ξ. Further, r restricts to (φ′)−1 over A′. Thus r(Z) = U , and φ′ : U → Z is an isomorphism. A ×A Pm (c). A point (x, y) ∈ Pn A corresponds to a pair of linear ideals n ⊂ C[x0, . . . , xn] and m ⊂ C[y0, . . . , ym]. Let C[(xiyj)i,j] ⊂ C[(xi)i][(yj)j] be the homogeneous coordinate ring of Σn,m. It is clear that the ideal defining φ(x, y) = {x} × Pm ∩ Pn × {y} in C[xiyj] is generated by n1 · (y0, . . . , ym) + (x0, . . . , xn) · m1. Let (x, y) ∈ (X ×A Y )o, where x corresponds to the colength 1 ideal I ′ ⊆ I and y corresponds to J ′ ⊆ J . That I ′J +IJ ′ gives the ideal K′ ⊂ K corresponding to φ(x, y) follows from the previous paragraph, together with the fact that the isomorphism φ′ between (X ×A Y )o and Z is given by the Segre embedding. Since φ′ is an isomorphism, any ideal K′ corresponding to a point z ∈ Z may be written K′ = I ′J + IJ ′ for appropriate I ′, J ′. We thus have I ′ ⊆ (K′ : J ) ∩ I $ I. Since I ′ is colength 1, this implies that I ′ = (K′ : J ) ∩ I, as claimed. (cid:3) Corollary 3.3. Let X be a projective variety of dimension ≥ 2, let σ ∈ Autk(X), and let I be an ideal sheaf on X. Let In := II σ · · · I σn−1 . For all n ≥ 0, let an : Xn → X be the blowup of X at In. Then there are birational morphisms αn : Xn → Xn−1 (for n ≥ 1) and βn : Xn → Xn−1 (for n ≥ 2) so that the diagrams Xn ❇❇❇❇❇❇❇❇ an αn X Xn−1 and Xn βn / Xn−1 ②②②②②②②② an−1 an an−1 X / X σ commute. Proof. Let K := In. Let ζ : X ′ isomorphism θ : X ′ n−1 → Xn−1 so that the diagram n−1 → X be the blowup of X at I σ n−1. Since (Ip)σ ∼= Iσ−1(p), there is an X ′ n−1 ζ X θ σ Xn−1 an−1 / X commutes. Apply Lemma 3.2(a) with I = I σ n−1 and J = I1. We obtain a morphism γ : Xn → X ′ n−1 so that Xn commutes. Let βn := θγ : Xn → Xn−1. θ σ / Xn−1 an−1 / X γ X ′ n−1 "❋❋❋❋❋❋❋❋❋ an ζ X 8 / / / / / / / / / " / / Let αn be the morphism Xn → Xn−1 given by Lemma 3.2(a) with I = In−1 and J = I σn−1 1 . The diagram Xn αn / Xn−1 #❋❋❋❋❋❋❋❋❋ an an−1 X commutes, as required. (cid:3) We will frequently suppress the subscripts on the maps αn, an, etc., when the source and target are indicated. Note that the equation an = α1 ◦ · · · ◦ αn that follows from Corollary 3.3 may be written more compactly as a = αn : Xn → X. 4. Infinite blow-ups In this section, we prove some general properties of infinite blow-ups that will be useful when we consider moduli spaces of embedded point modules. Such infinite blow-ups can be handled in two ways: either as pro- objects in the category of schemes, or as stacks, via the (inverse) limits of such pro-objects in the category of spaces or of stacks. We've chosen to treat infinite blow-ups as the limits rather than as pro-objects. This is formally the correct choice, in the sense that the limit formally contains less information than the pro-object. We note that in our setting, we could also work with the pro-object with no difficulties; however, we have found the language of stacks more natural. We begin with some technical preliminaries on schemes and stacks. We will work with stacks in the fpqc (fid`element plat et quasi-compact) topology; the fpqc topology of schemes is discussed in Section 2.3.2 of [Vis05]. We are interested in a class of stacks that are apparently not algebraic, but for which a certain amount of algebraic geometry is still possible. More precisely, recall that a stack X is called algebraic if the diagonal morphism of X is representable, separated and quasi-compact; and it has an fppf atlas f : Z → X that is a scheme: that is, f is representable, faithfully flat, and finitely presented. By Artin's theorem [LMB00, Th´eor`eme 10.1] the second condition is equivalent to requiring the existence of a smooth, surjective and representable f . Our stacks are very similar to algebraic stacks, but it seems not to be possible to find a finite-type f for which Z is a scheme. On the other hand, we can find f for which Z is a scheme and f is fpqc -- and even formally ´etale -- so in some sense our stacks are the fpqc analogs of algebraic stacks. Definition 4.1. We will refer to a stack X for which the diagonal ∆ : X → X ×k X is representable, separated, and quasi-compact, and which admits a representable fpqc morphism Z → X from a scheme Z, as fpqc-algebraic. Note that "separated" and "quasi-compact" make sense for fpqc stacks by [EGA, IV.2 Proposition 2.7.1 and IV.2 Cor. 2.6.4]. Unfortunately, in this weaker setting, there are fewer notions of algebraic geometry that one can check fpqc-locally, and hence fewer adjectives that one can sensibly apply to fpqc-algebraic stacks. Still, one can make sense, for example, of representable morphisms being separated, quasi-separated, locally of finite type or of finite presentation, proper, closed immersions, affine, etc. by [EGA, IV.2 Proposition 2.7.1]. Recall [EGA, IV.4 Def. 17.1.1] that a morphism of schemes f : X → Y is formally ´etale if for every 0 ⊂ Y ′ defined by a nilpotent ideal and morphism Y ′ → Y , the map 0 , X) is bijective. By faithfully flat descent (see [Vis05]), the definition extends affine scheme Y ′, closed subscheme Y ′ HomY (Y ′, X) → HomY (Y ′ immediately to stacks in the ´etale, fppf, and fpqc topologies of schemes. We will say that an fpqc-algebraic stack X is noetherian if it admits an fpqc atlas Z → X by a noetherian scheme Z. Unfortunately, since fpqc morphisms need not be of finite type even locally, it does not seem to be possible to check this property on an arbitrary atlas Y → X . Suppose we have a sequence of schemes {Xn n ∈ N} and projective morphisms πn : Xn → Xn−1. We Xn. More precisely, we define the functor define the infinite blowup X∞ to be the presheaf of sets X∞ = lim ←− of points hX∞ : Schemesop −→ Sets. For each scheme A, let For each n, there is an induced map π : X∞ → Xn, where the target space Xn is indicated explicitly. hX∞ (A) =(cid:8)(ζn : A → Xn)n∈N(cid:12)(cid:12) πnζn = ζn−1(cid:9). 9 / # Proposition 4.2. Suppose that X := X0 is a variety of dimension ≥ 2 and that there are maps πn : Xn → Xn−1 as above. Then the stack X∞ is a sheaf in the fpqc topology. Further, suppose that the maps πn satisfy the following conditions: (i) For all n, π−1 n is defined at all but finitely many points of X o n−1. That is, the set of exceptional points of π−1 : X 99K X∞ is countable; let {zm}m∈N be an enumeration of this set. (ii) The set {zm} is critically dense. (iii) For all m, there is some n(m) so that, for n ≥ n(m), the map πn is a local isomorphism at all points in the preimage of zm. (iv) For all m ∈ N, there is an ideal sheaf Jm on X, cosupported at zm, so that Xn(m) is a closed subscheme of Proj Sym X Jm above a neighborhood of zm. That is, Xn(m) → X factors as Xn(m) cm / / Proj Sym X Jm pm / / X , where pm : Proj Sym X Jm → X is the natural map, and cm is a closed immersion over a neighborhood of zm. (v) There is some D ∈ N so that mD zmOX,zm ⊆ Jm ⊆ OX,zm for every m. Then: (1) The stack X∞ is fpqc-algebraic: (1a) X∞ has a representable, formally ´etale, fpqc cover by an affine scheme U → X∞. (1b) The diagonal morphism ∆ : X∞ → X∞ ×k X∞ is representable, separated, and quasicompact. (2) The morphism π : X∞ → X is quasicompact. (3) X∞ is noetherian as an fpqc-algebraic stack. Proof. Because any limit of an inverse system of sheaves taken in the category of presheaves is already a sheaf (cf. [Har77, Exercise II.1.12]), X∞ is a sheaf in the fpqc topology. Now assume that (i) -- (v) hold. For n ∈ N, let Wn be the scheme-theoretic image of cn. Let X ′ n := W0 ×X W1 ×X · · · ×X Wn−1. n : X ′ Let π′ loss of generality that Xn = X ′ n → X ′ n−1 be projection on the first (n − 1) factors. We first show that we can assume without n; that is, we claim that lim←−π′ X ′ n ∼= lim←−π Xn. Let k ∈ N. Let K(k) := max{k, n(0), . . . , n(k − 1)}. For each 0 ≤ m ≤ k − 1, there is a morphism XK(k) πK(k)−n(m) / Xn(m) cm / / Wm. Since these agree on the base, we obtain an induced φk : XK(k) → X ′ the inverse systems π and π′. Taking the limit, we obtain k. The φk are clearly compatible with φ : lim←− XK(k) → lim←− X ′ k. Now let Fk be the set of fundamental points of X 99K Xk. Let N (k) := k + max{m zm ∈ Fk}. We claim N (k) → Xk. There is certainly a rational map defined over X r {Fk}, since there there is a morphism ψk : X ′ Xk is locally isomorphic to X. Let zm ∈ Fk and let n′(m) := max{k, n(m)}. The rational map X ′ N (k) / Wm /❴❴❴ Xn′(m) / Xk is then defined over a neighborhood of zm. These maps clearly agree on overlaps, so we may glue to define ψk as claimed. Let be the limit of the ψk. It is clear that ψ = φ−1; note that by construction both N (k) and K(k) go to infinity as k → ∞. ψ : lim←− X ′ N (k) → lim←− Xk Going forward, we replace Xn by X ′ n. Thus let Yn := Proj Sym X Jn, and assume that there are closed immersions in : Xn → Y0 ×X · · · ×X Yn−1 so that the πn are given by restricting the projection maps. It suffices to prove the proposition in the case that X = Spec C is affine; note that we can choose an affine subset of X that contains all zn. Let Jn ⊆ C be the ideal cosupported at zn so that (Jn)zn = Jn. Let mp denote the maximal ideal of C corresponding to p. 10 / / / / Claim 4.3. There is an N such that every ideal Jm (m ∈ N) is generated by at most N elements. To prove the claim, embed X in an affine space, i.e. choose a closed immersion X ⊆ Aℓ. Then each point of Aℓ, hence a fortiori each point of X, is cut out scheme-theoretically by ℓ elements of C, and the power of the maximal ideal mD elements of C. Now Jm contains mD zm appearing in hypothesis (v) of the proposition is generated by N0 := (cid:0)D+ℓ−1 ℓ−1 (cid:1) zm, and dim(Jm/m D zm) ≤ dim C/m D zm ≤ dim k[u1, . . . , uℓ]/(u1, . . . , uℓ)D =: N1. Thus Jm is generated by at most N := N0 + N1 elements. We continue with the proof of the proposition: (1a) To construct an affine scheme U with a representable, formally ´etale morphism U → X∞ we proceed as follows. For each n, and for each 1 ≤ i ≤ N , we choose hypersurfaces Dn,i = V (dn,i) ⊂ X, with the following properties: (A) For all n, the elements dn,1, . . . , dn,N generate Jn. (B) For all n, i, the hypersurface Dn,i does not contain any irreducible component Z of a hypersurface Dm,j with m < n or m = n and j < i. (C) For all n and each m 6= n, zm /∈ Dn,i for any i. We can make such choices because k is uncountable and X is affine (note that in order to satisfy (B), the choice of each Dn,i will depend on finitely many earlier choices). For each N -tuple of positive integers (n1, . . . , nN ), let Z(n1,...,nN ) := Dn1,1 ∩ · · · ∩ DnN ,N . Note that zm /∈ Z(n1,...,nN ) unless (n1, . . . , nN ) = (m, m, . . . , m) by property (C). Note also that, since Z(n1,...,nN ) is a union of intersections of pairwise distinct irreducible hypersurfaces, it has codimension at least 2 in X. Z(n1,...,nN ). This is a countable union of irreducible subsets of X of codimension Now let Z (1) := [(n1,...,nN ) at least 2. We may choose one point lying on each component of Z (1) r {zm}m∈N. Now, for each n, choose a hypersurface Dn,N +1 such that zn ∈ Dn,N +1; that the local ideal of Dn,N +1 at zn is contained in Jn; and that Dn,N +1 avoids all the (countably many) chosen points of components of Z (1) and all zm, m 6= n. Then for each n, Z (1) ∩ Dn,N +1 is a countable union of irreducible algebraic subsets of codimension at least 3 (it is a union of proper intersections of Dn,N +1 with irreducible subsets of codimension at least 2). Let Z (2) :=[n (Z (1) ∩ Dn,N +1). Repeating the previous construction with Z (2), we get hypersurfaces Dn,N +2 such that each Z (2) ∩Dn,N +2 is a countable union of irreducible subsets of codimension at least 4. Iterating, we eventually define hyper- surfaces Dn,N +i, i = 1, . . . , d with the following properties: (A′) For all m ∈ N, there is a scheme-theoretic equality Spec(C/Jm) = Dm,1 ∩ · · · ∩ Dm,N +d. (B′) For every sequence (n1, . . . , nN +d) of positive integers, we have a set-theoretic equality Dn1,1 ∩ · · · ∩ DnN +d,N +d =(zm if (n1, . . . , nN +d) = (m, . . . , m). otherwise. ∅ (C) For all n and each m 6= n, zm /∈ Dn,i for any i. m denote the proper transform of Dn,i. By construction, For 0 ≤ n ≤ m − 1, we abusively let eDn,i ⊂ X o eDn,1 ∩ · · · ∩ eDn,N +d = ∅. For each m ∈ N, the map C N +d → Jm, ei 7→ dm,i induces a closed immersion Ym → PN +d−1 . Let X Vm,i ⊆ Ym be the open affine given by (ei 6= 0). Note that Vm,i ∩ X o denotes the closure in Xm of the preimage in Xm of the generic point of X). Let m r eDm,i (recall that here X o The Un,i are open and affine. Since Dm,i 6∋ zn for m 6= n, the setSi Un,i includes all irreducible components Un,i := Xn ∩ (V0,i ×X V1,i ×X · · · ×X Vn−1,i). of Xn except possibly for X o m = X o n. But m X o n r N +d[i=1 Un,i =\i n−1[m=0 eDm,i =[m \i eDm,i = ∅. 11 Thus the Un,i are an open affine cover of Xn. Since πnUn,i πn(Un,i) ⊆ Un−1,i. Writing Ci := lim −→ is obtained by base extension from the affine morphism Vn−1,i → X, it is affine, and Um,i; all the Ui are affine Cm,i and Ui := Spec Ci, we get Ui = lim ←− schemes. By construction we obtain induced maps Ui → X∞. Let U :=Gi Claim 4.4. The induced morphism U → X∞ is representable and formally ´etale. Ui. Since each map Ui → X∞ is a limit of formally ´etale morphisms, each is itself formally ´etale. We must show that if T is a scheme equipped with a morphism T → X∞, then T ×X∞ U → T is a scheme over T . Each morphism T ×Xm Um,i → T is an affine open immersion since the morphisms Um,i → Xm are affine (T ×Xm Um,i) → T is an inverse limit of schemes affine over T open immersions. Hence the morphism lim ←− and thus is itself a scheme affine over T (see [EGA, IV, Proposition 8.2.3]). The claim now follows from: Lemma 4.5. For any scheme T equipped with a morphism T → X∞, we have T ×X∞ Ui ∼= lim ←− Um,i). (T ×Xm (cid:3) Claim 4.6. The map U → X∞ is surjective. Surjectivity for representable morphisms can be checked locally on the target by [LMB00, 3.10]. Thus, we may change base along a map T → X∞ from a scheme T ; and, taking a point that is the image of a map Spec(K) → T where K is a field containing k, it suffices to find Spec(K) → U making (4.7) Spec(K) / U commute. Spec(K) / X∞ Thus, suppose we are given a map Spec(K) → X∞; let yn denote its image (i.e. the image of the unique point of Spec(K)) in Xn. Let In denote the (finite) set of those i so that yn ∈ Un,i. Since the Un,i cover Xn, The maps Spec(K) → Um,i0 for m ≫ 0 define a map f : Spec(K) → Ui0 = lim ←− the map in the top row of (4.7) to be f gives the desired commutative diagram. This proves the claim. each In is nonempty; further, In ⊆ In−1. The intersection Tn In is thus nonempty and contains some i0. have R =Gi,j Returning to the proof of Proposition 4.2(1a), let R := U ×X∞ U . If we define Rij := Ui ×X∞ Uj, then we Rij . Note that Rij is a scheme affine over Ui by the previous paragraph. Since affine schemes are quasicompact, this proves that the morphism U → X∞ is quasicompact. Furthermore, O(Rij ) is a localization of Ci (obtained by inverting the images of the elements dn,j); so Rij → Ui is flat. We have already proved that U → X∞ is surjective, so we conclude that U → X∞ is faithfully flat. It follows that U → X∞ is fpqc using [Vis05, Proposition 2.33(iii)]. This completes the proof of (1a). Um,i0 ⊂ U , and thus defining (1b) The diagonal ∆ : X∞ → X∞ ×k X∞ is the inverse limit of the diagonals ∆n : Xn → Xn ×k Xn. Similarly to Lemma 4.5, if V → X∞ ×k X∞ is any morphism from a scheme V , we get X∞ ×X∞×kX∞ V ∼= Xn ×Xn×kXn V . Since each Xn is separated over k, each morphism Xn ×Xn×kXn V → V is a closed lim ←− immersion; hence X∞ ×X∞×kX∞ V → V is a closed immersion. This proves (1b). (2) Again, we may assume that X is affine. Then, as above, we have an fpqc cover U p −→ X∞ by an affine scheme U . Since an affine scheme is quasicompact and a continuous image of a quasicompact space is quasicompact, X∞ is quasicompact, as desired. (3) By our definition, it suffices to prove that U is noetherian, or, equivalently, that each Ci is a noetherian ring. This follows as in [ASZ99, Theorem 1.5]. We will need: Lemma 4.8. Let A be a commutative noetherian ring, and let J be an ideal of A with a resolution Am M / / An / J / 0. (Here M is an n × m matrix acting by left multiplication.) Let A′ := A[t1, . . . , tn−1]/(t1, . . . , tn−1, 1)M . Let P ′ be a prime of A′, and let P := P ′ ∩ A. If P and J are comaximal, then P A′ = P ′. 12 / / / / Note that A′ is the coordinate ring of a chart of Proj SymA J. Proof. We may localize at P , so without loss of generality J = A. Then A′ ∼= A[g−1] for some g ∈ A. The result follows. (cid:3) We return to the proof of (3). It suffices to show, by [Eis95, Exc. 2.22], that each prime of Ci is finitely some n ∈ N so that if m ≥ n, then Jm and P are comaximal. It follows from Lemma 4.8 that Pm = PnCm,i generated. Let eP 6= 0 be a prime of Ci. Let P := eP ∩ C, and let Pn := eP ∩ Cn,i. By critical density, there is for m ≥ n. So eP =S Pm = PnCi. This is finitely generated because Cn,i is noetherian, so Pn is finitely Proposition 4.2 is now proved. generated. (cid:3) Corollary 4.9. Let X be a projective variety, let σ ∈ Autk(X), and let L be a σ-ample invertible sheaf on X. Let P be a 0-dimensional subscheme of X, all of whose points have critically dense σ-orbits. Let In := IP I σ . Let an : Xn → X be the blowup of X at In, as in Corollary 3.3. Let αn : Xn → Xn−1 be given by Corollary 3.3. Then the limit P · · · I σn−1 P is a noetherian fpqc-algebraic stack. X∞ := lim←− Xn Proof. This follows immediately from Proposition 4.2. (cid:3) 5. Moduli schemes for truncated point modules Let X be a projective variety, let σ ∈ Autk(X), and let L be a σ-ample invertible sheaf on X. Let P be a 0- dimensional subscheme of X, all of whose points have critically dense σ-orbits. We define S := S(X, L, σ, P ) and S := S(X, L, σ, P ), as in Section 2. As usual, we assume that S is generated in degree one. In this section, we construct moduli schemes of truncated point modules over S and S. In the next section, we compare them. We begin by constructing moduli schemes for shifted point modules for an arbitrary connected graded noetherian algebra generated in degree 1, generalizing slightly results of [ATV90] and [RS09]. Let C be any commutative k-algebra. Recall that we use subscript notation to denote changing base. Thus if R is a k-algebra, we write RC := R ⊗k C. We write XC := X ×k Spec C. Recall that a C-point module (over R) is a graded factor M of RC so that Mi is rank 1 projective for i ≥ 0. An ℓ-shifted C-point module (over R) is a factor of (RC )≥ℓ that is rank 1 projective in degree ≥ ℓ. A truncated ℓ-shifted C-point module of length m is a factor module of (RC )≥ℓ so that Mi is rank 1 projective over C for ℓ ≤ i ≤ ℓ + m − 1 and Mi = 0 for i ≥ ℓ + m. Since these modules depend on a finite number of parameters, they are clearly parameterized up to isomorphism by a projective scheme. For fixed ℓ ≤ n, we denote the ℓ-shifted length (n − ℓ + 1) point scheme of R by ℓYn. A point in ℓYn gives a surjection R≥ℓ → M (up to isomorphism), or equivalently a submodule of R≥ℓ with appropriate Hilbert series. Thus we say that ℓYn parameterizes embedded (shifted truncated) point modules. The map M 7→ M/Mn induces a morphism χn : ℓYn → ℓYn−1. For later use, we explicitly construct a projective embedding of ℓYn. Proposition 5.1. (cf. [ATV90, section 3]) Let R be a connected graded k-algebra generated in degree 1. (1) For all ℓ ≤ n ∈ N, there is a closed immersion ℓΠn : ℓYn → P((R⊗ℓ 1 )∨) × P(R∨ 1 )×(n−ℓ). (2) Fix ℓ ≤ n and let π : P((R⊗ℓ 1 )∨) × P(R∨ 1 )×(n−ℓ) → P((R⊗ℓ 1 )∨) × P(R∨ 1 )×(n−ℓ−1) be projection onto the first n − ℓ factors. Then the diagram (5.2) ℓΠn ℓYn P((R⊗ℓ 1 )∨) × P(R∨ 1 )×(n−ℓ) χn π ℓYn−1 ℓΠn−1 / P((R⊗ℓ 1 )∨) × P(R∨ 1 )×(n−ℓ−1) 13 / / / commutes. Proof. (1) Let T = T •(R1) denote the tensor algebra on the finite-dimensional k-vector space R1. We identify T1 canonically with R1 and Tℓ with R⊗ℓ 1 . Given an element f ∈ Tn, we get an (ℓ, 1, . . . , 1)-form 1 )×(n−ℓ) −→ k ℓ ) × (P(T ∨ 1 )×(n−ℓ). ℓ × (T ∨ ef : T ∨ Y ({efi}) ⊆ P(T ∨ by pairing with f . The map is k-multilinear, hence ef defines a hypersurface Y (ef ) in P(T ∨ More generally, given a collection {fi} of elements of Tn, we get a closed subscheme ℓ ) × (P(T ∨ 1 )×(n−ℓ). Let I be the kernel of the natural surjection T ։ R. Then the above construction gives a closed subscheme ℓ ) × (P(T ∨ Y (eIn) ⊆ P(T ∨ 1 )×(n−ℓ). We claim that Y (eIn) is naturally isomorphic to ℓYn. Let C be a commutative k-algebra and let RC = R ⊗k C, TC = T ⊗k C with the gradings induced from R (respectively T ). Suppose that α : (RC )≥ℓ → M is an embedded ℓ-shifted truncated C-point module of length n − ℓ + 1. We write α : (TC )≥ℓ → M for the composite of the two surjections. Assume that M = ⊕n i=ℓmi · C is a free graded C-module on generators mi. Then α determines C-linear maps aj : T1 ⊗k C → C for 1 ≤ j ≤ n − ℓ by mℓ+j−1x = mℓ+jaj(x) for x ∈ T1 ⊗k C, and a C-linear map b : Tℓ ⊗k C → C by α(y) = mℓb(y) for y ∈ Tℓ ⊗k C. Since M is a (shifted, truncated) point module, hence generated in degree ℓ, these maps are surjective. Hence they determine a morphism Π(α) = (b, a1, . . . , an−ℓ) : Spec(C) −→ P(T ∨ ℓ ) × (P(T ∨ 1 )×(n−ℓ). It follows immediately that the above construction defines a morphism Π from the moduli functor of a scheme, hence a sheaf in the fpqc topology, Π induces a morphism, which we denote by ℓΠn, from the We see immediately from the construction that if f ∈ In ⊗k C, then ef (Π(α)) = 0. In particular, Π(α) factors through Y (eIn). shifted truncated point modules with free (as C-modules) graded components to Y (eIn). Since the latter is moduli functor ℓYn for all shifted truncated C-point modules over R to Y (eIn). Claim 5.3. The morphism ℓΠn is an isomorphism: that is, Y (eIn) ∼= ℓYn represents the moduli functor of Proof of Claim. A morphism Spec(C) → Y (eIn) ⊆ P(T ∨ 1 )×(n−ℓ) gives a tuple (b, a1, . . . , an−ℓ) where each aj is a surjective C-linear map aj : T1 ⊗k C → Nj and each Nj is a finitely generated projective C- module of rank 1; and b : Tℓ ⊗k C → Mℓ is a surjective C-linear map onto a finitely generated projective C-module Mℓ of rank 1. embedded truncated ℓ-shifted C-point modules over R of length n − ℓ + 1. ℓ )×(P(T ∨ Assume first that Mℓ and each Nj is a free C-module, and choose basis elements. Define a TC-module j=ℓmj · C by mj−1x = mjaj−ℓ(x) for x ∈ T1 ⊗k C. Moreover, define a map α : (TC)≥ℓ → M by M = ⊕n map α factors through (RC )≥ℓ and makes M an ℓ-shifted truncated C-point module over R. Next, we observe that the functor Π (on shifted truncated point modules with free C-module components) give mutual inverses. This follows from the argument of [ATV90, 3.9], which uses only the freeness condition. In particular, the functor Π is injective. α(y) = mℓb(y) for y ∈ Tℓ and extending linearly. It is a consequence of the construction of Y (eIn) that the and the above construction (on maps Spec(C) → Y (eIn) for which the modules Nj and Mℓ are free C-modules) that is, for every morphism Spec(C) → Y (eIn), there is a faithfully flat morphism Spec(C ′) → Spec(C), and map Spec(C ′) → Y (eIn). But it is standard that such a homomorphism C → C ′ can be found that makes each Nj and Mℓ trivial, and now the construction of the previous paragraph proves the existence of the desired shifted truncated point module. This completes the proof of the claim. (cid:3) a shifted truncated C ′-point module with free C ′-module components, whose image under Π is the composite To prove that the sheafification ℓΠn is an isomorphism, then, it suffices to show that Π is locally surjective: Part (1) follows from the claim. (2) follows by construction. (cid:3) 14 Proposition 5.4. (cf. [ATV90, Proposition 3.6]) Let R be a connected graded k-algebra generated in degree one. Let n > ℓ and consider the truncation morphism χn : ℓYn → ℓYn−1. Let y ∈ ℓYn−1 and suppose that dim χ−1 a neighborhood of y. n (y) = 0. Then χ−1 n is defined and is a local isomorphism locally in Proof. We consider the commutative diagram (5.2) of Proposition 5.1(2). By Proposition 5.1(1) the hor- 1 )×(n−ℓ) are (ℓ, 1, . . . , 1)-forms and in particular are linear in the last coordinate, the fibers of χn are linear subspaces of P(T ∨ (cid:3) izontal maps are closed immersions. Since the defining equations of Y (eIn) ⊆ P(T ∨ 1 ). The result follows as in the proof of [ATV90, Proposition 3.6(ii)]. ℓ ) × (P(T ∨ Proposition 5.5. (cf. [RS09, Proposition 2.5]) Let R be a noetherian connected graded k-algebra generated in degree 1. For n > ℓ ≥ 0, define χn : ℓYn → ℓYn−1 as in the beginning of the section. Let n0 ≥ 0 and let {yn ∈ ℓYn n ≥ n0} be a sequence of (not necessarily closed) points so that χn(yn) = yn−1 for all n > n0. Then for all n ≫ n0 the fibre χ−1 is defined and is a local isomorphism at yn−1. n (yn−1) is a singleton and χ−1 n Proof. This follows as in the proof of [RS09, Proposition 2.5], using Proposition 5.4 instead of [ATV90, Proposition 3.6(ii)]. (cid:3) We are interested in studying the limit ℓY∞ := lim←− ℓYn; however, we first study the point schemes of S. That is, for n ≥ ℓ ≥ 0, it is clear that we may also define a scheme that parameterizes factor modules M of S≥ℓ so that, as graded OX -modules, M ∼= kxtℓ ⊕ · · · ⊕ kxtn for some x ∈ X. We say that x is the support of M. We denote this ℓ-shifted length (n − ℓ + 1) truncated point scheme of S by ℓZn. More formally, a Spec(C)-point of ℓZn will be a factor module M of S≥ℓ ⊗k C which is isomorphic as a graded OXC -module to a direct sum Pℓ ⊕ · · · ⊕ Pn, where each Pi is a coherent OXC -module that is finite over C (in the sense that its support in XC is finite over Spec(C)) and is a rank one projective C-module (which is well defined because of the finite support condition). We let Zn := 0Zn be the unshifted length n + 1 point scheme of S. For all n > ℓ ≥ 0, there are maps φn : ℓZn → ℓZn−1 defined by M 7→ M/Mn. If ℓ = 0 and M is a truncated point module of length n over S, then M[1]≥0 is also cyclic (since S is generated in degree 1) and so is a factor of S, in a unique way up to scalar. This induces a map It is clear that ψn and φn map relevant components to relevant components. ψn : Zn → Zn−1 defined by M 7→ M[1]≥0. Lemma 5.6. Let fn : ℓZn → X be the map that sends a module M to its support. The diagrams (5.7) and (5.8) commute. ℓZn f ①①①①①①①① b❋❋❋❋❋❋❋❋❋ f φ ℓZn−1 f =φn X 0Zn ψ 0Zn−1 f =φn−1 15 / X σ / X b / / Proof. It is clear by construction that if M is a shifted truncated point module and M′ is a further factor of M, then M and M′ have the same support. Thus (5.7) commutes. Let M be a truncated point module corresponding to a point z ∈ 0Zn. Let x := f (z). By [KRS05, Lemma 5.5], we have M[1]n ∼= (Mn+1)σ−1 ∼= (kx)σ−1 ∼= kσ(x). Thus f ψ(z) = σf (z), as claimed, and (5.8) commutes. (cid:3) Recall that Sn = InLn. Proposition 5.9. For n ≥ 0, let Xn be the blowup of X at In, and let αn, βn : Xn → Xn−1 be as in Corollary 3.3. Then for all n > ℓ ≥ 0 there are isomorphisms jn : Xn → ℓZ o n ⊆ ℓZn so that the diagrams jn Xn αn / ℓZn and Xn jn / 0Zn φn βn ψn Xn−1 / ℓZn−1 jn−1 Xn−1 / 0Zn−1 jn−1 commute. Proof. We will do the case that ℓ = 0; the general case is similar. Let Zn := 0Zn. For 0 ≤ i ≤ n, let Wi = Proj Sym X Ii be the scheme parameterizing colength 1 ideals inside Ii, and let ci : Xi → Wi be the map from Lemma 3.1. Let be the composition rn : Xn → X × W1 × · · · × Wn αn×αn−1×···×1 Xn / X × X1 × · · · × Xn c0×···×cn / X × W1 · · · × Wn. Since this is the composition of the graph of a morphism with a closed immersion, it is also a closed immersion and is an isomorphism onto its image. Now, a point in Zn corresponds to an ideal J ⊂ S so that the factor is a truncated point module of length n + 1, and there is thus a canonical closed immersion δn : Zn → X × W1 × · · · Wn. The map δn sends a graded right ideal J of S to the tuple (J0, J1, . . . , Jn). Conversely, a point (J0, . . . , Jn) ∈ X × W1 × · · · × Wn is in Im(δ) if and only if we have JiS σi j ⊆ Ji+j for all i + j ≤ n. It follows from Lemma 3.2(c) that Im(rn) ⊆ Im(δn). Since rn and δn are closed immersions and Xn is reduced, we may define jn = δ−1 Let U := X r Cosupp In. Then f −1 n n are defined on U , and the diagram n rn : Xn −→ Zn. and a−1 Xn X × W1 × · · · × Wn δn rn 7♥♥♥♥♥♥♥♥♥♥♥♥♥ gPPPPPPPPPPPPPP a−1 n jn U gPPPPPPPPPPPPP 7♥♥♥♥♥♥♥♥♥♥♥♥♥♥ f −1 n Zn commutes. Since rn and δn are closed, rn(Xn) = rn(a−1 n (U )) = rna−1 n (U ) = δnf −1 n (U ) = δn(f −1 n (U )) = δn(Z o n). Therefore, jn is an isomorphism to Z o n. 16 / / / / / / / / 7 g g 7 Let q : X × W1 × · · · × Wn → X × W1 × · · · × Wn−1 be projection onto the first n factors. Consider the diagram Xn αn rn X × W1 × · · · × Wn q δn Zn φn Xn−1 rn−1 / X × W1 × · · · × Wn−1 Zn−1. δn−1 From the definitions of rn and δn we see that this diagram commutes; since jn = δ−1 n rn, the diagram jn Xn αn Zn φn Xn−1 / Zn−1 jn−1 commutes. Let X ′ n := Im(rn) ⊂ X × W1 × · · · × Wn, and let β′ proof of Corollary 3.3 shows that if(cid:0)J0, J1, . . . , Jn(cid:1) ∈ X ′ (5.10) β′ n(cid:16)(cid:0)J0, J1, . . . , Jn(cid:1)(cid:17)= (F0, . . . , Fn−1) ∈ X ′ n−1, n → X ′ n : X ′ n, then its image under β′ n is n−1 be the map induced from βn. The Now let J be the ideal defining a truncated point module of length n+1. By abuse of notation, we think of J as a point in Zn. Let M := S/J . Then ψ(J )i = (AnnS(M1))i, for 0 ≤ i ≤ n−1. Thus I1(ψ(J )i)σ ⊆ Ji+1 ∩ Ii. If J ∈ Im(jn), then we have equality by the computation in (5.10). Thus where Fi := (Ji+1 : I1)σ−1 ∩ Ii. or ψ(J )i ⊆(cid:0)Ji+1 : I1(cid:1)σ−1 the diagram Xn rn βn jn X ′ n β ′ n δ−1 n Zn ψn Xn−1 rn−1 / X ′ n−1 δ−1 n−1 / / Zn−1 jn−1 commutes. (cid:3) To end this section, we construct stacks ℓZ∞ and ℓY∞ that are fine moduli spaces for (shifted) embedded point modules and give some of their properties. A version of the following result was known long ago to M. Artin; however, it does it seem to have appeared in the literature. Theorem 5.11. Fix ℓ ∈ N. For n ≥ ℓ, let Xn be the blowup of X at In. Let ℓYn be the moduli space of ℓ-shifted length (n − ℓ + 1) point modules over S. Let ℓZn be the moduli space of ℓ-shifted length (n − ℓ + 1) point modules over S. Define the morphisms χn : ℓYn → ℓYn−1, φn : ℓZn → ℓZn−1, and αn : Xn → Xn−1 as above. Let ℓZ∞ := lim←− φn ℓZn, ℓY∞ := lim←− χn ℓYn, and X∞ := lim←− αn Xn. Then the stack ℓY∞ is a sheaf in the fpqc topology and is a fine moduli space for ℓ-shifted embedded point modules over S. The stack ℓZ∞ is noetherian fpqc-algebraic and is a fine moduli space for ℓ-shifted embedded point modules over S. The relevant component of ℓZ∞ is isomorphic to X∞. Proof. We suppress the subscript ℓ in the proof. For n ≥ ℓ, let Fn be the moduli functor for truncated ℓ-shifted point modules over S, so Yn ∼= Fn. Define a contravariant functor F : Affine schemes −→ Sets Spec C 7→(cid:8)Embedded ℓ-shifted C-point modules over S(cid:9). 17 / / o o / o o / / / / / * * / / / 3 3 By descent theory, F is a sheaf in the fpqc topology. More precisely, recall that quasicoherent sheaves form a stack in the fpqc topology (see [Vis05, Section 4.2.2]); consequently the (graded) quotients of S≥ℓ form a sheaf of sets in the fpqc topology. Moreover, as in the first paragraph of Section 4.2.3 of [Vis05], those quotients of S≥ℓ that are S-module quotients form a subsheaf in the fpqc topology; this subsheaf is F . It is formal that F is isomorphic to the functor hY∞ . Likewise, Z∞ parameterizes ℓ-shifted point modules over S. We show that (i) -- (v) of Proposition 4.2 apply to Z∞. We have Sn = InLn; let Pn ⊂ X be the subscheme defined by In. Consider the maps Zn φn / Zn−1 fn−1 fn / X from Lemma 5.6. Now, f −1 n is defined away from Pn, and ∪nPn is a countable critically dense set. Thus (i), (ii) hold. Let x ∈S Pn. As the points in P have infinite orbits, there is some m ∈ N so that x 6∈ σ−n(P ) for all n ≥ m. Let zm ∈ Zm with fm(zm) = x, corresponding to a right ideal J ⊆ S≥ℓ with S≥ℓ/J ∼=Lm j=ℓ Ox. Let J ′ := J≤m · S. For any j ≥ 0 we have (S σm )x, and so J ′ gives the unique preimage of zm in Z∞. A similar uniqueness holds upon base extension, so the scheme-theoretic preimage of zm in Z∞ is a k-point, and Φ−1 m is defined and is a local isomorphism at zm. For j ≥ ℓ, let Wj := Proj Sym X In. As in the proof of Proposition 5.9, we may regard Zn as a closed subscheme of Wℓ × · · · × Wn, and (iv) and (v) follow from this and the fact that the orbits of points in P are infinite. By Proposition 4.2, then, Z∞ is a noetherian fpqc-algebraic stack. )x = (Lσm j j Consider the morphisms jn : Xn ∼=→ Z o in that proposition gives an induced isomorphism j : X∞ → Z o ∞ ⊆ Z∞. n ⊆ Zn from Proposition 5.9. Commutativity of the first diagram (cid:3) 6. Comparing moduli of points In this section we prove that ℓY∞ is also noetherian fpqc-algebraic, and that, at least for sufficiently large ℓ, the stacks ℓZ∞ and ℓY∞ are isomorphic. In the following pages, we will always use the following notation. We write a commutative k-algebra C as p : k → C, to indicate the structure map explicitly. We write XC := X ⊗k Spec C. We abuse notation and let the projection map 1 ⊗ p : XC → X also be denoted by p. We let q : XC → Spec C be projection on the second factor. Suppose that p : k → C is a commutative k-algebra, and y : Spec C → X is a C-point of X. Then y determines a section of q, which we also call y. This is a morphism y : Spec C → XC . We define Iy ⊆ OXC to be the ideal sheaf of the corresponding closed subscheme of XC. We define Oy := OXC /Iy. We use the relative regularity results from Section 2 to study the pullbacks of the sheaves Sn to XC. Fix a very ample invertible sheaf OX (1) on X, which we will use to measure regularity. Lemma 6.1. Let p : k → C be a commutative noetherian k-algebra. Then {p∗Sn}n≥0 is a right ample sequence on XC . Proof. Let F be a coherent sheaf on XC. By [RS07, Corollary 3.14], limn→∞ reg(Sn) = −∞. Thus limn→∞ reg(p∗Sn) = −∞ by Lemma 2.9. Since each Sn is invertible away from a dimension 0 set, p∗Sn is invertible away from a locus of relative dimension 0. By Proposition 2.13 F ⊗XC p∗Sn is 0-regular for n ≫ 0. Theorem 2.10 shows that (1), (2) of Definition 2.7 apply. (cid:3) We now prove a uniform regularity result for certain subsheaves of a pullback of some Sn. Lemma 6.2. There exists m ≥ 0 so that the following holds for any n ≥ m: for any commutative noetherian k-algebra p : k → C, for any C-point y of X, and for any coherent sheaf K on XC so that Iyp∗Sn ⊆ K ⊆ p∗Sn, then K is 0-regular. In particular, K is globally generated and R1q∗K = 0. Proof. Let D be the constant from Proposition 2.13, and let r := reg(OX ). By [RS07, Corollary 3.14], we have limn→∞ reg(Sn) = −∞. Let m be such that for all n ≥ m, reg(Sn) ≤ −r − D − 1. We claim this m satisfies the conclusions of the lemma. Fix a commutative noetherian k-algebra p : k → C and a C-point y of X. We first claim that reg(Iy) ≤ r + 1. To see this, let i ≥ 1 and consider the exact sequence Ri−1q∗OXC (r + 1 − i) α→ Ri−1q∗Oy(r + 1 − i) → Riq∗Iy(r + 1 − i) → Riq∗OXC (r + 1 − i). 18 / 2 2 / If i ≥ 2, then The last term vanishes, as OXC is (r + 1)-regular by Lemma 2.9 and Theorem 2.10(1). Ri−1q∗Oy(r + 1 − i) = 0 for dimension reasons, so Riq∗Iy(r + 1 − i) = 0. On the other hand, if i = 1, then because OXC (r) is 0-regular, by Lemma 2.11 α is surjective. Again, Riq∗Iy(r + 1 − i) = 0. Thus Iy is (r + 1)-regular as claimed. Let n ≥ m. By Lemma 2.9, reg(p∗Sn) = reg(Sn). Note that Iy and p∗Sn are both locally free away from a set of relative dimension 0. Thus the hypotheses of Proposition 2.13 apply, and by that result we have reg(Iy ⊗XC p∗Sn) ≤ reg(Iy) + reg(p∗Sn) + D ≤ r + 1 + D + reg(p∗Sn) = r + 1 + D + reg(Sn). Our choice of n ensures this is non-positive. In particular, Iy ⊗XC p∗Sn is 0-regular. Let Iyp∗Sn ⊆ K ⊆ p∗Sn. There is a natural map f : Iy ⊗XC p∗Sn → K given by the composition Iy ⊗XC p∗Sn → Iy · p∗Sn ⊆ K. The kernel and cokernel of f are supported on a set of relative dimension 0, and it is an easy exercise that K is therefore also 0-regular. By Theorem 2.10, K is globally generated and R1q∗K = 0, as claimed. (cid:3) Definition 6.3. We call a positive integer m satisfying the conclusion of Lemma 6.2 a positivity parameter. The proof of Lemma 6.2 shows that if we are willing to replace L by a sufficiently ample invertible sheaf, we may in fact assume that m = 1 is a positivity parameter. (By [Kee00, Theorem 1.2], the existence of a σ-ample sheaf means that any ample invertible sheaf is σ-ample.) Corollary 6.4. Let p : k → C be a noetherian commutative k-algebra. Let m be a positivity parameter (Definition 6.3) and let n ≥ m. (1) If J ⊂ p∗Sn is a sheaf on XC so that p∗Sn/J has support on XC that is finite over Spec(C) and is a rank 1 projective C-module, then q∗J is a C-submodule of q∗p∗Sn = Sn ⊗ C such that the cokernel is rank 1 projective. (2) If K $ J ⊂ p∗Sn are sheaves on XC so that p∗Sn/J is a rank 1 projective C-module, then q∗K $ q∗J . (3) If J , J ′ ⊆ p∗Sn are sheaves on XC so that p∗Sn/J and p∗Sn/J ′ are rank 1 projective C-modules, then q∗J = q∗J ′ if and only if J = J ′. Proof. (1) Let x ∈ Spec C be a closed point. Consider the fiber square Xx XC (cid:3) q {x} / Spec C. Let Jx := J Xx . Since p∗Sn/J is flat over Spec C, we have Jx ⊆ p∗SnXx (Sn ⊗k k(x))/Jx ∼= Ox. By our choice of n, therefore, H 1(Xx, Jx) = 0. ∼= Sn ⊗k k(x). Further, Now J is the kernel of a surjective morphism of flat sheaves and so is flat over Spec C. Since H 1(Xx, Jx) = 0, by the theorem on cohomology and base change [Har77, Theorem III.12.11(a)] we have R1q∗J ⊗C k(x) = 0. The C-module R1q∗J thus vanishes at every closed point and is therefore 0. The complex 0 → q∗J → q∗p∗Sn → q∗(p∗Sn/J ) → 0 is thus exact. By assumption (1), q∗(p∗Sn/J ) is a rank 1 projective C-module. Since cohomology commutes with flat base change [Har77, Proposition III.9.3], we have q∗p∗Sn ∼= H 0(X, Sn) ⊗k C = Sn ⊗k C. (2). Since m is a positivity parameter, J is globally generated, and it follows immediately that q∗K 6= q∗J . (3). From (2) we have q∗(J ∩ J ′) = q∗J ⇐⇒ J ∩ J ′ = J ⇐⇒ J ⊆ J ′. It follows from our assumptions that this occurs if and only if J = J ′. Further, q∗J = q∗(J ∩ J ′) = q∗(J ) ∩ q∗J ′ ⇐⇒ q∗J ⊆ q∗J ′. From (1) we obtain that this is equivalent to q∗J = q∗J ′. (cid:3) We now apply these regularity results to show that ℓY∞ ∼= ℓZ∞ for ℓ ≫ 0. We will need the following easy lemma: 19 / / / We thus have ker s ⊆Tk Lemma 6.5. Let A, B be commutative noetherian local k-algebras with residue field k. Let s : A → B If s∗ : Homalg(B, C) → Homalg(A, C) is surjective for all finite-dimensional be a local homomorphism. commutative local k-algebras C, then s is injective. Proof. Let m be the maximal ideal of A and let n be the maximal ideal of B. Let f ∈ ker s. Suppose first that there is some k so that f ∈ mk−1 r mk. Let C := A/mk and let π : A → C be the natural map. Then C is a finite-dimensional artinian local k-algebra. Now, π(f ) 6= 0 but s(f ) = 0. Thus π 6∈ Im(s∗), a contradiction. mk. By the Artin-Rees lemma, ker s = 0. (cid:3) Proposition 6.6. Let m be a positivity parameter, and let n ≥ ℓ ≥ m. Let ℓZn be the ℓ-shifted length (n − ℓ + 1) point scheme of S, with truncation morphism φn : Zn → Zn−1 as in Proposition 5.9. Let ℓYn be the ℓ-shifted length (n − ℓ + 1) point scheme of S, with truncation morphism χn : ℓYn → ℓYn−1 as in Theorem 5.11. Let ℓZ∞ := lim←− ℓZn and let ℓY∞ := lim←− Yn. Then the global section functor induces a closed immersion sn : ℓZn → ℓYn so that the diagram (6.7) commutes. ℓZn φn sn ℓYn χn ℓZn−1 sn−1 / ℓYn−1 Proof. Let p : k → C be a commutative noetherian k-algebra. Let p : XC → X and q : XC → Spec C be the two projection maps, as usual. Note that if J ⊂ p∗S≥ℓ is the defining ideal of an ℓ-shifted length (n − ℓ + 1) truncated C-point module over S, then by Corollary 6.4(1), q∗J is the defining ideal of an ℓ-shifted length (n − ℓ + 1) truncated C-point module over S. Thus sn is well-defined. By Corollary 6.4(3), sn is injective on k-points and on k[ǫ] points. It is standard (cf. the proof of [Har95, Theorem 14.9]) that, because sn is projective, sn is a closed immersion. That (6.7) commutes is immediate. (cid:3) We will see that ℓY∞ and ℓZ∞ are isomorphic. It does not seem to be generally true that ℓYn and ℓZn are isomorphic; but we will see that there is an induced isomorphism of certain naturally defined closed subschemes. Let ℓ ∈ N. For any n, let Φn : ℓZ∞ → ℓZn and Υn : ℓY∞ → ℓYn be the induced maps. For any n ≥ ℓ, define ℓZ ′ n ⊆ ℓZn to be the image of Φn : ℓZ∞ → ℓZn. That is, ℓZ ′ n = \i≥0 Im(φi : ℓZn+i → ℓZn). Since ℓZn is noetherian and the φn are closed, this is a closed subscheme of ℓZn, equal to Im(φk : ℓZn+k → ℓZn) for some k. Similarly, let ℓY ′ n and ℓY∞ = lim←− ℓY ′ n. n and ℓY ′ We refer to ℓZ ′ n are truncations of honest (shifted) point modules. n = Im(Υn : ℓY∞ → ℓYn). It is clear that ℓZ∞ = lim←− ℓZ ′ n as essential point schemes, since modules in ℓZ ′ n and ℓY ′ Theorem 6.8. Let m be a positivity parameter, and let n ≥ ℓ ≥ m. (1) The morphism sn : ℓZn → ℓYn defined in Proposition 6.6 induces an isomorphism of essential point schemes / ℓY ′ n. (2) The limit s : ℓZ∞ → ℓY∞ is an isomorphism of stacks. s′ n : ℓZ ′ n ∼= / Proof. Since the subscript ℓ will remain fixed, we suppress it in the notation. Let s′ from commutativity of (6.7) that s′ n(Z ′ n) ⊆ Y ′ n. n := snZ ′ n . It follows We next prove (2). The limit s = lim←− sn is clearly a morphism of stacks -- that is, a natural transformation of functors. Let C be a commutative finite-dimensional local k-algebra. We will show that s is bijective on C-points. Let y ∈ Y∞(C) be a C-point of Y∞, which by Theorem 5.11 corresponds to an exact sequence 0 → J → (SC )≥ℓ → M → 0, 20 / / / where M is an ℓ-shifted SC-point module. Since p∗Sj is globally generated for j ≥ m, we have Jip∗S σi that M := p∗S≥ℓ/J is an ℓ-shifted p∗S-point module: that is, that there is a C-point y of X so that Mn ∼= Oy ⊗ p∗Ln for all n ≥ ℓ. For i ≥ ℓ, let Ji ⊆ p∗Si be the subsheaf generated by Ji ⊆ q∗p∗Si. Let J := Li≥ℓ Ji. We will show coherent right module over the bimodule algebra S ′ := OXC ⊕Lj≥m p∗Sj. Further, each Mj is clearly torsion over X, as Spec C is 0-dimensional. As in [RS07, Lemma 4.1(1)], it follows from critical density of the orbits of the points in P that there are a coherent X-torsion sheaf F on XC and n0 ≥ ℓ so that Mj ∼= F ⊗XC p∗Lj for j ≥ n0. By critical density again, there is n1 ≥ n0 so that j ⊆ Ji+j for i ≥ ℓ, j ≥ m. Thus M is a for j ≥ n1. This implies that for j ≥ n1 and k ≥ 1, we have Supp(F ) ∩ p−1(cid:0)σ−j(P )(cid:1) = ∅ Mj · p∗S σj k = Mj ⊗XC p∗S σj k = Mj ⊗XC p∗Lσj k ∼= Mj+k and Jj (p∗S σj k ) = Jj+k. (In particular, J≥n1 and M≥n1 are right p∗S-modules.) By Lemma 6.1, {p∗S σj we have q∗(Jip∗S σi There is thus n2 ≥ n1 so that Jj = q∗Jj for j ≥ n2. k }k≥0 is right ample on XC . As in [RS09, Lemma 9.3], for any i ≥ ℓ and k ≫ 0 k ) = q∗Ji+k ⊇ Ji+k. k ) = Ji(SC )k ⊆ Ji+k. For i ≥ n1 and for k ≥ 1, we have q∗(Jip∗S σi It follows from Lemma 6.1 that there is n3 ≥ n2 so that the top row of 0 / q∗Jj / (SC )j / q∗Mj / 0 Jj q∗(F ⊗XC p∗Lj) is exact for j ≥ n3. Thus q∗(F ⊗XC p∗Lj) ∼= (SC )j/Jj ∼= C for j ≥ n3. Since {p∗Lj} is right ample, this implies that F ∼= Oy for some C-point y of X. For ℓ ≤ i ≤ n3, let J ′ i := Ji + Iyp∗Si. By choice of ℓ, J ′ i is 0-regular. Thus the rows of the commutative diagram (6.9) 0 0 / (SC )i / Ji ⊆ / C α / q∗J ′ i / q∗p∗Si / q∗(p∗Si/J ′ i ) / 0 / 0 are exact, and α is therefore surjective. This shows that, as a (C ∼= Oy)-module, p∗Si/J ′ i is cyclic. i ). For i ≫ 0, p∗Si/Ji ∼= Oy ⊗XC p∗Li is killed by Iy. Thus for i ≫ 0 we have Let N := AnnC (p∗Si/J ′ i )y = (Ji)y + (Iyp∗Si)y = (Ji)y, and (J ′ (Ji+j )y ⊇ (Ji)y(p∗S σi j )y = (J ′ i )y(p∗S σi j )y ⊇ N (p∗Si)y(p∗S σi j )y = N (p∗Si+j )y. As AnnC (p∗Si+j /Ji+j) = 0 for j ≫ 0, we must have N = 0. Thus in fact p∗Si/J ′ i ∼= Oy ∼= C. Looking again at (6.9), we see that α is an isomorphism, and so Ji = q∗J ′ i . As J ′ i is 0-regular, it is globally generated: in other words, Ji = J ′ i . We still need to show that J is a p∗S-module. Let i ≥ ℓ and suppose that Jip∗S σi 1 1 )/Ji+1 is a nonzero submodule of p∗Si+1/Ji+1 ∼= Oy. Since (Ji+1 + Jip∗S σi (Ji+1 + Jip∗S σi for all j ≫ 0, a similar argument to the last paragraph but one produces a contradiction. 6⊆ Ji+1. Then ⊆ Ji+j+1 1 )S σi+1 j Thus p∗S≥ℓ/J is an ℓ-shifted C-point module for S, and q∗J = J. This shows that s : Z∞ → Y∞ induces a surjection on C-points. It follows from Corollary 6.4(3) that s is injective on C-points. Consider the commutative diagram (6.10) Z∞ s Y∞ Φn Z ′ n Υn / Y ′ n. s′ n 21 / / / / / / / / / / / / / / / Let x ∈ X be a k-point. There is some N so that for n ≥ N , Φn is a local isomorphism at all points in the preimage of x. We claim that for n ≥ N , in fact all maps in (6.10) are local isomorphisms at all points in the preimage of x. In particular, s is an isomorphism in the preimage of x; since x was arbitrary, s is therefore an isomorphism of stacks. So it suffices to prove the claim. We may work locally. If z ∈ Z∞ is a point lying over x, let zn, y, yn n,zn . Let A := OY∞,y and let n, respectively. Let B := OZ∞,z ∼= OZ ′ n, Y∞, and Y ′ be the images of z in Z ′ A′ := OY ′ n,yn. We have a commutative diagram of local homomorphisms B B s# A Υ# n A′. (s′ n)# As Υn is scheme-theoretically surjective, Υ# isomorphic to a local ring of some Y ′ is injective. Thus (s′ isomorphism; thus all maps in (6.10) are local isomorphisms above x. This proves the claim, as required. It follows from Proposition 5.5 that A is m and in particular is a noetherian k-algebra. By Lemma 6.5, s# n)# is also surjective and thus an n)# is injective. As sn is a closed immersion, (s′ n is injective. (1). Consider the diagram (6.10). By Proposition 6.6, sn : Zn → Yn is a closed immersion. Thus the restriction s′ n → Y ′ On the other hand, Y ′ n : Z ′ Υns = s′ theoretically surjective closed immersion is an isomorphism. nΦn is scheme-theoretically surjective, so s′ n is also a closed immersion. n is the scheme-theoretic image of Υn and s is an isomorphism. Thus the composition n is scheme-theoretically surjective. But a scheme- (cid:3) Of course, the defining ideal of a 1-shifted point module also defines a 0-shifted point module, so if the positivity parameter m = 1, the conclusions of Theorem 6.8 in fact hold for m = 0. In this situation, we will refer to m = 0 as a positivity parameter, by slight abuse of notation, since we need only the isomorphism ℓY∞ ∼= ℓZ∞ for ℓ ≥ m in the sequel. Corollary 6.11. Let ℓ ∈ N and let ℓY∞ be the moduli stack of embedded ℓ-shifted S-point modules, as above. Then ℓY∞ is a noetherian fpqc-algebraic stack. Proof. We know that ℓY∞ is a sheaf in the fpqc topology by Theorem 5.11. Let m be a positivity parameter. If ℓ ≥ m, then ℓY∞ ∼= ℓZ∞ is noetherian fpqc-algebraic by Theorem 5.11. Suppose then that ℓ < m. Let T : ℓY∞ → mY∞ be the morphism defined by T (M ) := M≥m. It is straightforward that T is a morphism of functors, and that the product Φm × T : ℓY∞ → ℓYm ×X (mY∞) is a closed immersion. Since mY∞ is noetherian by the first paragraph, it has an fpqc cover U → mY∞ by a noetherian affine scheme U . This cover can clearly be lifted and refined to induce an fpqc cover V → ℓYm ×X (mY∞) where V is a noetherian affine scheme. But then the Cartesian product V ′ ℓY∞ (cid:3) Φm×T V / ℓYm ×X (mZ∞) gives an fpqc cover V ′ → Y∞. Since Φm × T is a closed immersion, so is V ′ → V . Thus V ′ is isomorphic to a closed subscheme of V and is noetherian and affine. (cid:3) Let m be a positivity parameter and let ℓ ≥ m. We note that the relevant component of ℓY∞ ∼= ℓZ∞ is the component containing the k(X)-point corresponding to the generic point module k(X)zℓ ⊕ k(X)zℓ+1 ⊕ · · · , which is isomorphic to (Qgr(S))≥ℓ. 22 o o o o O O / / / 7. A coarse moduli space for point modules In this section, we consider point modules up to module isomorphism in qgr-S, and show that the scheme X is a coarse moduli scheme for this functor. We define the following maps. For any ℓ, let Φ : ℓZ∞ → X be the map induced from the fn. Let Ψ : 0Z∞ → 0Z∞ be the map induced from ψn : 0Zn → 0Zn−1. Taking the limit of (5.8), we obtain that ΦΨ = σΦ : 0Z∞ → X. For any noetherian commutative k-algebra C, there is a graded (OXC , σ×1)-bimodule algebra SC given by pulling back S along the projection map XC → X. Taking global sections gives a functor Gr-SC → Gr-SC . If C = k, this induces an equivalence qgr-SC → qgr-SC by Theorem 2.8. In order to avoid the issues involved with extending this result to bimodule algebras over arbitrary base schemes, we work instead with point modules in Qgr-SC and Qgr-SC . Let ℓ ≥ m where m is a positivity parameter (Definition 6.3) and let F be the moduli functor of (embedded) ℓ-shifted point modules over S, as in the previous section. Define an equivalence relation ∼ on F (C) by saying that M ∼ N if (their images) are isomorphic in Qgr-SC. Define a contravariant functor G : Affine schemes −→ Sets by sheafifying, in the fpqc topology, the presheaf Gpre of sets Spec C 7→ F (C)/ ∼. Let µ : F → G be the natural map. Likewise, let F ∼= ℓZ∞ be the moduli functor of ℓ-shifted point modules over S, and let G := F / ∼, as above. Let µ : F → G be the natural map. We recall that an : Xn → X is the blowup of X at In, and that by Corollary 3.3 there are morphisms α : Xn+1 → Xn that intertwine with the maps an. We briefly discuss point modules over local rings. We note that if C is a local ring of a point of Z o ∞ ∼= X∞ with maximal ideal m, then the map hZ∞(C) → F (C) has a particularly simple form. Let ζ : Spec C → X∞ be the induced morphism. By critical density, there is some n ≥ m so that ζn(m) is not a fundamental point of any of the maps αi : Xi+n → Xn, for any i > 0. Let xn := ζn(m). Define a−1 n (Sj ) := a−1 n (Ij) ⊗Xn a∗ nLj . Then a−1 n (Sj) is flat at xn for all j. Let M :=Mj≥0 a−1 n (Sj )xn . Then M is flat over C and is the C-point module corresponding to ζ. The C-action on M is obvious; to define the S-action on M, let x := an(xn). Then there are maps Mj ⊗k Si(X) / Mj ⊗k S σj i (X) / Mj ⊗OX,x (S σj i )x / a−1 n (Sj+i)xn . This gives a right S-action on M; by letting C act naturally on the left and identifying C with Cop we obtain an action of SC on M . If C is a local ring, we do not know if SC is necessarily noetherian. However, C-point modules in Qgr-SC are well-behaved, as follows. Lemma 7.1. Let C be a commutative noetherian local k-algebra. Let N and M be ℓ-shifted C-point modules, with M ∼= N in Qgr-SC . Then for some k we have M≥k ∼= N≥k. Proof. The torsion submodules of M and N are trivial. Thus HomQgr-SC (M, N ) = lim−→ HomGr-SC (M ′, N ), where the limit is taken over all submodules M ′ ⊆ M with M/M ′ torsion. If M ∼= N in Qgr-SC, then there is some submodule M ′ ⊆ M so that M/M ′ is torsion, and so that there is a homomorphism f : M ′ → N so that ker f and N/f (M ′) are torsion. Since M and N are torsion-free, we must have ker f = 0. Thus it suffices to show that if N is an ℓ-shifted C-point module and M ⊆ N is a graded submodule with T := N/M torsion, then Tn = 0 for n ≫ 0. Let L be the residue field of C. By assumption, N is C-flat. Thus there is an exact sequence 0 → TorC 1 (T, L) → ML → NL → TL → 0. 23 / / / By [ASZ99, Theorem 5.1], the algebra RL is noetherian. Thus NL and TL are also noetherian. Since TL is torsion, it is finite-dimensional. Thus for n ≫ 0, we have (TL)n = (Tn) ⊗C L = 0. By Nakayama's Lemma, Tn = 0. (cid:3) Lemma 7.2. Let C be a noetherian local ring, and let M, N be C-point modules over S, corresponding to morphisms fM, fN : Spec C → 0Z∞. If N ∼ M, then there is some k so that ΨkfM = ΨkfN . Proof. Let m be a positivity parameter and let M := s(M≥m) and N := s(N≥m). Then M ∼ N ; by Lemma 7.1 there is some k, which we may take to be at least m, so that M≥k ∼= N≥k. Since kZ∞ ∼= kY∞, we have M≥k ∼= N≥k. Thus the modules M[k]≥0 ∼= N [k]≥0 correspond to the same point of 0Z∞; that is, ΨkfM = ΨkfN . (cid:3) We now show that X is a coarse moduli space for 0Z∞/ ∼. In fact, we prove this result in greater generality, to be able to analyze the spaces ℓY∞. Proposition 7.3. Let Z∞ := 0Z∞. Let V∞ be a closed algebraic substack of Z∞ so that X∞ ⊆ V∞ ⊆ Z∞, Vn where Vn ⊂ Zn is a closed subscheme that maps into Vn−1 under Zn → Zn−1 and assume that V∞ = lim ←− for all n. Then X is a coarse moduli space for V∞/ ∼. More precisely, let H be the image of V∞ under µ : Z∞ → G. Then: (1) The morphism Φ : V∞ → X factors via V∞ (2) Every morphism H → A where A is a scheme (of finite type) factors uniquely through H t−→ X. µ −→ H t−→ X. Proof. (1) It suffices to prove that if C is a commutative noetherian ring and M ∼ N are C-point modules over S, corresponding to maps fM, fN : Spec C → Z∞, then we have ΦfM = ΦfN : Spec C → X. To show this, it suffices to consider the case that C is local. By Lemma 7.2, ΨkfM = ΨkfN for some k. Thus ΦfM = σ−kΦΨkfM = σ−kΦΨkfN = ΦfN , as required. on overlaps Ui ∩ Uj. (2) Let ν : H → hA be a natural transformation for some scheme A. For all n ∈ N, let Pn be the subscheme C := OX,x. The induced map Spec C → X∞ → V∞ gives a C-point module Mx as described above; its ∼-equivalence class is an element of H(C). Applying ν, we therefore have a morphism Spec C → A. This extends to a morphism gx : Ux → A for some open subset Ux of X. It follows from critical density of the of X defined by In. Fix any closed point x ∈ X rS Pn; some such x exists since k is uncountable. Let orbits of points in P that X rS Pn is quasi-compact. Thus we may take finitely many Ux, say U1, . . . , Uk, that cover X rS Pn, with maps gi : Ui → A. These maps all agree on the generic point of X and so agree Ui, and let g : U → A be the induced map. Then X r U ⊆ S Pn is a closed subset of X, and so X r U = {z1, . . . , zr} for some z1, . . . , zr ∈ S Pn. Let n be such that for any i > 0, the map φi : Zn+i → Zn is a local isomorphism at all points in the preimage of {z1, . . . , zr}. Let Φn : Z∞ → Zn be the map induced from the φm. There is an induced map f −1 n (U ) ∩ Vn → U → A. Further, for every y ∈ V∞ r Φ−1(U ), the map Spec OV∞,y → V∞ → H → A factors through V∞ → Vn, as Φn is a local isomorphism at y. We thus obtain morphisms Spec OVn,y → A for all y ∈ Vn. Using these, we may extend g to give a morphism θ : Vn → A so that the diagram Let U := k[i=1 H ν A µ V∞ Φn Vn fn U θ >⑤⑤⑤⑤⑤⑤⑤⑤ g commutes. 24 / / / / ✤ ✤ ✤ > We claim that θ contracts each of the loci f −1 n (zj) ∩ Vn to a point. To see this, let x, y ∈ Φ−1(zj) ∩ V∞, corresponding to maps fx, fy : Spec k → V∞. We must show that θΦnfx = θΦnfy. Since for k ≫ 0, σk(zj) is not inS Pn, Φ is a local isomorphism at Ψk(Φ−1(zj)). We have ΦΨkfx = σkΦfx = σkΦfy = ΦΨkfy and so Ψkfx = Ψkfy. Therefore, µfx = µfy, and so θΦnfx = νµfx = νµfy = θΦnfy, as we wanted. The morphism θ : Vn → A thus factors set-theoretically to give a map from X to A. Since X is smooth at all zi by critical density of the orbits of the zi, it is well known that θ also factors scheme-theoretically. Consequently, we have the morphism X → A that we sought. This proves Proposition 7.3. (cid:3) Theorem 7.4. Fix a positivity parameter m (Definition 6.3) and let ℓ ≥ m. Then X is a coarse moduli space for G = F/ ∼. Proof. As above, we denote the functor of ℓ-shifted point modules over S modulo ∼ by G. By Theorem 6.8(2), it is enough to show that X is a coarse moduli space for G. Let L : ℓZ∞ → 0Z∞ be the map that sends M 7→ M[ℓ]. Notice that if M, N are ℓ-shifted point modules, then M ∼ N if and only if M[ℓ] ∼ N [ℓ]. That is, if we let G′ be the functor of (unshifted) point modules over S modulo ∼, then L induces an inclusion G → G′ so that the diagram ℓZ∞ L / 0Z∞ µ G µ / G′ L commutes. Let Vn := Im(ℓZℓ+n → 0Zn) and V∞ := lim ←− Vn, so V∞ = L(ℓZ∞). Note that L is injective on X∞ ⊆ ℓY∞. Thus V∞ satisfies the hypotheses of Proposition 7.3, and so X is (cid:3) a coarse moduli scheme for H = µ(V∞) ∼= µ(ℓZ∞) ∼= G. References [ASZ99] M. Artin, L. W. Small, and J. J. Zhang, Generic flatness for strongly noetherian algebras, J. Algebra 221 (1999), no. 2, 579 -- 610. 12, 24 [ATV90] M. Artin, J. Tate, and M. Van den Bergh, Some algebras associated to automorphisms of elliptic curves, The Grothendieck Festschrift Vol. I, Progr. Math., vol. 86, Birkhauser Boston, Boston, 1990, pp. 33 -- 85. 1, 13, 14, 15 [AV90] M. Artin and M. Van den Bergh, Twisted homogeneous coordinate rings, J. Algebra 133 (1990), no. 2, 249 -- 271. 3 [AZ01] M. Artin and J. J. Zhang, Abstract Hilbert schemes, Algebr. Represent. Theory 4 (2001), no. 4, 305 -- 394. 1 [Eis95] D. Eisenbud, Commutative algebra with a view toward algebraic geometry, Graduate Texts in Mathematics, vol. 150, Springer-Verlag, New York, 1995. 13 [Gab62] P. Gabriel, Des cat´egories ab´eliennes, Bull. Soc. Math. France 90 (1962), 323 -- 448. 4 [EGA] A. Grothendieck, ´El´ements de g´eom´etrie alg´ebrique, IV, Inst. Hautes ´Etudes Sci. Publ. Math., 20 (1964), 24 (1965), 28 (1966), 32 (1967). 9, 12 [Har77] R. Hartshorne, Algebraic geometry, Graduate Texts in Mathematics, vol. 52, Springer-Verlag, New York, 1977. 8, 10, 19 [Har95] Joe Harris, Algebraic geometry, Graduate Texts in Mathematics, vol. 133, Springer-Verlag, New York, 1995, A first course, Corrected reprint of the 1992 original. 20 [HL97] Daniel Huybrechts and Manfred Lehn, The geometry of moduli spaces of sheaves, Aspects of Mathematics, E31, Friedr. Vieweg & Sohn, Braunschweig, 1997. 3 [Kee00] D. S. Keeler, Criteria for σ-ampleness, J. Amer. Math. Soc. 13 (2000), no. 3, 517 -- 532. 5, 19 , Ample filters and Frobenius amplitude, http://arXiv.org/abs/math/0603388, 2006. 6 [Kee06] [Kle90] Steven L. Kleiman, Multiple-point formulas. II. The Hilbert scheme, Enumerative geometry (Sitges, 1987), Lecture Notes in Math., vol. 1436, Springer, Berlin, 1990, pp. 101 -- 138. 7 [KRS05] D. S. Keeler, D. Rogalski, and J. T. Stafford, Naıve noncommutative blowing up, Duke Math. J. 126 (2005), no. 3, 491 -- 546. 1, 2, 3, 4, 5, 16 [LMB00] G´erard Laumon and Laurent Moret-Bailly, Champs alg´ebriques, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics], vol. 39, Springer-Verlag, Berlin, 2000. 2, 9, 12 25 / / [Laz04] R. Lazarsfeld, Positivity in algebraic geometry. I. Classical setting: line bundles and linear series, Ergebnisse der [NS] Mathematik und ihrer Grenzgebiete. 3. Folge., vol. 48, Springer-Verlag, Berlin, 2004. 5 T. A. Nevins and S. J. Sierra, Canonical birationally commutative factors of noetherian graded algebras, in preparation. 3 [Rog04] D. Rogalski, Generic noncommutative surfaces, Adv. Math. 184 (2004), no. 2, 289 -- 341. 1 [RS07] D. Rogalski and J. T. Stafford, Naıve noncommutative blowups at zero-dimensional schemes, J. Algebra 318 (2007), no. 2, 794 -- 833. 1, 6, 7, 18, 21 [RS09] , A class of noncommutative projective surfaces, Proc. Lond. Math. Soc. (3) 99 (2009), no. 1, 100 -- 144. 13, 15, 21 [RZ08] D. Rogalski and J. J. Zhang, Canonical maps to twisted rings, Math. Z. 259 (2008), no. 2, 433 -- 455. 1, 3 [Sie11] [SV01] S. J. Sierra, Geometric idealizer rings, Trans. Amer. Math. Soc. 363 (2011), 457 -- 500. 3 J. T. Stafford and M. Van den Bergh, Noncommutative curves and noncommutative surfaces, Bull. Amer. Math. Soc. 38 (2001), 171 -- 216. 1 [Van96] M. Van den Bergh, A translation principle for the four-dimensional Sklyanin algebras, J. Algebra 184 (1996), no. 2, 435 -- 490. 3, 4, 5 [Vis05] Angelo Vistoli, Grothendieck topologies, fibered categories and descent theory, Fundamental algebraic geometry, Math. Surveys Monogr., vol. 123, Amer. Math. Soc., Providence, RI, 2005, pp. 1 -- 104. 9, 12, 18 Department of Mathematics, University of Illinois at Urbana-Champaign, Urbana, IL 61801 USA E-mail address: [email protected] School of Mathematics, University of Edinburgh, Edinburgh EH9 3JZ, United Kingdom. E-mail address: [email protected] 26
1212.0774	1	1212	2012-12-04T16:03:47	The Tate-Hochschild cohomology ring of a group algebra	[ "math.RA", "math.KT", "math.RT" ]	We show that the Tate-Hochschild cohomology ring $HH^(RG,RG)$ of a finite group algebra $RG$ is isomorphic to a direct sum of the Tate cohomology rings of the centralizers of conjugacy class representatives of $G$. Moreover, our main result provides an explicit formula for the cup product in $HH^(RG,RG)$ with respect to this decomposition. As an example, this formula helps us to compute the Tate-Hochschild cohomology ring of the symmetric group $S_3$ with coefficients in a field of characteristic 3.	math.RA	math	THE TATE-HOCHSCHILD COHOMOLOGY RING OF A GROUP ALGEBRA VAN C. NGUYEN Abstract. We show that the Tate-Hochschild cohomology ringdHH explicit formula for the cup product indHH (RG, RG) of a finite group algebra RG is isomorphic to a direct sum of the Tate cohomology rings of the cen- tralizers of conjugacy class representatives of G. Moreover, our main result provides an (RG, RG) with respect to this decomposition. As an example, this formula helps us to compute the Tate-Hochschild cohomology ring of the symmetric group S3 with coefficients in a field of characteristic 3. ∗ ∗ 1. Introduction The theory of group cohomology is a well-studied yet ongoing research area. It has many applications to other areas such as representation theory, algebraic geometry, and commutative algebra. It is well-known that the Hochschild cohomology of a group algebra agrees with its usual cohomology, in the following sense. For an arbitrary commutative ring R and a group G, if M is an RG-bimodule, then it may be regarded as a left RG-module via the diagonal action g · m = gmg−1, where g ∈ G, m ∈ M . Conversely, any left RG-module M may be considered as an RG-bimodule by letting RG act trivially on the right. Together with the Eckmann-Shapiro Lemma, this shows that the Hochschild cohomology ring, HH∗(RG, M ) := RG⊗RRGop(RG, M ) with coefficients in an RG-bimodule M , is the same as the usual RG(R, M ) with coefficients in M under the diagonal action. In particular, by considering RG as its own bimodule, HH∗(RG, RG) is iso- morphic to H∗(G, RG), where RG is a left RG-module via conjugation. From this identification and the Eckmann-Shapiro Lemma, one can prove that HH∗(RG, RG) may be decomposed as a direct sum of the cohomology of the centralizers of conjugacy class representatives of G ([6], Theorem 2.11.2). In 1999, Siegel and Witherspoon then described a formula for the products in HH∗(RG, RG) in terms of this additive decomposition [11]. When G is abelian, Cibils and Solotar proved that the Hochschild cohomology ring of G is (isomorphic to) the tensor product over R of RG and its usual cohomology ring [9]. Ln≥0 Extn group cohomology ring, H∗(G, M ) := Ln≥0 Extn In 1952, John Tate introduced a group cohomology theory that is based on complete res- olutions, and hence, expands the study of group cohomology to negative degrees [12]. Tate cohomology exploits the similarities between the usual homology and cohomology of a finite group G. A lot of study about this new cohomology has been done and can be found in ([7], Ch. VI) or ([8], Ch. XII). In particular, in 1992, Benson and Carlson proved that for a finite group G and a field k of characteristic p > 0, if the depth of H∗(G, k) is greater than one, products of elements in negative cohomology are zero [2]. More recently, Bergh and Jorgensen combined the 1 2 VAN C. NGUYEN notions of Hochschild cohomology and Tate cohomology to form the Tate-Hochschild cohomol- ogy [3]. In this paper, we explore the structure of the Tate-Hochschild cohomology of a group algebra and generalize some known results about group cohomology to its negative degrees. The material is organized as follows. In Section 2, we recall the definition and properties of Tate cohomology. Let M be a left RG-module. Then for all n ∈ Z: The readers who are familiar with Tate cohomology can skip this section. In Section 3, we specialize the definition of Tate-Hochschild cohomology in [3] to the group algebra. Let M be an RG-bimodule. Then for all n ∈ Z: n bH (G, M ) :=dExt n RG(R, M ). In this section, we also briefly present a cup product that makes the Tate-Hochschild cohomology n (RG, M ) :=dExt dHH (RG, RG) =Mn∈ZdHH dHH ∗ n RG⊗RRGop(RG, M ). n (RG, RG) ∗ ∗ ∗ ∗ become a graded ring [10]. to its Tate cohomology using a similar argument as in the usual cohomology. In Section 4, we reduce the Tate-Hochschild cohomology of RG In particular, (G, RG), where RG is a left RG-module via conjugation. In addition, we state known Tate cohomology relations with subgroups, which will be useful in proving our main result. In Section 5, we introduce a generalization of the Tate- Hochschild cohomology ring of RG by letting another finite group H act on G and consider (H, RG). Let g1, . . . , gt ∈ G be representatives of the orbits of the action of H on G. Let (H, RG) (RG, RG) is isomorphic to bH we show that dHH bH Hi decomposes as: (H, RG) ∼=Mi bH := StabH(gi) = {h ∈ H hgi = gi} be the stabilizer of gi. We show that bH In the main Theorem 5.5, we describe the multiplicative structure of bH observe that the Tate-Hochschild cohomology ring bH (RG, RG) ∼= bH of bH Moreover, the product formula reduces the computation of products indHH Solotar's result in [9] is also generalized to the isomorphism dHH (H, RG) by giving it a product formula in terms of this additive decomposition. Our work, which uses mainly the idea from [11], is a straightforward generalization of the usual group cohomology results. We (G, RG) is a special case (H, RG) by letting H = G act on itself by conjugation. Working with this generalization, we prove that the Tate-Hochschild cohomology of G decomposes as a direct sum of the Tate cohomology of the centralizers of conjugacy class representatives of G with coefficients in R. (RG, RG) to prod- ucts within the Tate cohomology rings of certain subgroups of G. When G is abelian, Cibils and (G, R) in Proposition 5.1. In Section 6, we describe the Tate-Hochschild cohomology ring of the sym- metric group on three elements S3 over a field k of characteristic 3, utilizing the product formula in Theorem 5.5. (RG, RG) ∼= RG ⊗RbH (Hi, R). ∗ bH ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ 2. Tate Cohomology of a Group Algebra Throughout this paper, we let G be a finite group and R be the ring of integers Z or a field k of characteristic p > 0 such that p divides the order of G. All rings and algebras are assumed to possess a unit; all modules are assumed to be left modules unless stated otherwise; and THE TATE-HOCHSCHILD COHOMOLOGY RING OF A GROUP ALGEBRA 3 tensor products will be over R unless stated otherwise. If R = k is a field, then kG is a finite dimensional, symmetric, and self-injective algebra over k ([5], Prop. 3.1.2), and hence, projective kG-modules are the same as injective kG-modules. If G is acting on a set X, then we denote the action gx = gxg−1, for all g ∈ G and x ∈ X. Let M and N be left RG-modules. Then for any g ∈ G, m ∈ M , n ∈ N , and f ∈ HomR(M, N ), we observe some basic facts: • M ⊗ N is a left RG-module via g · (m ⊗ n) = (g · m) ⊗ (g · n), • HomR(M, N ) is a left RG-module via (g · f )(m) = g(f (g−1 · m)), • We may regard left RG-modules as right RG-modules via m · g = g−1 · m. The Tate cohomology for RG is defined in both positive and negative degrees using the following general resolution: Definition 2.1. Let R be a two-sided Noetherian ring. A complete resolution of a finitely generated R-module M is an exact complex P = {{Pi}i∈Z, di : Pi → Pi−1} of finitely generated projective R-modules such that: (1) The dual complex HomR(P, R) is also exact (2) There exists a projective resolution Q ε→ M of M and a chain map P ϕ → Q where ϕn is bijective for n ≥ 0 and 0 for n < 0. We explicitly construct an RG-complete resolution of R as follows. Let · · · d3−→ P2 d2−→ P1 d1−→ P0 ε−→ R → 0 be an RG-projective resolution of R, where each Pi is a finitely generated projective RG-module. Apply HomR(−, R) to get a dual sequence: 0 → R → HomR(P0, R) → HomR(P1, R) → HomR(P2, R) → · · · , which is an exact sequence of RG-modules, since R ∼= HomR(R, R) and HomR(−, R) is an exact functor when R = k is a field (the case R = Z is shown in [8], Prop. XII.3.3). Splicing these two sequences together, one forms a doubly infinite sequence: · · · d3−→ P2 d2−→ P1 d1−→ P0 → HomR(P0, R) → HomR(P1, R) → · · · We use the notation P−(n+1) := HomR(Pn, R). By definition, the above sequence is an RG- complete resolution of R: P : · · · d3−→ P2 d2−→ P1 d1−→ P0 → P−1 → P−2 → · · · Let M be any (left) RG-module. Apply HomRG(−, M ) to P and take the homology of this new complex, we obtain the Tate cohomology for RG: n RG(R, M ) = Hn(HomRG(P, M )), for all n ∈ Z. n bH (G, M ) :=dExt Observe that in our context, naturally, the Tate (co)homology is independent of the complete resolution of R ([1], Theorem 5.2 and Lemma 5.3). One can see this by applying a complete chain map between two complete resolutions of R in both positive and negative degrees. This is the generalized Comparison Theorem on complete resolutions ([7], Prop. VI.3.3). We note that the Tate cohomology of RG obtains a multiplicative structure as described in ([7], Section VI.5) and ([8], Sections XII.4 and XII.5). In particular, if M is a ring on which G 4 VAN C. NGUYEN (G, M ) becomes an associative Moreover, for a left RG-module M , we recall the following properties of Tate cohomology ([7], ∗ acts by automorphisms: g · (m1m2) = (g · m1)(g · m2), then bH graded ring. It is graded-commutative, in the sense that, for α ∈ bH αβ = (−1)ijβα ([8], Props. XII.5.2 and XII.5.3). When M = R, we denote bH (a) For all n > 0, bH (G, M ) ∼= Hn(G, M ). Section VI.5): n i j ∗ (G, M ) and β ∈ bH (G) := bH This follows from the construction of a complete resolution of R. The positive-degree component of a complete resolution of R arises from a projective resolution of R which is also used to form the usual cohomology groups Hn(G, M ). (G, M ), ∗ (G, R). 0 (G, M ) is the dually defined submodule of H0(G, M ). These follow from the construction of complete resolutions. (G, M ) is a quotient of H0(G, M ). −1 (b) The group bH (c) The group bH (d) For all n < −1, we have isomorphisms: bH explicitly by: τ (f ⊗RG m)(p) = Pg∈G f (g−1p)gm, where m ∈ M, p ∈ Pt, and f ∈ For any left finitely generated RG-module Pt, the map τ : HomR(Pt, R) ⊗RG M → It is defined HomR(Pt, R). With t = n < −1, Pn := HomR(P−(n+1), R). Since P−(n+1) is finitely generated, Hom(Pn, R) = Hom(Hom(P−(n+1), R), R) ∼= P−(n+1). Hence, by the duality isomorphism, we have P−(n+1) ⊗RG M ∼= HomRG(Pn, M ), which induces the desired isomorphism on homology. HomRG(Pt, M ) is an RG-module isomorphism ([8], Section XII.3 (5)). (G, M ) ∼= H−(n+1)(G, M ). n (e) If 0 → M → M ′ → M ′′ → 0 is a short exact sequence of (left) RG-modules, then there is a doubly infinite long exact sequence of Tate cohomology groups: n n · · · → bH (G, M ) → bH n (G, M ′) → bH (G, M ′′) → bH This is true since the spliced complex P consists of projective modules, so the short exact sequence of modules 0 → M → M ′ → M ′′ → 0 gives rise to a short exact sequence of complexes: n+1 (G, M ) → · · · 0 → HomRG(P, M ) → HomRG(P, M ′) → HomRG(P, M ′′) → 0 whose cohomology long exact sequence is the desired long exact sequence ([1], Prop. 5.4). (f) If (Nj)j∈J is a finite family of (left) RG-modules and (Mi)i∈I is any family of RG- modules, then there are natural isomorphisms, for all n ∈ Z: n RG(Mj∈J dExt RG(N,Yi∈I dExt Nj, M ) ∼=Yj∈JdExt Mi) ∼=Yi∈IdExt n n RG(Nj, M ) n RG(N, Mi). The idea of this proof is similar to that of the usual Extn RG, using the analogous relation for Hom ([1], Prop. 5.7). In 2011, Bergh and Jorgensen introduced the notion of Tate-Hochschild cohomology which is the extended Hochschild cohomology using complete resolutions [3]. One natural question THE TATE-HOCHSCHILD COHOMOLOGY RING OF A GROUP ALGEBRA 5 to ask is whether the Tate-Hochschild cohomology shares the same properties as those of the usual Hochschild cohomology. In the following sections, we will examine the Tate-Hochschild cohomology of RG and its ring structure. 3. Tate-Hochschild Cohomology of a Group Algebra The opposite algebra of RG, denoted by RGop, is RG with the same addition but with multiplication performed in the reverse order. Let RGe = RG ⊗ RGop be the enveloping algebra of RG. Since G is finite, RGe is two-sided Noetherian and Gorenstein of Gorenstein dimension 0. Let M be an RG-bimodule. M may be regarded as a left RGe-module by setting (a⊗b)·m = amb, where a ∈ RG, b ∈ RGop, and m ∈ M . In particular, we shall regard RG as a left RGe-module. For any integer n ∈ Z, the n-th Tate-Hochschild cohomology of RG is defined as: n dHH (RG, M ) :=dExt n RGe(RG, M ), where the dExt functor is taken using an RGe-complete resolution of RG. As RGe is a two-sided Noetherian and Gorenstein ring, Theorems 3.1 and 3.2 in [1] guarantee that every finitely gener- ated RGe-module admits a complete resolution. Hence, we obtain an RGe-complete resolution X for RG. The n-th Tate-Hochschild cohomology group of RG is the n-th homology group of the complex HomRGe(X, M ). Remark. The Tate-Hochschild cohomology groups of RG agree with the usual Hochschild cohomology groups in all positive degrees: n (RG, M ) ∼= HHn(RG, M ), for all n > 0. Let M and N be RG-bimodules. Then M ⊗RG N is also an RG-bimodule, which can be considered as a left RGe-module via (a ⊗ b) · (m ⊗RG n) = am ⊗RG nb, where a ∈ RG, b ∈ RGop, m ∈ M , and n ∈ N . There is a cup product on Tate-Hochschild cohomology: dHH i dHH j i+j (RG, M ⊗RG N ), Xi ⊗RG Xj, for all n ∈ Z. (RG, N ) →dHH chain complex of RGe-modules. Lemma 6.5 in [10] shows the existence of a complete diagonal which we describe as follows. Let X be any RGe-complete resolution of RG. The complete (RG, M ) ⊗dHH tensor product Xb⊗RGX is obtained by defining: (Xb⊗RGX)n = Yi+j=n By the same argument as in ([7], Section VI.5) or ([10], Section 6.1), Xb⊗RGX is an acyclic approximation chain map Γ : X → Xb⊗RGX that preserves the augmentation. v : C ′ → B′ of degree j, there is a map ub⊗v : Cb⊗C ′ → Bb⊗B′ of degree i + j defined by: (ub⊗v)n = Yr+s=n Let f ∈ HomRGe(Xi, M ) represent an element of dHH represent an element ofdHH Given graded modules B, B′, C, C ′ and module homomorphisms u : C → B of degree i and (−1)rjur ⊗ vs : Yr+s=n (RG, M ) and let g ∈ HomRGe(Xj, N ) s → Yr+s=n Cr ⊗ C ′ Br+i ⊗ B′ s+j. j (RG, N ). Then: i f ⌣ g = (fb⊗g) ◦ Γ ∈ HomRGe (X, M ⊗RG N ) 6 VAN C. NGUYEN (RG, M ⊗RG N ). One can check that this product is independent of X and Γ and satisfies the usual cup product properties. When M = N = RG, this cup product (RG, RG) the structure of an associative graded ring, as RG ∼= RG ⊗RG RG. i+j represents an element ofdHH gives dHH ∗ 4. Reduction to Tate Cohomology and Relations with Subgroups We show here that the Tate-Hochschild cohomology of RG can be reduced to its Tate coho- mology. We begin with a lemma which is based on the original Eckmann-Shapiro Lemma but is generalized to a complete resolution: Lemma 4.1 (Eckmann-Shapiro, ([10], Lemma 7.1) or ([7], VI.5.2)). Let H be a subgroup of a finite group G, M be a left RH-module and N be a left RG-module. Consider N ↓G H = N to be a left RH-module via restriction of the action, and let M ↑G H := RG ⊗RH M denote the induced RG-module where G acts on the leftmost factor by multiplication. Then for all n ∈ Z, there is an isomorphism of abelian groups: n RH (M, N ↓G n RG(M ↑G H , N ). For any subgroup H ⊆ G, every RH-complete resolution gives rise to an RG-complete reso- lution by inducing the modules (induction takes projectives to projectives, and exact sequences to exact sequences). Since coinduction is the same as induction in the finite group case ([7], Prop. III.5.9), the proof for this lemma goes through in the present context and follows from the Nakayama relations ([5], Prop. 2.8.3). We have RG ∼= RGop as algebras via g 7→ g−1, and RG ⊗ RG ∼= R(G × G) as algebras. As a result, RGe = RG ⊗ RGop ∼= R(G × G). Here, RG is a left R(G × G)-module by the two-sided action: (g1, g2)·g = g1gg−1 δ(G) = R(G×G)⊗Rδ(G) R on the cosets of the diagonal δ(G) = {(g, g) g ∈ G}. Moreover, any RG-bimodule M may be considered as a left RG-module with action given by g · m = gmg−1, for all g ∈ G and m ∈ M . It follows from this discussion and Lemma 4.1 that: 2 . So RG is just the permutation module R ↑G×G dExt H ) ∼=dExt ∗ ∗ dHH (RG, M ) =dExt ∗ R(G×G)(RG, M ) ∗ R(G×G)(R ↑G×G RG(R, M ↓G×G δ(G) ) ∗ δ(G) , M ) (G, M ). RGe(RG, M ) ∼=dExt ∼=dExt ∼=dExt ∼= bH ∗ The Tate-Hochschild cohomology of G with coefficients in the bimodule M is just the Tate cohomology of G with coefficients in M under the diagonal action. When M is an algebra itself with compatible G-action, this is in fact an algebra isomorphism as it respects the cup products defined for Tate and Tate-Hochschild cohomology ([10], Theorem 7.2). In particular, when M = RG with G-action via conjugation, Therefore, all properties of bH Tate-Hochschild cohomology. ∗ (G, RG). (4.2) (G, RG) that we observed in Section 2 transfer to those for the ∗ dHH ∗ (RG, RG) ∼= bH Let H be a subgroup of G. By restricting the action, any RG-module N may be regarded as an RH-module and any RG-complete resolution P of R may also be considered as an RH-complete THE TATE-HOCHSCHILD COHOMOLOGY RING OF A GROUP ALGEBRA 7 resolution. Sections XII.8 and XII.9 in [8] show that there are maps in the Tate cohomology with properties analogous to those in the usual group cohomology: • The restriction map: which is induced from the inclusion HomRG(P, N ) ⊂ HomRH(P, N ). • The corestriction map (or transfer): which is given on the cochain level by defining: resG corG ∗ ∗ H :bH (G, N ) → bH H :bH (H, N ) → bH H f )(p) =Xg∈G (corG ∗ (H, N ), ∗ (G, N ), gf (g−1p), where G denotes a set of left coset representatives of H in G, f ∈ HomRH (Pi, N ), and p ∈ Pi. One can check that this definition is independent of the choice of the representatives g ∈ G. • Moreover, for any g ∈ G, there is an isomorphism: defined on the cochain level as (g∗f )(p) = g(f (g−1p). ∗ ∗ g∗ : bH (H, N ) → bH (gHg−1, N ) = bH ∗ (gH, N ) We shall list some properties of these maps without proving them. The proof goes through using similar arguments as in [8]. The readers can refer to [8] for more details. ∗ ∗ H ◦ resG K ◦ resG H ◦ corH Proposition 4.3 ([8], Section XII.8 (4)-(14) and Section XII.9 (4)). Let K ⊆ H ⊆ G be sub- (G, Ni), groups, and N1, N2 be RG-modules which may be regarded as RH-modules. Let αi ∈bH βi ∈bH (H, Ni), and gi ∈ G, for i = 1, 2. Then the maps defined above satisfy: 1g∗ 2 = (g1g2)∗ H = (G : H)1 H = resG K K = corG K (1) g∗ (2) g∗ = 1, if g ∈ H (3) corG (4) resH (5) corG (6) g∗ ◦ resH (7) g∗ ◦ corH (8) resG (9) corG (10) corG (11) g∗(β1 ⌣ β2) = (g∗β1) ⌣ (g∗β2) (12) Let H, K ⊆ G be subgroups and N be an RG-module which may be regarded as an RH gH gK ◦g∗ gH gK ◦g∗ H (α1 ⌣ α2) = (resG H(β1 ⌣ resG H(resG H α1 ⌣ β2) = α1 ⌣ (corG H β1) ⌣ α2 H β2) K = res K = cor H α2) = (corG H α1) ⌣ (resG H α2) ∗ ∗ H :bH corK (H, N ) → bH K∩ xH(res xH K∩ xH (x∗β)), (K, N ) is given by: (or RK)-module. The map resG K ◦ corG resG K(corG H(β)) =Xx∈D ∗ where β ∈ bH [x∈D KxH is a disjoint union. (H, N ) and D is a set of double coset representatives such that G = 8 VAN C. NGUYEN 5. Generalized Additive Decomposition ∗ dHH (RG, RG) ∼= bH ∗ ∗ dHH ∗ ∗ For the rest of this paper, we let H be another finite group which acts as automorphisms on G. Via this action, RG becomes an RH-module. The multiplication map RG ⊗ RG → RG is an (H, RG) := ∗ RH (R, RG) by compositing with the cup product. We will study the additive decomposition (G, RG) is a RH-module homomorphism. Hence, it induces the ring structure on cohomology bH dExt of this ring bH (RG, RG) ∼= bH special case of bH Proposition 5.1. If H acts trivially on G, then bH (RG, RG) ∼= RG ⊗RbH Proof. If G is abelian, H = G acting on itself by conjugation yields the trivial action. Hence, the second statement follows from the first statement and the isomorphism (4.2): (H, RG). The Tate-Hochschild cohomology ring bH (H, RG) by letting H = G act on itself by conjugation. (H, RG) ∼= RG ⊗R bH R-algebras. In particular, if G is abelian, then (H, R) as graded (G, R). ∗ ∗ ∗ ∗ ∗ ∗ (G, RG) ∼= RG ⊗RbH ∗ (G, R). To prove the first statement, let ε : P → R be an RH-complete resolution of R. Since H acts trivially on G, RG is a trivial RH-module and is free as an R-module. We claim that γ : RG ⊗ HomRH (P, R) → HomRH (P, RG) is an isomorphism. It can be seen by sending g ⊗ f 7→ γ(g ⊗ f ) = F , where F (p) = f (p)g, for f ∈ HomRH (Pi, R), p ∈ Pi, and g ∈ G. It is easy to check that F ∈ HomRH(P, RG) and it is a cocycle when f is. Hence, passing to the homology, γ induces an isomorphism of graded R-modules: The definition of cup product corresponds to this map, making γ∗ a ring homomorphism. (cid:3) ∗ γ∗ : RG ⊗RbH (H, R) → bH ∗ (H, RG). This proposition helps us to find and study the structure of the Tate-Hochschild cohomology ring of a finite abelian group algebra, given its Tate cohomology ring. For example, knowing the Tate cohomology of a cyclic group G, (see [8], Section XII.7), one can easily compute its Tate-Hochschild cohomology by applying Proposition 5.1. Now we return to the general case where H acts on G non-trivially and G is not necessarily abelian. Let g1, . . . , gt ∈ G be representatives of the orbits of the action of H on G. Let Hi := StabH(gi) = {h ∈ H hgi = gi} be the stabilizer of gi. For any g ∈ G, there are two R(StabH(g))-module homomorphisms: If V is any subgroup of StabH (g), then these maps induce maps on cohomology: θg : R → RG via r 7→ rg, πg : RG → R via Xa∈G (V, R) → bH g :bH (V, RG) → bH g :bH π∗ θ∗ ∗ ∗ ∗ raa 7→ rg. (V, RG), ∗ (V, R), is covariant in the second argument. The following properties of θ∗ g and π∗ g will help ∗ since dExt us in proving the main result. Lemma 5.2. Let h ∈ H and a, b ∈ G. THE TATE-HOCHSCHILD COHOMOLOGY RING OF A GROUP ALGEBRA 9 (a) If V is a subgroup of StabH(a), then h∗◦θ∗ a = θ∗ (b) Suppose V ⊆ StabH(a) ∩ StabH (b) and α, β ∈ bH θ∗ a(α) ⌣ θ∗ (c) Suppose V ′ ⊆ V ⊆ StabH(a). Then θ∗ (d) If V ⊆ StabH (a) ∩ StabH (b), then π∗ b (β) = θ∗ a and π∗ a ◦ θ∗ and δa,b is the Kronecker delta. ∗ ha◦h∗ as maps from bH (V ). Then: ∗ (hV, RG). ∗ (V ) to bH ab(α ⌣ β). a commute with resV V ′ and corV V ′. b = δa,b1, where 1 is the identity map on bH ∗ (V ) Proof. Lemma 5.2 in [11] showed these properties in positive degrees. We generalize the proof to negative degrees and present it on the cochain level. The desired results are induced on cohomology. i (a) Let P be an RV -complete resolution of R, f ∈ HomV (Pi, R) be a cocycle representing (V ), and p ∈ Pi. Then Then on the cochain level: h∗(θaf )(p) = f (h−1p)(ha) = θha(h∗(f ))(p). an element of bH (b) Let m : RG ⊗ RG → RG be the multiplication map and Γ : P → Pb⊗P be a complete di- agonal approximation map. Let f, q ∈ HomV (P, R) represent α, β ∈ bH m ◦ ((θa ◦ f )b⊗(θb ◦ q)) ◦ Γ = m ◦ (θa ⊗ θb) ◦ (fb⊗q) ◦ Γ = θab ◦ (fb⊗q) ◦ Γ, (V ′, RG), q ∈ HomV ′(Pi, R) represent an element of bH V ′)(f )(p) = πa Xv∈V /V ′ (c) Let P be an RV -complete resolution of R which can also be regarded as an RV ′-complete resolution by restricting the action. Let f ∈ HomV ′(Pi, RG) represent an element of b (β) and the right side represents θ∗ where the left side represents θ∗ πv−1 a(f (v−1p)) (V ′), and p ∈ Pi. (V ), respectively. ab(α ⌣ β). a(α) ⌣ θ∗ bH (π∗ a corV ∗ i i (πa ◦ f )(v−1p) = (corV V ′ π∗ a)(f )(p), since v ∈ V ⊆ StabH(a), we have v−1 a = a, and V acts trivially on R. Similarly, (vf (v−1p)) =Xv =Xv q(v−1p) =Xv =Xv (θ∗ a corV V ′)(q)(p) = θa Xv∈V /V ′ θv−1 a(q(v−1p)) (θa ◦ q)(v−1p) = (corV V ′ θ∗ a)(q)(p). The other cases follow similarly by commutativity between πa, θa and the inclusion map ι : HomV (P, N ) ֒→ HomV ′(P, N ), where N = RG or R. (d) Let r ∈ R. πa(θb(r)) = πa(rb) =(r, 0, if a = b else. (cid:3) (H, RG) be defined as the composition ψi = gi. We describe the additive decomposition that is generalized from the usual cohomology For all i = 1, 2, . . . , t, let ψi : bH corH Hi ([6], Theorem 2.11.2): ◦θ∗ (Hi, R) → bH ∗ ∗ 10 VAN C. NGUYEN ∗ ∗ ∗ ∗ (H, RG). (H, RG). (ζ))i, is an gi ◦ resH Hi (Hi, R), sending ζ 7→ (π∗ Proof. For i = 1, 2, . . . , t, let Mi be the free R-module generated by elements of the orbit = RH ⊗RHi R given (H, RG) → Mi bH Lemma 5.3. The map bH isomorphism of graded R-modules, for ζ ∈ bH ψi(α) ∈bH containing gi. Then RG =Li Mi. There is an isomorphism Mi → R ↑H by r(hgi) 7→ h ⊗ r. It induces an isomorphism in cohomology bH (H, Mi) ∼=bH Since dExt is additive, bH (H, RG) ∼= LibH we have bH (H, Mi). Apply the Eckmann-Shapiro Lemma 4.1, (Hi, R). One can also check directly that the maps given in the statement of the lemma are inverses of each other by taking their compositions and applying Proposition 4.3 and Lemma 5.2 to show that their compositions are the identity maps. (cid:3) Its inverse sends α ∈ bH (H, RG) ∼=LibH (H, R ↑H Hi ). (Hi, R) to Hi ∗ ∗ ∗ ∗ ∗ ∗ ∗ Remark. If H = G acts on itself by conjugation, then Mi is the free R-module generated by the conjugacy class of gi. Mi is isomorphic to R ↑G CG(gi), where CG(gi) is the centralizer of gi. Therefore, the isomorphism in Lemma 5.3 gives an additive decomposition of the Tate-Hochschild cohomology of G as a direct sum of the Tate cohomology of the centralizers of conjugacy class representatives of G with coefficients in R: result which provides a formula for products in bH Theorem 5.5. Let α ∈bH (Hi) and β ∈ bH ψi(α) ⌣ ψj(β) =Xx∈D ψk(corHk ∗ ∗ (Hj). Then yHi V V (res ∗ y∗α ⌣ res yxHj V (yx)∗β)) where D is a set of double coset representatives for Hi\H/Hj, k = k(x) and y = y(x) are chosen to satisfy (5.4), and V = V (x) = yxHj ∩ yHi ⊆ Hk. ∗ dHH (RG, RG) ∼=Mi bH ∗ (CG(gi), R). In 1999, Siegel and Witherspoon showed that there is a product formula for the usual Hochschild cohomology of G in terms of a similar additive decomposition ([11], Theorem 5.1). We will de- (H, RG) with respect to the isomorphism in Lemma 5.3. The argument ∗ will be analogous to that in [11] but is generalized to the Tate cohomology. scribe products in bH Fix i, j ∈ {1, 2, . . . , t}. Let D be a set of double coset representatives for Hi\H/Hj. For each x ∈ D, there is a unique k = k(x) such that gk = ygi yxgj (5.4) for some y ∈ H. One can expand the action on the right hand side and get gk = y(gi xgj) showing that gk is just a representative of an orbit of the action of H on the double coset HixHj. Moreover, k is independent of the choice of double coset representative x. The set of all y satisfying (5.4) is also a double coset. To see this, let us fix y = y(x) for which (5.4) holds. Let y′ ∈ H be another element such that gk = y′ yxgj implies y′ xgj) = gk showing h ∈ Hk = StabH(gk). On the other hand, if h ∈ Hk, then let y′ = hy ∈ H, we have hygi y′xgj. Then y′ xgj). Let h = y′y−1. We have hgk = h(y(gi yxgj) = hgk = gk. Putting together, we have shown: gi xgj)) = y′ y′xgj = gk = ygi hyxgj = h(ygi xgj) = y(gi (gi (gi gi {y′ ∈ Hgk = y′ gi y′xgj} = Hky = Hky(xHj ∩ Hi) ∈ Hk\H/(xHj ∩ Hi), where the last equality follows from (5.4) and yxHj ∩ yHi ⊆ Hk. We can now prove our main (H, RG) with respect to Lemma 5.3. THE TATE-HOCHSCHILD COHOMOLOGY RING OF A GROUP ALGEBRA 11 Proof. By Lemma 5.3, ψi(α) ⌣ ψj(β) = corH = corH Hi(θ∗ Hi(θ∗ corH giα) ⌣ corH giα ⌣ resH Hi(θ∗ Hj (θ∗ Hi corH giα ⌣ corHi gj β), by definition of ψi, ψj Hj θ∗ gj β), by Prop. 4.3 (9) xHj ∩Hi res xHj xHj ∩Hi x∗θ∗ gj β), by Prop. 4.3 (12) corH Hi(corHi xHj ∩Hi (resHi xHj ∩Hi θ∗ giα ⌣ res xHj xHj ∩Hi x∗θ∗ gj β)), by Prop. 4.3 (10) corH xHj ∩Hi(resHi xHj ∩Hi θ∗ giα ⌣ res xHj xHj ∩Hi x∗θ∗ gj β), by Prop. 4.3 (5) corH xHj ∩Hi θ∗ gi xgj (resHi xHj ∩Hi α ⌣ res xHj xHj ∩Hi x∗β), by Lemma 5.2 (a)-(c) ψkπ∗ gk resH Hk (corH xHj ∩Hi θ∗ gi xgj (resHi xHj ∩Hi α ⌣ res xHj xHj ∩Hi x∗β)), by the isomorphism in Lemma 5.3 yxHj ∩ yHi V ′ gk corHk V ′ res ψkπ∗ y∗θ∗ gi xgj (resHi xHj ∩Hi α ⌣ res xHj xHj ∩Hi x∗β), by Prop. 4.3 (12), where y runs over a set of representatives for Hk\H/xHj ∩ Hi and V ′ = Hk ∩ yxHj ∩ yHi, ψk corHk V ′ π∗ gkθ∗ ygi yxgj res yxHj ∩ yHi V ′ y∗(resHi xHj ∩Hi α ⌣ res xHj xHj ∩Hi x∗β), =Xx∈D =Xx∈D =Xx∈D =Xx∈D =Xk Xx =Xk Xx,y =Xk Xx,y =Xx∈D by Lemma 5.2 (a),(c) yHi V ′ y∗α ⌣ res ψk corHk V ′ (res yxHj V ′ (yx)∗β), by Prop. 4.3 (1), (4), (6) and Lemma 5.2 (d). By Lemma 5.2 (d), the only terms that can be non-zero in the next to last step are those for which gk = ygi yxgj. We have seen in the discussion prior to this theorem that each x determines a unique k and double coset Hky(xHj ∩Hi) for which this holds. Therefore, we may take y = y(x) and yxHj ∩ yHi ⊆ Hk. Hence, V ′ = V = yxHj ∩ yHi. (cid:3) Remark. Since the cup product is well-defined and unique ([8], Theorem XII.5.1), the sum in the statement of the theorem is independent of the choice of x and y. One can see this directly by replacing y with hy, for some h ∈ Hk. By Prop. 4.3 (6), (7), and (11), h∗ respects the cup product and commutes with the restriction and corestriction maps. Moreover, since Hk acts trivially on its own cohomology, any term of the sum in the theorem is unchanged by replacing y with hy. If x is multiplied on the right by an element of Hj, the terms are unchanged for similar reasons. If x is replaced by hx, for some h ∈ Hi, then we must replace y with yh−1 so that (5.4) holds: (yh−1)gi (yh−1)(hx)gj = ygi yxgj = gk, and the terms remain unchanged. (H, RG) is an algebra monomorphism that is induced by the algebra homomorphism R → RG mapping r 7→ r1. Alternatively, by letting We observe that when i = 1, ψ1 : bH (H, R) → bH ∗ ∗ 12 VAN C. NGUYEN i = j = 1 in Theorem 5.5, we see that ψ1 respects the cup product: ψ1(α) ⌣ ψ1(β) = ψ1(α ⌣ β), ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ (Hi) bH (H, RG) (H)-modules: (H)-module via restriction. As a consequence, we obtain: (H)-module with (Hi) may also be regarded as an (H). Hence, via ψ1, we may view bH (H)). Each bH ∼=−→Mi bH (H, RG) as a (left) bH where α, β ∈ bH action via multiplying (on the left) by ψ1(bH bH Corollary 5.6. The isomorphism in Lemma 5.3 is an isomorphism of graded bH Proof. For i = 1, let α ∈ bH where the left hand side is considered as action of bH (H) on bH (H) on each bH action of bH orem 5.5 gives a formula for the multiplicative structure of dHH As noted in the remark following Lemma 5.3, when H = G acts on itself by conjugation, The- (CG(gi), R) in terms of this decomposition. (RG, RG) to products within the Tate cohomology rings of certain subgroups of G. In the next section, we will show a basic non-abelian example that demonstrates the usefulness of Theorem 5.5. (RG, RG) ∼=LibH It reduces the computation of products in dHH (H, RG) that corresponds to the (Hj) on the right hand side, via the isomorphism in Lemma 5.3. (cid:3) (H) and β ∈ bH (Hj). Theorem 5.5 reduces to: ψ1(α) ⌣ ψj(β) = ψj(resH Hj (α) ⌣ β). ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ 6. The symmetric group on three elements ∗ ∗ ∗ ∗ ∗ (Hi) := bH product formula given in Theorem 5.5. (N ) is of the form Λ(w1)⊗k k[w2, w−1 Λ(w1) is the exterior k-algebra on the element w1 of degree 1 and k[w2, w−1 the elements w2 of degree 2 and w−1 and w2w−1 conjugation. Without loss of generality, we choose conjugacy class representatives g1 = 1, g2 = a, and g3 = b whose centralizers are H1 = G, H2 = {1, a, a2} =: N , and H3 = {1, b}, respectively. (Hi, k) and the (Hi, k). Since the characteristic of k is 3 and N is cyclic (N ) is periodic by ([8], Theorem XII.11.6) and ([4], (4.1.3)). 2 ], where 2 ] is generated by 2 of degree −2, subject to the graded-commutative relations 2 = 1. By a similar computation, because the characteristic of k does not divide the Let k be a field of characteristic 3. Let G = S3 =(cid:10)a, b a3 = 1 = b2, ab = ba2(cid:11) act on itself by We will find the Tate-Hochschild cohomology ring of kG using elements of bH Let us examine each ringbH of order 3, the cohomology ring bH By direct computation from Section XII.7 in [8],bH order of H3 = {1, b}, we find that bH follows from ([8], Theorem XII.11.6) and ([4], (4.1.3)) that the Tate cohomology ring bH periodic and Noetherian. One can directly compute bH following the discussion in ([8], Section XII.10), we see that for any G-module M , bH We now compute the Tate cohomology ring of G = S3 with coefficients in k. N is a normal It is easy to check that G It (G) is (G) by using an N -complete resolution of k, imposing on it an action of Z2 to make it become a G-complete resolution of k, computing the Tate cohomology groups from that resolution, and studying their products. Alternatively, (G, M ) is subgroup of G. The quotient group G/N is (isomorphic to) Z2. is isomorphic to a semidirect product N ⋊ Z2 and every abelian subgroup of G is cyclic. (H3) = 0 in all degrees. As a result, bH (H3) = 0. n ∗ ∗ ∗ ∗ THE TATE-HOCHSCHILD COHOMOLOGY RING OF A GROUP ALGEBRA 13 ∗ ∗ ∗ ∗ ∗ ∗ ∗ 2, w−2 ∼= Λ(w1w2) ⊗k k[w2 (G, M, p), where bH (G, k, 3) ∼=hbH (G, M ) and p runs through all the prime divisors of G = 6. Here, M = k is a field of characteristic 3, so (N ) (G) is of the form Λ(x) ⊗k k[z, z−1], where x and z are of degrees 3 and 4, respectively, subject to the graded-commutative relations. (G, M, p) is the p-primary component of bH a direct sum of bH only the 3-primary component is non-zero. By ([8], Theorem XII.10.1), G/N operates on bH (N )iG/N (G) = bH and so bH 2 ]. Therefore, bH By the decomposition Lemma 5.3, bH (G) ⊕bH is an algebra monomorphism, we may identify any element of bH elements. Then the Tate-Hochschild cohomology dHH (N ) as graded k-modules. We then can define elements of the Tate-Hochschild cohomology ring of kG as follows. Since ψ1 (G) with its image under ψ1. 2 ). For simplification, we will use Theorem 6.1. Let k be a field of characteristic 3 and S3 be the symmetric group on three (kS3, kS3) of S3 is generated as an algebra of degrees 3, 4, −4, 0, 1, 2, and −2, respectively, subject Let Ei = ψi(1), Wi = ψ2(wi), for i = 1, 2, and W −1 C := E2 + 1 in the following theorem. by elements x, z, z−1, C, W1, W2, and W −1 to the following relations: (G, kG) ∼= bH 2 = ψ2(w−1 2 ∗ ∗ ∗ ∗ ∗ ∗ xW1 = 0, xW2 = zW1, z−1W1 = (xz−1)W −1 2 , W 2 2 = zC, W −2 C 2 = CW −1 2 = z−1C, W1W2 = xC, W1W −1 2 = CWi = 0 (i = 1, 2), 2 = xz−1C, together with the graded-commutative relations. In particular, the algebra monomorphism ψ1 : (S3, kS3) induces an isomorphism modulo radicals. generated by E2, W1, W2 and W −1 5.5. Moreover, we will check that these generators satisfy the following conditions: . This follows from the discussion after the proof of Theorem (G, kG) is a graded-commutative k-algebra whose underlying k- (kG, kG) ∗ (N ). Here, bH (G) is a graded subalgebra of dHH (G)-submodule ofdHH (N )) is a gradedbH ∗ ∗ ∗ ∗ ∗ ∗ (kG, kG) (G)-module with action via resG N : x · w1 = w1w2w1 = w2 x · w2 = w1w2w2 = (−1)2w2w1w2 = (−1)2w2w2w1 = z · w1, x · w−1 1w2 = 0, 2 = w1w2w−1 2 = w1, ∗ ∗ ∗ ∗ ∗ ∗ ∗ 2 (G) ⊕bH (N )) as an bH (S3) → bH bH Proof. dHH (kG, kG) ∼= bH module is isomorphic to bH generated by x, z and z−1. Additionally, ψ2(bH (1) action on ψ2(bH (2) every product in ψ2(bH bH Therefore, it is clear thatdHH N : bH (N ) is an bH check that bH (G) → bH 2 ∗ ∗ (G)-module, and ∗ ∗ (G)-linear combination of the images under ψ2 of the generators of bH (N )) can be expressed as the sum of an element of bH (kG, kG) is generated as a k-algebra by x, z, z−1, E2, W1, W2, and W −1 , subject to these conditions. The first line of the relations in the statement of the theorem satisfies the first condition. The second and third lines satisfy the second condition. We will check each of them in detail. ∗ 2, and z−1 7→ w−2 2 , is injective. We also observe that by graded-commutativity of the Tate cohomology ring, every element of odd degree has square 0. In particular, w1w1 = −w1w1 implies w2 1 = 0. One can (N ), which sends x 7→ w1w2, z 7→ w2 The restriction resG (G) and a (N ). ∗ ∗ ∗ 14 VAN C. NGUYEN 2 = w2, z · w2 = w2 z · w−1 2 = w2 z−1 · w1 = w−1 z−1 · w2 = w−1 z−1 · w−1 2w2 = w3 2, 2w−1 2 w−1 2 w−1 2 )3. 2 = (w−1 2 w1 = (−1)−2w−1 2 w2 = w−1 2 , 2 w1w−1 2 = (−1)−2w1w−1 2 w−1 2 = (xz−1) · w−1 2 , ∗ ∗ Therefore, as anbH (G)-module,bH x · w1 = 0, x · w2 = z · w1, and z−1 · w1 = (xz−1) · w−1 mapping through ψ2, we obtain the first line of the relations. (N ) is generated by 1, w1, w2 and w−1 2 , subject to the relation 2 . By the isomorphism in Lemma 5.3 and To check the second and third lines of the relations, we recall the fact that the submodule of invariants (kG)G is the center Z(kG) of the group algebra kG, which is generated by conjugacy class representatives of G. Therefore, we may identify the degree-0 Tate-Hochschild cohomology (G, kG) is a quotient of H0(G, kG). Under this identification, Ei corresponds to (a quotient of) the sum of the group elements conjugate to gi. In particular, with a quotient of Z(kG), asdHH (kG, kG) ∼= bH 0 0 E2 2 = (a + a−1)2 = a2 + 2 + a−2 = a−1 − 1 + a = E2 − 1 in characteristic 3, which implies C 2 = (E2 + 1)2 = E2 2 + 2E2 + 1 = 3E2 = 0. For the rest of the relations, we use the product formula in Theorem 5.5. Let α and β be ψ2(α) ⌣ ψ2(β) = ψ2(b∗(αβ)) + ψ1(corG N (αb∗(β))). (N ). By checking on the definition of b∗ and the degrees and b∗(wi) = −wi, for i = 1, 2. Moreover, as there are ∗ (G), we have corG 2 ) = 0. Similarly, by checking on the cochain level and using Lemma 4.3 (10), for all n ∈ Z, we obtain: (N ) → bH N (w2) = corG N (w1) = corG N (w−1 ∗ ∗ (N ), we have: elements of bH Recall that b∗ :bH (bN ) = bH no degree-1, 2 and −2 elements in bH of wi, we see that b∗(w−1 2 ) = −w−1 2 ∗ ∗ corG N (wn Hence, using Lemma 4.3 (10) again, −zn/2, 2 ) =(0, 2 ) =(−xz(n−1)/2, 0, n is odd n is even. n is odd n is even. corG N (w1wn Let α = 1 and β = w1, using the product formula in Theorem 5.5, we obtain: ψ2(1) ⌣ ψ2(w1) = ψ2(b∗(w1)) + ψ1(corG N (b∗(w1))) = ψ2(−w1) + 0 = −W1. So CW1 = (E2 + 1)W1 = E2W1 + W1 = −W1 + W1 = 0. Similarly, let α = 1 and β = w2 or 2 , we show that CW2 = 0 = CW −1 w−1 Let α = β = w2, we have: . This proves the second line of the relations. 2 W 2 2 = ψ2(w2) ⌣ ψ2(w2) = ψ2(b∗(w2 = ψ2(resG = z ⌣ ψ2(1) + z N z ⌣ 1) + ψ1(z) 2)) + ψ1(corG N (w2b∗(w2))) THE TATE-HOCHSCHILD COHOMOLOGY RING OF A GROUP ALGEBRA 15 Similarly, for α = β = w−1 2 , we acquire that W −2 Let α = w1 and β = w−1 2 : = zE2 + z = zC. 2 = z−1C. W1W −1 2 = ψ2(w1) ⌣ ψ2(w−1 2 ) = ψ2(b∗(w1w−1 2 )) + ψ1(corG N (w1b∗(w−1 2 ))) N xz−1 ⌣ 1) + ψ1(xz−1) = ψ2(resG = xz−1 ⌣ ψ2(1) + xz−1 = xz−1E2 + xz−1 = xz−1C. ∗ ∗ C 2 = 0 = W 2 This implies that C, W1, W2, and W −1 2 = W 2 1 , W 3 Using the same argument, for α = w1 and β = w2, we obtain W1W2 = xC. Thus, we have (kG, kG). Furthermore, because the ring (G, kG) is graded-commutative, its nilpotent elements all lie in its radical. We observe that 2 = 0. are contained in the radical of the Tate-Hochschild (kG, kG). Consequently, modulo radicals, the algebra monomorphism (cid:3) found all necessary relations for the generators ofdHH bH cohomology ring dHH ψ1 :bH (S3) →dHH 2 W2 = zCW2 = 0, and (W −1 (kS3, kS3) induces an isomorphism. 2 W −1 2 = z−1CW −1 )3 = W −2 References 2 2 ∗ ∗ ∗ [1] Luchezar L. Avramov and Alex Martsinkovsky, "Absolute, relative, and Tate cohomology of modules of finite Gorenstein dimension," Proc. London Math. Soc. (3) 85 (2002), 393-440. [2] D. J. Benson and J. F. Carlson, "Products in negative cohomology," J. Pure & Applied Algebra, 82 (1992), 107-129. [3] P. A. Bergh and D. A. Jorgensen, "Tate-Hochschild homology and cohomology of Frobenius algebras," to appear in Journal of Noncommutative Geometry, (2011). [4] D. J. Benson, "Commutative Algebra in the Cohomology of Groups," Trends in Commutative Algebra, MSRI Publications, 51 (2004), 1-50. [5] D. J. Benson, Representations and Cohomology I: Basic representation theory of finite groups and associative algebras, Cambridge Studies in Advanced Mathematics, Cambridge University Press, 30 (1991). [6] D. J. Benson, Representations and Cohomology II: Cohomology of groups and modules, Cambridge Studies in Advanced Mathematics, Cambridge University Press, 31 (1991). [7] K. S. Brown, Cohomology of Groups, Graduate texts in mathematics, 87 Springer-Verlag, 1982. [8] H. Cartan and S. Eilenberg, Homological Algebra, Princeton University Press, Princeton, NJ, 1956. [9] C. Cibils and A. Solotar, "Hochschild cohomology algebra of abelian groups," Arch. Math, 68 (1997), 17-21. [10] V. C. Nguyen, "Tate and Tate-Hochschild Cohomology for finite dimensional Hopf Algebras," (submitted) arXiv:1209.4888. [11] S. F. Siegel and S. J. Witherspoon, "The Hochschild cohomology ring of a group algebra," Proc. London Math. Soc. 79 (1999), 131-157. [12] J. Tate, "The higher dimensional cohomology groups of class field theory," Ann. of Math. (2) 56 (1952), 294-297. Department of Mathematics, Texas A&M University, College Station, TX 77843 E-mail address: [email protected]
1112.1877	1	1112	2011-12-08T16:48:54	Clifford algebras of $p$-central sets	[ "math.RA" ]	A generalization of the term "generalized Clifford algebras" (as appears in papers on advances in applied Clifford algebras) is introduced. This algebra is studied by means of structure theory of central simple algebras. A graph theoretical approach is proposed for studying the generating set of this algebra in case where the prime number under discussion is three. Finally, it is shown how to obtain solutions in to the equation $\alpha Y^3=\alpha X_1^3+\beta X_2^3+\alpha^2 \beta^2 X_3^3$ in $\mathbb{Z}[\rho]$ where $\rho$ is the primitive third root of unity.	math.RA	math	CLIFFORD ALGEBRAS OF P -CENTRAL SETS ADAM CHAPMAN∗ c e D 8 ] . A R h t a m [ 1 v 7 7 8 1 . 2 1 1 1 : v i X r a Abstract. A generalization of the term "generalized Clifford algebras" (as appears in papers on advances in applied Clifford algebras) is introduced. This algebra is studied by means of struc- ture theory of central simple algebras. A graph theoretical approach is proposed for studying the generating set of this algebra in case where the prime number under discussion is three. Finally, it is shown how to obtain solutions in to the equation αY 3 = αX 3 3 in Z[ρ] where ρ is the primitive third root of unity. 2 + α2β2X 3 1 + βX 3 Key words. Clifford algebras, Diophantine equations AMS subject classifications. 15A66, 16K20, 11D25 1. Introduction. 1.1. The Clifford algebra of a p-central set. Let F be a field containing a primitive pth root of unity ρ, and A be an associative F -algebra. A finite subset B = b1, . . . , bn ⊆ A consisting of invertible elements is called a p-central set1 if 1. For any 1 ≤ k ≤ n, bp 2. For any 1 ≤ m < k ≤ n, bmbk = ρdm,kbkbm The Clifford algebra of B, denoted by C(B), k = αk ∈ F . is defined to be F [x1, . . . , xn : xp k = αp k, xmxk = ρcm,k xkxm. This generalizes the definition of a generalized Clifford algebra as appears in papers on advances in applied Clifford algebras, where all dm,k-s are equal to one. (See for example [4]) The subalgebra F [b1, . . . , bn] is obviously a homomorphic image of C(B). Since the Clifford algebra is depended only on the choice of α-s and c-s, we can define the Clifford algebra according to these elements instead of referring to a specific p-central set B. In [6, Vol. II, pp. 248-251] it is explained that F [b1, . . . , bn] decomposes as a tensor product of p-cyclic algebras over some extension of F . Furthermore, a way of how to obtain this decomposition is suggested: One should change the p-generating set by replacing some bi with bibk j (for some 1 ≤ k ≤ p − 1) until he receives a set ∗Department of Mathematics, Bar Ilan University, Ramat-Gan, Israel ([email protected]). Ph.D student under the supervision of Prof. Uzi Vishne 1This term was introduced by Rowen in [6, Vol II, pp. 248-251] 1 2 of "best possible form", i.e. a set whose center is the biggest possible, say m. Then this algebra decomposes as a tensor product of n−m p-cyclic algebras and the field extension generated by the center of the generating set of best possible form. 2 It does not say, however, how can one obtain this p-central set of best possible form. In Section 2, a detailed arithmetical method is given for completely analyzing the structure of C(B) (and consequently the structure of F [b1, . . . , bn]). 1.2. Coherent p-sets. A subspace V = F x1 +. . .+F xn ⊂ A is called p-central if each v ∈ V satisfies vp ∈ F . If its basis {x1, . . . , xn} forms a p-central set then we call it a coherent p-central set. Remark 1.1. It should be mentioned that if {x1, . . . , xn} is a coherent p-central set and xi commutes with xj then F xi = F xj. Consequently, if this set spans a p-central space of dimension n then it contains no commuting pair of elements. Question 1.2. When is a p-central set coherent? For p = 2 it is always true, i.e. every 2-central set is coherent. (See [5] for further details) In order to answer this question we turn to graph theory. A directed graph is a pair of sets (V, E). V is the set of vertices and E is the set of edges. Every edge is an ordered pair (a, b) where a, b ∈ V . A cycle is a sequence of vertices a1, a2, . . . , an with no repetitions such that (a1, a2), . . . , (an−1, an), (an, a1) ∈ V . Later in this paper will be shown how coherent 3-central sets have unique graph- ical form. 1.3. Solving a Diophantine equation. In [1] it is shown how Clifford algebras 1 + βX 2 2 . can be used for solving quadratic Diophantine equations such as αY 2 = αX 2 Here we show how a similar technique provides solutions to the cubic Diophantine equation αY 3 = αX 3 3 in Z[ρ] (for p = 3). 1 + βX 3 2 + α2β2X 3 For the case of α = β = 1, these solutions are completely distinct from the known integral solutions, such as Euler's (See [3, Vol. II, pp. 550-561]). 2. The structure of the Clifford algebra of a p-central set. Let C = F [x1, . . . , xn : xp In particular, B = {x1, . . . , xn} is a p-central set and it generates C therefore we shall call it a generating p-central k, xmxk = ρcm,k xkxm]. k = αp 3 . . .  set. We build a matrix as follows: MB =   ∈ Mn(Z/pZ) and  a vector v = [α1, . . . , αn]. As a matrix in Mn(Z/pZ), MB is skew-symmetric, and therefore similar to some block matrix B = (⊕m k=1H) ⊕ 0(n−2m)×(n−2m) where H = 0 (cid:21). Consequently there exists an orthogonal matrix D ∈ Mn(Z/pZ) for which (cid:20) 0 −1 c1,n ... cn,n c1,1 ... cn,1 . . . . . . 1 M ′ = DM Dt. Without changing the algebra C, we can change its generating p-central set B = {x1, . . . , xn} by replacing each xi with xdi,1 and obtain an altered generating p-central set B ′. This change affects the matrix MB who is in turn replaced with MB ′ = DMBDt, where D = (di,j ). 1 xdi,2 . . . xdi,n n 2 change of generating p-central sets {x1, . . . , xn} → {xd1,1 On the other hand, any change of matrices MB → DMBDt can be obtained by a . . . xdn,n }. In particular, because there exists an orthogonal matrix D such that M ′ = is the matrix related to the generating p-central set B ′ = . . . xdn,n , . . . , xdn,1 , . . . , xdn,1 . . . xd1,n DMBDt, M ′ = MB ′ {xd1,1 . . . xd1,n n n n 1 1 1 1 n }. . . . xdi,n 2 Denoting yi = xdi,1 1 xdi,2 for each 1 ≤ i ≤ n, the algebra C is isomorphic to F [y1, y2]⊗ . . .⊗ F [y2m−1, y2m] ⊗ F [y2m+1, . . . , yn]. For each 1 ≤ k ≤ m, F [y2k−1, y2k] is the p-cyclic algebra (αd2k−1,1 )p,F , and F [y2m+1, . . . , yn : i = αdi,1 yp n ∀2m + 1 ≤ i ≤ n] is the commutative ring (maybe a field, but not necessarily). . . . αd2k−1,n . . . αd2k,n . . . αdi,n , αd2k,1 n n n 1 1 1 The algebra C is obviously Azumaya and each simple homomorphic image of it is of degree pm and exponent either p or 1. 3. Coherent 3-central sets. Let A be a central simple algebra over the field F containing a primitive 3rd root of unity ρ. All the discussion below takes place in this algebra. According to [2, Corollary 2.2], a set {x1, . . . , xn} spans a 3-central spaces if and only if every subset of cardinality three {xi, xj, xk} spans a 3-central space. Therefore we will start with the set of cardinality 3. Let {x, y, z} span a 3-central space such that x ∗ y ∗ z = 0. Now, y = y1 + y2 and z = z1 + z2 such that yix = ρixyi and zix = ρixzi. Consequently x ∗ y ∗ z = x((1 − ρ2)(y1z2 + z1y2) + (1 − ρ)(y2z1 + z2y1)) = 0, which means that y1z2 + z1y2 = ρ(y2z1 + z2y1). If {x, y, z} is also a 3-central set then either y1 = 0 or y2 = 0 and either z1 = 0 4 or z2 = 0. Henceforth (up to change of order of the elements) yx = ρxy, zy = ρyz and zx = ρxz. Question 3.1. What happens if we do not require x ∗ y ∗ z = 0? For a 3-central set {x, y, z} spanning a 3-central space there are two options: 1. x ∗ y ∗ z = 0 which is the case we already dealt with. 2. x ∗ y ∗ z 6= 0, but still x ∗ y ∗ z ∈ F . Because {x, y, z} is a 3-central set, x∗ y ∗ z = axyz where a ∈ F . Consequently z ∈ F x−1y−1. Since multiplying an element by a central element does not change any of the characteristic we are dealing with here, we shall say that in this case z = x−1y−1. Before we proceed, there are two relevant matrices to each 3-central set: Given a 3-central set B = {x1, . . . , xn}, there is the matrix MB as in Section 2. This matrix has an underlying directed graph (B, EB) such that (xi, xj ) ∈ E ⇔ ai,j = 1. Disregarding the orientation, this graph is complete. Every cycle of length 3 in this graph is of Case (2), i.e. three elements whose product is central. Proposition 3.2. Let (xi, xj, xk) be a cycle, then for every xl outside this cycle, either (xl, xi), (xl, xj ), (xl, xk) ∈ EB or (xi, xl), (xj , xl), (xk, xl) ∈ EB . Proof. If (xl, xi), (xj , xl) ∈ EB then (xl, xi, xj ) is a cycle, and so xl = xk (up to multiplication by a central element), which is not true. Consequently if (xl, xi) ∈ EB then also (xl, xj) ∈ EB and inductively also (xl, xk) ∈ EB. Similarly, if (xj , xl) ∈ EB then also (xi, xl), (xk, xl) ∈ EB. Proposition 3.3. No cycle shares a vertex with another. Proof. Let (xi, xj, xk) and (xi, xr, xs) be two cycles. If x = r then k = s which makes them the same. So, we assume that j 6= r and k 6= s. If (xj , xs) ∈ EB then (xi, xj , xs) is a cycle, and then s = k, i.e. a contradiction. Similarly we will have a contradiction if (xs, xj) ∈ EB. Proposition 3.4. There are no cycles of length greater than 3. Proof. Let x1, . . . , xr be a cycle of length r ≥ 4. Let i be the maximal index for which (x1, xi) ∈ EB. It exists because (x1, x2) ∈ EB and it is no more than r − 1 because (xr, x1) ∈ EB. Now, (xi+1, x1) ∈ EB. Therefore (x1, xi, xi+1) is a cycle. If i ≥ 3 then xi−1 is outside this cycle, therefore (xi−1, x1) ∈ EB. Let j be the minimal index for which (xj+1, x1) ∈ EB. j 6= i, and still (x1, xj ) ∈ EB. Therefore we have another cycle (x1, xj, xj+1) which shares a vertex with the previous one, and that is impossible. If i = 2 then x4 is outside of the circle and since (x3, x4) ∈ EB, we have 5 (x1, x4) ∈ EB which contradicts the maximality of i. Given (B, EB) we can diminish it as follows: Of each cycle (xi, xj, xk), we omit one of the vertices arbitrarily from B and its related edges from EB. Finally we are left with a graph (B, EB) with no cycles. Now, the algebra generated by B is isomorphic to a tensor product of ⌊ ♯B 2 ⌋, and since the algebra generated by B is the same as the algebra generated by B, we obtain the following easy result: Corollary 3.5. The maximal coherent 3-central set in a tensor product of m cyclic algebras of degree 3 is of cardinality 3m + 1. 4. Solving a Diophantine equation. Let A be a central simple Q[ρ]-algebra containing two 3-central elements x, y ⊆ satisfying yx = ρxy, x3 = α and y3 = β, where α, β ∈ Z[√3, i]. (ax + by)x(ax + by)−1 = (a3α + b3β)−1(ax + by)x(a2x2 − ρ−1abxy + b2y2) = (a3α + b3β)−1((a3α + ρb3β)x + (1 − ρ−1)ba2αy + (1 − ρ)ab2x2y2) The set {x, y, x2y2} is a coherent 3-central set. In particular x ∗ y ∗ (x2y2) = xyx2y2 + x3y3 + yx3y2 + yx2y2x + x2y2xy + x2y3x = −3ραβ. Consequently, α = ((ax + by)x(ax + by)−1)3 = (a3α + b3β)−3((a3α + ρb3β)3α + (1 − ρ−1)3a6b3α3β + (1 − ρ)3a3b6α2β2 − 9ρa3b3α(a3α + ρb3β)αβ), which means that α(a3α + b3β)3 = α(a3α + ρb3β)3 + 3(1 − ρ)β(a2bα)3 + 3(1 − ρ−1)α2β2(ab2)3 This means that for any a, b ∈ O, Y = a3α + b3β, X1 = a3α + ρb3β, X2 = a2bα and X3 = ab2 are solutions in Z[ρ] (the ring of integers of the field Q[ρ]) to the equation αY 3 = αX 3 1 + 3(1 − ρ)βX 2 2 + 3(1 − ρ−1)α2β2X 2 3 . Since α can be replaced with 3(1 − ρ)γ we actually have solutions in Z[ρ] to any equation of the form γY 3 = γX 3 3 . Now, 27 = 33, and so we have solutions to the equation γY 3 = γX 3 3 for any γ and β. In this case the solution will be Y = 3(1−ρ)a3γ +b3β, X1 = 3(1−ρ)a3γ +ρb3β, X2 = 3(1−ρ)a2bγ and X3 = 3ab2. 2 + 27γ2β2X 3 2 + γ2β2X 3 1 + βX 3 1 + βX 3 By replacing b with (1− ρ)c our solutions to γY 3 = γX 3 2 + γ2β2X 3 1 = a3γ − ρ2βc3, X2 = γ(1 − ρ)a2c and X3 = (1 − ρ)ac2. Y = a3γ − ρβc3, X 3 1 + βX 3 3 become 6 REFERENCES [1] G. Arag´on-Gonz´alez and J. L. Arag´on and M. A. Rodr´ıguez-Andrade. Solving some quadratic Diophantine equations with Clifford algebra. Adv. Appl. Clifford Algebr., 21 no. 2:259-272, 2011. [2] A. Chapman and U. Vishne. Clifford Algebras of Binary Homogenous Forms submitted to J. of Alg. [3] L. E. Dickson. History of the Theory of Numbers Carnegie Institution of Washington, Wahs- ington, 1919 [4] C. Ko¸c. C-lattices and decompositions of generalized Clifford algebras. Adv. Appl. Clifford Algebr., 20 no. 2:313-320, 2010. [5] T.Y. Lam The Algebraic Theory of Quadratic Forms W. A. Benjamin, Inc., 1973 [6] L. Rowen. Ring Theory. Academic Press, New York, 1988
1807.07642	1	1807	2018-07-17T13:16:22	Explicit inverse of nonsingular Jacobi matrices	[ "math.RA", "math.CO", "math.NA" ]	We present here the necessary and sufficient conditions for the invertibility of tridiagonal matrices, commonly named Jacobi matrices, and explicitly compute their inverse. The techniques we use are related with the solution of Sturm-Liouville boundary value problems associated to second order linear difference equations. These boundary value problems can be expressed throughout a discrete Schr\"odinger operator and their solutions can be computed using recent advances in the study of linear difference equations. The conditions that ensure the uniqueness solution of the boundary value problem lead us to the invertibility conditions for the matrix, whereas the solutions of the boundary value problems provides the entries of the inverse matrix.	math.RA	math	Explicit inverse of nonsingular Jacobi matrices A.M. Encinas and M.J. Jim´enez Departament de Matem`atiques, UPC, BarcelonaTech, Spain Abstract We present here the necessary and sufficient conditions for the invertibility of tridiagonal matrices, commonly named Jacobi matrices, and explicitly compute their inverse. The techniques we use are related with the solution of Sturm– Liouville boundary value problems associated to second order linear difference equations. These boundary value problems can be expressed throughout a dis- crete Schrodinger operator and their solutions can be computed using recent advances in the study of linear difference equations. The conditions that ensure the uniqueness solution of the boundary value problem lead us to the invert- ibility conditions for the matrix, whereas the solutions of the boundary value problems provides the entries of the inverse matrix. Keywords: tridiagonal matrices, second order linear difference equations, Sturm–Liouville boundary value problems, discrete Schrodinger operator, Chebyshev functions and polinomyals 2010 MSC: 15B99, 31E05, 39A06 1. Preliminaries If we consider n ∈ N\ {0}, the set Mn(R) of real matrices of size n× n, and the sequences a = {a(k)}n+1 k=0 ⊂ R, then the Jacobi matrix associated with a, b and c is J(a, b, c) ∈ Mn+2(R) given k=0 ⊂ R, and c = {c(k)}n+1 k=0 ⊂ R, b = {b(k)}n+1 Preprint submitted to arXiv July 23, 2018 by J(a, b, c) =  b(0) −a(0) −c(0) 0 b(1) −a(1) −c(1) b(2) ... 0 ... 0 0 0  0 ... 0 0 0 0 ··· ··· ··· . . . 0 ... ··· b(n) ··· −c(n) 0 0 0 ...   −a(n) b(n + 1) (1) Jacobi matrices appear frequently in both general Mathematics and Applied Mathematics, see [1]. As in this reference, we have chosen to write down the coefficients outside the main diagonal with negative sign. This is only a suitable 5 convention, motivated by the existing relationship between Jacobi matrices and Schrodinger operators on a path, that we will use to analyze the invertibility of the Jacobi matrix. We must make also some assumptions about the coefficients of the matrix to avoid trivial situations or problems reducible to others with a minor order. Therefore, we will require a(k) 6= 0 and c(k) 6= 0, k = 0, . . . , n; since, in other case, J(a, b, c) is a reducible matrix and hence the inversion 10 problem leads to the invertibility of a matrix of lower size. Moreover, the values of the coefficients for the sequences a and c at n + 1 have no influence in the analysis of the matrix (1), since these coefficients do not appear in it. So, without loss of generality, we can impose a(n + 1) = c(n) and c(n + 1) = a(n). 15 In the sequel, we also assume that 00 = 1 and the usual convention that empty sums and empty products are defined as 0 and 1, respectively. The matrix J(a, b, c) is invertible if and only if for each f ∈ Rn+2 there exists k = 1, . . . , n, (2) Moreover, when this happens u is the unique solution of (2). We can rec- ognize in the previous equations the structure of a boundary value problem associated with a second order linear difference equation with coefficients a, b, c 2 u ∈ Rn+2 such that J(a, b, c)u = f; which is equivalent to b(0)u(0) − a(0)u(1) = f (0), −a(k)u(k + 1) + b(k)u(k) − c(k − 1)u(k − 1) = f (k), −c(n)u(n) + b(n + 1)u(n + 1) = f (n + 1).   and data f or, equivalently, with a Schrodinger operator Lq with potential q on the path I = {0, . . . , n + 1}. Specifically, if I = {1, . . . , n}, δ(I) = {0, n + 1} and C(I) is the vector space of real functions defined on I, the Schrodinger operator with potential q ∈ C(I) on the path I is the linear operator Lq : C(I) −→ C(I) defined as ◦ Lq(u)(0) = a(0)(cid:0)u(0) − u(1)(cid:1) + q(0)u(0), Lq(u)(k) = a(k)(cid:0)u(k) − u(k + 1)(cid:1) + c(k − 1)(cid:0)u(k) − u(k − 1)(cid:1) + q(k)u(k), Lq(u)(n + 1) = c(n)(cid:0)u(n + 1) − u(n)(cid:1) + q(n + 1)u(n + 1), ◦ I, k ∈   ◦ where q ∈ C(I) is defined as q(0) = b(0) − a(0), q(k) = b(k) − a(k) − c(k − 1), I and q(n + 1) = b(n + 1)− c(n). Identifying C(I) with Rn+2, and using this k ∈ functional notation, Equation (2) is equivalent to the equation Lq(u) = f on I; that is, to the Sturm–Liouville value problem Lq(u) = f on ◦ I, Lq(u)(0) = f (0) and Lq(u)(n + 1) = f (n + 1), (3) where the identities Lq(u) = f on δ(I) play the role of boundary conditions, ◦ ◦ whereas the equation Lq(u) = f on I is named the Schrodinger equation on I with data f . 20 Therefore, the Jacobi matrix J(a, b, c) is invertible if and only if the linear operator Lq is invertible. In terms of the boundary value problem, the invert- ibility conditions for J(a, b, c) are exactly the same conditions to ensure that the boundary value problem is regular; that is, it has a unique solution for each given data and, hence, the computation of the inverse of J(a, b, c) can be re- 25 duced to the calculus of this solution. Implicitly or explicitly, determining the solutions for initial or final value problems for the Schrodinger equation is the strategy followed to achieve the inversion of tridiagonal matrices, see for instance [2, 3, 4, 5, 6, 7, 8, 9]; but either the general case is not analyzed, the explicit expressions of these solutions are not obtained, or the expressions obtained are 30 excessively cumbersome. 3 35 40 2. Initial value problems It is well-known that every initial value problem for the Schrodinger equation ◦ I has a unique solution. Specifically, given f ∈ C(I) and m = 0, . . . , n, for on any α, β ∈ R there exists a unique u ∈ C(I) satisfying Lq(u) = f on ◦ I and u(m) = α, u(m + 1) = β. In particular, when m = n, the above problem is also known as final value problem. ◦ ◦ If S denotes the set of solutions of the homogeneous Schrodinger equation I - that is Lq(u) = 0 on I - then S is a vector space such that dimS = 2; on while for any f ∈ C(I), the set S(f ) of solutions of the Schrodinger equation on I with data f satisfies S(f ) 6= ∅ and given u ∈ S(f ), it is verified S(f ) = u + S. Given u, v ∈ C(I), their Wronskian or Casoratian, see [10], is w[u, v] ∈ C(I) ◦ defined as w[u, v](k) = det  u(k) v(k) u(k + 1) v(k + 1)   = u(k)v(k+1)−v(k)u(k+1), 0 ≤ k ≤ n, ◦ and as w[u, v](n + 1) = w[u, v](n). The Wronskian is a skew–symmetric bilinear form and and given u, v ∈ S, either w[u, v] = 0 or w[u, v] 6= 0 for any k ∈ I∪{0}. Moreover, u and v are linearly independent if and only if their Wronskian is non null and then {u, v} form a basis of S. The Green's function of the Schrodinger equation on ◦ I is the function g ∈ C(I × I), defined for any s ∈ I as g(·, s), the unique solution of the ini- tial value problem with conditions g(s, s) = 0 and g(s + 1, s) = − , when 0 ≤ s ≤ n, and as the unique solution of the initial value problem with condi- tions g(n + 1, n + 1) = 0 and g(n, n + 1) = when s = n + 1. Notice a(s) 1 1 for any s = 0, . . . , n. Therefore, if for any s = 0, . . . , n 1 that g(s, s + 1) = we consider u = g(·, s) and v = g(·, s + 1), then c(s) a(n + 1) w[u, v](s) = g(s, s)g(s + 1, s + 1) − g(s + 1, s)g(s, s + 1) = 1 a(s)c(s) 4 which implies that {g(·, s), g(·, s + 1)} is a basis of S. Moreover, for any f ∈ C(I) and m = 0, . . . , n, the function u ∈ C(I) given by u(k) = max{k,m} Xs=min{k,m}+1 g(k, s)f (s), k ∈ I is the unique solution of the initial value problem Lq(u) = f on u(m + 1) = 0. ◦ I, and u(m) = It will be very useful to introduce the companion function defined as ρ(k) = k−1 Ys=0 a(s) c(s) , k = 0, . . . , n + 1. Notice that ρ(0) = 1. 45 Remembering the assumption a(k), c(k) 6= 0, 0 ≤ k ≤ n, it is easy to prove that ρ(k)a(k) = ρ(k + 1)c(k). Moreover, the companion function verifies the following meaningful result. Proposition 2.1. Given u, v ∈ C(I), then a(k)w[u, v](k) = c(k − 1)w[u, v](k − 1) for any k ∈ ◦ I . Therefore, the multiplication of functions ρaw[u, v] is constant in I and is zero if and only if u and v are linearly dependent. 50 3. Regular Sturm–Liouville boundary value problems A boundary condition at 0 is a linear function c : C(I) −→ R of the form c(u) = αu(0) + βu(1) + γu(n) + δu(n + 1), and a boundary condition at n + 1 is a linear function d : C(I) −→ R of the form d(u) = αu(0)+ βu(1)+γu(n)+δu(n+1). The pair (c, d) is named Sturm–Liouville conditions if γ = δ = α = β = 0, see [11]. Therefore, defining the pair of Sturm–Liouville conditions (c1, c2) as c1(u) = Lq(u)(0) = b(0)u(0) − a(0)u(1), c2(u) = Lq(u)(n + 1) = −c(n)u(n) + b(n + 1)u(n + 1), 5 and according to Equation (3), we must consider the Sturm–Liouville boundary value problem (Lq, c1, c2); that is, for any f ∈ C(I), we should determine if there exists u ∈ C(I) such that Lq(u) = f on ◦ I, c1(u) = f (0) and c2(u) = f (n + 1). The boundary value problem (Lq, c1, c2) is called homogeneous when f = 0. We are only interested in regular problems; that is, in those boundary value problems with a unique solution. For the resolution of this sort of boundary value problems, we use the so–called resolvent kernel, see [12, Sections 2 and 55 3], and the process of determining the resolvent kernel always depends on an appropriate choice of solutions of the corresponding homogeneous Schrodinger equation. We reproduce here some of the main results of the above–mentioned work of the authors, essential for the main result developed in the next section. Therefore, for more details or to check out proofs that are not included on the 60 present section, see [12]. If g is the Green function of the Schrodinger equation on ◦ I, the value Da,b,c = c1(cid:0)g(·, 0)(cid:1)c2(cid:0)g(·, 1)(cid:1) − c2(cid:0)g(·, 0)(cid:1)c1(cid:0)g(·, 1)(cid:1) encompasses information of both the Schrodinger equation on ◦ I and the pair of boundary conditions (c1, c2). In fact, we next show that it plays a fundamental role in the analysis of the Sturm–Liouville problem. Definition 3.1. The boundary value problem (Lq, c1, c2) is called regular if the solution of the corresponding homogeneous problem is unique, and so the null 65 one. Proposition 3.2. The following assertions are equivalent: (i) The boundary value problem (Lq, c1, c2) is regular. (ii) For any f ∈ C(I) the corresponding boundary value problem has a solution 70 (and hence a unique solution). (iii) Da,b,c 6= 0. 6 Proof. If z1 = g(·, 0) and z2 = g(·, 1), then {z1, z2} form a basis of solutions of the homogeneous Schrodinger equation Lq(u) = 0 on I. If given f ∈ C(I) ◦ we consider a solution y of the Schrodinger equation with data f on I, the expression u = αz1 + βz2 + y where α, β ∈ R, determines all the solutions of the ◦ Schrodinger equation on I. Therefore, u = αz1 + βz2 + y denotes a solution of ◦ the boundary value problem Lq(u) = f on ◦ I, c1(u) = f (0) and c2(u) = f (n + 1), if and only if α and β are solutions of the linear system   c1(z1) c1(z2) c2(z1) c2(z2)     α β   =   f (0) − c1(y) f (n + 1) − c2(y) .   When f goes over C(I), then the right term of the previous system goes over the whole R2. Therefore, the system has a solution for any f ∈ C(I) if and only if the coefficient matrix is non–singular and, hence, the system has a unique solution. 75 As the homogeneous system associated with the previous one determines the solutions of the homogeneous boundary value problem, the problem is regular if the homogeneous system has as its unique solution the null one. Therefore, (i) and (ii) are equivalent and, in addition, the coefficient matrix is non–singular and it implies that its determinant is different from 0. Hence, (i) and (iii) are 80 equivalent. In the sequel, for any s ∈ I, we denote by εs ∈ C(I) the Dirac function at s. Therefore εs(s) = 1 and εs(k) = 0, when k 6= s. Definition 3.3. Let (Lq, c1, c2) be a regular boundary value problem. We call resolvent kernel of the boundary value problem to Ra,b,c : I × I −→ R charac- terized by Lq(cid:0)Ra,b,c(·, s)(cid:1) = εs on for any s ∈ I. ◦ I, c1(cid:0)Ra,b,c(·, s)(cid:1) = εs(0), c2(cid:0)Ra,b,c(·, s)(cid:1) = εs(n + 1) Notice that for any s ∈ I, Ra,b,c(·, s) is the unique solution of the Sturm- Liouville problem for the data εs and hence it makes sense when the boundary 85 7 value problem is regular. The role of the resolvent kernel is showed in the following result. Proposition 3.4. If the boundary value problem (Lq, c1, c2) is regular and Ra,b,c is the resolvent kernel, then for any f ∈ C(I) the function u(k) = Xs∈I Ra,b,c(k, s) f (s), k ∈ I, is the unique solution of the boundary value problem with data f , i.e. Lq(u) = f on ◦ I, c1(u) = f (0), c2(u) = f (n + 1). Definition 3.5. We call fundamental solutions of the homogeneous Schrodinger ◦ equation on I, related to the boundary conditions c1 and c2 or, simply, funda- mental solutions, to Φa,b,c, Ψa,b,c ∈ C(I), the unique solutions of the homoge- neous Schrodinger equation on ◦ I determined respectively by the conditions Φa,b,c(0) = a(0), Φa,b,c(1) = b(0), Ψa,b,c(n) = b(n + 1), Ψa,b,c(n + 1) = c(n). 90 Notice that Φa,b,c is the solution of a initial value problem, whereas Ψa,b,c is the solution of a final value problem. The reason to choose these definitions for the fundamental solutions is shown in the following result. Proposition 3.6. If Φa,b,c and Ψa,b,c are the fundamental solutions of the ho- mogeneous Schrodinger equation on ◦ I, related to the boundary conditions c1 and c2, then c1(Φa,b,c) = c2(Ψa,b,c) = 0, c2(Φa,b,c) = a(0)c(0)Da,b,c. Moreover, c1(Ψa,b,c) = c(0)a(n)ρ(n)Da,b,c = w[Ψa,b,c, Φa,b,c](0). Proof. Consider {u, v} the basis of solutions of the homogeneous Schrodinger equation satisfying u(0) = 1, u(1) = 0, v(0) = 0 and v(1) = 1; that is, u = c(0)g(·, 1) and v = −a(0)g(·, 0). Moreover, w[u, v](0) = 1. 95 If we prove that Φa,b,c = c1(u)v − c1(v)u and Ψa,b,c = a(0)−1a(n)ρ(n)(cid:0)c2(v)u − c2(u)v(cid:1), 8 then, clearly, c1(Φa,b,c) = c2(Ψa,b,c) = 0, c2(Φa,b,c) = a(0)c(0)Da,b,c and c1(Ψa,b,c) = a(0)−1a(n)ρ(n)c2(Φa,b,c) = c(0)a(n)ρ(n)Da,b,c. Moreover, w[Φa,b,c, Ψa,b,c](0) = a(0)−1a(n)ρ(n)(cid:0)c1(v)c2(u) − c1(u)c2(v)(cid:1) = −c(0)a(n)ρ(n)Da,b,c. To end the proof, let us consider the functions z = c1(u)v − c1(v)u z = a(0)−1a(n)ρ(n)(cid:0)c2(v)u − c2(u)v(cid:1). Then z(0) = −c1(v) = a(0), z(1) = c1(u) = b(0) and on the other hand, and z(n) = a(0)−1a(n)ρ(n)(cid:0)c2(v)u(n) − c2(u)v(n)(cid:1) = b(n + 1)a(0)−1a(n)ρ(n)w[u, v](n) = b(n + 1)a(0)−1a(0)ρ(0)w[u, v](0) = b(n + 1), z(n + 1) = a(0)−1a(n)ρ(n)(cid:0)c2(v)u(n + 1) − c2(u)v(n + 1)(cid:1) = c(n)a(0)−1a(n)ρ(n)w[u, v](n) = c(n)a(0)−1a(0)ρ(0)w[u, v](0) = c(n). The uniqueness of the solution of any initial value problem concludes that z = Φa,b,c and z = Ψa,b,c. Corollary 3.7. The boundary value problem (Lq, c1, c2) is regular if and only if the fundamental solutions are a basis of solutions of the homogeneous Schrodinger 100 equation on ◦ I. The next step in this section is to obtain the resolvent kernel for a reg- ular boundary value problem with Sturm-Liouville conditions in terms of the fundamental solutions, see [12] for its proof. Theorem 3.8. The Sturm–Liouville boundary value problem (Lq, c1, c2) is regu- lar if and only if b(0)Ψa,b,c(0) 6= a(0)Ψa,b,c(1) or, equivalently, iff c(n)Φa,b,c(n) 6= b(n + 1)Φa,b,c(n + 1) and its resolvent kernel is determined by Φa,b,c(min{k, s})Ψa,b,c(max{k, s}) a(0)(cid:2)b(0)Ψa,b,c(0) − a(0)Ψa,b,c(1)(cid:3) Ra,b,c(k, s) = ρ(s), 9 for any k, s = 0, . . . , n + 1. 105 Finally, let us remind that the boundary conditions associated with the Jacobi matrix were c1(u) = Lq(u)(0) and c2(u) = Lq(u)(n + 1), so the boundary value problem (Lq, c1, c2) associated with the inversion of that matrix is the Poisson equation Lq(u) = f on I. Applying now Theorem 3.8 to this equation, we obtain the fundamental result for the inversion of Jacobi matrices. Corollary 3.9. The Schrodinger operator Lq is invertible if and only if b(0)Ψa,b,c(0) 6= a(0)Ψa,b,c(1) and, moreover, given f ∈ C(I), (Lq)−1(f )(k) = Xs∈I Φa,b,c(min{k, s})Ψa,b,c(max{k, s}) a(0)hb(0)Ψa,b,c(0) − a(0)Ψa,b,c(1)i ρ(s)f (s), 110 for any k = 0, . . . , n + 1. 4. The inverse of a Jacobi matrix The invertibility conditions of the Jacobi matrix J(a, b, c) described in Equa- tion (1), as well as determining its inverse J−1 = R = (rij ) in terms of the solutions Φa,b,c and Ψa,b,c of the Schrodinger equation, are described in Corol- lary 3.9. So, to obtain the explicit values of the entries of R, the next step is to compute explicitly the functions Φa,b,c and Ψa,b,c, that can be seen as the solutions of an initial and a final value problem respectively, associated with the second order linear difference equation with coefficients a, b and c that cor- responds to the Schrodinger equation. To compute these solutions we will use recent advances in the study of difference equations developed by the authors in [13]. In particular, in Section 7 of this work it has been proved that the solution of any initial value problem for a second order difference equation with any data f , can be expressed as a linear combination of the functions Pk(x, y) called k-th Chebyshev functions and defined for any x, y ∈ C(Z) as ⌊ k 2 ⌋ P0(x, y) = 1, P−1(x, y) = 0 and Pk(x, y) = 10 Xm=0 (−1)m Xα∈ℓm k x ¯αyα, k ≥ 1. (4) 115 120 We reproduce here some brief explanations about the notation involved in Equa- tion (4), for the sake of completeness. The parameter α = (α1, . . . , αp) is a binary multi–index of order p; i.e. α is a p–tuple α = (α1, . . . , αp) ∈ {0, 1}p, αj ≤ p. Given α ∈ {0, 1}p and a func- and its length is defined as α = tion a ∈ C(Z), we consider the value aα = a(j)αj . Given p ∈ N \ {0}, we denote by i1, . . . , im the indices such that 1 ≤ i1 < ··· < im ≤ p and αij = 1, j = 1, . . . , m. We just need to consider the binary multi–indexes α of order p in Pj=1 Qj=1 p p the set ℓp defined as (i) ℓ0 (ii) ℓ1 p = {α : α = 0} = {(0, . . . , 0)}, for p ∈ N \ {0}, p = {α : αp = 0 and α = 1}, for p ≥ 2, p = {α : αp = 0, p ≥ 4 and m = 2, . . . ,⌊ p 2⌋. Finally, ¯α is the binary multi–index of the same order as α defined by α = m and ij+1 − ij ≥ 2, j = 1, . . . , m − 1}, for (iii) ℓm ¯αij = ¯αij +1 = 0, j = 1, . . . , m, and ¯αi = 1 otherwise. The name of Chebyshev function for (4) is justified due to its relation with the usual Chebyshev polynomials of second kind, since Pk(x, y) can be identified with them when x and y are constant sequences. In that case, P0(x, y) = 1 and ⌊ k 2 ⌋ Pk(x, y) = P−1(x, y) = 0 and moreover, since #ℓm k = (cid:0)k−m (−1)m(cid:18)k − m m (cid:1) for any k ∈ N∗, we obtain that m (cid:19)xk−2mym. Clearly, for any k ≥ −1 and any constant sequence x, we have m (cid:19)(2x)k−2m, (−1)m(cid:18)k − m Uk(x) = Pk(2x, 1) = Xm=0 ⌊ k 2 ⌋ Xm=0 that is known as the standard k–th Chebyshev polynomial of second kind, see [14] and also [11, 15]. Definitely, for constant sequences x and y, it is satisfied Pk(x, y) = y k 2 ⌊ k 2 ⌋ Xm=0 (−1)m(cid:18)k − m m (cid:19)(cid:18) x √y(cid:19)k−2m = y k 2 Uk(cid:18) x 2√y(cid:19) , k ≥ 1. 11 Now we are ready to compute the basis of solutions {Φa,b,c(k), Ψa,b,c(k)} of the homogeneous Schrodinger equation Lq(u) = 0 on I applying the results showed in [13] on second order difference equations, that is through a linear combination of the Chebyshev functions Pk(b, ac) and Pk(bm, amcm), where a, b, c ∈ C(I) are the coefficients of the second order difference equation associ- ated to the Schrodinger equation, and given a ∈ C(Z) and m ∈ N, the function am corresponds to the m–shift of a, so am = a(k + m). We must consider, for this first result and most of those that will appear from now on, the functions ΦJ , ΨJ ∈ C(I) defined as ΦJ(0) = 1, ΦJ(k) = b(0)Pk−1(b, ac) − a(0)c(0)Pk−2(b1, a1c1), k = 1 . . . , n + 1, ΨJ(k) = b(n + 1)Pn−k(bk, akck) − a(n)c(n)Pn−k−1(bk, akck), k = 0, . . . , n, ΨJ(n + 1) = 1, and the value DJ = b(0)hb(n + 1)Pn(b, ac) − a(n)c(n)Pn−1(b, ac)i − a(0)c(0)hb(n + 1)Pn−1(b1, a1c1) − a(n)c(n)Pn−2(b1, a1c1)i. Lemma 4.1. For any k = 0, . . . , n + 1, it is satisfied that Φa,b,c(k) = a(0)(cid:16) k−1 Ys=0 a(s)(cid:17)−1 ΦJ(k) and Ψa,b,c(k) = c(n)(cid:16) n Ys=k c(s)(cid:17)−1 ΨJ(k) and, moreover, b(0)Ψa,b,c(0) − a(0)Ψa,b,c(1) = DJ(cid:16) n−1 Ys=0 c(s)(cid:17)−1 . Proof. Applying [13, Theorems 4.3 and 7.4], we have that Φa,b,c is a linear combination of the Chebyshev functions {Pk−2(b, ac), Pk−1(b1, a1c1)} and, in addition, Ψa,b,c is a linear combination of {Pn−k−1(bk, akck), Pn−k(bk, akck)}. To obtain all the results, we must just to impose the conditions Φa,b,c(0) = a(0), Φa,b,c(1) = b(0), Ψa,b,c(n) = b(n + 1), Ψa,b,c(n + 1) = c(n). 12 125 Before showing the main result for the explicit inversion of a Jacobi matrix, we add a previous result extracted from [7, Theorem 3.3], that allow us to compute also the determinant of the inverse matrix. Lemma 4.2. If R = (rij ) ∈ Mm(R) is an irreducible and invertible matrix, the following statements are equivalents: (i) There exists a diagonal and invertible matrix H = (hj) such that RH−1 is a Green's matrix; that is, there exist v, w, h ∈ Rm, where hj 6= 0, j = 1, . . . , m, such that rij = hjvmin{i,j}wmax{i,j} =   viwjhj, vjhjwi; si i ≤ j, si i ≥ j, that is, R = v1 h1v1 v1 v2 v1 v2 v3 h1v1 h2v2 ... ... ... h1v1 h2v2 h3v3   ··· ··· ··· . . . ··· v1 v2 v2 ... vm   ◦   w2 w3 ... wm w2 ... wm h1w1 h2w2 h3w3 h2w2 h3w3 ··· ··· ··· . . . h3w3 ... wm ··· hnwn hmwm hmwm ... hmwm .   130 (ii) R−1 is a tridiagonal and irreducible matrix. Moreover, det R = h1v1wm m Ys=2 hs(vsws−1 − vs−1ws). Theorem 4.3. The matrix J(a, b, c) is invertible if and only if DJ 6= 0, and in that case, the entries of its inverse R are explicitly given by rks = 1 DJ Moreover, (cid:16) s−1 Qj=k (cid:16) k−1 Qj=s a(j)(cid:17)ΦJ(k)ΨJ (s), c(j)(cid:17)ΦJ(s)ΨJ(k),   if 0 ≤ k ≤ s ≤ n + 1, if 0 ≤ s ≤ k ≤ n + 1. det R = −D−1 J . 13 Proof. The first part is consequence of Corollary 3.9 and taking into account the identities from Lemma 4.1. Then, for any k, s = 0, . . . , n + 1, we obtain min{k,s}−1 ρ(s)(cid:16) n−1 Qs=0 a(s)(cid:17)(cid:16) Qs=1 c(s)(cid:17) Qs=max{k,s} n−1 = c(s)(cid:17) a(0)(cid:16) s−1 Qj=k Qj=s k−1 a(j), si k ≤ s, c(j), si k ≥ s.   To prove the formula for the determinant of R, we apply Lemma 4.2 with hj = hρ(j), h = a(0)−1D−1 J implies n−1 Qs=0 c(s), vj = Φa,b,c c1,c2(j) and wj = Ψa,b,c c1,c2(j), which hs(vsws−1 − vs−1ws) = −hρa,c(s)w[Φa,b,c c1 ,c2 , Ψa,b,c c1,c2](s − 1) = −hc(s − 1)−1a(s − 1)ρa,c(s − 1)w[Φa,b,c = −hc(s − 1)−1a(0)w[Φa,b,c = hc(s − 1)−1a(0)DJ(cid:16) n−1 Qs=0 c(s)(cid:17)−1 c1,c2 , Ψa,b,c c1,c2](0) = c(s − 1)−1, c1,c2 , Ψa,b,c c1,c2](s − 1) for any s = 1, . . . , n + 1. Therefore det R = h0v0wn+1 n+1 Ys=1 c(s − 1)−1 = −D−1 J . Although the expression of the inverse of J(a, b, c) in terms of solutions of initial and final value problems is well known, see [3, 6], the above–explained proposal has the novelty of computing such solutions explicitly. On the other 135 hand, the formula for the determinant of R appears to be new, probably because this is the first study on the inversion of matrices from an algebraic point of view, particularly based on the properties of difference equations. We end this section particularizing the last results for a Jacobi matrix J(a, b, c) with constant diagonals except for the first and the last row, that is a(j) = α 6= 0, b(j) = β, j = 1, . . . , n, and c(j) = γ 6= 0, j = 0, . . . , n−1, and also for the easiest case when J(a, b, c) is also a Toeplitz matrix, so then has the three main diagonals completely constant. In both cases, the Schrodinger equation 14 corresponds to a second order linear difference equation with constant coeffi- cients (in the first case, the first and the last row are related to the boundary conditions), so its solution can be expressed in terms of Chebyshev polynomials, a known result that can be consulted in [14, Theorem 2.4] or [15, Theorem 2.4]. Of course, this result coincides with the one showed below when we use Cheby- shev functions Pk(x, y) valued in constant sequences x(j) = x and y(j) = y 6= 0, j = 1, . . . , n, so then Equation (4) become Chebyshev polynomials of second kind, P−1(x, y) = 0, P0(x, y) = 1 and Pk(x, y) = y If we consider q = β 2√αγ , then ΦJ (0) = 1, k 2 Uk(cid:16) x 2√y(cid:17). ΦJ (k) = (√αγ)k−2hb(0)√αγ Uk−1(q) − a(0)γUk−2(q)i, ΨJ (k) = (√αγ)n−k−1hb(n + 1)√αγ Un−k(q) − c(n)αUn−k−1(q)i, k = 1 . . . , n + 1, k = 0, . . . , n, ΨJ (n + 1) = 1, and DJ = dJ(√αγ)n−2 where dJ = b(0)√αγhb(n + 1)√αγ Un(q) − c(n)αUn−1(q)i − a(0)γhb(n + 1)√αγ Un−1(q) − c(n)αUn−2(q)i. (5) (6) The next result corresponds to the first case, a Jacobi matrix with con- stant diagonals except for the first and the last row, and is a straightforward 140 consequence of Theorem 4.3 using Equations (5) and (6). Corollary 4.4. If a(j) = α 6= 0, b(j) = β, j = 1, . . . , n, c(j) = γ 6= 0, j = 0, . . . , n − 1, then J(a, b, c) is invertible if and only if dJ 6= 0, and in that 15 case the entries of its inverse R are explicitly given by rks = 1 dJ (√αγ)n−2 Moreover, a(0)αs−1ΦJ(0)ΨJ(s), αs−kΦJ(k)ΨJ(s), γk−sΦJ(s)ΨJ (k), c(n)γn−sΦJ(s)ΨJ(n + 1), si 0 = k ≤ s ≤ n + 1, si 1 ≤ k ≤ s ≤ n + 1, si 0 ≤ s ≤ k ≤ n, si 0 ≤ s ≤ k = n + 1.   det R = − 1 dJ(√αγ)n−2 . Finally, the two last results showed above correspond to Jacobi and Toeplitz matrices. Corollary 4.5. If αγ 6= 0, the Jacobi and Toeplitz matrix of size n + 2 J(α, β, γ) = 0 ··· β −α ··· ··· . . .  β −α −γ 0 −γ ... ... 0 0 β ... 0 0 0 0  0 0 0 0 0 0 ... ... β −α β   ··· ··· −γ is invertible if and only if β 6= 2√αγ cos(cid:18) kπ n + 3(cid:19) , k = 1, . . . , n + 2, and then, the entries of the inverse of J(α, β, γ) are explicitly given by αs−k(√αγ)k−s−1Uk(q)Un−s+1(q), γk−s(√αγ)s−k−1Us(q)Un−k+1(q), if 0 ≤ k ≤ s ≤ n + 1, if 0 ≤ s ≤ k ≤ n + 1, rks = Un+2(q) where q = 1   2√αγ β . Moreover, det R = 1 √αγ)(n+2)Un+2(q) . 16 Proof. All the results are consequence of Theorem 4.3 by imposing in Equa- tions (5) and (6) the identities a(0) = −α, b(0) = b(n + 1) = β, c(n) = −γ. Then, dJ = −α2γ2Un+2(q), 145 so dJ 6= 0 if and only if q is not a zero of the polymonial Un+2(x); that is, if and only if q 6= cos(cid:0) kπ n+3(cid:1), k = 1, . . . , n + 2, see [16]. Moreover, the expression for the determinant follows. On the other hand, ΦJ(k) = (√αγ)kUk(q), ΨJ(k) = −(√αγ)n−k+1Un−k+1(q) for any k = 0, . . . , n + 1, that leads to the given expressions for the inverse entries. A more detailed proof of the above result for Jacobi and Toeplitz matrices 150 can be consulted in [12]. Besides, the expression obtained for the matrix inverse of this kind of matrices coincides with that published by Fonseca and Petronilho in [2, Corollary 4.1] and [3, Equation 4.26]. Corollary 4.6. If α 6= 0, the symmetric Jacobi and Toeplitz matrix of order n + 2 J(α, β) = 0 ··· β −α ··· ··· . . .  β −α −α 0 −α ... ... 0 0 β ... 0 0 0 0  0 0 0 0 0 0 ... ... β −α β   ··· ··· −α is invertible if and only if β 6= 2α cos(cid:18) kπ n + 3(cid:19) , k = 1, . . . , n + 2, 17 and then, the entries of the inverse of J(α, β) are explicitly given by rks = Moreover, Umin{k,s}(cid:0) β 2α(cid:1)Un−max{k,s}+1(cid:0) β 2α(cid:1) αUn+2(cid:0) β 2α(cid:1) , k, s = 0, . . . , n + 1. det R = 1 α(n+2)Un+2(cid:0) β 2α(cid:1) . The expression for the inverse of a symmetric Jacobi and Toeplitz matrix is well–known, see for instance [2, Corollary 4.2] and the references of this article. 155 This work has been partly supported by the Spanish Program I+D+i (Mi- nisterio de Econom´ıa y Competitividad) under projects MTM2014-60450-R and MTM2017-85996-R. References References 160 [1] G. Meurant, A review on the inverse of symmetric tridiagonal and block tridiagonal matrices, SIAM J. Matrix Anal. Appl. 13 (1992) 707–728. doi:10.1137/0613045. [2] C. M. da Fonseca, J. Petronilho, Explicit inverse of some tridiagonal matrices, Linear Algebra Appl. 325 (2001) 7–21. 165 doi:10.1016/S0024-3795(00)00289-5. [3] C. M. da Fonseca, J. Petronilho, Explicit inverse of a tridi- agonal k–Toeplitz matrix, Numer. Math. 100 (2005) 457–482. doi:10.1007/s00211-005-0596-3. [4] M. A. El-Shehawey, G. A. El-Shreef, A. S. Al-Henawy, Analytical inversion 170 of general periodic tridiagonal matrices, J. Math. Anal. Appl. 345 (2008) 123–134. doi:10.1016/j.jmaa.2008.04.002. 18 [5] F. P. Gantmacher, M. G. Krein, Oscillation matrices and kernels and small vibrations of mechanical systems, AMS Chelsea Publishing, Providence, RI, 2002 (Translation based on the 1941 Russian original). 175 [6] R. K. Mallik, The inverse of a tridiagonal matrix, Linear Algebra Appl. 325 (2001) 109–139. doi:10.1016/S0024-3795(00)00262-7. [7] J. J. McDonald, R. Nabben, M. Neumann, H. Scheider, M. J. Tsatsomeros, Inverse tridiagonal Z-Matrices, Linear Multilinear Algebra 45 (1998) 75–97. doi:10.1080/03081089808818578. 180 [8] R. A. Usmani, Inversion of Jacobi's tridiagonal matrix, Computers Math. Applic. 27 (1994) 59–66. doi:10.1016/0898-1221(94)90066-3. [9] R. A. Usmani, Inversion of a tridiagonal Jacobi matrix, Linear Algebra Appl. 212/213 (1994) 413–414. doi:10.1016/0024-3795(94)90414-6. [10] R. P. Agarwal, Difference equations and inequalities, Marcel, 2000. 185 [11] E. Bendito, A. Carmona, A. M. Encinas, Eigenvalues, eigenfunctions and Green's functions on a path via Chebyshev polynomials, Appl. Anal. Dis- crete Math. 3 (2009) 282–302. doi:10.2298/AADM0902282B. [12] A. M. Encinas, M. J. Jim´enez, Explicit inverse of a tridiago- nal (p, r)–Toeplitz matrix, Linear Algebra Appl. 542 (2018) 402–421. 190 doi:10.1016/j.laa.2017.06.010. [13] A. M. Encinas, M. J. Jim´enez, Second order linear differ- ence equations, J. Diff. Eq. Appl. 24 (3) (2018) 305–343. doi:10.1080/10236198.2017.1408608. [14] D. Aharanov, A. Beardon, K. Driver, Fibonacci, Chebyshev and 195 orthogonal polynomials, Am. Math. Mon. 112 (2005) 612–630. doi:10.2307/30037546. 19 [15] A. M. Encinas, M. J. Jim´enez, Floquet theory for second order lin- ear difference equations, J. Diff. Eq. and App. 22 (2016) 353–375. doi:10.1080/10236198.2015.1100609. 200 [16] J. C. Mason, D. C. Handscomb, Chebyshev Polynomials, Chapman & Hall/CRC, 2003. 20
1707.03128	1	1707	2017-07-11T05:00:40	The Hilbert series and $a$-invariant of circle invariants	[ "math.RA", "math.AC", "math.CO" ]	Let $V$ be a finite-dimensional representation of the complex circle $\mathbb{C}^\times$ determined by a weight vector $\mathbf{a}\in\mathbb{Z}^n$. We study the Hilbert series $\operatorname{Hilb}_{\mathbf{a}}(t)$ of the graded algebra $\mathbb{C}[V]^{\mathbb{C}_{\mathbf{a}}^\times}$ of polynomial $\mathbb{C}^\times$-invariants in terms of the weight vector $\mathbf{a}$ of the $\mathbb{C}^\times$-action. In particular, we give explicit formulas for $\operatorname{Hilb}_{\mathbf{a}}(t)$ as well as the first four coefficients of the Laurent expansion of $\operatorname{Hilb}_{\mathbf{a}}(t)$ at $t=1$. The naive formulas for these coefficients have removable singularities when weights pairwise coincide. Identifying these cancelations, the Laurent coefficients are expressed using partial Schur polynomial that are independently symmetric in two sets of variables. We similarly give an explicit formula for the $a$-invariant of $\mathbb{C}[V]^{\mathbb{C}_{\mathbf{a}}^\times}$ in the case that this algebra is Gorenstein. As an application, we give methods to identify weight vectors with Gorenstein and non-Gorenstein invariant algebras.	math.RA	math	THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS L. EMILY COWIE, HANS-CHRISTIAN HERBIG, DANIEL HERDEN, AND CHRISTOPHER SEATON Abstract. Let V be a finite-dimensional representation of the complex circle C× determined by a weight vector a ∈ Zn. We study the Hilbert series Hilba(t) of the graded algebra C[V ]C× a of polynomial C×-invariants in terms of the weight vector a of the C×-action. In particular, we give explicit formulas for Hilba(t) as well as the first four coefficients of the Laurent expansion of Hilba(t) at t = 1. The naive formulas for these coefficients have removable singularities when weights pairwise coincide. Identifying these cancelations, the Laurent coefficients are expressed using partial Schur polynomial that are independently symmetric in two sets of variables. We similarly give an explicit formula for the a-invariant of C[V ]C× a in the case that this algebra is Gorenstein. As an application, we give methods to identify weight vectors with Gorenstein and non-Gorenstein invariant algebras. Contents Introduction 1. Acknowledgements 2. Background and notation 3. Computation of the Hilbert series 4. An algorithm to compute the Hilbert series in the generic case 5. Partial Laurent-Schur polynomials 6. The coefficients of the Laurent series 7. The a-invariant and the Gorenstein property Appendix A. The Laurent expansion of the Hilbert series of a Cohen-Macaulay algebra References 1 2 3 4 8 9 13 20 23 25 1. Introduction Let V be a finite-dimensional representation of a complex reductive group G and let R = C[V ]G denote the algebra of G-invariant polynomials. It is well known that this algebra is a finitely generated graded algebra R = ⊕∞ m=0Rm such that R0 = C. The Hilbert series of R is the generating function HilbR(t) = dimC(Rm) tm. ∞Xm=0 The Hilbert series is known to be rational with a pole at t = 1 of order dim(R), the Krull dimension of R. It therefore admits a Laurent expansion of the form (1.1) HilbR(t) = ∞Xm=0 γm(R)(1 − t)m−dim R, see [7, Proposition 1.4.5 and Lemma 1.4.6]. The a-invariant a(R) of R is defined to be the degree of HilbR(t), i.e. the degree of the numerator minus the degree of the denominator. 2010 Mathematics Subject Classification. Primary 13A50; Secondary 13H10, 05E05. Key words and phrases. Hilbert series, circle invariants, a-invariant, Gorenstein ring, Schur polynomial. C.S. was supported by the E.C. Ellett Professorship in Mathematics. 1 2 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON The Hilbert series HilbR(t) contains important imformation about the algebra R and is a relatively accessible quantity. It is used in constructive invariant theory, e.g., when computing generators and relations for R (see for example [7, Section 2.6] and [17, Chapter 2]). Similarly, the coefficients γm(R) are often of significance. For instance when G is a finite group, γ0(R) = 1/G and γ1(R) determines the number of pseudoreflections in G; see [22, Lemma 2.4.4] or [4, Sections 2.4 and 2.6]. When the coefficient field is Fp, an analogous formula for γ1(Fp[V ]G) in terms of the stabilizers of hyperplanes in V was conjectured by Carlisle and Kropholler and proven by Benson and Crawley -- Boevey [5]; see [4, Sections 2.6 and 3.13]. When R = C[V ]G is the invariant ring of a reductive group G, less is known about the meaning of the γm(R). When G = SL2(C) and V is irreducible, Hilbert gave an explicit formula for γ0(R). Popov has shown [17, 3.3 Theorem 5] that if G is connected and semisimple, then for all but finitely many representations V , 2γ1(C[V ]G)/γ0(C[V ]G) = dim G. See also [7, Remark 4.6.16] and the references given there. In this paper, we investigate the Hilbert series and first four Laurent coefficients when G is a circle C×. The techniques and results are parallel to those in [11] where Hilbert series of symplectic circle quotients have been studied. The main observation in these calculations is that the naive formulas for the Laurent coefficients have removable singularities when certain weights pairwise coincide. By construction, the expressions for the γm's are invariant with respect to permutations of the positive weights and permutations of the negative weights. After removing the singularities in the expressions, the Laurent coefficients are written in terms of a kind of partial Schur polynomial of the weights, which are independently symmetric in the positive and negative weights. These calculations lead us in particular to a neat formula for the a-invariant of Gorenstein circle invariants; see Corollary 7.1. The phenomenon of removable singularities in the formulas for Laurent coefficients appears to be a general feature of calculations of Laurent coefficients in the case of reductive G and deserves further attention. Part of the motivation for studying the Hilbert series of C[V ]C× is related to the question of which representations of C× have Gorenstein invariant rings. Because the canonical module of a ring of torus invariants is determined explicitly in terms of covariants [6, Theorem 6.4.2], it is known that C[V ]C× is Gorenstein for some but not all representations. If the representation is unimodular, then the Gorenstein property holds [6, Corollary 6.4.3], but unlike the case of finite groups that contain no pseudoreflections [24, 25] this condition is not necessary. This approach to the Gorenstein ques- tion, however, requires the computation of invariants and covariants, which can be computationally expensive. Because C[V ]C× is a normal domain by [6, Proposition 6.4.1], and the Hochster-Roberts Theorem [13, 14] implies that C[V ]C× is Cohen-Macaulay, the Gorenstein property can be determined from HilbC[V ]C× (t) alone using the characterization of Stanley [21] recalled in Equation (2.2) below. Our understanding of HilbC[V ]C× (t) gives a method of checking the Gorenstein condition without com- puting the invariants. In some cases, we can resolve this question without computing HilbC[V ]C× (t), see Corollaries 7.4, 7.6, and 7.7. The contents of this paper are as follows. After reviewing the necessary background and establishing notation in Section 2, we compute the Hilbert series in Section 3; see Theorem 3.3. As in [11], particular attention is paid to the case of a degenerate representation in which negative weights appear with multiplicity. This formula suggests an algorithm for computing the Hilbert series as a rational function in the generic case, which we describe in Section 4. We introduce a generalization of Schur polynomials in Section 5 in order to give explicit descriptions of the first few γm in Section 6. In Section 7, we discuss the Gorenstein property of C[V ]G in this case and give an explicit formula for the a-invariant when this property holds. In Appendix A, we demonstrate how the Laurent coefficients of a Cohen- Macaulay domain are related to the degrees of the elements of a Hironaka decomposition. Herbig and Seaton would like to thank Baylor University, and Herden and Seaton would like to thank the Instituto de Matem´atica Pura e Aplicada (IMPA) for hospitality during work contained Acknowledgements THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS 3 here. This paper developed from Cowie's senior seminar project in the Rhodes College Department of Mathematics and Computer Science, and the authors gratefully acknowledge the support of the department and college for these activities. We benefited from lecture notes of Leonid Petrov and follow his terminology for the Laurent-Schur polynomials. 2. Background and notation Let V be a finite-dimensional unitary C×-module. Choosing a basis for V with respect to which the C×-action is diagonal, we can describe the action with a weight vector a = (a1, a2, . . . , an) ∈ Zn. Specifically, for w ∈ C× and z ∈ V with coordinates z = (z1, z2, . . . , zn), we have w · (z1, z2, . . . , zn) = (wa1 z1, wa2 z2, . . . , wan zn). To indicate the representation explicitly, we use the notation C[V ]C× nomials in z1, . . . , zn that are invariant under this C×-action. Similarly, let Hilba(t) := HilbC[V ]C denote the Hilbert series of C[V ]C× a . a to denote the algebra of poly- (t) × a Recall that a representation V is stable if it contains an open set consisting of closed orbits. If V is not stable, then we may replace V with a stable sub-representation without changing the invariants [26], so we will assume stability with no loss of generality. Similarly, the existence of a trivial subrepresentation has the trivial effect of multiplying Hilba(t) by 1/(1−t), so we will frequently assume with no loss of generality that V C× a = {0}, i.e. that 0 does not appear as a weight. Finally, by taking the quotient of C× by the kernel of the action on V , which does not change the invariants, we often assume that V is faithful. It is easy to see that the C×-module V with weight vector a is stable if and only if a contains both positive and negative weights. Let k be the number of negative weights; we will assume the weights are ordered such that ai < 0 for i ≤ k and ai > 0 for i > k. The hypothesis that V is faithful corresponds to gcd(a1, . . . , an) = 1. We say the weight vector a is generic if ai 6= aj for i 6= j with i, j ≤ k and degenerate other- wise. That is, degenerate weight vectors are those that have repeated negative weights. Note that these properties are not invariants of the representation; the weight vectors a and −a correspond to equivalent representations though it is clearly possible that a is generic while −a is degenerate. By the Molien-Weil formula [7, Equation (4.6.2)], the Hilbert series Hilba(t) is given by (2.1) Hilba(t) = Zz∈S1 dz z detV (1 − tz) . This will be the starting point of our computations, which follow the idea of [7, Section 4.6.4]. As explained in the introduction, these computations were conducted for cotangent-lifted representations, those with weight vectors of the form (a,−a), in [11]; though the regular functions on the symplectic quotient were the focus of that paper, the computations corresponding to the usual invariant ring are equivalent by [11, Proposition 2.1]. By [20, Proposition (5.8)], up to tensoring with C, the invariant ring corresponding to (a,−a) is equal to the ring of real invariants of the action with weight vector a. Hence, this paper constitutes a generalization of these results to arbitrary representations. If R is a Cohen-Macaulay ring, then the a-invariant a(R) was defined in [9, Definition 3.1.4] to be the negative of the least degree of a generator of the canonical module of R. This is equal to the degree of HilbR(t), i.e. the degree of the numerator minus the degree of the denominator, and this is sometimes taken as an extension of the definition of the a-invariant to any positively-generated algebra over a field; see [6, Theorem 4.4.3 and Definition 4.4.4]. If R is a Cohen-Macaulay normal domain, then by [21, Theorem 4.4], R is Gorenstein if and only if (2.2) If R is Gorenstein, then (2.3) HilbR(1/t) = (−1)dim Rt−a(R) HilbR(t). 2γ1(R) γ0(R) = −a(R) − dim(R) 4 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON where the γm(R) denote the Laurent coefficients of HilbR(t) as in Equation (1.1). See [18, Equation (3.32)], and note that q in that reference denotes −a(R). 3. Computation of the Hilbert series In this section, we compute the Hilbert series Hilba(t) of the invariants associated to a weight vector a. We first consider the case that a is generic in Proposition 3.1. Though this is a special case of Theorem 3.3 below, we include the brief proof, as it illustrates the fundamental idea behind the other computations in this section. Proposition 3.1. Let a = (a1, . . . , an) ∈ Zn be a weight vector for an action of C× on Cn. Assume that ai < 0 for i ≤ k and ai > 0 for i > k. Assume further that a is generic (ai 6= aj for i 6= j and i, j ≤ k) and stable (0 < k < n). Then (3.1) Hilba(t) = kXi=1 Xζ−ai =1 −ai nQj=1 j6=i 1 . 1 − ζaj t(ai−aj )/ai Proof. By the Molien -- Weyl formula, Equation (2.1), we have Hilba(t) = 1 2πiZS1 dz j=1(1 − zaj t) zQn = 1 2πiZS1 z−1−Pk j=1 aj dz . kQj=1 (z−aj − t) (1 − zaj t) nQj=k+1 From the latter expression, we see that for fixed t with t < 1, the poles inside the unit circle occur when z−aj = t for 1 ≤ j ≤ k. Fix an i with 1 ≤ i ≤ k and an −aith root of unity ζ0, and let t−1/ai be defined with respect to a fixed, suitably chosen branch of the log function. We then express Hilba(t) = (3.2) = 1 2πiZS1 2πiZS1 1 z−ai−1dz (z−ai − t) nQj=1 j6=i (1 − zaj t) (z − ζ0t−1/ai ) Qζ−ai =1 ζ6=ζ0 z−ai−1dz (z − ζt−1/ai ) . (1 − zaj t) nQj=1 j6=i Hence, the residue at ζ0t−1/ai is given by (ζ0t−1/ai)−ai−1 Res z=ζ0t−1/ai f (z, t) = = where we simplify using ζ6=ζ0 Qζ−ai =1 nQj=1 −ai j6=i (ζ0t−1/ai − ζt−1/ai ) 1 (1 − ζaj 0 t(ai−aj )/ai) (1 − t(ζ0t−1/ai )aj ) nQj=1 j6=i , Yζ−ai =1 ζ6=ζ0 (ζ0t−1/ai − ζt−1/ai ) = (ζ0t−1/ai)−ai−1 Yζ−ai =1 ζ6=1 (1 − ζ) = ζ−1 0 t(ai+1)/ai(−ai). Summing the residues over all choices of i ≤ k and ζ0, yields the claim. (cid:3) THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS 5 We now consider the same computation in the case that a is degenerate. The change occurs in the decomposition of the integrand given in Equation (3.2). If ai = aj for some j (which must be ≤ k), then the factor (1 − zaj t) also vanishes at z = ζ0t−1/ai, which is no longer a simple pole of the integrand. We give two ways of approaching this computation. The first, yielding Proposition 3.2, is simply to compute each such residue. This results in Equation (3.3) which, while not particularly elucidating, is useful for calculations of specific examples. The second method, yielding Theorem 3.3, is to introduce new variables so that each pole remains simple. This result essentially demonstrates that, treating the weights ai in Equation (3.2) as real variables except in choosing the root of unity ζ, the apparent singularities that occur in factors of the form 1 − ζaj t(ai−aj )/ai when ζ = 1 and ai = aj are in fact removable, and the Hilbert series at degeneracies is the extension of this continuous function of the weights at these removable singularities. We begin with the first method. To simplify the argument, we use a slightly different notation to index the weights. Proposition 3.2. Let (a, b) be a weight vector for an action of C× on Cn, where a1, . . . , a1, a2, . . . , a2, . . . , b = (b1, . . . , bm), representation is stable (k > 0 and m > 0); note that the bi need not be distinct. Then i=1 ri. We assume that each ai < 0, that each bi > 0, and that the 1 (ri − 1)! Fi,a,b(z, t), . (1 − zbj t) mQj=1 (1 − zbj t) . r1 a = ( } each ri ≥ 1, and n = m +Pk z { (3.3) where (3.4) Hilb(a,b)(t) = Fi,a,b(z, t) = { r2 z } kXi=1 Xζ−ai 0 =1 rk ∂ri−1 ak, . . . , ak), z { } ∂zri−1(cid:12)(cid:12)(cid:12)(cid:12)z=ζ0t−1/ai kQj=1 z−riai−1 j6=i Qζ−ai =1 ζ6=ζ0 (z − ζt−1/aj )ri (1 − zaj t) Proof. By the Molien -- Weyl formula, Hilb(a,b)(t) = = 1 2πiZS1 2πiZS1 1 dz j=1(1 − zaj t)rj mQj=1 z−1−Pk j=1 rj aj dz (z−aj − t)rj (1 − zbj t) mQj=1 zQk kQj=1 As in the proof of Proposition 3.1, for fixed t with t < 1, the poles inside the unit circle occur when z−aj = t for some j. Fix an i with 1 ≤ i ≤ k and an −aith root of unity ζ0, and let t−1/ai be defined with respect to a suitable fixed branch of the log function. Then we can express Hilb(a,b)(t) as 1 2πiZS1 z−riai−1 dz (z − ζ0t−1/ai )ri Qζ−ai =1 ζ6=ζ0 (z − ζt−1/ai )ri (1 − zaj t)rj kQj=1 j6=i . (1 − zbj t) mQj=1 Noting that the factor Fi,a,b(z, t) of the integrand, as defined in Equation (3.4), is holomorphic at t = ζ0t−1/ai , the residue at t = ζ0t−1/ai is given by the (ri − 1)st coefficient of the Taylor series of this function at z = ζ0t−1/ai, i.e. Res z=ζ0t−1/ai f (z, t) = ∂ri−1 ∂zri−1(cid:12)(cid:12)(cid:12)(cid:12)z=ζ0t−1/ai 1 (ri − 1)! Fi,a,b(z, t). 6 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON Summing over each pole completes the proof. (cid:3) We now turn to the second method for dealing with degeneracies and prove the following, the main result of this section. Theorem 3.3. Let a = (a1, . . . , an) ∈ Zn be a weight vector for an action of C× on Cn. Assume that ai < 0 for i ≤ k and ai > 0 for i > k, and moreover that a is stable (0 < k < n). Then (3.5) Hilba(t) = lim c→a c = (c1, . . . , cn) ∈ Rn. kXi=1 Xζ−ai =1 −ci nQj=1 j6=i 1 , 1 − ζaj t(ci−cj )/ci Note that we may consider c = (c1, . . . , ck, ak+1, . . . , an), i.e. take ci = ai for i > k as, in the computation below, we require ci 6= cj for i 6= j only when i, j ≤ k. Proof. We consider the computation of the residue at a point satisfying t = z−ai where ai < 0 is a degenerate weight. For simplicity, we relabel the weights to express the weight matrix as (a, . . . , a, a1, . . . , ak−q, b1, . . . , bm) where a < 0 occurs q times, ai < 0 for each i, and a 6= ai for each i. We do not require that the ai nor the bi are distinct. Consider the integral (3.6) 1 2πiZS1 z qQj=1 (1 − zaxj ) dz (1 − zaj yj) k−qQj=1 . (1 − zbj wj ) mQj=1 xj,yj,wj < 1 and the xj are assumed distinct. Let x = (x1, . . . , xq) ∈ Cq, Here, y = (y1, . . . , yk−q) ∈ Ck−q, and w = (w1, . . . , wm) ∈ Cm. These additional variables are intro- duced to force each pole to be simple; the idea is to compute this integral and then take the limit as x → ∆qt := (t, . . . , t) ∈ Cq, y → ∆k−qt := (t, . . . , t) ∈ Ck−q, and w → ∆mt := (t, . . . , t) ∈ Cm. contained in the domain of this branch. Let Fix a branch of the logarithm near t. We will assume throughout that each xi, yi, and wi is F (x, y, w, z) := 1 . qQj=1 As a function of z, the poles of F (x, y, w, z) for z < 1 occur at z = ζx−1/a a (−a)th root of unity or z = ηy−1/ai where 1 ≤ i ≤ q and ζ is where 1 ≤ i ≤ k − q and η is a (−ai)th root of unity. Hence, (1 − zbj wj ) (1 − zaj yj) (1 − zaxj ) k−qQj=1 mQj=1 z i i the integral in Equation (3.6) is given by Xζa=1 qXi=1 Res z=ζx−1/a i F (x, y, w, z) + k−qXi=1 Xηai =1 Res z=ηy−1/ai i F (x, y, w, z). Fix a (−a)th root of unity ζ0. We claim that Ra(x, y, w, ζ0) := Res z=ζ0x−1/a i F (x, y, w, z) qXi=1 admits an analytic continuation whose domain includes x1 = x2 = ··· = xq = t. To see this, fix an i, and then we express F (x, y, w, z) as (z − ζ0x−1/a i ) Qζ−a=1 ζ6=ζ0 (z − ζx−1/a i z−a−1 ) qQj=1 j6=i (1 − zaxj ) k−qQj=1 (1 − zaj yj) . (1 − zbj wj ) mQj=1 THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS 7 Hence we have a simple pole at ζ0x−1/a steps as in the proof of Proposition 3.1, we express the residue Resz=ζ0x−1/a i . Computing the reside and simplifying following the same F (x, y, w, z) as i −a (xi − xj ) qQj=1 j6=i k−qQj=1 xq−1 i (1 − ζaj 0 x−aj /a i (1 − ζbj 0 x−bj /a i . wj) yj) mQj=1 Summing over i = 1, . . . , q and combining the result into a single rational expression, we express i=1 Resz=ζ0x−1/a i F (x, y, w, z) as Ra(x, y, w, ζ0) =Pq qPi=1 (−1)q−ixq−1 −a Q1≤j<ℓ≤q i Q1≤j<ℓ≤q j,ℓ6=i (xℓ − xj ) ℓ6=i k−qQj=1 qQℓ=1 k−qQj=1 (1 − ζaj qQℓ=1 (xℓ − xj) (1 − ζaj 0 x−aj /a ℓ (1 − ζbj 0 x−bj /a ℓ wj) . 0 x−aj /a ℓ yj) 0 x−bj /a ℓ wj ) yj) ℓ6=i qQℓ=1 mQj=1 mQj=1 (1 − ζbj qQℓ=1 We claim that the numerator of Ra(x, y, w, ζ0) is alternating in the x1, . . . , xq. To see this, define αi(x, y, w, ζ0) := i Y1≤j<ℓ≤q (xℓ − xj) (−1)q−ixq−1 (1 − ζaj so that the numerator of Ra(x, y, w, ζ0) is equal to Pq k−qYj=1 qYℓ=1 j,ℓ6=i ℓ6=i i=1 αi(x, y, w, ζ0). It is easy to see that for a transposition σ ∈ Sq, αi(xσ(1), . . . , xσq, y, w, ζ0) = −ασ(i)(x1, . . . , xq, y, w, ζ0), implying that the numerator of Ra(x, y, w, ζ0) is alternating. It follows that there is a S(x, y, w, ζ0), symmetric in the x1, . . . , xq, such that Ra(x, y, w, ζ0) is equal to 0 x−aj /a ℓ yj) qYℓ=1 ℓ6=i mYj=1 (1 − ζbj 0 x−bj /a ℓ wj) a qQℓ=1 k−qQj=1 (1 − ζaj 0 x−aj /a ℓ yj) S(x, y, w, ζ0) (1 − ζbj 0 x−bj /a ℓ . wj ) qQℓ=1 mQj=1 Hence, as long as the yj are chosen so that ζaj yj 6= 1, the singularities at xi = xj are removable. Now that we have determined that the limit limx→∆qt Ra(x, y, w, ζ0) exists, we will carry out a sequence of parameterizations to compute its value. Let t = (t1, . . . , tq) and s = (s1, . . . , sq) and express xj = ttj , yj = t, and wj = t to express limx→∆qt Ra(x, y, w, ζ0) as 0 x−aj /a ℓ lim t→∆q1 qXi=1 −a (1 − ttj −ti ) j6=i qQj=1 qXi=1 qXi=1 a ti j6=i qQj=1 qQj=1 j6=i = lim t→∆q1 s→∆q1 = lim t→∆q1 1 (1 − ζaj 0 t(a−tiaj )/a) k−qQj=1 (1 − ζbj 0 t(a−tibj )/a) mQj=1 1 −a si (1 − t(tj −ti)/sj ) (1 − ζaj 0 t(a−tiaj )/a) k−qQj=1 k−qQj=1 1 (1 − ζbj 0 t(a−tibj )/a) , mQj=1 mQj=1 (1 − t(tj −ti)/tj ) (1 − ζaj 0 t(a−tiaj )/a) (1 − ζbj 0 t(a−tibj )/a) 8 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON where in the last step we set si = ti for each i. Doing some elementary algebra and setting ci = a/ti for i = 1, . . . , q, we obtain lim c→∆qa qXi=1 ci qQj=1 j6=i 1 (1 − t(ci−cj )/ci) k−qQj=1 (1 − ζaj 0 t(ci−aj )/ci) . (1 − ζbj 0 t(ci−bj )/ci) mQj=1 Applying this computation to each of the poles and returning to the notation for a in the statement of the theorem, we obtain that the sum of the residues of all poles in the unit circle is given by kXi=1 Xζai =1 ci kQj=1 j6=i 1 (1 − ζaj t(ci−cj )/ci) . (1 − ζbj t(ci−bj )/ci) mQj=1 It remains only to show that we may exchange the limit with the integral. But this follows from a simple application of the Dominated Convergence Theorem, where we note that the integrand is continuous and hence bounded on the circle. (cid:3) 4. An algorithm to compute the Hilbert series in the generic case In this section, we outline how Proposition 3.1 can be turned into an algorithm to compute the Hilbert series Hilba(t) associated to a generic weight vector. This algorithm is very similar to that described in [11, Section 4], so we give a brief summary and refer the reader to that reference for more details. m=0 Gmtm and a positive integer a, the operator Ua Recall that for a formal power series G(t) =P∞ is given by (UaG)(t) := G(a)(t) := Gmatm. ∞Xm=0 By [11, Lemma 4.1], if G(t) is the power series of a rational function, then G(a)(t) is as well rational. Now, start with a generic weight vector (a1, . . . , an) with ai < 0 for i ≤ k and ai > 0 for i > k. For i = 1, . . . , k, define fΦi(t) := nQj=1 j6=i 1 , 1 − t(ai−aj )/ai which we note is analytic at t = 0. Using the expression of Ua in terms of averaging over roots of unity described in [11, Section 4], Equation (3.1) can be written as Hilba(t) = kXi=1 1 −ai Xζ−ai =1fΦi(ζt) = kXi=1 (fΦi)(−ai)(t). terms of the form (1 − ts), we replace each (1 − ts) in the denominator with ExpressingfΦi(t) as a rational function P (t)/Q(t) where P (t) is a monomial and Q(t) is a product of This yields the denominator of (fΦi)(ai)(t). Then we determine the Taylor series of (fΦi)(ai)(t) using the U(ai) operator and multiply these out to determine the numerator of (fΦi)(ai)(t). By Kempf's bound [15, Theorem 4.3], dim P (t) ≤ dim Q(t) so that we need only compute the Taylor series of (fΦi)(ai)(t) up to degree deg(Q). Then the Hilbert series is given by the sum of the (fΦi)(ai)(t) for 1 ≤ i ≤ k. This algorithm has been implemented on Mathematica [28] and is available from the authors upon request. (1 − tlcm (−ai,s))gcd (−ai,s). (5.1) Su(x, y) = 1 V(x) V(y) xu 1 xn−2 1 xn−3 1 ... x1 1 ··· ··· ··· ··· ··· xu k xn−2 k xn−3 k ... xk 1 0 yn−2 1 yn−3 1 ... y1 1 ··· ··· ··· ··· ··· 0 yn−2 m yn−3 m ... ym 1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) . (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS 9 5. Partial Laurent-Schur polynomials In this section, we introduce a family of polynomials, independently symmetric in two sets of variables, that will be useful for describing the Laurent coefficients of the Hilbert series of circle invariants and, in particular, removing singularities in their descriptions. For a set of k indeterminates x = (x1, . . . , xk), we let V(x) denote the Vandermonde determinant V(x) = Y1≤i<j≤k (xi − xj ). We now state the following. Definition 5.1. Let x = (x1, . . . , xk) and y = (y1, . . . , ym) be two sets of indeterminates with n = k + m, and let u ≤ n − 2 be an integer. We define the partial Laurent-Schur polynomial Su(x1, . . . , xk, y1, . . . , ym) = Su(x, y) to be To indicate the connection with ordinary Schur polynomials, let λ be a partition of length n, i.e. λ ∈ Zn with λ1 ≥ λ2 ≥ ··· ≥ λn ≥ 0. Recall that the alternant associated to λ in the indeterminates x = (x1, . . . , xn) is the determinant Aλ(x) = Aλ1,...,λn (x1, . . . , xn) =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) xλ1 1 xλ2 1 ... xλn 1 xλ1 2 xλ2 2 ... xλn 2 ··· ··· ··· xλ1 n xλ2 n ... xλn n . (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) sλ(x) = Aλ+δn (x) V(x) = Aλ+δn (x) Aδn (x) , The Schur polynomial associated to λ is the symmetric polynomial where δn = (n − 1, n − 2, . . . , 0). The fact that Aλ(x) is obviously alternating in the xi implies that the polynomial Aλ(x) is divisible by V(x) and hence sλ(x) is a polynomial. If we allow the λi to take negative values, then λ is called a signature, and sλ(x), defined in the same way, is the Laurent-Schur polynomial associated to λ. See [16, I.3] or [19, 4.4 -- 6] for more details. We now indicate the motivation for Definition 5.1. In the computations of the Laurent coefficients of the Hilbert series Hilba(t) of circle invariants in Section 6, we will frequently run into rational functions of the form (5.2) (−1)i−1xu (xp − xq) (−1)i−1xu (xp − xq) kPi=1 = i Q1≤p<q≤n p,q6=i V(x1, . . . , xn) kPi=1 i Q1≤p<q≤n p,q6=i (xp − xq) Q1≤p<q≤n where u ≤ n − 2 is an integer. Usingb∗ to denote removed columns, we can express kXi=1 (−1)i−1xu i Y1≤p<q≤n p,q6=i (xp − xq) = kXi=1 (−1)i−1xu i xn−2 1 xn−3 1 ... x1 1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ··· ··· [ xn−2 i [ xn−3 i ... ··· ··· ··· ··· ··· ··· bxi b1 xn−2 n xn−3 n ... xn 1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 10 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON to see that the numerator of Equation (5.2) is the first k terms of the cofactor expansion of the alternant Au,n−2,n−3,...,0(x1, . . . , xn) along the first row which again coincides with the determinant (5.3) xu 1 xn−2 1 xn−3 1 ... x1 1 ··· ··· ··· ··· ··· xu k xn−2 k xn−3 k ... xk 1 0 xn−2 k+1 xn−3 k+1 ... xk+1 1 ··· ··· ··· ··· ··· 0 xn−2 n xn−3 n ... xn 1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) . (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Relabeling variables to account for the missing symmetries, it is obvious that Equation (5.3) corre- sponds to the determinant given in Equation (5.1). While this determinant is not divisible by the full Vandermonde determinant V(x1, . . . , xn), it is divisible by the partial Vandermonde determinants in the first k and last n − k variables in Equation (5.1). The purpose of this section is to give explicit descriptions of the quotient polynomials. To facilitate this, we introduce the following notation. If λ is a partition with λ1 ≤ n, we let λc,n denote the complement of λ in n, i.e. the partition that contains one instance of each element of n that does not appear in λ. If λ is a partition with λ1 ≤ n containing the integer u, let λc,n For a positive integer n, let n = {1, 2, . . . , n} and n = {0, 1, 2, . . . , n}. u denote the partition λc,n with the entry u added. We then have the following. Theorem 5.2. Let x = (x1, . . . , xk) and y = (y1, . . . , ym) be two sets of indeterminants with n = k + m, and let u ≤ n − 2 be an integer. If u ≥ 0, then (5.4) Su(x, y) = (−1)k(k+1)/2+n(k−1)+u Xn−2≥λ1>···>λk≥0 λi=u (−1)kλk+isλ−δk (x)sλc,n−2 u (y), −δm where the sum is over all partitions λ of length k with λ1 ≤ n − 2 such that λi = u for some i. If u < 0, then (5.5) Su(x, y) = (−1)k(k−1)/2+n(k−1) Xn−2≥λ1>···>λk−1≥0 (−1)kλks(λ,u)−δk (x)sλc,n−2−δm(y), where the sum is over all partitions of λ of length k − 1 with λ1 ≤ n − 2. In either case, Su(x, y) is homogeneous of degree (m − 1)(k − 1) + u. Note that if u ≥ 0, then via row reduction, it is obvious that we can express (5.6) Su(x, y) = 1 V(x) V(y) xu 1 xn−2 1 ... xu+1 1 0 xu−1 1 ... 1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ··· ··· ··· ··· ··· ··· xu k xn−2 k ... xu+1 k 0 xu−1 k ... 1 0 yn−2 1 ... yu+1 1 yu 1 yu−1 1 ... 1 ··· ··· ··· ··· ··· ··· 0 yn−2 m ... yu+1 m yu m yu−1 m ... 1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) , u ≥ 0. THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS 11 Proof of Theorem 5.2. We first consider the case 0 ≤ u ≤ n − 2. For P ⊆ n, let P denote the cardinality of P and kPk the sum of the elements of P . Let 0 0 Z = xu 1 xn−2 1 ... xu+1 1 0 xu 2 xn−2 2 ... xu+1 2 0 xu−1 1 ... x1 1 xu−1 2 ... x2 1  ··· ··· ··· ··· ··· ··· ··· xu k xn−2 k ... xu+1 k 0 xu−1 k ... xk 1 yn−2 1 ... yu+1 1 yu 1 yu−1 1 ... y1 1 ··· ··· ··· ··· ··· ··· ··· yn−2 m ... yu+1 m yu m yu−1 m ... ym 1  be the matrix whose determinant appears in Equation (5.6). For P, Q ⊆ n such that P = Q, let Z P,Q denote the cofactor of Z formed by the determinant of the submatrix with rows in P and columns in Q. Note that Z P,k = 0 for each P ⊆ n with P = k that contains the element n − u, and similarly Z nrP,nrk = 0 for each P ⊆ n with P = k that does not contain 1. Hence, the Laplace expansion of the determinant along the first k columns, see [10, Theorem 13.8.1], is given by Su(x, y) = 1 V(x) V(y) X1∈P ⊆n,n−u6∈P P =k (−1)kkk+kP kZ P,kZ nrP,nrk. To express the minors Z P,k and Z nrP,nrk in terms of alternants, consider a choice of P with elements 1 = p1 < p2 < ··· < pk. Then Z P,k = (−1)i−1Aλ(x) where λ is a partition of length k, n − 2 ≥ λ1 ≥ ··· ≥ λk ≥ 0, and λi = u. In particular, λ is given by transposing the first entry of (u, n − p2, n − p3, . . . , n − pk) so that the result is decreasing, hence kPk = n(k − 1) − kλk + u + 1. Similarly, Z nrP,nrk = Aλc,n−2 is the complement of λ in n − 2 with the element u added. Therefore, we can express (y), where λc,n−2 u u Su(x, y) = (−1)k(k+1)/2+n(k−1)+u Xn−2≥λ1>···>λk≥0 = (−1)k(k+1)/2+n(k−1)+u Xn−2≥λ1>···>λk≥0 λi=u λi=u (−1)kλk+i Aλ(x) V(x) (y) Aλc,n−2 u V(y) (−1)kλk+isλ−δk (x)sλc,n−2 u (y), −δm yielding Equation (5.4). Now suppose u < 0 and set Z = xu 1 xn−2 1 ... x1 1  xu 2 xn−2 2 ... x2 1 ··· ··· ··· ··· xu k xn−2 k ... xk 1 0 yn−2 1 ... y1 1 ··· ··· ··· ··· 0 yn−2 m ... ym 1 .  We again consider the Laplace expansion of det Z along the first k columns, noting that Z nrP,nrk = 0 for each P ⊆ n with P = k that does not contain 1, and express Su(x, y) = 1 V(x) V(y) X1∈P ⊆n P =k (−1)kkk+kP kZ P,kZ nrP,nrk. 12 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON Considering a choice of P with elements 1 = p1 < p2 < ··· < pk, in this case, Z P,k = (−1)k−1A(λ,u)(x) where λ = (n − p2, . . . , n − pk), and kPk = n(k − 1) − kλk + 1. Thus Su(x, y) = (−1)k(k+1)/2+n(k−1)−k = (−1)k(k−1)/2+n(k−1) Xn−2≥λ1>···>λk−1≥0 Xn−2≥λ1>···>λk−1≥0 Aλc,n−2 (y) (−1)kλk A(λ,u)(x) (−1)kλks(λ,u)−δk (x)sλc,n−2−δm(y), V(x) V(y) yielding Equation (5.5). That Su(x, y) is homogeneous of degree (n − 2)(n − 1) + u − (k − 1)k/2 − (m − 1)m/2 = (m − 1)(k − 1) + u is easily observed from Equation (5.1), completing the proof. (cid:3) Recall that the Schur polynomial sλ(x) associated to a partition λ of length n in the indeterminates x = (x1, . . . , xn) can be represented as (5.7) sλ(x) =XT xT , where the sum ranges over all semistandard Young tableaux T of shape λ. We include a similar combinatorial interpretation of Theorem 5.2. First, we consider the case 0 ≤ u ≤ n − 2. Corollary 5.3. If 0 ≤ u ≤ n − 2, then Su(x, y) = (−1)k(k+1)/2+n(k−1)+uXT (−1)kλk+i(x, y)T , where the sum is over all tableaux T formed as follows: (1) Start with a Young diagram of shape (n − 2, n − 3, . . . , u + 1, u, u, u − 1, . . . , 1, 0), i.e. n rows consisting of each integer from 0 to n − 2 with u appearing twice. (2) Label k of the rows x and the remaining m = n − k rows y, where the upper u row must be labeled x while the lower u row is labeled y. Note that λ corresponds to the shape of the diagram consisting of the rows labeled x and λc,n−2 is the shape of the diagram of rows labeled y. Define i by letting the u row labeled x appear as the ith row labeled x from the top. u (3) Starting from the bottom row of length 0, for each j, delete j− 1 boxes from the jth row labeled x and delete j − 1 boxes from the jth row labeled y. (4) Fill in the rows labeled x with x1, . . . , xk so that the rows labeled x form a semistandard Young tableau of shape λ− δk, and similarly fill in the rows labeled y with y1, . . . , ym so that the rows labeled y form a semistandard Young tableau of shape λc,n−2 − δm. As usual, (x, y)T denotes the product of the entries of T . u Proof. This is an immediate consequence of interpreting Equation (5.4) in the context of Equa- tion (5.7). Note in particular, that we have multiplicativity sλ−δk (x)sλc,n−2 u −δm (y) = XS xS! XT yT! =XS,T xSyT , where S ranges over all tableaux of shape λ − δk and T over all tableaux of shape λc,n−2 u − δm. (cid:3) The case u < 0 can be described by a variation of this construction. Corollary 5.4. If u < 0, then Su(x, y) = (−1)k(k−1)/2+n(k−1) kYi=1 xu i ·XT (−1)kλk(x, y)T , where the sum is over all tableaux T formed as follows: (1) Start with a Young diagram of shape (n − 2, n− 3, . . . , 1, 0), i.e. n− 1 rows consisting of each integer from 0 to n − 2. THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS 13 the Young diagram labeled x and λc,n−2 is the shape of the Young diagram labeled y. (2) Label k − 1 of the rows x and the remaining m = n − k rows y. Note that λ is the shape of (3) Add −u − 1 boxes to each row labeled x. Then, starting from the bottom row of length 0, for each j, delete j − 1 boxes from the jth row labeled x and delete j − 1 boxes from the jth row labeled y. (4) Fill in the rows labeled x with x1, . . . , xk so that the rows labeled x form a semistandard Young tableau of shape λ − δk−1 − u − 1, and similarly fill in the rows labeled y with y1, . . . , ym so that the rows labeled y form a semistandard Young tableau of shape λc,n−2 − δm. As usual, (x, y)T denotes the product of the entries of T . Proof. This is proven similarly to Corollary 5.3. Note in particular that for λ = (λ1, . . . , λn) and λ − u = (λ1 − u, . . . , λn − u), we have a shifting rule for alternants ··· ··· xλ1−u 1 xλ2−u 1 xλ1−u 2 xλ2−u 2 xλ1−u n xλ2−u n = nYi=1 xu i · ... ... ... = nYi=1 xu i · Aλ−u(x). xλ1 1 xλ2 1 ... xλn 1 Aλ(x) =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) As a consequence, xλ1 2 xλ2 2 ... xλn 2 ··· ··· ··· xλ1 n xλ2 n ... xλn n (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) xλn−u 1 xλn−u 2 ··· xλn−u n (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) kYi=1 s(λ,u)−δk (x) = xu i · s(λ−u,0)−δk (x) allows us to interpret generalized Laurent-Schur polynomials in the context of classical Schur polyno- mials. (cid:3) 6. The coefficients of the Laurent series Let a = (a1, . . . , an) be the weight vector for an effective, stable action of C× on Cn, where ai < 0 for i ≤ k and ai > 0 for i > k. In this section, we detail a computation of the first few coefficients γm(a) of the Laurent series of Hilba(t) at t = 1. Using Equation (3.5), we consider the Laurent coefficients of the expression (6.1) Ha,c(t) = kXi=1 Xζ−ai =1 −ci nQj=1 j6=i 1 , 1 − ζaj t(ci−cj )/ci where c = (c1, . . . , cn) ∈ Rn with ci < 0 for i ≤ k and ci > 0 for i > k. As we will see, the resulting expressions for the γm(Ha,c(t)) will be continuous functions of the ci with poles occurring only at ci = cj where i ≤ k and j > k and hence not in the domain under consideration. Noting that Ha,c(t) has a pole at t = 1 of order n − 1 (occurring in each term with ζ = 1), we can express γm(Ha,c(t)) = Ha,c(t)(t − 1)m−n dt 1 2π√−1ZC where C is a positively oriented curve about 1. Hence, as Ha,c(t) and the expressions we will derive for γm(Ha,c(t)) are analytic (up to removable singularities) in a punctured neighborhood of t = 1, an application of the Dominated Convergence Theorem implies that lim c→a γm(Ha,c(t)) = γm(a). See [11, Secion 5.2] for more details in the case of cotangent-lifted representations; the argument applies without change to our setting. Remark 6.1. Our expressions for the γm(a) will frequently involve the partial Schur polynomials Su(a). We will always understand this notation to mean that the weight vector a is split into two different sets of indeterminates, the negative weights in the first set and the positive weights in the 14 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON second. This slight abuse of notation will be particularly convenient when we consider polynomials of the form Su(aj), where aj denotes the weight vector a with the jth entry removed, as it will allow us to avoid using separate notation for the cases when j ≤ k and j > k. Theorem 6.2. Let a = (a1, . . . , an) be the weight vector for an effective action of C× on Cn. Assume that ai < 0 for i ≤ k and ai > 0 for i > k, and moreover that a is stable (1 < k < n). Then dim(C[Cn]C× a ) = n − 1, and (6.2) γ0(a) = In particular, γ0(a) 6= 0. When a is generic, this can be expressed as (6.3) γ0(a) = . −Sn−2(a) (ap − aq) kQp=1 nQq=k+1 kXi=1 −an−2 nQj=1 (ai − aj) j6=i . i Proof. Examining Equation (6.1), we see that each term of Ha,c(t) has a pole order of at most n − 1, with this maximum obtained at the terms with ζ = 1. Therefore, γ0(Ha,c(t)) is the degree 1− n term in the Laurent series of (6.4) kXi=1 −ci nQj=1 j6=i 1 1 − t(ci−cj )/ci = kXi=1 1 −ci nYj=1 j6=i 1 1 − t(ci−cj )/ci . Using the fact that the Laurent series of 1/(1 − tc) at t = 1 begins c2 − 1 (6.5) 24c (1 − t)−1 + (1 − t) + c2 − 1 12c c − 1 2c 1 c + 1 1 − tc = we have that (1 − t)2 + O(cid:0)(1 − t)3(cid:1), γ0(Ha,c(t)) = kXi=1 , i −cn−2 nQj=1 (ci − cj) j6=i from which Equation (6.3) follows. To express this as a single rational function, we simplify (6.6) γ0(Ha,c(t)) = = i (−1)icn−2 nQj=i+1 Q1≤p<q≤n (cj − ci) kXi=1 i−1Qj=1 kPi=1 (−1)icn−2 Q1≤p<q≤n p,q6=i i (cp − cq) (ci − cj) (cp − cq) , where we recognize the numerator as a cofactor expansion of the form described in Equation (5.2). Therefore, γ0(Ha,c(t)) = −Sn−2(c) , (cp − cq) kQp=1 nQq=k+1 where we note that the singularities at ci = cj for i, j ≤ k or i, j > k have been removed. Taking the limit as c → a completes the proof of Equation (6.2). THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS 15 order of Hilba(t) at t = 1 is equal to dim(Cn//C× By [26, Remark 2], the assumption of stability implies that dim(Cn//C× a ) = n − 1. As the pole a ) by [7, Lemma 1.4.6], it follows that γ0(a) 6= 0. (cid:3) Before stating the next result, we introduce some additional notation. For each j, let aj ∈ Zn−1 denote the weight vector a with aj removed and let gj := gcd aj = gcd{ai : i 6= j}. For indeterminates x = (x1, . . . , xn) and 1 ≤ j ≤ n, we let Ej (x) = X1≤i1≤···≤ij ≤n xi1 ··· xij denote the elementary symmetric polynomial of degree j. Theorem 6.3. Let a = (a1, . . . , an) be the weight vector for an effective action of C× on Cn. Assume that ai < 0 for i ≤ k and ai > 0 for i > k, and moreover that a is stable (1 < k < n). Then (6.7) γ1(a) = E1(a)Sn−3(a) − Sn−2(a) + 2 kQp=1 nQq=k+1 (ap − aq) nXj=1(cid:18) gj − 1 2 (cid:19) γ0(aj). When a is generic, this can be expressed as (6.8) γ1(a) = kXi=1 nXj=1 j6=i 2 an−3 i aj (ai − aℓ) nQℓ=1 ℓ6=i + kXi=1 nXj=1 j6=i (cid:18) gj − 1 2 (cid:19) . i −an−3 nQℓ=1 (ai − aℓ) ℓ6=i,j Proof. Considering Equation (6.1), a term of Ha,c(t) contributes to the degree 2 − n term of the Laurent series if it has a pole order of either n − 1 or n − 2. The former case corresponds to terms with ζ = 1, while the latter corresponds to choices of i and aith root of unity ζ such that ζaj = 1 for all but one j 6= i. We first consider the former. of the Laurent series of the expression in Equation (6.4) for the terms with ζ = 1 is given by Using the Laurent series in Equation (6.5) and the Cauchy product formula, the degree 2 − n term (6.9) kXi=1 1 −ci nXj=1 j6=i = kXi=1 nXj=1 j6=i 2 cn−3 i cj . (ci − cℓ) nQℓ=1 ℓ6=i Combining via the same process as in Equation (6.6) above yields We now consider the terms in Ha,c(t) with a pole order of n − 2, which correspond as noted above to choices of i and ζ such that ζaj = 1 for all but one j 6= i. This clearly occurs only if there is a  ci ℓ6=i,j nYℓ=1 2(ci − cj) (ci − cℓ)(cid:0)(ci − cj)/ci − 1(cid:1)ci Q1≤p<q≤n 2 Q1≤p<q≤n (−1)i−1cn−3 (cp − cq) − (cp − cq) (cp − cq) p,q6=i nPj=1 j6=i cj i Q1≤p<q≤n p,q6=i 2 Q1≤p<q≤n (cp − cq) kPi=1 (−1)i−1cn−3 i E1(c) kPi=1 E1(c)Sn−3(c) − Sn−2(c) . 2 kQp=1 nQq=k+1 (cp − cq) (6.10) (6.11) = = = kPi=1 (−1)i−1cn−2 i Q1≤p<q≤n p,q6=i (cp − cq) 16 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON j such that gj 6= 1 and ζ is a non-unit gjth root of unity. Hence, for such a choice of i and j, the corresponding terms of Ha,c(t) are Xζgj =1 ζ6=1 1 −ci(1 − ζaj t(ci−cj )/ci) . (1 − t(ci−cℓ)/ci ) nQℓ=1 ℓ6=i,j Using Equation (6.5) and noting that gcd(aj, gj) = gcd(a1, . . . , an) = 1, the first term of the Laurent series of this expression is Xζgj =1 ζ6=1 (6.12) −ci(1 − ζaj ) cn−2 i nQℓ=1 ℓ6=i,j (ci − cℓ) = = −ci 1 1 − ζ ζ6=1 ℓ6=i,j cn−2 i (ci − cℓ) Xζgj =1 nQℓ=1 2 (cid:19) , (ci − cℓ)(cid:18) gj − 1 −cn−3 nQℓ=1 i ℓ6=i,j where the sum over ζ is computed using [8, Corollary 3.2]. Along with taking the limit as c → a as described above, combining Equations (6.9) and (6.12) yields Equation (6.8). Then Equation (6.7) follows from Equation (6.11) and comparing Equa- tion (6.12) with Equation (6.3) for γ0(aj). (cid:3) The approach used in Theorems 6.2 and 6.3 above can be used to compute γm(a) for m > 1, and the results can similarly be expressed in terms of the Su(a). However, in these cases, one meets sums of rational expressions over roots of unity as in Equation (6.12) that are not readily computable using the results of [8]. Recall [3, (1.13)] that a Fourier-Dedekind sum is a sum of the form σr(a2, . . . , an; a1) = 1 a1 Xζa1 =1 ζ6=1 ζr (1 − ζaj ) nQj=2 where each a2, . . . , an is coprime to a1; see also [2, 23]. The expressions below for γm(a) with m > 1 involve nontrivial "partial" Fourier -- Dedekind sums that bear a resemblance to Ramanujan's sum, i.e. sums over a subset of nonunit roots of unity, with the coprime conditions on the ai relaxed. For example, we have the following, whose proof involves tedious computations but requires no more ingredients than those in the proof of Theorem 6.3. For each j 6= ℓ, let aj,ℓ ∈ Zn−2 denote the weight vector a with aj and aℓ removed and let gj,ℓ := gcd aj,ℓ = gcd{ai : i 6= j, ℓ}. THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS 17 Theorem 6.4. Let a = (a1, . . . , an) be the weight vector for an effective action of C× on Cn. Assume that ai < 0 for i ≤ k and ai > 0 for i > k, and moreover that a is stable (1 < k < n). Then γ2(a) = 5E1(a)Sn−3(a) − (E2(a) + E1(a)2)Sn−4(a) − 4Sn−2(a) 12 kQp=1 nQq=k+1 (ap − aq)   1 − g2 12 j + nXj=1 Sn−3(aj) − ajSn−4(aj) kQp=1 p6=j nQq=k+1 q6=j (ap − aq) 4 + nXj=1 + X1≤j<ℓ≤n gj − 1 E1(aj)Sn−4(aj) − Sn−3(aj) q6=j p6=j Sn−4(aj,ℓ) (ap − aq) kQp=1 nQq=k+1 (ap − aq) Xζgj,ℓ =1 nQq=k+1 q6=j,ℓ ζgj 6=1,ζgℓ 6=1 kQp=1 p6=j,ℓ   1 (1 − ζaj )(1 − ζaℓ ) . When a is generic, this can be expressed as γ2 = kXi=1 (ai − aj) an−4 i j6=i 12 nQj=1 nXj=1 kXi=1 kXi=1 X1≤j<ℓ≤n  j6=i j,ℓ6=i + +   nXj=1 j6=i (2ai − aj)aj − 3 nXℓ=1 gj − 1 + 2 ℓ6=i,j (1 − g2 12 j )an−4 i (ai − aj) ℓ6=i,j (ai − aℓ) nQℓ=1 (ai − ap) Xζgj,ℓ =1 −an−4 nQp=1 i p6=i,j,ℓ ζgj 6=1,ζgℓ 6=1 nX1≤j<ℓ≤n j,ℓ6=i ajaℓ an−4 i aℓ (ai − ap) 2 nQp=1 p6=i,j  1 (1 − ζaj )(1 − ζaℓ ) . For the entertainment of the reader, we also state the following. Continuing the same notational convention as above, for each distinct j, ℓ, p, let aj,ℓ,p ∈ Zn−3 denote the weight vector a with aj, aℓ, and ap removed and let gj,ℓ,p := gcd aj,ℓ,p = gcd{ai : i 6= j, ℓ, p}. 18 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON Theorem 6.5. Let a = (a1, . . . , an) be the weight vector for an effective action of C× on Cn. Assume that ai < 0 for i ≤ k and ai > 0 for i > k, and moreover that a is stable (1 < k < n). Then 24 kQp=1 (ap − aq) nQq=k+1 kQp=1 nQq=k+1 p6=j q6=j kQp=1 p6=j nQq=k+1 q6=j (ap − aq)    γ3 = −6Sn−2(a) + (8E1(a))Sn−3(a) − (3E2(a) + 2E1(a)2)Sn−4(a) + E1(a)E2(a)Sn−5(a) + nXj=1(cid:18) 1 − gj 24 (cid:19) 4Sn−3(aj) − (5E1(aj))Sn−4(aj) + (E2(aj) + E1(aj)2)Sn−5(aj) (ap − aq) −2Sn−3(aj) + (2aj + E1(aj))Sn−4(aj) − ajE1(aj)Sn−5(aj) g2 j − 1 24 + nXj=1   + X1≤j<ℓ≤n Xζgj,ℓ =1 ζgj 6=1,ζgℓ 6=1 1 (1 − ζaj )(1 − ζaℓ) E1(aj,ℓ)Sn−5(aj,ℓ) − Sn−4(aj,ℓ) 2 kQp=1 p6=j,ℓ nQq=k+1 q6=j,ℓ (ap − aq) + X1≤j<ℓ≤n Xζgj,ℓ =1 (cid:18) ζaj (1 − ζaj )2(1 − ζaℓ)(cid:19) Sn−4(aj,ℓ) − ajSn−5(aj,ℓ) (ap − aq) ζgj 6=1,ζgℓ 6=1 kQp=1 nQq=k+1 (1 − ζaj )(1 − ζaℓ )2(cid:19) Sn−4(aj,ℓ) − aℓSn−5(aj,ℓ) ζaℓ p6=j,ℓ q6=j,ℓ (ap − aq) +(cid:18) p6=j,ℓ kQp=1 nQq=k+1 (ap − aq) Xζgj,ℓ,p =1 q6=j,ℓ ζgj,ℓ 6=1,ζgj,p 6=1 ζgℓ,p 6=1 + X1≤j<ℓ<p≤n −Sn−5(aj,ℓ,p) kQp=1 p6=j,ℓ,p nQq=k+1 q6=j,ℓ,p  1 (1 − ζaj )(1 − ζaℓ )(1 − ζap ) . THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS 19 When a is generic, ajaℓ(aℓ − 2ai) + nPj=1 j6=i (ai − aq) aℓap i (ai − aq) + nXℓ=1 ℓ6=i,j i −an−5 nQp=1 12 p6=i,j aiaj(2ai − aj)  aℓ(aℓ − 2ai) (ai − ap) γ3 = kXi=1 ℓ6=i,j nPℓ=1 nQq=1 q6=i an−5 j6=i 24 j,ℓ,p6=i gj − 1 3ajaℓap + nPj=1 i  P1≤j<ℓ<p≤n   nX1≤ℓ<p≤n −an−5 nQq=1 (ai − aj)−ai +  nPℓ=1 nQℓ=1 (ai − aℓ) nXj=1 an−5 i ℓ,p6=i,j q6=i,j ℓ6=i,j j6=i 2 4 ℓ6=i,j kXi=1 + + g2 j − 1 24 aℓ   + kXi=1 X1≤j<ℓ≤n j,ℓ6=i Xζgj,ℓ =1 ζgj 6=1,ζgℓ 6=1 1 (1 − ζaj )(1 − ζaℓ ) nXp=1 p6=i,j,ℓ 2  an−5 i ap q6=i,j,ℓ (ai − aq) nQq=1 (1 − ζaj )(1 − ζaℓ)2(cid:19) ζaℓ(ai − aℓ)  + an−5 i (ai − ap) Xζgj,ℓ =1 Xζgj,ℓ,p =1 nQp=1 kXi=1 X1≤j<ℓ<p≤n p6=i,j,ℓ j,ℓ,p6=i ζgj 6=1,ζgℓ 6=1 ζgj,ℓ 6=1,ζgj,p 6=1 ζgℓ,p 6=1 + (cid:18) ζaj (ai − aj) (1 − ζaj )2(1 − ζaℓ ) + −an−5 (1 − ζaj )(1 − ζaℓ )(1 − ζap ) i . (ai − aq) nQq=1 q6=i,j,ℓ,p As complicated as the expressions in Theorems 6.4 and 6.5 appear, it is important to note that they become vastly simpler when one imposes mild coprime hypotheses on the weights. For instance, if we assume that no set of n− 2 weights has a nontrivial common divisor, then each gj = gj,ℓ = 1. Hence in both expressions for γ2, all but the first line vanishes. In particular, the generalized Fourier-Dedekind sum no longer appears. The same holds for γ3 if we assume that no set of n− 3 weights has a common divisor. To indicate one application of these formulas, note that the γm(A) can be expressed in terms of the degrees of the elements of a Hironaka decomposition of A; see Appendix A. Hence, Theorems 6.2, 6.3, 6.4, and 6.5 can be used to determine bounds on these degrees. To illustrate this in a simple case, assume a is a generic weight vector with k = 1. Examining Equation (6.3), it is easy to see that γ0(a) ≤ 1/2. From Equation (A.1), it then follows that the number of module generators in the Hironaka decomposition, so-called secondary invariants, is bounded by half the product of the degrees of elements of a homogeneous system of parameters, the primary invariants. While we will not pursue this line of reasoning further here, it would be interesting to conduct numerical experiments to identify 20 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON if such comparisons yield strong restrictions on the number and degrees of primary and secondary invariants. We end this section with illustrations of Theorems 6.2 and 6.3 for small values of n. Example 6.6. Suppose n = 2. Then stability implies k = 1 so that a = (a1, a2) with a1 < 0 and a2 > 0. Then Theorems 6.2, 6.3, 6.4, and 6.5 yield γ0(a) = −1 and γ3(a) = , γ2(a) = , γ1(a) = , . 1 + a1 − a2 2(a1 − a2) 1 − (a1 − a2)2 12(a1 − a2) 1 − (a1 − a2)2 24(a1 − a2) a1 − a2 Example 6.7. If n = 3, multiplying by −1 if necessary, we may assume k = 1. Then a = (a1, a2, a3) with a1 < 0 and a2, a3 > 0. The first two Laurent coefficients are in this case given by −a1 , and γ1(a) = 2a1 + (a3 − a1) gcd(a1, a2) + (a2 − a1) gcd(a1, a3) . 2(a1 − a2)(a1 − a3) Of course, if a1 is assumed coprime to both a2 and a3, this simplifies to γ1(a) = (a2 + a3)/(2(a1 − a2)(a1 − a3)). In this case, g1,2 = a3, so unless each weight is ±1, γ2(a) always involves sums of the γ0(a) = (a1 − a2)(a1 − a3) formP 1/((1 − ζai )(1 − ζaj )). Example 6.8. If n = 4, then up to multiplying by −1, k is either 1 or 2. If k = 1, then −a2 1 , and (a1 − a2)(a1 − a3)(a1 − a4) a1(cid:0)3a1 − (a1 − a4) gcd(a1, a2, a3) − (a1 − a3) gcd(a1, a2, a4) − (a1 − a2) gcd(a1, a3, a4)(cid:1) 2(a1 − a2)(a1 − a3)(a1 − a4) . γ0(a) = γ1(a) = If k = 2, then γ0(a) = γ1(a) = a1a2(a3 + a4) − (a1 + a2)a3a4 (a1 − a3)(a1 − a4)(a2 − a3)(a2 − a4) 3(cid:0)(a1 + a2)a3a4 − a1a2(a3 + a4)(cid:1) 2(a1 − a3)(a1 − a4)(a2 − a3)(a2 − a4) + , and a3(a1 − a4)(a2 − a4) gcd(a1, a2, a3) + (a1 − a3)(a2 − a3)a4 gcd(a1, a2, a4) 2(a1 − a3)(a1 − a4)(a2 − a3)(a2 − a4) a1(a2 − a3)(a2 − a4) gcd(a1, a3, a4) + a2(a1 − a3)(a1 − a4) gcd(a2, a3, a4) . 2(a1 − a3)(a1 − a4)(a2 − a3)(a2 − a4) − 7. The a-invariant and the Gorenstein property Recall from Section 2 that the a-invariant of C[Cn]C× a is given by the degree of Hilba(t), i.e. the degree of the numerator minus the degree of the denominator (also equal to the pole order at infinity). Hence, the a-invariant can be computed for specific a using the algorithm described in Section 4. However, if C[Cn]C× a is Gorenstein, then we may use Equation (2.3) and Theorems 6.2 and 6.3 to give an explicit formula for a(C[Cn]C× a ) in terms of the weights as follows. Corollary 7.1. Let a = (a1, . . . , an) be the weight vector for an effective action of C× on Cn. Assume that ai < 0 for i ≤ k and ai > 0 for i > k, and moreover that a is stable (1 < k < n). If the invariant ring C[Cn]C× a is Gorenstein, then the a-invariant is given by (7.1) a(C[Cn]C× a ) = 1 Sn−2(a)E1(a)Sn−3(a) + kXj=1 nXj=k+1 + (1 − gj)Sn−2(aj) (1 − gj)Sn−2(aj) (aj − aq) nYq=k+1 (ap − aj) − n, kYp=1 THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS 21 where gj and aj are defined as in Section 6. In particular, if every collection of n − 1 weights has no nontrivial common factor, then (7.2) a(C[Cn]C× a ) = E1(a)Sn−3(a) Sn−2(a) − n. When a is generic, we can express Equation (7.1) as (7.3) a(C[Cn]C× a ) = kPi=1 (−1)i−1an−3 i (ap − aq) nPj=1 Q1≤p<q≤n Q1≤p<q≤n kPi=1 (−1)ian−2 p,q6=i p,q6=i j6=i i aj − (gj − 1)(ci − cj) . (ap − aq) The proof of Equation (7.1) is a simple computation using Equations (6.2) and (6.7); Equation (7.3) is similarly derived using Equations (6.3), (6.10), and (6.12). The fact that Sn−2(a) 6= 0 so that this expression is defined is a consequence of the fact that γ0 6= 0, see Theorem 6.2. Example 7.2. Suppose n = 2 so that a = (a1, a2) with a1 < 0 and a2 > 0. Then using Example 6.6, 2γ1(a) γ0(a) = a2 − a1 − 1 so that if C[Cn]C× a is Gorenstein, a(C[Cn]C× a ) = − 2γ1(a) γ0(a) − dim(C[Cn]C× a ) = a1 − a2. Example 7.3. If n = 3 and k = 1 so that a = (a1, a2, a3) with a1 < 0 and a2, a3 > 0, then using Example 6.7, we have 2γ1(a) γ0(a) = − 2a1 + (a3 − a1) gcd(a1, a2) + (a2 − a1) gcd(a1, a3) a1 so that if C[Cn]C× a is Gorenstein a(C[Cn]C× 2γ1(a) γ0(a) − dim(C[Cn]C× Of course, as the a-invariant a(C[Cn]C× a ) = − a ) = (a3 − a1) gcd(a1, a2) + (a2 − a1) gcd(a1, a3) a1 . a ) and dimension dim(C[Cn]C× a ) are always integers, an immediate consequence of Equation (2.3) is the following. Corollary 7.4. Let a = (a1, . . . , an) be the weight vector for an effective action of C× on Cn. Assume that ai < 0 for i ≤ k and ai > 0 for i > k, and moreover that a is stable (1 < k < n). If the invariant ring C[Cn]C× a is Gorenstein, then 2γ1(a)/γ0(a) ∈ Z. The converse of Corollary 7.4 is false, which we illustrate with the following. Example 7.5. Let a = {−1,−2, 1, 14}. Using the algorithm described in Section 4, one computes that Hilba(t) = 1 + t3 + t6 + 2t9 + t10 + t11 + 2t12 + t13 + t14 + t15 , (1 − t2)(1 − t8)(1 − t15) with γ0(a) = 1/20 and γ1(a) = 3/40 so that 2γ1(a)/γ0(a) = 3. However, if C[Cn]C× then Equation (2.3) would imply that a(C[Cn]C× Hilba(t). Moreover, by inspection, Hilba(t) does not satisfy Equation (2.2) for any integer a. a were Gorenstein, a ) = 6, which does not coincide with the degree of 22 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON For a specific a with small weights, the most direct way of determining whether C[Cn]C× Though we have not identified a counterexample to the converse of Corollary 7.4 with n = 3, they seem to be common when n = 4 and k = 2; other examples include (−1,−2, 4, 8); (−1,−2, 5, 6); (−1,−3, 1, 27); (−1,−3, 2, 9); (−1,−3, 3, 9); (−1,−3, 4, 6); (−1,−3, 12, 23); and (−1,−4, 2, 2). a is Goren- stein using the results in this paper is to compute Hilba(t) using the algorithm of Section 4 and testing to see if it satisfies Equation (2.2). However, as the weights become large, Corollary 7.4 can be a sur- prisingly useful way of quickly determining that the invariant ring associated to a weight vector a is not Gorenstein. For instance, for the weight vector a = (−501, 500, 503), a quick computation by hand using Corollary 7.4 demonstrates that C[Cn]C× a is not Gorenstein, as 2γ1(a)/γ0(a) = −1003/501 /∈ Z. However, our implementation of the algorithm in Section 4 took over six hours on a desktop PC to compute Hilb(−501,500,503)(t) and yield the same conclusion. We conclude this section with some illustrations of how Theorem 3.3 can be used to identify classes of weight vectors with Gorenstein invariants. We first give a quick proof of the known fact that when n = 2, the invariant ring C[Cn]C× Corollary 7.6. Suppose a = (a1, a2) ∈ Z2 with a1 < 0 and a2 > 0. Then C[Cn]C× ring generated by xa2 a is always polynomial; see [27]. a is a polynomial 2 . In particular, C[Cn]C× a is Gorenstein with a-invariant a1 − a2. 1 xa1 Proof. Assume for simplicity that gcd(a1, a2) = 1, and then by Theorem 3.3, 1 Hilba(t) = Xζ−a1 =1 a1 Xζ−a1 =1 = −1 −a1(1 − ζa2 t(a1−a2)/a1) 1 1 − ζt(a1−a2)/a1 , where the second equation is by reordering terms, as ζa2 simply permutes the set of a1st roots of unity. By [8, Theorem 3.1], this is equal to 1/(1 − ta2−a1). Then as xa2 a -invariant, it then generates C[Cn]C× (cid:3) a , and the result follows. is clearly C× 1 xa1 2 Finally, for general n, we indicate a large class of a for which C[Cn]C× a is Gorenstein. Corollary 7.7. Let a ∈ Zn with each ai 6= 0 and suppose k = 1 so that a1 < 0 and ai > 0 for i > 1. j=2 aj, then C[Cn]C× a is Gorenstein with a-invariant (Pn j=1 aj)/a1 − n. Proof. If k = 1, then the weight vector is always generic, and Theorem 3.3 yields If a1 divides Pn Applying Equation (2.2), we have Hilba(1/t) = Xζ−a1 =1 Hilba(t) = Xζ−a1 =1 −a1 1 . (1 − ζaj t(a1−aj )/a1 ) nQj=2 1 (1 − ζaj t−(a1−aj )/a1 ) −a1 nQj=2 j=2 aj /a1 Xζ−a1 =1 = tn−1−Pn j=2 aj ζ− Pn (ζ−aj t(a1−aj )/a1 − 1) . Reordering terms by replacing ζ with ζ−1 yields Hilba(1/t) = (−1)n−1tn−1−(Pn −a1 nQj=2 j=2 aj )/a1 Xζ−a1 =1 −a1 j=2 aj ζPn (1 − ζaj t(a1−aj )/a1 ) , nQj=2 THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS which, if a1 dividesPn j=2 aj, is equal to (−1)n−1tn−(Pn j=1 aj )/a1 Hilba(t). 23 (cid:3) Note that the condition of Corollary 7.7 is sufficient though not necessary: the ring of invariants corresponding to the weight vector a = (−3, 1, 3) has Hilbert series 1 and hence is Gorenstein with a-invariant −6. (1 − t2)(1 − t4) Appendix A. The Laurent expansion of the Hilbert series of a Cohen-Macaulay algebra Let A = ⊕∞ i=0Ai be a Cohen-Macaulay algebra over the field K, let d = dim(A), and let x1, x2, . . . , xd be a homogeneous system of parameters with respective degrees α1, α2, . . . , αd. Furthermore, let β1, β2, . . . , βr be the degrees of a basis of the free K[x1, x2, . . . , xd]-module A. Such a choice of homo- geneous system of parameters and module generators is called a Hironaka decomposition. The Hilbert series of A can be written as HilbA(t) = ∞Xm=0 dimK(Am) tm = Pr Qd i=1 tβi , j=0(1 − tαj ) see [18, Equation 3.28] or [22, Corollary 2.3.4]. Our aim in this section is to elaborate a formula for the Laurent expansion HilbA(t) = γm(A)(1 − t)m−d ∞Xm=0 using symmetric functions in the α1, α2, . . . , αd and β1, β2, . . . , βr, respectively. For the αi's we use the elementary symmetric functions, which we denote ek := Ek(α1, α2, . . . , αd) = while for the β's, we use the power sums X0≤j1<j2<···<jk≤d rXi=1 βk i , αj1 αj2 ··· αjd , k ≥ 0, 0 ≤ k ≤ r. pk := Pk(β1, β2, . . . , βr) = It is well-known [18, Equations 3.29 and 3.30] that (A.1) γ0(A) = r ed , and γ1(A) = r 2 (e1 − d) − p1 . ed In order to deduce a generalization of these formulas, we recall the definition of the Todd polynomials tdj using the generating function dYi=1 xαi 1 − e−xαi =: ∞Xj=0 = 1 + tdj(e1, e2, . . . , ed) xj e1 2 x + e2 1 + e2 12 x2 + e1e2 24 x3 + −e4 1 + 4e2 1e2 + e1e3 + 3e2 720 2 − e4 x4 + ··· , see [12, Section 1.7]. Substituting x = − log t we find ed(− log t)d dYi=1 1 1 − tαi = ∞Xj=0 tdj(e1, e2, . . . , ed) (− log t)j. 24 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON For k ∈ Z, we define λm by the expansion (− log t)k =: ∞Xm=0 λm(k)(1 − t)m+k and observe that λm(k) is a polynomial in k of degree m. The first few λm(k) are λ0(k) = 1, λ1(k) = k 2 , λ2(k) = k 24 (3k + 5), λ3(k) = k 48 (k2 + 5k + 6), λ4(k) = k 5760 (15k3 + 150k2 + 485k + 502). Note that the λm(k) are determined implicitly by the recursion (m + k)λm(k) − kλm(k − 1) = (m + k − 1)λm−1(k), λ0(k) = 1, λm(0) = 0 for m > 1. Defining we find ϕm := ϕm(e1, e2, . . . , ed) = The first few ϕm are ϕ0 = 1, ϕ1 = dYi=1 1 1 − tαi = 1 ed 1 2 (e1 − d), ϕ2 = λm−k(k − d) tdk(e1, e2, . . . , ed), mXk=0 ∞Xm=0 12(cid:18)e2 + e2 1 ϕm(e1, e2, . . . , ed) . (1 − t)d−m 1 − 3(d − 1)e1 + d 2 2 (3d − 5)(cid:19) , (cid:19) . ϕ3 = 1 24(cid:18)e2e1 − (d − 2)(e2 + e2 1) + d − 1 2 (3d − 8)e1 − d(d − 2)(d − 3) In order to incorporate the numerator Pi tβi, we recall the definition of the Stirling numbers of the first kind s(m, k) in terms of falling factorials x(x − 1)(x − 2)··· (x − m + 1) = s(m, k)xk, mXk=0 (−1)j(cid:18)βi j(cid:19)(1 − t)j , see [1, Section 24.1.3]. Using tβi = rXi=1 we arrive at rXi=1 βiXj=0 rXi=1(cid:0)1 − (1 − t)(cid:1)βi = ϕmPr i=1 tβi (1 − t)d−m 1 ed HilbA(t) = ∞Xm=0 ∞Xm=0 ∞Xℓ=0 ∞Xℓ=0Pℓ = = = 1 ed 1 ed 1 ed (1 − t)d−m ϕm 1 (1 − t)d−ℓ (−1)j j=0 j(cid:19)(1 − t)j j(cid:19)ϕℓ−j (−1)j(cid:18)βi rXi=1 ∞Xj=0 (−1)j(cid:18)βi βiXj=0 rXi=1 j! ϕℓ−jPj (1 − t)d−ℓ k=0 s(j, k)pk . THE HILBERT SERIES AND a-INVARIANT OF CIRCLE INVARIANTS 25 The terms corresponding to ℓ = 0, 1 reproduce the formulas in Equation (A.1). We evaluate the terms corresponding to ℓ = 2: γ2(A) = = 1 ed(cid:18)rϕ2 − ϕ1p1 + 12ed(cid:18)r(cid:18)e2 + e2 1 1 2 (p1 + p2)(cid:19) 1 − 3(d − 1)e1 + d 2 (3d − 5)(cid:19) + 6p2 + 6p1(d − 1 − e1)(cid:19) , and the terms corresponding to ℓ = 3: γ3(A) = = (2p1 − 3p2 + p3)(cid:19) r 1 (p2 − p1) − ϕ1 2 ed(cid:18)rϕ3 − ϕ2p1 + 24ed(cid:18)e2e1 − (d − 2)(e2 + e2 12ed(cid:18)e2 + e2 1 − 3(d − 1)e1 + 1) + p1 − 1 6 d − 1 2 (3d − 8)e1 − d 2 (3d − 5)(cid:19) + 1 4ed d(d − 2)(d − 3) 2 (cid:19) (e1 − d)(p2 − p1) − 1 6ed (2p1 − 3p2 + p3). References 1. Milton Abramowitz and Irene A. Stegun, Handbook of mathematical functions with formulas, graphs, and mathe- matical tables, National Bureau of Standards Applied Mathematics Series, vol. 55, For sale by the Superintendent of Documents, U.S. Government Printing Office, Washington, D.C., 1964. 2. Matthias Beck, Ricardo Diaz, and Sinai Robins, The Frobenius problem, rational polytopes, and Fourier-Dedekind sums, J. Number Theory 96 (2002), no. 1, 1 -- 21. 3. Matthias Beck and Sinai Robins, Computing the continuous discretely, second ed., Undergraduate Texts in Math- ematics, Springer, New York, 2015, Integer-point enumeration in polyhedra, With illustrations by David Austin. 4. D. J. Benson, Polynomial invariants of finite groups, London Mathematical Society Lecture Note Series, vol. 190, Cambridge University Press, Cambridge, 1993. 5. D. J. Benson and W. W. Crawley-Boevey, A ramification formula for Poincar´e series, and a hyperplane formula for modular invariants, Bull. London Math. Soc. 27 (1995), no. 5, 435 -- 440. 6. Winfried Bruns and Jurgen Herzog, Cohen-Macaulay rings, Cambridge Studies in Advanced Mathematics, vol. 39, Cambridge University Press, Cambridge, 1993. 7. Harm Derksen and Gregor Kemper, Computational invariant theory, Invariant Theory and Algebraic Transforma- tion Groups, I, Springer-Verlag, Berlin, 2002, Encyclopaedia of Mathematical Sciences, 130. 8. Ira M. Gessel, Generating functions and generalized Dedekind sums, Electron. J. Combin. 4 (1997), no. 2, Research Paper 11, approx. 17 pp. (electronic), The Wilf Festschrift (Philadelphia, PA, 1996). 9. Shiro Goto and Keiichi Watanabe, On graded rings. I, J. Math. Soc. Japan 30 (1978), no. 2, 179 -- 213. 10. David A. Harville, Matrix algebra from a statistician's perspective, Springer-Verlag, New York, 1997. 11. Hans-Christian Herbig and Christopher Seaton, The Hilbert series of a linear symplectic circle quotient, Exp. Math. 23 (2014), no. 1, 46 -- 65. 12. F. Hirzebruch, Topological methods in algebraic geometry, Third enlarged edition. New appendix and transla- tion from the second German edition by R. L. E. Schwarzenberger, with an additional section by A. Borel. Die Grundlehren der Mathematischen Wissenschaften, Band 131, Springer-Verlag New York, Inc., New York, 1966. 13. M. Hochster, Rings of invariants of tori, Cohen-Macaulay rings generated by monomials, and polytopes, Ann. of Math. (2) 96 (1972), 318 -- 337. 14. Melvin Hochster and Joel L. Roberts, Rings of invariants of reductive groups acting on regular rings are Cohen- Macaulay, Advances in Math. 13 (1974), 115 -- 175. 15. George Kempf, The Hochster-Roberts theorem of invariant theory, Michigan Math. J. 26 (1979), no. 1, 19 -- 32. 16. I. G. Macdonald, Symmetric functions and Hall polynomials, second ed., Oxford Classic Texts in the Physical Sciences, With contribution by A. V. Zelevinsky and a foreword by Richard Stanley, Reprint of the 2008 paperback edition. 17. V. L. Popov, Groups, generators, syzygies, and orbits in invariant theory, Translations of Mathematical Mono- graphs, vol. 100, American Mathematical Society, Providence, RI, 1992, Translated from the Russian by A. Martsinkovsky. 18. V. L. Popov and `E. B. Vinberg, Invariant theory, Algebraic geometry. IV, Encyclopaedia of Mathematical Sciences, vol. 55, Springer-Verlag, Berlin, 1994, Linear algebraic groups. Invariant theory, A translation of ıt Algebraic geometry. 4 (Russian), Akad. Nauk SSSR Vsesoyuz. Inst. Nauchn. i Tekhn. Inform., Moscow, 1989 edited by A. N. Parshin and I. R. Shafarevich, pp. vi+284. 19. Bruce E. Sagan, The symmetric group, second ed., Graduate Texts in Mathematics, vol. 203, Springer-Verlag, New York, 2001, Representations, combinatorial algorithms, and symmetric functions. 26 L. E. COWIE, H.-C. HERBIG, D. HERDEN, AND C. SEATON 20. Gerald W. Schwarz, Lifting smooth homotopies of orbit spaces, Inst. Hautes ´Etudes Sci. Publ. Math. (1980), no. 51, 37 -- 135. 21. Richard P. Stanley, Hilbert functions of graded algebras, Advances in Math. 28 (1978), no. 1, 57 -- 83. 22. Bernd Sturmfels, Algorithms in invariant theory, Texts and Monographs in Symbolic Computation, Springer-Verlag, Vienna, 1993. 23. Emmanuel Tsukerman, Fourier-Dedekind sums and an extension of Rademacher reciprocity, Ramanujan J. 37 (2015), no. 2, 421 -- 460. , Certain invariant subrings are Gorenstein. II, Osaka J. Math. 11 (1974), 379 -- 388. 24. Keiichi Watanabe, Certain invariant subrings are Gorenstein. I, Osaka J. Math. 11 (1974), 1 -- 8. 25. 26. David L. Wehlau, A proof of the Popov conjecture for tori, Proc. Amer. Math. Soc. 114 (1992), no. 3, 839 -- 845. 27. 28. Wolfram Research, Mathematica edition: Version 7.0, (2008), http://www.wolfram.com/mathematica/. , When is a ring of torus invariants a polynomial ring?, Manuscripta Math. 82 (1994), no. 2, 161 -- 170. Department of Mathematics, 303 Lockett Hall, Louisiana State University, Baton Rouge, LA 70803 E-mail address: [email protected] Departamento de Matem´atica Aplicada, Av. Athos da Silveira Ramos 149, Centro de Tecnologia - Bloco C, CEP: 21941-909 - Rio de Janeiro, Brazil E-mail address: [email protected] Department of Mathematics, Baylor University, One Bear Place #97328, Waco, TX 76798-7328, USA E-mail address: Daniel [email protected] Department of Mathematics and Computer Science, Rhodes College, 2000 N. Parkway, Memphis, TN 38112 E-mail address: [email protected]
1907.00947	1	1907	2019-07-01T17:33:50	Automorphism-Liftable Modules	[ "math.RA" ]	In this paper, we describe all automorphism-liftable torsion modules over non-primitive hereditary Noetherian prime rings. We also study automorphism-liftable non-torsion modules over not necessarily commutative Dedekind prime rings	math.RA	math	Automorphism-Liftable Modules A.A. Tuganbaev National Research University "MPEI" Lomonosov Moscow State University e-mail: [email protected] Abstract. In this paper, we describe all automorphism-liftable torsion mod- ules over non-primitive hereditary Noetherian prime rings. We also study automorphism-liftable non-torsion modules over not necessarily commuta- tive Dedekind prime rings. The work is supported by Russian Scientific Foundation, project 16-11-10013. Key words: automorphism-liftable module, hereditary Noetherian prime ring, torsion module 2000 MATHEMATICS SUBJECT CLASSIFICATION 16D40; 16D80; 16N80 1 Introduction All considered rings are associative and contain the non-zero identity element. Writing expressions of the form "A is a non-primitive ring or a Noetherian ring" we mean that A is not a right and left primitive ring or the both AA and AA are Noetherian. 1.1. endomorphism-liftable, and quasi-projective modules. Automorphism-liftable, strongly automorphism-liftable, (resp., said to be automorphism-liftable1 A module M is strongly automorphism-liftable) if for any epimorphism h : M → ¯M and every au- tomorphism ¯f of the module ¯M , there exists an endomorphism (resp., auto- morphism) f of the module M with ¯f h = hf . A module M is said to be endomorphism-liftable if for any epimorphism h : M → ¯M and every endomorphism ¯f of the module ¯M , there exists an endomorphism f of the module M with ¯f h = hf . In the above definitions, without loss of generality, we can assume that ¯M is an arbitrary factor mod- ule of the module M and h : M → ¯M is the natural epimorphism. 1The notion of an automorphism-liftable module is dual to the notion of an automorphism-extendable module studued in [14] and [15]. A module M is said to be automorphism-extendable if every automorphism of any its submodule can be extended to an endomorphism of the module M . 1 is clear It endomorphism-liftables modules are automorphism-liftable. strongly automorphism-liftable modules and all that all Endomorphism-liftable modules and Abelian groups were studied in many papers under various names; e.g., see [3], In particular, endomorphism-liftable Abelian groups were studied in [6] and [3]; endomorphism-liftable modules over non-primitive hereditary Noethe- rian prime rings were studied in [11], [12]. [12], [10], [6], [11], [15]. 1.2. Automorphism-liftable Z-modules. Any quasi-cyclic Abelian group Z(p∞) (automorphism-liftable) non-quasi-projective Z-module. is an endomorphism-liftable The ring of integers Z is an automorphism-liftable Z-module which is not Indeed, let ¯f be an automorphism of the strongly automorphism-liftable. simple Z-module Z/5Z such that ¯f multiplies all elements of this module by 3. Since the only non-identity automorphism of the module ZZ coincides with the multiplication by −1, the projective module ZZ is not strongly automorphism-liftable. i.e., strongly automorphism-liftable Z-modules. A.P.Mishina [6] completely described strongly automorphism-liftable Abelian groups, It follows from this description that strongly automorphism-liftable Z-modules are torsion automorphism-liftable Z-modules. In [1], A.N.Abyzov and T.C.Quynh de- scribed torsion automorphism-liftable2 Z-modules; it follows from this de- scription and results of A.P.Mishina [6] that the strongly automorphism- liftable Z-modules coincide with torsion automorphism-liftable Z-modules. 1.3. Non-primitive hereditary Noetherian prime rings. The ring Z is a very partial case of a non-primitive hereditary Noetherian prime ring.3 A ring A is said to be right bounded (resp., left bounded) if every its essential right (resp., left) ideal contains a non-zero ideal of the ring A. If A is a non-primitive hereditary Noetherian prime ring, then A is not a right or left Artinian; see [4]. Every hereditary Noetherian prime ring A is a primitive ring or a bounded ring and if A is a primitive bounded ring, then A is a simple Artinian ring; see [4]. Let A be a Noetherian prime ring. It is well known that the ring A has the simple Artinian classical ring of fractions Q. An ideal B of the ring A is called an invertible ideal if there exists a subbimodule B−1 of the bimodule AQA such that BB−1 = B−1B = A. The maximal elements of the set of all 2In this paper, automorphism-liftable modules are called dually automorphism- extendable. Automorphism-liftable modules are also studied in [7]. 3A module is said to be hereditary if all its submodules are projective. 2 proper invertible ideals of the ring A are called maximal invertible ideals of the ring A. The set of all maximal invertible ideals of the ring A is denoted by P(A). If P ∈ P(A), then the submodule {m ∈ M mP n = 0, n = 1, 2, . . .} is called the P -primary component of the module M; it is denoted by M(P ). If M = M(P ) for some P ∈ P(A), then M is called a primary module or a P -primary module. In connection with 1.2, we prove Theorem 1.4 which is the first main re- sult of this paper. This theorem generalizes the description of torsion automorphism-liftable Z-modules from [1] for the case of singular modules over non-primitive hereditary Noetherian prime rings.4 1.4. Theorem. If A is a non-primitive hereditary Noetherian prime ring and M is a singular right A-module, then M is automorphism-liftable if and only if every P -primary component5 M(P ) of the module M is either a projective A/r(M(P ))-module or a uniserial injective module M(P ) such that all proper submodules are cyclic and form a countable chain 0 = x0R ( x1R ( . . . , all subsequent factors of this chain are simple modules and there exists a pos- itive integer n such that xkA ∼= xk+n/xnA and M(P )/xkA ∼= M(P )/xk+nA for all k = 0, 1, 2, . . .. 1.5. Remark. In [12], arbitrary endomorphism-liftable modules over non- primitive hereditary Noetherian prime rings are described. 1.6. Non-primitive Dedekind prime rings. A hereditary Noetherian prime ring A is called a Dedekind prime ring (see [5, §5.2]) if every its non- zero ideal is invertible in the simple Artinian ring of fractions of the ring A. Any hereditary Noetherian prime PI ring is a bounded ring. In particular, all commutative Dedekind domains (e.g., the ring Z) and matrix rings over com- mutative Dedekind domains are non-primitive Dedekind prime rings. Other examples of Dedekind prime rings are given in [5, § 5.2, § 5.3]. The second main result of this paper is Theorem 1.7, where the description of singular automorphism-liftable modules from Theorem 1.4 is specified in the case, where A is a non-primitive Dedekind prime ring. 1.7. Theorem. If A is a non-primitive Dedekind prime ring and M is a singular right A-module, then the module M is automorphism-liftable if 4Over any Noetherian prime ring A, the singular modules coincide with the torsion modules, where a module MA is said to be singular (resp., torsion if the annihilator of every its element is an essential right ideal of the ring A (resp., contains a non-zero-divisor of the ring A). 5Necessary definitions are given at the end of Introduction. 3 and only if every P -primary component M(P ) of the module M is either the direct sum of isomorphic cyclic uniserial modules of finite length or a uniserial injective module M(P ) such that all proper submodules are cyclic and form a countable chain 0 = x0A ( x1A ( . . . such that all subsequent factors of this chain are isomorphic simple modules and M(P )/xkA ∼= M(P )/xk+1A for all k = 0, 1, 2, . . .. 1.8. Theorem [8]. If R is a non-primitive Dedekind prime ring with simple Artinian ring of fractions Q, then the following conditions are equivalent. 1) QR is a quasi-projective module. 2) R = Dn, where D is a local principal right (left) domain which is complete in the topology defined by powers of the Jacobson radical J(D). In this case, R is a Dedekind prime ring, J(R) is a maximal ideal, and RQ is a quasi-projective module. A ring R is called a special Dedekind prime ring if R satisfies Theorem 1.8. The third main result of this paper is Theorem 1.9, where all automorphism- liftable A-modules are described in the case, where A is a non-primitive Dedekind prime ring with 1 2 . 1.9. Theorem. Let A be a non-primitive Dedekind prime ring with 1 2 ∈ A. A right A-module M is automorphism-liftable if and only if one of the following three conditions holds: a) If M is a torsion module, then M is endomorphism-liftable if and only if every primary component of the module M is either an indecomposable injective module or a projective A/r(M)-module. b) If M is a mixed module, then M is endomorphism-liftable if and only if M = T ⊕F , where T is a torsion injective module such that all primary com- ponents are indecomposable and F is a finitely generated projective module. c) If M is a torsion-free module, then M is endomorphism-liftable if and only if either M is projective or A is a special ring with classical ring of fractions Q and M = E ⊕ F , where E is a minimal right ideal of the ring Q, F is a finitely generated projective module, and n is a positive integer. We give some necessary definitions and notation. Let A be a ring and M a right A-module. We denote by J(A) the Jacobson radical of A. We denote by r(X) the annihilator in the ring A of the subset X of the module M. 4 We denote by T (M) the set of all elements of M whose annihilators contain a non-zero-divisor; the set T (M) is called the torsion part of the module M. If T (M) = 0 (resp., 0 6= T (M) 6= M), then M is said to be torsion-free module (resp., mixed). Every module is a torsion module, or a torsion-free module, or a mixed module. A module M is said to be projective with respect to the module N (or N - projective) if for any epimorphism h : N → ¯N and every homomorphism ¯f : M → ¯N, there exists a homomorphism f : M → N with hf = ¯f . Thus, quasi-projective modules coincide with the modules which are projective with respect to itself. Clearly, in the definition of the relative projectivity, it is suf- ficient to consider only the case, where h : N → ¯N is the natural epimorphism of the module N onto its arbitrary factor module ¯N . A module is said to be uniserial if any two of its submodules are comparable with respect to inclusion. For a module M, a submodule X of M is said to be essential (in M) if X has the non-zero intersection with any non-zero submodule of the module M. A module M is said to be finite-dimensional if M does not contain an infinite direct sum of non-zero submodules. 2 Proof of Theorem 1.4 and Theorem 1.7 2.1. Lemma. If A is a ring and M is a right A-module, then M is an automorphism-liftable (resp., idempotent-liftable, quasi-projective,) A- module if and only if M is an automorphism-liftable (resp., idempotent- liftable; quasi-projective) A/r(M)-module. Lemma 2.1 is directly verified. 2.2. Lemma. Let M = ⊕i∈IMi be a module and N = ⊕i∈I (N T Mi) for any submodule N of the module M. Then: 1) Hom(Mi, Mj) = 0 for any i 6= j in I (therefore, all the Mi are fully invariant submodules in M); 2) if N, P and Q are three submodules of the module M, then the relation N = P + Q is equivalent to the property that N T Mi = P T Mi + Q T Mi for all i ∈ I; 3) the module M is automorphism-liftable if and only if each of the modules Mi are automorphism-liftable. Proof. 1) and 2). The assertions are proved in [13, Lemma 2.1(1),(2)]. 3. The assertion follows from 1) and 2). (cid:3) 5 2.3. Lemma [1, Proposition 6, Lemma 2]. If an automorphism-liftable module M is the direct sum of some modules X and Y , then X, Y are automorphism-liftable modules which are projective with respect to each other. 2.4. Lemma. If A is a non-primitive hereditary Noetherian prime ring and M is a primary right A-module, then the following conditions are equivalent. 1) M is an automorphism-liftable module; 2) M is an endomorphism-liftable module; 3) for any direct decomposition M = M1 ⊕ M2, the module M1 is projective with respect to the module M2; 4) M is a projective A/r(M)-module or an indecomposable injective A- module. Proof. The equivalence of conditions 2), 3) and 4) is proved in [12, Lemma 13]. 2) ⇒ 1). The assertion is always true. 1) ⇒ 2). The assertion follows from Lemma 2.4. (cid:3) 2.5. Theorem. Let A be a non-primitive hereditary Noetherian prime ring, M a torsion right A-module, {Mi} the set of all primary components of the module M. Then the module M is automorphism-liftable if and only if every primary component Mi of the module M is a projective A/r(M)-module or an indecomposable injective A-module. Proof. For an arbitrary torsion A-module M and every its submodule N, we have N = ⊕i∈I (N T Mi); see [13, Lemma 2.2(1)]. Therefore, our assertion follows from Lemma 2.4. (cid:3) 2.6. Remark. If A is a non-primitive hereditary Noetherian prime ring, then the structure of indecomposable injective torsion A-modules is known; e.g., see [8] and [9]. Namely, the indecomposable injective torsion A-modules coincide with the primary modules M such that all proper submodules of M are cyclic uniserial primary modules and form a countable chain 0 = x0R ( x1R ( . . . , all subsequent factors of this chain are simple modules and there exists a pos- itive integer n such that xkA ∼= xk+n/xnA and M(P )/xkA ∼= M(P )/xk+nA for all k = 0, 1, 2, . . .. If A is a non-primitive Dedekind prime ring, then n = 1. 2.7. Remark. If A is a non-primitive Dedekind prime ring and M is a torsion A-module, then M is quasi-projective if and only if every primary 6 component of the module M is a direct sum of isomorphic cyclic modules of finite length. [9, Theorem 15] 2.8. Completion of the proof of Theorem 1.4 and Theorem 1.7. Theorem 1.4 and Theorem 1.7 follow from Theorem 2.5, Remark 2.6 and Remark 2.7. (cid:3) 3 Proof of Theorem 1.9 3.1. Idempotent-liftable modules and π-projective modules. A module M is said to be idempotent-liftable if for any epimorphism h : M → ¯M and every idempotent endomorphism ¯f of the module ¯M , there exists an endomorphism f of the module M with ¯f h = hf . Without loss of generality, we can assume that ¯M is an arbitrary factor module of the module M and h : M → ¯M is the natural epimorphism. A module M is said to be π-projective if for any its submodules X and Y with X + Y = M, there exist endomorphisms f and g of the module M with f + g = 1M , f (M) ⊆ X and g(M) ⊆ Y (see [16, p.359]). The idempotent-liftable modules coincide with the π-projective modules; see Lemma 3.2. It is clear that every endomorphism-liftable module is idempotent-liftable. If A is a uniserial principal right (left) ideal domain with division ring of frac- tions Q which is not complete with respect to the J(A)-adic topology, then Q is an idempotent-liftable A-module which is not endomorphism-liftable. The class of all idempotent-liftable modules contains all modules such that all their factor modules are indecomposable, all projective modules, all uniserial modules, and all local modules. In particular, all cyclic modules over local rings and all free modules are idempotent-liftable. 3.2. Lemma. For a module M, the following conditions are equivalent. 1) M is an idempotent-liftable module. 2) M is a π-projective module. Proof. 1) ⇒ 2). Let M = X + Y be an idempotent-liftable module, N = X ∩ Y , ¯M = M/N, ¯X = X/N, ¯f , ¯g1 natural projections of the module ¯M = ¯X ⊕ ¯Y onto the components ¯X, ¯Y , respectively, and h the natural epimorphism from the module M onto ¯M . Since M is an idempotent-liftable module, there are two endomorphisms f , g1 of the module M such that ¯f h = hf , ¯g1h = hg1. It is easy to see that f M ⊆ X, g1M ⊆ Y . Since f + g1 coincides with the identity mapping on ¯M , we have (1M −f −g1)M ⊆ N. We 7 set g = lM − f . Then f M ⊆ X, gM ⊆ (1 − f − g1)M + g1M ⊆ N + Y = Y . 2) ⇒ 1). Let ¯M = M/N be an arbitrary factor module of the module M, h : M → ¯M the natural epimorphism, ¯f an idempotent endomorphism of the module ¯M , ¯X = X/N = ¯f ( ¯M), ¯Y = Y /N = (1 − ¯f )( ¯M ), where X, Y are complete pre-images of the modules ¯X and ¯Y in M, respectively. Then ¯M = ¯X ⊕ ¯Y , M = X + Y and N = X ∩ Y . Since M is a π-projective module, there exist homomorphisms f : M → X and g : M → Y with f + g = 1M . Then ¯f h = hf and M is an idempotent-liftable module. (cid:3) 1 2 3.3. Lemma. If A is a ring with and M is a right A-module, then every idempotent endomorphism of the module M is the sum of two automor- phisms of the module M; in particular, every automorphism-liftable (resp., automorphism-extendable6) right A-module is a idempotent-liftable (resp., idempotent-extendable7). Proof. Let f be an idempotent endomorphism of the module M. Then M = X ⊕ Y , where X = f (M) and Y = (1M − f )(M). We denote by u an automorphism of the module M such that u(x + y) = x − y for x ∈ X, y ∈ Y . Then f = 1 2 · u are automorphisms of the module M. (cid:3) 2 · u, where 1 2 · 1M + 1 2 · 1M and 1 3.4. Theorem [13, Theorem 1]. A module M over a non-primitive Dedekind prime ring A is π-projective if and only if one of the following three conditions holds: a) M is a torsion module such that every primary component is either an indecomposable injective module or direct sum of isomorphic cyclic modules of finite length; b) M = T ⊕ F , where T is a non-zero injective torsion module such that every primary component is an indecomposable module and F is a non-zero finitely generated projective module; c) M is a projective module or there exist two positive integers k and n such that the ring A is isomorphic to the ring of all k × k matrices Dk over some uniserial principal right (left) ideal domain D, M = X ⊕ Y , X is the finite direct sum of non-zero injective indecomposable torsion-free modules X1, . . . , Xn, Y is a finitely generated projective module, and either n = 1 or n ≥ 2 and D is a complete domain. 3.5. Theorem [12]. Let A be a non-primitive hereditary Noetherian prime 6A module M is said to be automorphism-extendable if every automorphism of any its submodule can be extended to an endomorphism of the module M . 7A module M is said to be idempotent-extendable if every idempotent endomorphism of any its submodule can be extended to an endomorphism of the module M . 8 ring and M a right A-module. a) If M is a torsion module, then M is endomorphism-liftable if and only if every primary component of the module M is either an indecomposable injective module or a projective A/r(M)-module. b) If M is a mixed module, then M is endomorphism-liftable if and only if M = T ⊕F , where T is a torsion injective module such that all primary com- ponents are indecomposable and F is a finitely generated projective module. c) If M is a torsion-free module, then M is endomorphism-liftable if and only if either M is projective or A is a special ring with classical ring of fractions Q and M = En ⊕ F , where E is a minimal right ideal of the ring Q, F is a finitely generated projective module, and n is a positive integer. 3.6. Theorem [11, Theorem 2]. If A is a non-primitive hereditary Noethe- rian prime ring, then a right A-module M is quasi-projective if and only if either M is a torsion module and every its primary component is a projective A/r(M)-module, or M is projective, or A is a special ring and M = E ⊕ F , where E is an injective finite-dimensional torsion-free module and F is a finitely generated projective module. 3.7. Lemma. If M is an automorphism-liftable module with local endo- morphism ring End M, then M is an endomorphism-liftable module. Proof. Let h : M → ¯M be an epimorphism and ¯f an endomorphism of the module ¯M. If ¯f is an automorphism of the module ¯M , then it follows from the assumption that there exists an endomorphism f of the module M that ¯f h = hf . We assume that ¯f is not an automorphism of the module M. By assump- tion, the ring End M is local. Therefore, 1 ¯M − ¯f is an automorphism of the module ¯M . Since M is an automorphism-liftable module, there exists an endomorphism g of the module M such that (1 ¯M − ¯f )h = hg. We denote by f the endomorphism 1 − g of the module M. Since h − 1 ¯M h = 0, we have hf = h − hg = h − 1 ¯M h + ¯f h = ¯f h. Therefore, M is an endomorphism-liftable module. (cid:3) 3.8. Theorem. A module M over a non-primitive Dedekind prime ring A is an automorphism-liftable, idempotent-liftable module if and only if one of the following three conditions holds: a) If M is a torsion module, then M is a endomorphism-liftable if and only if every primary component of the module M is either an indecomposable injective module or a projective A/r(M)-module. 9 b) If M is a mixed module, then M is a endomorphism-liftable if and only if M = T ⊕ F , where T -- torsion injective module such that all primary com- ponents are indecomposable and F is a finitely generated projective module. c) If M is a torsion-free module, then M is endomorphism-liftable if and only if either M is projective or A is a special ring with classical ring of fractions Q and M = E ⊕ F , where E is a minimal right ideal of the ring Q, F is a finitely generated projective module, and n is a positive integer. Proof. If one of the conditions a, b or c holds, then it follows from Theorem 3.5 that M is an endomorphism-liftable module. In particular, M is an automorphism-liftable, idempotent-liftable module. Now let M be an automorphism-liftable, idempotent-liftable module. By Lemma 3.2, M is a π-projective module. By Theorem 3.4, either one of the conditions a and b of our theorem holds or the following condition holds: there exist two positive integers k and n such that the ring A is isomorphic to the ring of all k ×k matrices Dk over some uniserial Noetherian domain D, M = X ⊕ Y , X is a finite direct sum of non-zero injective indecomposable torsion-free modules X1, . . . , Xn, Y is a finitely generated projective module and either n = 1 or n ≥ 2 and D is a complete domain. If n ≥ 2 and D is a complete domain, then A is a special ring, which is required. Now we assume that n = 1. Then A = D is a uniserial principal right (left) ideal domain with classical division ring of fractions Q = E. Since QA = EA is a direct summand of the automorphism-liftable module M, we have that QA is an automorphism-liftable module. Since the ring End QA isomorphic to the division ring Q, we have that QA is an endomorphism-liftable module, by Lemma 3.7. By Theorem 3.5, A is a special ring. (cid:3) 3.9. Completion of the proof of Theorem 1.9. By Lemma 3.3, ev- ery automorphism-liftable right A-module is an idempotent-liftable module. Therefore, Theorem 1.9 follows from Theorem 3.8. (cid:3) 4 Remarks and Open Questions 4.1. There are automorphism-liftable modules which are not endomorphism- liftable; see [2, Example 5.1] 4.2. Let A be a non-primitive Dedekind prime ring. ∈ A, then the automorphism-liftable A-modules coincide with the endomorphism-liftable A-modules by Theorems 1.9 and 3.5. Are there automorphism-liftable A- If 1 2 10 modules which are not endomorphism-liftable? 4.3. Let F be a field, A = F [[x]] be the formal power series ring, and let M = F ((x)) be the Laurent series ring. Then A is a commutative non- primitive Dedekind domain and M is a strongly automorphism-liftable A- module which is not singular. 4.4. For a non-primitive Dedekind prime ring A, describe automorphism- liftable A-modules and strongly automorphism-liftable A-modules. The an- swer to this question is related to the study invertible elements of the ring A and automorphisms of cyclic A-modules. 4.5. By Lemma 3.3, every automorphism-liftable A-module is idempotent- liftable provided A contains which are not idempotent-liftable? 1 2 . Are there automorphism-liftable modules References [1] Abyzov A.N., Quynh T.C. Lifting of automorphisms of factor modules // Commun. Algebra. - 2018. V. 46, no. 11. -- P. 5073-5082. [2] Guil Asensio P.A., Quynh T.C., Srivastava A.K. // Additive unit struc- ture of endomorphism rings and invariance of modules. -- Bull. Math. Sci. -- 2017. -- Vol. 7. -- P. 229-246. [3] Janakiraman S. Skew projective Abelian groups // Indag. Math. -- 1973. V.76, no. 3. -- P.233 -- 236. [4] Lenagan T.H., Bounded hereditary Noetherian prime rings // J. London Math. Soc. -- 1973. -- Vol. 6. -- P. 241 -- 246. [5] McConnell J. C., Robson J. C. Noncommutative Noetherian Rings. New York: Wiley-Interscience, 1987. [6] Mishina A. P. On automorphisms and endomorphisms of Abelian groups // Moscow University Mathematics Bulletin. -- 1972. -- no. 1. -- P. 62 -- 66. [7] Selvaraj C., Santhakumar A. S. Automorphism liftable modules // Com- ment. Math. Univ. Carolin. -- 2018. -- V. 59, no. 1. -- 35-44. [8] Singh S., Quasi-injective and quasi-projective modules over hereditary Noetherian prime rings // Canad. J. Math. -- 1974. -- Vol. 26, no. 5. -- P. 1173 -- 1185. 11 [9] Singh S. Modules over hereditary Noetherian prime rings // Can. J. Math. 1975. V. 27, No. 4. P. 867 -- 883. [10] Tuganbaev A. A. The structure of modules close to projective modules // Sbornik: Mathematics -- 1979. -- V. 35, no. 2. -- P. 219-228. [11] Tuganbaev A. A. Quasi-projective modules // Sib. math. j. 1980. Vol. 21, no. 3. P. 446 -- 450. [12] Tuganbaev A. A. Semiprojective modules // Sib. math. j. 1980. Vol. 21, no. 5. P. 725 -- 728. [13] Tuganbaev A. A. Modules over bounded Dedekind prime rings // Sbornik: Mathematics -- 2001. -- V. 192, no. 5. -- P. 705-724. [14] Tuganbaev A. A. Automorphisms of submodules and their extensions // Discrete Math. Appl. -- 2013. -- Vol. 23, no. 1. -- P. 115-124. [15] Tuganbaev A. A. Automorphism-extendable and endomorphism- extendable modules // J. Math. Sci. (New York) -- To appear. [16] Wisbauer R. Foundations of Module and Ring Theory. Philadelphia: Gordon and Breach, 1991. 12
1303.2683	1	1303	2013-03-11T20:48:39	Inequalities for generalized minors	[ "math.RA" ]	It is a classical result that the absolute value of any $k$-minor of an $r\times s$ real or complex matrix is bounded by the product of its first $k$ singular values. We generalize this statement to the context of real or complex simple Jordan pairs with generalized minors given by Jordan algebra determinants.	math.RA	math	INEQUALITIES FOR GENERALIZED MINORS BENJAMIN SCHWARZ Abstract. It is a classical result that the absolute value of any k- minor of an r × s real or complex matrix is bounded by the product of its first k singular values. We generalize this statement to the context of real or complex simple Jordan pairs with generalized minors given by Jordan algebra determinants. Introduction The goal of this paper is a generalization of the following statement. Let A be a real or complex r × s matrix, r ≤ s, with sin- gular values σ1 ≥ ⋯ ≥ σr ≥ 0. Then for 1 ≤ k ≤ r the absolute value of any k-minor of A is bounded by the product σ1⋯σk. This estimate can be proved using the Cauchy -- Binet formula, see e.g. [2, Theorem 4.1]. We generalize this statement to the context of Jordan theory. Let (V, V ) be a real or complex simple Jordan pair with positive involution. Then any element z ∈ V admits a singular value decomposition, and to any tripotent e ∈ V there is associated a unital Jordan algebra [e] ⊆ V and a Jordan algebra determinant ∆e. Theorem. Let (V, V) be a real or complex simple Jordan pair with positive involution and rank r. Let e ∈ V be a tripotent of rank k, and z ∈ V be an element with singular values σ1 ≥ ⋯ ≥ σr ≥ 0. Then ∆e(z) ≤ σ1⋯σk. Our proof is analytic and quite elementary. Let us describe the main idea in the classical context with V = k e∈ k r×s, k= R or C. Then an element r×s is tripotent if it satisfies the identity e= ee∗e, and it turns out that the corresponding Jordan algebra determinant ∆e is given by ∆e(z)= Det(1r +(e − z)e∗) (z ∈ k r×s). Moreover, the set Sk of all rank-k tripotents forms a real smooth com- r×s. Now the theorem follows from plain analysis pact submanifold of k of the smooth map f ∶ Sk → R given by f(e) = ∆e(z)2 with fixed z ∈ k 2010 Mathematics Subject Classification. Primary 15A45; Secondary 17C50, r×s. 15A18. Key words and phrases. Generalized minor, determinant, singular value decom- position, Jordan pair, Jordan algebra. 1 2 BENJAMIN SCHWARZ We note that choosing the tripotent e appropriately, ∆e(z) coincides with a given k-minor of z, see Example 1.3 for details. Therefore, the theorem indeed generalizes the classical statement. In the first section, we discuss generalized minors in the context of simple Jordan pairs over arbitrary fields of characteristic ≠ 2 in some detail. We note that even though the basic definition of generalized minors (resp. the corresponding Jordan algebra determinants) is widely known, our results seem to be new. In Section 2 we specialize to the case of real or complex simple Jordan pairs, prove our main theorem, and provide an application involving some representation theory on the space of polynomials on V , see Corollary 2.4. 1. Generalized minors In this section we consider Jordan pairs without fixing an involution. Let(V +, V −) be a finite dimensional simple Jordan pair over a field k of characteristic ≠ 2, and with quadratic maps Q± ∶ V ± → Hom(V ∓, V ±). As usual, we omit the signs and simply write Qz ∶= Q±(z) for z ∈ V ±. Moreover, for x, z ∈ V ± and y ∈ V ∓ the Jordan triple product {x, y, z} and the operators Dx,y and Qx,z are defined by polarization of Qx, {x, y, z} ∶= Dx,yz ∶= Qx,zy ∶= Qx+zy − Qxy − Qzy. 2 (e) ⊕ V ± 1 (e) ⊕ V ± 0 (e) with V ± The results of this section are independent of the choice of k. We refer to [5, 6] for a detailed introduction to Jordan pairs. We briefly recall some basic notions necessary for our purposes. An element e =(e+, e−) in (V +, V −) is idempotent, if e+ = Qe+e− and e− = Qe−e+. The corresponding Peirce decomposition is given by ν (e) ∶={z ∈ V ± De±,e∓z = ν z} . V ± = V ± For the following fix σ ∈ {+, −}. The Peirce 2-space V σ 2 (e) coin- cides with the principal inner ideal [eσ] ∶= Qeσ V −σ generated by eσ. Moreover, [eσ] forms a unital Jordan algebra with product x ○ y ∶= 2{x, e−σ, y} and unit element eσ, see also Remark 1.2. By definition, e of [eσ] is the exact denominator e(eσ) = 1. As usual, we e to a polynomial on all of V σ by firstly projecting onto [eσ] 1 (e) ⊕ V σ the Jordan algebra determinant ∆σ of the rational map z ↦ z−1, normalized to ∆σ expand ∆σ (along V σ this expansion is also denoted by ∆σ associated to e. 0 (e)) and then evaluating ∆σ e. By abuse of notation, e and called the generalized minor There is also a determinant attached to the Jordan pair (V +, V −), which we describe next. A pair (x, y) ∈ V σ × V −σ is quasi-invertible, if the Bergman operator Bx, y ∶= Id −Dx,y + QxQy is invertible. In this 1 case, xy ∶= B−1 x, y(x − Qxy) INEQUALITIES FOR GENERALIZED MINORS 3 is the quasi-inverse of (x, y). The exact denominator ∆ ∶ V + × V − → k of the rational map(x, y) ↦ xy, normalized to ∆(0, 0)= 1, is the Jordan pair determinant (also often called the generic norm, see [5, § 16.9]). There is a simple relation between Jordan algebra determinants and the Jordan pair determinant, which we already noted in [9] for the spe- cial case of complex simple Jordan pairs. To the best of our knowledge this relation has not been stated elsewhere. Proposition 1.1. Let e = (e+, e−) be an idempotent of (V +, V −). Then (1.1) ∆+ e(x)= ∆(e+ − x, e−), ∆− e(y)= ∆(e+, e− − y) decomposition with respect to e. By definition, ∆+ it suffices to prove the first formula in (1.1). If the Jordan pair is the if e+ is the unit element of J, then (1.1) is an immediate consequence of [5, § 16.3(ii)]. We reduce the general case to this Jordan algebra case. for all x∈ V +, y ∈ V −. Proof. Due to the duality of the Jordan pairs (V +, V −) and (V −, V +), one associated to a unital Jordan algebra J, i.e., (V +, V −)=(J, J), and Let x = x2 + x1 + x0 be the decomposition of x according to the Peirce e(x2). On the other hand, due to [5, § 3.5] the relation ∆(u, Qvw) = ∆(w, Qvu) holds for all u, w ∈ V +, v ∈ V −, so it follows that ∆(e+ − x, e−)= ∆(e+ − x2, e−) since e− = Qe−e+ and Qe−x = Qe−x2. Therefore it suffices to assume x= x2. We may consider ([e+], [e−]) as a subpair of (V +, V −). [e+]×[e−]. Moreover, due to [5, § 1.11] we may identify ([e+], [e−]) with the Jordan pair(J, J) with J = [e+] via the Jordan algebra isomorphism Qe− ∶ [e+] → [e−]. Now we are in the Jordan algebra case, and (1.1) Remark 1.2. We note that the Jordan algebra structure on [eσ] = 2 (e) is independent of e−σ, since for x= Qeσ u in [eσ], the fundamental Then its Jordan pair determinant coincides with the restriction of ∆ to V σ formula for the quadratic map yields follows from [5, § 16.3(ii)]. e(x) = ∆+ (cid:3) x2 = Qxe−σ = QQeσ ue−σ = Qeσ QuQeσ e−σ = Qeσ Queσ, 1 (e) and V σ e ∶ [eσ] → k does not and polarization of x2 also shows that x ○ y is independent of e−σ. It follows that the Jordan algebra determinant ∆σ depend on e−σ. However, its expansion to V σ depends on e−σ since the Peirce spaces V σ 0 (e) are dependent on e−σ. Example 1.3. Consider the simple Jordan pair(k s×r) with r ≤ s and quadratic maps given by Qxy = xyx. Then, idempotents are pairs of matrices (e+, e−) satisfying e+e−e+ = e− and e−e+e− = e+, and the Jordan pair determinant is given by ∆(x, y) = Det(1r − xy). For 1 ≤ k ≤ r and tuples I =(i1, . . . , ik), J =(j1, . . . , jk) with 1≤ i1 < ⋯< ik ≤ r r×s, k 4 BENJAMIN SCHWARZ and 1≤ j1 < ⋯< jk ≤ s let e+ be the r × s-matrix defined by if (i, j)=(iℓ, jℓ) for some 1≤ ℓ≤ k, (1.2) (e+)ij ∶=1 0 else, and set e− ∶= e⊺ to show that (e+, e−) is an idempotent, and the generalized minor +, the transpose matrix of e+. Then, it is straightforward ∆+ e(z) = Det(1r −(e+ − z)e−) (z ∈ k r×s) coincides with the usual minor corresponding to rows and columns given by I and J, respectively. Proposition 1.4. Let e, c∈(V +, V −) be idempotents. If [e+]= [c+], then (i) ∆+ (ii) ∆− (iii) ∆+ e(x) = ∆+ e(y)= ∆− c(e+) ⋅ ∆− e(c+) ⋅ ∆+ e(c−) ⋅ ∆− c(e−)= 1. c(x) for all x∈ [e+], c(y) for all y ∈ V −, If [e−]= [c−], the same formulas hold when + and − are interchanged. Proof. The first formula is well-known from the theory of mutations of Jordan algebras, see e.g. [1, V.§ 3]. The second formula needs different arguments, since [e−] might differ from [c−]. We claim that (e+, e− −c−) = c+. In this case, (ii) follows is quasi-invertible with quasi-inverse ee−−c− from standard identities of the Jordan pair determinant [5, § 16.11], + ∆− e(y)= ∆(e+, e− − y) = ∆(e+, e− − c− + c− − y) = ∆(e+, e− − c−)∆(ee−−c− e(c−) ∆(c+, c− − y) = ∆− e(c−) ∆− = ∆− c(y) . + , c− − y) In order to show quasi-invertibility of (e+, e− − c−), consider the decom- position c− = c2 ⊕ c1 ⊕ c0 of c− according to the Peirce decomposition of V − with respect to e. Since c is an idempotent, the Peirce rules [5, § 5.4] yield the following relations: Qc+c2 = c+ , Qc2c+ ⊕{c2, c+, c1} ⊕ Qc1c+ = c2 ⊕ c1 ⊕ c0 . Comparing the components of the Peirce spaces in the second identity, we conclude that the pair (c+, c2) is also idempotent with [c+] = [e+] and [c2] = [e−]. By assumption, c+ is invertible in the Jordan algebra [e+]. Therefore, c2 is invertible in the Jordan algebra [e−], and it follows that (e− − c2, e+) is quasi-invertible with quasi-inverses (e− − c2)e+ = c−1 2 − e−, INEQUALITIES FOR GENERALIZED MINORS 5 where c−1 2 is the inverse of c2 in [e−], see [5, § 3.1]. Now recall that Jordan algebra inverses satisfy a−1 = P −1 operator corresponding to a∈ [e−]. Here, Pa = QaQe+[e−], so we obtain a a, where Pa is the quadratic Due to the symmetry formula [5, § 3.3] for quasi-inverses it follows that c−1 2 =(Qc2Qe+[e−])−1c2 = Qe−(Qc2[c2])−1c2 = Qe−c+. = e+ + Qe+(e− − c2)e+ = e+ + Qe+(Qe− c+ − e−)= c+. ee−−c2 + ∆− the same quasi-inverse. Finally, the shifting formula [5, § 3.5] yields that (e+, e− − c−) is quasi- invertible if and only if (e+, e− − c2) is quasi-invertible, and both have It remains to show (iii). Since ∆(Quv, w) = ∆(Quw, v) for all u∈ V +, v, w ∈ V −, Proposition 1.1 yields the relation e(c−)= ∆+ Since Qe+c− = e2 + in the Jordan algebra [c+], the second term on the c(e+)2. Moreover, setting x= e+ in (i) yields e(c+)−1, so (1.3) implies (iii). By duality of the Jordan c(e+) = ∆+ pairs (V +, V −) and (V −, V +) the same statement holds when + and − e(Qe+c−). Now applying (i), we obtain c(Qe+c−). right hand side becomes ∆+ ∆+ e(c−)= ∆+ e(c+) ⋅ ∆+ (1.3) ∆− (cid:3) are interchanged. Remark 1.5. We note that Proposition 1.4(i) does not necessarily hold for all x ∈ V σ, as the following calculation illustrates. As in Ex- ample 1.3 consider the Jordan pair (k r×s). Let e = (e+, e−) be r×s, k defined by B e− =A−1 0 0 , e+ =A 0 k×k is invertible, and B ∈ k C CAB , where A∈ k k×(s−k), C ∈ k Any such pair of matrices is an idempotent of (k k×k . the same principal inner ideal in k [e+]=a 0 0 0 a∈ k r×s, (r−k)×k are arbitrary. r×s, k r×s), and yields The generalized minors are given by ∆+ k×k, and cA−1 + dC c d)∈ k r×s with a∈ k 1 + cB + dCAB e(x) = Det(1r −(e+ − x)e−)= DetaA−1 + bC −AB + aB + bCAB where x=( a b e(y)= Det(1r − e+(e− − y))= DetAα −A(B − β) = Det A ⋅ Det α where y = α β e(x) simplifies to Det A−1 ⋅ Det a only if x∈ [e+] or B = 0 and C = 0, so Proposition 1.4(i) fails for x≠ [e+], e.g. in the case where B ≠ 0 and c=(( 1 0 0 0)). s×r with α ∈ k 0 0)),( 1 0 k×k. We see that ∆+ γ δ ∈ k 1r−k ∆− 0 6 BENJAMIN SCHWARZ Remark 1.6. Strictly speaking, Proposition 1.4(ii) is not needed for the purpose of this paper. However we find it worthwhile to include this formula, since it is a rather suprising identity, in particular considered in contrast to Proposition 1.4(i) with its restricted domain. 2. Inequalities for generalized minors In this section, let (V, V) be a simple Jordan pair over k= R or k= C with positive involution ϑ ∶ V → V . For convenience, we set ¯z ∶= ϑ(z) for z ∈ V . Recall that an element e ∈ V is a tripotent, if e ∶= (e, ¯e) is an idempotent. All notions and results of the last section also apply to this idempotent, and since e uniquely determines the idempotent, we simplify the notation and set Vν(e) ∶= V + ν (e), ∆e(x) ∶= ∆+ e(x), a sum of two non-zero strongly orthogonal tripotents. Any tripotent is the sum of primitive tripotents, and the number of summands is for ν = 2, 1, 0 and x∈ V . Two tripotents e, c∈ V are strongly orthogonal, if e∈ V0(c) (or equiv- alently c ∈ V0(e)), and e is called primitive, if it cannot be written as called its rank. A frame is a maximal system (e1, . . . , er) of primitive Due to [6, §3.12] any element z ∈ V admits a singular value decom- strongly orthogonal tripotents. Here, r is independent of the choice of the frame, and called the rank of the Jordan pair V . position, i.e., (2.1) z = σ1e1 + ⋯ + σrer uniquely determined real numbers, called the singular values of z. (1.2) is in fact a tripotent. The singular value decomposition (2.1) co- s×r. Then, an element e ∈ k r×s discussed in our previous ex- amples, a positive involution is given by the Hermitian transpose, r×s is a tripotent if and only where (e1, . . . , er) is a frame of tripotents and σ1 ≥ ⋯ ≥ σr ≥ 0 are Example 2.1. In the case V = k z ↦ z∗ ∈ k if e = ee∗e, i.e., if e is a partial isometry. The element constructed in incides with the usual one. Indeed, let z = U1ΣU2 be the usual singular value decomposition with U1 ∈ Ur(k), U2 ∈ Us(k) and diagonal matrix r×s with entries σ1 ≥ ⋯ ≥ σr ≥ 0. For 1 ≤ k ≤ r set ei ∶= U1EiU2 Σ ∈ k where Ei denotes the matrix with 1 at the (i, i)'th position and 0 else- where. Then (e1, . . . , er) is a frame of tripotents, and z = σ1e1 +⋯+σrer. Theorem 2.2. Let e ∈ V be a tripotent of rank k, and z ∈ V be an element with singular values σ1 ≥ ⋯≥ σr ≥ 0. Then ∆e(z)≤ σ1⋯σk . INEQUALITIES FOR GENERALIZED MINORS 7 Before proving this, we determine the derivative of the Jordan pair determinant ∆. Since (V, V) is assumed to be simple, it follows from [5, § 17.3] that ∆ is irreducible and satisfies Det Bx, y = ∆(x, y)p, (2.2) (2.3) where p is a structure constant of (V, V), and Det denotes the standard determinant of the Bergman operator Bx, y ∈ End(V). Recall that τ ∶ V × V → k, (x, y) ↦ Tr Dx,y is a non-degenerate pairing, called the trace form of (V, V). Lemma 2.3. The derivative of the Jordan pair determinant at (x, y)∈ V × V along (u, v)∈ V × V is given by d(u,v)∆(x, y)= − 1 p ∆(x, y)τ(u, yx) + τ(xy, v) . (2.4) Proof. We note that is suffices to prove (2.4) for generic(x, y)∈ V ×V , so we may assume that ∆(x, y)≠ 0, or equivalently that Bx, y is invertible. Taking derivatives on both sides of (2.2) yields p ∆(x, y)p−1 d(u,v)∆(x, y)= Det Bx, y TrB−1 x, yd(u,v)Bx, y , and hence d(u,v)∆(x, y)= 1 p ∆(x, y) TrB−1 x, yd(u,v)Bx, y . TrB−1 decomposition of V with respect to e in the following way: Recall that rank k. It is known [6, §§ 5.6, 11.12] that Sk is a compact submanifold, Dx,v − QxQy,v, see the appendix of [6]. This completes the prove. (cid:3) Since d(u,v)Bx, y = −Du,y + Qx,uQy − Dx,v + QxQy,v, it follows that x, yd(u,v)Bx, y= − Tr Du,yx − Tr Dxy ,v = −τ(u, yx) − τ(xy, v) , where we used the relations Du,yxBx, y = Du,y − Qx,uQy and Bx, yDxy,v = Proof of Theorem 2.2. Let Sk ⊆ V denote the subset of tripotents of and the tangent space TeSk at e ∈ Sk is given in terms of the Peirce the map x ↦ x# ∶= Qe ¯x defines an involution on V2(e) with eigenspace decomposition V2(e)= A(e) ⊕ B(e) where Then TeSk = B(e) ⊕ V1(e). Moreover, recall that the decomposition V = A(e) ⊕ B(e) ⊕ V1(e) ⊕ V0(e) is orthogonal with respect to the positive definite inner product(u, v) ∶= τ(u, ¯v) on V . For fixed z ∈ V , we determine the maximum value of the map f ∶ Sk → R, f(e) ∶=∆e(z)2. that the derivative of f at e∈ Sk along u= u2 + u1 ∈ TeSk is given by A(e) ∶=x∈ V2(e) x= x# , B(e) ∶=x∈ V2(e) x= −x# . Due to Lemma 1.1, it is clear that f is smooth, and Lemma 2.3 implies duf(e)= − 2 p f(e) ⋅ Reτ(u,(¯e)e−z) + τ((e − z)¯e, ¯u) . 8 BENJAMIN SCHWARZ The symmetry formula for quasi-inverses [5, § 3.3] yields (¯e)e−z = ¯e + Q¯e(e − z)¯e, and since τ(x, Qyz)= τ(z, Qyx) it follows that τ(u, ¯ee−z)= τ(u, ¯e + Q¯e(e − z)¯e)= τ(u, ¯e) + τ((e − z)¯e, Q¯eu). We thus obtain duf(e)= − 2 p f(e) ⋅ Re τ((e − z)¯e, ¯u1) , since e ∈ A(e) ⊥ TeSk and Q¯eu + ¯u = ¯u1. Therefore, f attains its maximum value at e only if τ((e − z)¯e, ¯u1) = 0 for all u1 ∈ V1(e), i.e., only if(e−z)¯e = x2 +x0 ∈ V2(e)⊕V0(e). In this case, the shifting formula for quasi-inverses [5, § 3.5] yields e − z =(x2 + x0)−¯e = x−¯e 2 + x0 ∈ V2(e) ⊕ V0(e) , so it follows that z = z2+z0 ∈ V2(e)⊕V0(e). Now, strong orthogonality of V2(e) and V0(e) implies that the singular value decomposition of z splits into the corresponding decompositions of z2 and z0, so {σ1, . . . , σr} = {µ1, . . . , µk} ∪{ν1, . . . , νr−k}, where the µi (resp. νi) are the singular values of z2 (resp. z0). We claim that ∆e(z) = ∏ µi. f attains its maximum value if µi = σi for i = 1, . . . , k, and we are finished. To evaluate ∆e(z) first recall that ∆e(z) = ∆e(z2). Let z2 = µ1c1 + ⋯ + µkck denote the spectral decomposition of z2. We note that this is not necessarily the spectral decomposition of z2 ∈ [e] in the sense of Jordan algebras [3, III.1.1], since the sum c ∶= c1 + ⋯ + ck might differ from e. However, since [c] = [e], Proposition 1.4(i) yields ∆e(z2) = ∆c(e)−1∆c(z2). Now, due to Proposition 1.1 and [5, § 16.15] If so, then we obtain ∆c(z2)= ∆(c − z2, ¯c)=M(1 −(1 − µi) ⋅ 1)=M µi. Finally, recall that ∆(u, ¯v) = ∆(v, ¯u) for all u, v ∈ V , where α denotes complex conjugation of α ∈ k. Therefore, Proposition 1.4(iii) yields ∆c(e) = 1, and we conclude that ∆e(z2) = ∏ µi. This completes the (cid:3) proof. We finally give an application of Theorem 2.2 involving some repre- mial maps on V . Let L denote the identity component of the structure sentation theory. Consider a complex simple Jordan pair (V, V) with involution of rank r, and let P(V) denote the space of complex polyno- group associated to (V, V), which consists of linear maps h ∈ GL(V) satisfying the relation h{x, y, z} = {hx, h−#y, hz} where h−# denotes subgroup K ∶= L ∩ U(V), where U(V) denotes unitary operators with respect to the inner product (u, v)= τ(u, v) on V . The induced action of K on polynomials yields a decomposition of P(V) into irreducible the inverse of the adjoint of h with respect to the trace form τ defined in (2.3). Then L is a reductive complex Lie group with maximal compact components. Due to Hua, Kostant, Schmid [4, 7], this decomposition INEQUALITIES FOR GENERALIZED MINORS 9 is multiplicity free, P(V)= ࣷm≥0Pm(V), and the irreducible components can be parametrized by tuples m = (m1, . . . , mr) of integers satisfying m1 ≥ ⋯ ≥ mr ≥ 0 (corresponding to certain highest weights). As an application of Theorem 2.2, we obtain a growth condition for polynomials in each component. ⋯σmr . r 1 Corollary 2.4. Let z ∈ V be an element with singular values σ1 ≥ ⋯≥ σr ≥ 0. For any p∈Pm(V) there exists C > 0 such that Proof. Let (e1, . . . , er) be a frame of tripotents. Recall from [10] that the following polynomial map is a highest weight vector of Pm(V) (for p(z)≤ C ⋅ σm1 an appropriate choice of a Borel subgroup of L), pm(z) ∶= ∆ǫ1(z)m1−m2 ⋅ ∆ǫ2(z)m2−m3 ⋯∆ǫr−1(z)mr−1−mr ⋅ ∆ǫr(z)mr where ǫk ∶= e1 + ⋯ + ek. For pm, Theorem 2.2 immediately yields For general p ∈ Pm(V), irreducibility implies that there are ci ∈ C and ki ∈ K, such that pm(z)≤ σm1 1 ⋯σmr . r p(z)= sQi=1 ci ⋅ pm(kiz) Since singular values are invariant under the action of K, this proves (cid:3) our statement with C ∶= ∑ici. We refer to [8, Theorem 2.2] for an application of this growth con- dition. The main advantage of this estimate is the K-invariance of the right hand side. References 1. H. Braun and M. Koecher, Jordan-Algebren, Grundlehren der mathematischen Wissenschaften, vol. 128, Springer-Verlag, Berlin -- New York, 1966. 2. N.G. de Bruijn, Inequalities concerning minors and eigenvalues, Nieuw Archief v. Wiskunde IV (1956), no. 3, 18 -- 35. 3. J. Faraut and A. Koranyi, Function spaces and reproducing kernels on bounded symmetric domains, J. Funct. Anal. 88 (1990), 64 -- 89. 4. L.K. Hua, Harmonic analysis of functions of several complex variables in the classical domains, Translations of Mathematical Monographs, vol. 6, American Mathematical Society, Providence, R.I., 1963. 5. O. Loos, Jordan Pairs, Lecture notes in Mathematics, vol. 460, Springer-Verlag, Berlin-New York, 1975. 6. , Bounded symmetric domains and Jordan pairs, Lecture notes, Univer- sity of California, Irvine, 1977. 7. W. Schmid, Die Randwerte holomorpher Funktionen auf hermitesch sym- metrischen Räumen, Inv. math. 9 (1969), 61 -- 80. 8. B. Schwarz, Existence of nearly holomorphic sections on compact Hermitian symmetric spaces, preprint, available at arXiv.org. 10 BENJAMIN SCHWARZ 9. B. Schwarz and H. Seppänen, Symplectic branching laws and Hermitian sym- metric spaces, to appear in Trans. Amer. Math. Soc. 10. H. Upmeier, Jordan algebras and harmonic analysis on symmetric spaces, Amer. J. Math. 108 (1986), no. 1, 1 -- 25. Benjamin Schwarz, Universität Paderborn, Fakultät für Elektrotech- nik, Informatik und Mathematik, Institut für Mathematik, Warburger Str. 100, 33098 Paderborn, Germany E-mail address: [email protected]
1602.05646	1	1602	2016-02-18T01:27:39	Recollements, sinks elimination and Leavitt path algebras	[ "math.RA" ]	For Leavitt path algebras, we show that whereas removing sources from a graph produces a Morita equivalence, removing sinks gives rise to a recollement situation. In general, we show that for a graph $E$ and a finite hereditary subset $H$ of $E^0$ there is a recollement $$\xymatrix{ L_K(E/\overline H) \rModd \ar[r] & \ar@<3pt>[l] \ar@<-3pt>[l] L_K(E) \rModd \ar[r] & \ar@<3pt>[l] \ar@<-3pt>[l] L_K(E_H) \rModd .}$$ We record several corollaries.	math.RA	math	RECOLLEMENTS, SINKS ELIMINATION AND LEAVITT PATH ALGEBRAS R. HAZRAT AND J. HUANG Abstract. For Leavitt path algebras, we show that whereas removing sources from a graph produces a Morita equivalence, removing sinks gives rise to a recollement situation. In general, we show that for a graph E and a finite hereditary subset H of E 0 there is a recollement LK (E/H) -Mod / LK (E) -Mod / LK (EH) -Mod . We record several corollaries. In this short note we record an application of recollements in naturally decomposing a Leavitt path algebra and gluing the pieces together. A rec- ollement (gluing) of abelian categories consists of three abelian categories A , B, C and six functors relating them as follows q A i p / B s j r / C , (1) such that (i) (s, j, r) and (q, i, p) are adjoint triples, i.e., s is left adjoint to j which is left adjoint to r, similarly for the second triple; (ii) the functors i, s, and r are fully faithful; (iii) Im(i) = Ker(j). In the setting of unital rings, the following is an archetype example of recollement: Let A be a ring with identity and e ∈ A an idempotent. Then there is a recollement situation A/AeA⊗A− eA⊗eAe− A/AeA -Mod inc / A -Mod e(−) / eAe -Mod . (2) HomA(A/AeA,−) HomeAe(Ae,−) To check the adjointness of (i) and the fully faithfullness of (ii), one uses the following general hom-tensor calculus (see [3, §20]): For a triple (RM, SWR, SN ) of modules over unital rings R and S, we have a natural isomorphism HomR(M, HomS(W, N )) ∼= HomS(W ⊗R M, N ). Furthermore, if M is finitely generated projective R-module then HomS(N, W ) ⊗R M ∼= HomS(N, W ⊗T M ). Finally for the triple (MR, S WR, SN ), where M is finitely generated projec- tive R-module, we have the natural isomorphism M ⊗R HomS(W, N ) ∼= HomS(Hom(M, W ), N ). 1 / o o o o / o o o o / o o o o / o o o o / o o o o / o o o o 2 R. HAZRAT AND J. HUANG For rings with local units, the calculus of module theory is similar to the rings with units. In particular the above isomorphisms are valid if R is a ring with local units and S is a unital ring (see for example [13, 49.1 -- 49.3]). Since eAe is a ring with unit, the recollement (2) can be extended for a ring A with local units. The concept of recollement has a geometric origin, and it was first ap- peared in the work of Beilinson, Bernstein and Delign [6] in the setting of triangulated categories and then in the setting of abelian categories in Fran- jou and Pirashvilli [7] motivated by MacPherson-Vilonen construction for the category of perverse sheaves (see [7, 9] for background on recollements, applications and further references). In this note we work with a directed graph E = (E0, E1, r, s) which consist of two sets E0, E1 and maps r, s : E1 → E0. We consider the Leavitt path algebra L(E) associated to E. For a background on Leavitt path algebras and the relevant terminologies see for example [2, 5, 11] and the references there. To state the main theorem we need to recall the notion of restriction and quotient of a graph. Let E be a graph and H a hereditary subset E0. We denote by EH the restriction graph (cid:16)H, {e ∈ E1 s(e) ∈ H}, r(EH )1, s(EH )1(cid:17). On the other hand, for a hereditary and saturated subset X, we denote by E/X the quotient graph (cid:16)E0 \ X, {e ∈ E1 r(e) /∈ X}, r(E/X)1 , s(E/X)1(cid:17) We are in a position to record our theorem. Theorem 1. Let E be a row finite graph, H a finite hereditary subset of E0 and let H be its hereditary saturated closure. Then there is a recollement of abelian categories LK(E/H) -Mod / LK(E) -Mod / LK(EH ) -Mod . Proof. Observe that since H is finite (3) L(EH ) = pHL(E)pH , where pH = Pv∈H v ∈ L(E) is an idempotent. is H = ∪∞ n=0Λn(H), where Λ0(H) = H and Next we observe that hHi = hHi. The hereditary saturated closure of H Λn(H) = (cid:8)y ∈ E0 s−1(y) 6= ∅, r(s−1(y)) ⊆ Λn−1(H)(cid:9) ∪ Λn−1(H), for n ≥ 1. Clearly hHi ⊆ hHi. We prove the converse by induction. For n = 0, Λ0(H) = H ⊆ hHi. Suppose Λn−1(H) ⊆ hHi and let y ∈ Λn(H). Then y = Pα∈s−1(y) αα∗. Since r(α) ∈ Λn−1(H) ⊆ hHi and hHi is a two- sided ideal, it follows that y ∈ hHi. Thus H = ∪∞ n=0Λn(H) ⊆ hHi. We have a natural isomorphism L(E)/hHi = L(E)/hH i ∼= L(E/H). (4) Now replacing (3) and (4) in recollement diagram 2, the theorem follows. (cid:3) / o o o o / o o o o RECOLLEMENTS AND LEAVITT PATH ALGEBRAS 3 It was observed in [2] that removing sources from a finite graph would give a Leavitt path algebra Morita equivalent to the one associated to the original graph. However removing sinks change the structure of the Leavitt path algebras substantially. As an example, consider the Topilz algebra, i.e., the Leavitt path algebra associated to the graph E : e • f / •. As soon as we remove the sink and the edge f attached to it, we are left with a loop whose Leavitt path algebra is the Laurent polynomial ring K[x, x−1]. Whereas removing sources gives a Morita equivalence, removing sinks cre- ates a recollement situation. Corollary 2. Let E be a finite graph and E a graph where all sources and sinks of E are removed. Then there is a recollement LK(E) -Mod / LK(E) -Mod / Lsinks K -Mod . (5) Proof. Let H be the set of all sinks in E, which is a hereditary set. One can observe that L(EH ) ∼= Lsinks K. On the other hand, E/H is a graph by repeatedly removing all sinks from E until there is no sinks exist. Indeed, if u is a sink in E/H, then r(s−1(u)) has to be in H, which gives u ∈ H, as H is saturated. But this is a contradiction. By [2, Propositons 1.4 and 3.1] removing sources from a graph, gives a Leavitt path algebra Morita equivalent to the original one. Thus repeating the source elimination give the Morita equivalence LK(E) ≈ LK(E/H). Now the corollary follows from Theorem 1. (cid:3) Example 3. (1) Let T = LK(cid:0) • we have the following recollement situation / • (cid:1) be the Topiliz algebra. By Corollary 2 K[x, x−1] -Mod / T / K -Mod Although T is indecomposable [4], but one can still break it down into less complicated algebras using recollement. (2) Let T = LK(cid:0) ment situation • / • (cid:1). Then we have the following recolle- K[x, x−1] -Mod / T / K[x, x−1] -Mod (3) In [1], Leavitt path algebras associated to finite acyclic graphs were classified. For such a graph, the Leavitt path algebra is a direct product of matrix algebras over the field K. Lemma 3.4 and Propo- sition 3.5 in [1] prove this by explicitly constructing an isomorphism between these two algebras. In Corollary 2, for acyclic graph E, we get that E is in fact empty and thus the left hand side of (5) is zero. Since in the general recollement situation (1), A is a Serre subcate- gory of B and C ≈ A /B, it follows that, in our setting (for B = 0) L(E) is Morita equivalent to Lsinks K, confirming the result of [1]. % % / / o o o o / o o o o % % / / o o o o / o o o o % % / y y / o o o o / o o o o 4 R. HAZRAT AND J. HUANG In a series of papers Rangaswamy [11, 12] and Ara-Rangaswamy [5] stud- ied simple modules of Leavitt path algebras. In [5] they proved that for a Leavitt path algebra L(E), where E is a finite graph, any simple module is of the form of a Chen simple module (i.e., arising from an infinite path) if and only if any vertex of E is the base of at most one cycle (these algebras have finite Gelfand-Kirillov dimension). Using the recollement and Kuhn's result [8] we give another characteri- sation of such algebras. In the following theorem we use a partial ordering defined in [5]. A pre-order ≥ on the set of cycles of a directed graph E is defined as follows: Let c1 and c2 be two cycles. We write c1 ≥ c2 if there is a path from a vertex of c1 to a vertex of c2. For graphs whose vertices are bases of at most one cycle, ≥ is a partial order. In Theorem 4 we work with minimal cycles with respect to ≥, i.e., cycles with no exist. In the following we will use Kuhn's result [8, Proposition 4.7], which states that for a recollement q A i p / B s j r / C , there is a bijection between isomorphism classes of simple objects nsimples in A oansimples in Co −→ nsimples in Bo. (6) Theorem 4. Let E be a finite graph whose vertices are base of at most one cycle. Then the isomorphism classes of simple (right) modules of LK (E) are in one to one correspondence with the set of sinks and the set A × B, where A is the set of all cycles in E and B is the set of all irreducible polynomials in K[x, x−1]. Proof. By Corollary 2 (and its proof), we can write LK(E) -Mod / LK(E) -Mod / Lsinks K -Mod . where E is a graph with no sinks whose vertices are base of at most one cycle. Now by (6) simple modules of LK (E) is in one to one correspondence with the sinks of E and the simple modules of L(E). On the other hand L(E) can be stratified by cycles, i.e., we have iterated recollement situations Bi / Bi−1 / Ci where 1 ≤ i ≤ k and k is the number of cycles in E. Here Bi = L(bi), Ci = L(ci), where b0 = E, bi = bi−1/ci, and ci is a minimal cycle in the graph bi−1 with respect to the ordering ≥. Again by (6) there is a bijection between isomorphism classes of simple objects nsimples in Bioansimples in Cio −→ nsimples in Bi−1o. (7) Since ci are cycles, L(ci) is Morita equivalent to the ring K[x, x−1]. Thus the simple modules of L(ci) is in one to one correspondence with the ir- reducible polynomials of K[x, x−1]. The theorem now follows from (7) by noting that L(bk) is also Morita equivalent to K[x, x−1]. (cid:3) / o o o o / o o o o / o o o o / o o o o / o o o o / o o o o RECOLLEMENTS AND LEAVITT PATH ALGEBRAS 5 We note that the above theorem can also be deduced from the results of papers [10, 5]. In [5, Theorem 1.1], a bijection between S, the set of non-isomorphic simple modules, and the set of primitive ideals of L(E) was established. On the other hand Corollary 3.14 in [10] implies there is a bijection between S and A × B, where A is the set of all cycles in E and B is the set of all irreducible polynomials in K[x, x−1]. References [1] G. Abrams, G. Aranda Pino, M. Siles Molina, Finite-dimensional Leavitt path alge- bras, J. Pure Appl. Algebra 209 (2007), no. 3, 753 -- 762. 3 [2] G. Abrams, A. Louly, E. Pardo, C. Smith, Flow invariants in the classification of Leavitt path algebras, J. Algebra 333 (2011) 202 -- 23. 2, 3 [3] F. Anderson, K. Fuller, Rings and categories of modules, second ed. GTM 13, Springer Verlag, 1992. 1 [4] A. Aranda Pino, A. Nasr-Isfahani Decomposable Leavitt path algebras for arbitrary graphs, Forum Math., In press (2015). 3 [5] P. Ara, K.M. Rangaswamy, Finitely presented simple modules over Leavitt path alge- bras, J. Algebra 417 (2014), 333 -- 352. 2, 4, 5 [6] A. Beilinson, J. Bernstein, P. Deligne, Faisceaux Pervers (Perverse sheaves), in: Anal- ysis and Topology on Singular Spaces, I, Luminy, 1981, Asterisque 100 (1982) 5 -- 171 (in French). 2 [7] V. Franjou, T. Pirashvili, Comparison of abelian categories recollements, Doc. Math. 9 (2004) 41 -- 56. 2 [8] N. Kuhn, Generic representations of the finite general linear groups and the Steenrod algebra: II, K-Theory, 8 (1994) 395 -- 428. 4 [9] C. Psaroudakis, Homological theory of recollements of abelian categories, Journal of Algebra 398 (2014) 63 -- 110. 2 [10] K.M. Rangaswamy, The theory of prime ideals of Leavitt path algebras over arbitrary graphs, Journal of Algebra 375 (2013) 73 -- 96. 5 [11] K.M. Rangaswamy, On simple modules over Leavitt path algebras, J. Algebra 423 (2015), 239 -- 258. 2, 4 [12] K.M. Rangaswamy, Leavitt path algebras with finitely presented irreducible represen- tations, J. Algebra 447 (2016), 624 -- 648. 4 [13] R. Wisbauer, Foundations of modules and ring theory, CRC Press, 1991. 2 Western Sydney University, Australia E-mail address: [email protected] Fujian Normal University, China E-mail address: [email protected]
1010.2419	1	1010	2010-10-12T15:26:29	$\delta$-derivations of simple Jordan algebras and superalgebras	[ "math.RA", "math.AC" ]	We describe non-trivial $\delta$-derivations of semisimple finite-dimensional Jordan algebras over an algebraically closed field of characteristic not 2, and of simple finite-dimensional Jordan superalgebras over an algebraically closed field of characteristic 0. For these classes of algebras and superalgebras, non-zero $\delta$-derivations are shown to be missing for $\delta\neq 0,{1}{2},1$, and we give a complete account of ${1}{2}$-derivations.	math.RA	math	δ-DERIVATIONS OF SIMPLE FINITE-DIMENSIONAL JORDAN ALGEBRAS AND SUPERALGEBRAS Ivan Kaygorodov Sobolev Inst. of Mathematics Novosibirsk, Russia [email protected] Keywords: δ-derivation, Jordan (super)algebra. Abstract: We describe non-trivial δ-derivations of semisimple finite-dimensional Jordan algebras over an algebraically closed field of characteristic not 2, and of simple finite-dimensional Jordan superalgebras over an algebraically closed field of characteristic 0. For these classes of algebras and superalgebras, non-zero δ-derivations are shown to be missing for δ 6= 0, 1 2, 1, and we give a complete account of 1 2-derivations. INTRODUCTION The notion of derivation for an algebra was generalized by many mathematicians along quite different lines. Thus, in [1], the reader can find the definitions of a derivation of a subalgebra into an algebra and of an (s1, s2)-derivation of one algebra into another, where s1 and s2 are some homomorphisms of the algebras. Back in the 1950s, Herstein explored Jordan derivations of prime associative rings of characteristic p 6= 2; see [2]. (Recall that a Jordan derivation of an algebra A is a linear mapping jd : A → A satisfying the equality jd(xy + yx) = jd(x)y + xjd(y) + jd(y)x + yjd(x), for any x, y ∈ A.) He proved that the Jordan derivation of such a ring is properly a standard derivation. Later on, Hopkins in [3] dealt with antiderivations of Lie algebras (for definition of an antiderivation, see [1]). The antiderivation, on the other hand, is a special case of a δ-derivation (cid:22) that is, a linear mapping µ of an algebra such that µ(xy) = δ(µ(x)y + xµ(y)), where δ is some fixed element of the ground field. Subsequently, Filippov generalized Hopkin's results in [4] by treating prime Lie algebras over an associative commutative ring Φ with unity and 1 2. It was proved that every prime Lie Φ-algebra, on which a non-degenerated symmetric invariant bilinear form is defined, has no non-zero δ-derivation if δ 6= −1, 0, 1 2-derivations were described for an 2, 1. In [4], also, 1 arbitrary prime Lie Φ-algebra A (cid:0) 1 6 ∈ Φ(cid:1) with a non-degenerate symmetric invariant bilinear form defined on the algebra. It was shown that the linear mapping φ : A → A is a 1 2-derivation iff φ ∈ Γ(A), where Γ(A) is the centroid of A. This implies that if A is a central simple Lie algebra over a field of characteristic p 6= 2, 3 on which a non-degenerate symmetric invariant bilinear form is defined, then every 1 2-derivation φ has the form φ(x) = αx, α ∈ Φ. At a later time, Filippov described δ-derivations for prime alternative and non-Lie Mal'tsev Φ-algebras with some restrictions on the operator ring Φ. In [5], for instance, it was stated that algebras in these classes have no non-zero δ-derivations if δ 6= 0, 1 In the present paper, we come up with an account of non-trivial δ-derivations for semisimple finite-dimensional Jordan algebras over an algebraically closed field of characteristic not 2, and for simple finite-dimensional Jordan superalgebras over an algebraically closed field of 2, 1. 1 2, 1, and we provide in a complete description of 1 characteristic 0. For these classes of algebras and superalgebras, non-zero δ-derivations are shown to be missing for δ 6= 0, 1 2 -derivations. The paper is divided into four parts. In Sec. 1, relevant definitions are given and known results cited. In Sec. 2, we deal with δ-Derivations of simple and semisimple finite- dimensional Jordan algebras. In Secs. 3 and 4, δ-derivations are described for simple finite- dimensional Jordan supercoalgebras over an algebraically closed field of characteristic 0. For some superalgebras, note, the condition on the characteristic may be weakened so as to be distinct from 2. A proof for the main theorem is based on the classification theorem for simple finite-dimensional superalgebras and on the results obtained in Secs. 3 and 4. 1. BASIC FACTS AND DEFINITIONS Let F be a field of characteristic p, p 6= 2. An algebra A over F is Jordan if it satisfies the following identities: xy = yx, (x2y)x = x2(yx). Jordan algebras arise naturally from the associative algebras. If in an associative algebra A we replace multiplication ab by symmetrized multiplication a ◦ b = 1 2(ab + ba) then we will face a Jordan algebra. Denote this algebra by A(+). Below are essential examples of Jordan algebras. (1) The algebra J(V, f ) of bilinear form. Let f : V × V −→ F be a symmetric bilinear form on a vector space V . On the direct sum J = F · 1 + V of vector spaces, we then define multiplication by setting 1 · v = v · 1 = v and v1 · v2 = f (v1, v2) · 1; under this multiplication, J = J(V, f ) is a Jordan algebra. If the form f is non-degenerate and dim V > 1, then the algebra J(V, f ) is simple. (2) The Jordan algebra H(Dn, J). Here, n > 3, D is a composition algebra, which is associative for n > 3, j : d → d is a canonical involution in D, and J : X → X is a standard involution in Dn. THEOREM 1.1 [6]. Every simple finite-dimensional Jordan algebra A over an algebraically closed field F of characteristic not 2 is isomorphic to one of the following algebras: (1) F · 1; (2) J(V, f ); (3) H(Dn, J). We recall the definition of a superalgebra. Let Γ be a Grassmann algebra over F , which is generated by elements 1, e1, . . . , en, . . . and is defined by relations e2 i = 0, eiej = −ejei. Products 1, ei1ei2 . . . eik , i1 < i2 < . . . < ik, form a basis for Γ over F . Denote by Γ0 and Γ1 the subspaces generated by products of even and odd lengths, respectively. Then Γ is represented as a direct sum of these subspaces, Γ = Γ0 + Γ1, with ΓiΓj ⊆ Γi+j(mod 2), i, j = 0, 1. In other words, Γ is a Z2-graded algebra (or superalgebra) over F . Now let A = A0 + A1 be any supersubalgebra over F . Consider a tensor product of F - algebras, Γ ⊗ A. Its subalgebra Γ(A) = Γ0 ⊗ A0 + Γ1 ⊗ A1 is called a Grassmann envelope for A. Let Ω be some variety of algebras over F . A Z2-graded algebra A = A0 + A1 is a Ω- superalgebra if its Grassmann envelope Γ(A) is an algebra in Ω. In particular, A = A0 ⊕ A1 is a Jordan superalgebra if its Grassmann envelope Γ(A) is a Jordan algebra. 2 In [7], it was shown that every simple finite-dimensional associative superalgebra over an algebraically closed field F is isomorphic either to A = Mm,n(F ), which is the matrix algebra Mm+n(F ), or to B = Q(n), which is a subalgebra of M2n(F ). Gradings of superalgebras A and B are the following: 0 D (cid:19)(cid:12)(cid:12)(cid:12)(cid:12) A0 = (cid:26)(cid:18) A 0 C 0 (cid:19)(cid:12)(cid:12)(cid:12)(cid:12) A1 = (cid:26)(cid:18) 0 B 0 A (cid:19)(cid:12)(cid:12)(cid:12)(cid:12) B0 = (cid:26)(cid:18) A 0 A ∈ Mm(F ), D ∈ Mn(F )(cid:27) , B ∈ Mm,n(F ), C ∈ Mn,m(F )(cid:27) , B 0 (cid:19)(cid:12)(cid:12)(cid:12)(cid:12) A ∈ Mn(F )(cid:27) , B1 =(cid:26)(cid:18) 0 B B ∈ Mn(F )(cid:27) . Let A = A0 + A1 be an associative superalgebra. The vector space of A can be endowed with the structure of a Jordan supersubalgebra A(+), by defining new multiplication as follows: a ◦ b = 1 2(ab + (−1)p(a)p(b)ba). In this case p(a) = i if a ∈ Ai. Using the above construction, we arrive at superalgebras Mm,n(F )(+), m > 1, n > 1; Q(n)(+), n > 2. Now, we define the superinvolution j : A → A. A graded endomorphism j : A → A is called a superinvolution if j(j(a)) = a and j(ab) = (−1)p(a)p(b)j(b)j(a). Let H(A, j) = {a ∈ A : j(a) = a}. Then H(A, j) = H(A0, j) + H(A1, j) is a subsuperalgebra of A(+). Below are superalgebras which are obtained from Mn,m(F ) via a suitable superinvolution: C D (cid:19), where (1) the Jordan superalgebra osp(n, m), consisting of matrices of the form(cid:18) A B 0 (cid:19); AT = A ∈ Mn(F ), C = Q−1BT , D = Q−1DT Q ∈ M2m(F ), and Q =(cid:18) 0 C D (cid:19), where (2) the Jordan superalgebra P (n), consisting of matrices of the form (cid:18) A B BT = −B, C T = C, and D = AT , with A, B, C, D ∈ Mn(F ). −Em Em THEOREM 1.2 [8, 9]. Every simple finite-dimensional non-trivial (i.e., with a non-zero odd part) Jordan superalgebra A over an algebraically closed field F of characteristic 0 is isomorphic to one of the following superalgebras: Mm,n(F )(+); Q(n)(+); osp(n, m); P (n); J(V, f ); Dt, t 6= 0; K3; K10; J(Γn), n > 1. The superalgebras J(V, f ), Dt, K3, K10, and J(Γn) will be defined below. Let δ ∈ F . A linear mapping φ of A is called a δ-derivation if φ(xy) = δ(xφ(y) + φ(x)y) (1) for arbitrary elements x, y ∈ A. The definition of a 1-derivation coincides with the conventional definition of a derivation. A 0-derivation is any endomorphism φ of A such that φ(A2) = 0. A non-trivial δ-derivation is a δ-derivation which is not a 1-derivation, nor a 0-derivation. Obviously, for any algebra, the multiplication operator by an element of the ground field F is a 1 2-derivation. We are interested in the behavior of non-trivial δ-derivations of semisimple finite-dimensional Jordan algebras over an algebraically closed field of characteristic not 2, and of simple finite-dimensional Jordan superalgebras over an algebraically closed field of characteristic 0. 3 2. δ-DERIVATIONS FOR SEMISIMPLE FINITE-DIMENSIONAL JORDAN ALGEBRAS In this section, we look at how non-trivial δ-derivations of simple finite-dimensional Jordan algebras behave over an algebraically closed field F of characteristic distinct from 2. As a consequence, we furnish a description of δ-derivations for semisimple finite-dimensional Jordan algebras over an algebraically closed field of characteristic not 2. THEOREM 2.1. Let φ be a non-trivial δ-derivation of a superalgebra A with unity e over a field F of characteristic not 2. Then δ = 1 2. Proof. Let δ 6= 1 2. Then φ(e) = φ(e · e) = δ(φ(e) + φ(e)) = 2δφ(e), that is, φ(e) = 0. Thus φ(x) = φ(x · e) = δ(φ(x) + xφ(e)) = δφ(x) for arbitrary x ∈ A. Contradiction. The theorem is proved. LEMMA 2.2. Let φ be a non-trivial 1 2 -derivation of a Jordan algebra A isomorphic to the ground field. Then φ(x) = αx, α ∈ F . Proof. Let e be unity in A. Then that is, φ(x) = αx, α ∈ F . The lemma is proved. φ(x) = 2φ(xe) − φ(x) = xφ(e), (2) LEMMA 2.3. Let φ be a non-trivial 1 2-derivation of an algebra J(V, f ). Then φ(x) = αx for α ∈ F . Proof. Let φ(e) = αe + v, where α ∈ F and v ∈ V . From (2), it follows that φ(x) = xφ(e) for any x ∈ J(V, f ). For w ∈ V , we then have αf (w, w)e + f (w, w)v = w2(αe + v) = φ(w2) = 1 2(wφ(w) + φ(w)w) = wφ(w) = w(w(αe + v)) = w(αw + f (v, w)e) = αf (w, w)e + f (w, v)w. As the result, f (w, w)v = f (w, v)w. Now, since w is arbitrary and dim(V ) > 1, we have v = 0. Thus φ(x) = αx for any x ∈ J(V, f ). The lemma is proved. LEMMA 2.4. Let φ be a non-trivial 1 2-derivation of an algebra H(Dn, J), n > 3. Then φ(x) = αx for α ∈ F . Proof. Relevant information on composition algebras can be found in [6]. Let φ(e) = αe+v, where v = Pi,j=1 From (2), for x ∈ H(Dn, J) arbitrary, we have xi,jei,j, x1,1 = 0, xi,j = xj,i, α ∈ F , xi,j ∈ D. x2 ◦ (αe + v) = φ(x2) = x ◦ φ(x) = x ◦ (x ◦ (αe + v)), x2 ◦ v = x ◦ (x ◦ v). (3) If we put x = ek,k we obtain n Pj=1 n n xk,jek,j + xk,kek,k + xk,kek,k + xi,kei,k), whence v = xi,iei,i. xi,kei,k = 2e2 k,k ◦v = 2ek,k ◦(ek,k ◦v) = 1 2 ( n xk,jek,j + n Pj=1 Pi=1 (en,k + ek,n) ◦ ((en,k + ek,n) ◦ n Pi=1 Pi=1 Pi=1 4 For x = en,k + ek,n substituted in (3), we have xn,nen,n + xk,kek,k = (en,k + ek,n)2 ◦ xi,iei,i = xi,iei,i) = (en,k + ek,n) ◦ 1 2 (xn,nek,n + xk,kek,n + xk,ken,k + xn,nen,k) = n Pi=1 1 2(xk,kek,k + xk,ken,n + xn,nek,k + xn,nen,n), which yields xn,n = xn−1,n−1 = . . . = x1,1 = 0 and v = 0. Consequently, φ(x) = αx for any x ∈ H(Dn, J). The lemma is proved. THEOREM 2.5. Let φ be a non-trivial δ-derivation of a simple finite-dimensional Jordan 2 and algebra A over an algebraically closed field F of characteristic distinct from 2. Then δ = 1 φ(x) = αx, α ∈ F . The proof follows from Theorems 1.1, 2.1 and Lemmas 2.2-2.4. THEOREM 2.6. Let φ be a non-trivial δ-derivation of a semisimple finite-dimensional n Jordan algebra A = Ai, where Ai are simple algebras, over an algebraically closed field of characteristic not 2. Then δ = 1 2, and for x = n n xi where xi ∈ Ai, we have φ(x) = αixi, Pi=1 Pi=1 Li=1 n Pk=1 αi ∈ F . Proof. Unity in Ak is denoted by ek. If xi ∈ Ai, then φ(xi) = x+ i ∈ Ai ek − ei and φ(ei) = ei+ + ei−, where ei+ ∈ Ai and ei− /∈ Ai. Then i + x− i , where x+ and x− i /∈ Ai. Put ei = 0 = φ(xi · ei) = δ(φ(xi) · ei + xi · φ(ei)) = δ((x+ i + xi · ei+), which i = 0. Consequently, the mapping φ is invariant on Ai. In virtue of Theorem 2.5, δ = 1 yields x− 2 and φ(xi) = αixi for some αi ∈ F defined for Ai with xi ∈ Ai arbitrary. It is easy to verify that i )ei + xi(ei+ + ei−)) = δ(x− i + x− the mapping φ, given by the rule φ(cid:18) n Pi=1 is proved. xi(cid:19) = n Pi=1 αixi, xi ∈ Ai, is a 1 2-derivation. The theorem 3. δ-DERIVATIONS FOR SIMPLE FINITE-DIMENSIONAL JORDAN SUPERALGEBRAS WITH UNITY In this section, all superalgebras but J(Γn) are treated over a field of characteristic not 2. The superalgebra J(Γn) is treated over a field of characteristic 0. Among the title superalgebras are Mm,n(F )(+), Q(n)(+), osp(n, m), P (n), J(V, f ), and J(Γn). Theorem 2.1 implies that these superalgebras all lack in non-trivial δ-derivations, for δ 6= 1 2. Therefore, we need only consider the case of a 1 2-derivation. LEMMA 3.1. Let φ be a non-trivial 1 2 -derivation of Mm,n(F )(+). Then φ(x) = αx for some α ∈ F . Proof. It is easy to see that, for 1 6 i, j 6 n + m, elements ei,j form a basis for the superalgebra Mm,n(F )(+). Let φ(ei,j) = αi,j k,lek,l, where αi,j k,l ∈ F , i, j = 1, . . . , n + m. If in (1) we put x = y = ei,i we arrive at m+n Pk,l=1 m+n Pk,l=1 αi,i k,lek,l = φ(ei,i) = φ(e2 i,i) = 1 2(ei,i ◦ φ(ei,i) + φ(ei,i) ◦ ei,i) = 1 2(cid:18)n+m Pl=1 αi,i i,lei,l + αi,i k,iek,i(cid:19) , n+m Pk=1 whence φ(ei,i) = αiei,i, where αi = αi,i i,i, i = 1, . . . , m + n. Substituting x = ei,j and y = ei,i, i 6= j, in (1), we obtain αi,j k,lek,l = φ(ei,j) = 2φ(ei,j ◦ ei,i) = 1 2(cid:18)αiei,j + m+n Pk,l=1 αi,j i,l ei,l + m+n Pl=1 αi,j k,iek,i(cid:19) . m+n Pk=1 5 Analyzing the resulting equalities, we conclude that αi,j and ej,j yields αi,j x ∈ Mn,m(F )(+). The lemma is proved. i,j = αi. A similar argument for ei,j i,j = αj. Since φ is linear, φ(e) = αe. Using (2) gives φ(x) = αx, for any LEMMA 3.2. Let φ be a non-trivial 1 2-derivation of Q(n)(+). Then φ(x) = αx, where α ∈ F . Proof. Clearly, ∆i,j = ei,j + en+i,n+j and ∆i,j = en+i,j + ei,n+j form a basis for the superalgebra Q(n)(+). On the basis elements, the following relations hold: ∆i,j ◦ ∆k,l = 1 2(δj,k∆i,l + δl,i∆k,j), ∆i,j ◦ ∆k,l = 1 2(δj,k∆i,l + δl,i∆k,j). Let φ(∆i,j) = αi,j k,l∆k,l + α∗i,j k,l ∆k,l. Put x = y = ∆i,i in (1). Then n Pk,l=1 αi,i k,l∆k,l + α∗i,i k,l ∆k,l = φ(∆i,i) = φ(∆2 i,i) = 1 2 (∆i,i ◦ φ(∆i,i) + φ(∆i,i) ◦ ∆i,i) = n Pk,l=1 n n Pk,l=1 Pk,l=1 2(cid:18) n Pl=1 1 n Pk,l=1 αi,i i,l∆i,l + αi,i k,i∆k,i + α∗i,i k,i ∆k,i + n Pk=1 n Pk=1 α∗i,i i,l ∆i,l(cid:19) . n Pl=1 Consequently, φ(∆i,i) = αi∆i,i + αi∆i,i, where αi = αi,i i,i and αi = α∗i,i i,i . If we substitute x = ∆i,i and y = ∆i,j, i 6= j, in (1) we obtain (αi,j k,l∆k,l + α∗i,j k,l ∆k,l) = φ(∆i,i) = 2φ(∆i,i ◦ ∆i,j) = 1 2(cid:18)αi∆i,j + αi∆i,j + αi,j i,l ∆i,l + n Pl=1 n Pk=1 αi,j k,i∆k,i + n Pl=1 α∗i,j i,l ∆i,l + α∗i,j k,i ∆k,i(cid:19) . n Pk=1 Hence αi,j i,j = αi, α∗i,j i,j = αi. A similar argument for ∆j,j and ∆i,j yields φ(∆i,j) = αi,j j,j∆j,j + αj∆i,j + α∗i,j j,j ∆j,j + αj∆i,j. These relations readily imply that αi = αj = α and αi = αj = β, that is, φ(∆i,i) = α∆i,i +β∆i,i. Clearly, φ(E) = αE + β∆, where E is unity in Q(n)(+), and ∆ = (ei,n+i + en+i,i). Suppose that β 6= 0 and φ(x) = αx + β∆ ◦ x is a 1 2-derivation. A mapping ψ : Q(n)(+) → Q(n)(+), for which ψ(x) = ∆ ◦ x, likewise is a 1 2(∆i,i − ∆j,j) = ψ(∆i,j ◦ ∆j,i) = 1 2((∆i,j ◦ ∆) ◦ ∆j,i + ∆i,j ◦ (∆j,i ◦ ∆)) = 0. On the other hand, ∆i,i − ∆j,j 6= 0. Consequently, β = 0, that is, φ(x) = αx. The lemma is proved. 2-derivation. Obviously, 1 LEMMA 3.3. Let φ be a non-trivial 1 2-derivation of osp(n, m). Then φ(x) = αx for some n Pi=1 α ∈ F . Proof. It is easy to see that E = n m ∆i + ∆j, where ∆j = en+j,n+j + en+m+j,n+m+j and ∆i = ei,i is unity in the supersubalgebra osp(n, m). Let Pi=1 Pj=1 φ(∆i) = n+2m Pk,l=1 αi k,lek,l, i = 1, . . . , n, φ(∆j) = βj k,lek,l, j = 1, . . . , m. n+2m Pk,l=1 6 If we put x = y = ∆i, i = 1, . . . , n, in (1) we obtain 1 2(φ(∆i) ◦ ∆i + ∆i ◦ φ(∆i)) = 1 i = 1, . . . , n. Put x = y = ∆i, i = 1, . . . , m, in (1). Then 2 n+2m Pk=1 αi k,iek,i + n+2m Pl=1 n+2m k,lek,l = φ(∆i) = φ(∆2 αi Pk,l=1 i,lei,l!, which yields φ(∆i) = αi∆i, i ) = αi n+2m Pk,l=1 βi k,n+iek,n+i + 1 2 n+2m Pk=1 k,lek,l = φ(∆i) = φ((∆i)2) = 1 βi 2(∆i ◦ φ(∆i) + φ(∆i) ◦ ∆i) = βi k,n+m+iek,n+m+i + n+2m Pk=1 βi n+i,len+i,l + n+2m Pl=1 βi n+m+i,len+m+i,l!. n+2m Pl=1 By the definition of osp(n, m), we have βi βi n+m+i,n+m+i. Thus φ(∆j) = βj∆j, j = 1, . . . , m. n+i,n+m+i = βi m+n+i,n+i = 0 and βi n+i,n+i = Let (ei,j + ej,i) ∈ osp(n, m), i, j = 1, . . . , n, and φ(ei,j + ej,i) = x = ei,j + ej,i and y = ∆i in (1) we arrive at γi,j k,lek,l. If we put 2m+n Pk,l=1 2m+n Pk,l=1 γi,j k,lek,l = φ(ei,j + ej,i) = 2φ((ei,j + ej,i) ◦ ∆i) = 1 2(cid:18)2m+n Pk=1 γi,j k,iek,i + 2m+n Pl=1 γi,j i,l ei,l + αi(ei,j + ej,i)(cid:19) . In view of the last relation, γi,j j,i = γi,j i,j = αi. Similar calculations for ei,j + ej,i and ∆j give γi,j j,i = γi,j i,j = αj. Ultimately, φ(∆i) = α∆i, i = 1, . . . , n. Let Eij = (en+i,n+j + en+m+j,n+m+i) ∈ osp(n, m), i, j = 1, . . . , m, and φ(Eij) = Put x = Eij and y = ∆i in (1); then ωi,j k,lek,l. 2m+n Pk,l=1 ωi,j k,lek,l = φ(Eij) = 2φ(Eij ◦ ∆i) = 1 ωi,j k,n+iek,n+i+ 2m+n ωi,j n+i,len+i,l + 2 2m+n Pk=1 Pl=1 k,n+m+iek,n+m+i + βiEij!. ωi,j ωi,j n+m+i,len+m+i,l + 2m+n Pl=1 2m+n Pk=1 Consequently, ωi,j n+i,n+j = ωi,j n+m+j,n+m+i = βi. A similar argument for Eij and ∆j shows that ωi,j n+i,n+j = ωi,j n+m+j,n+m+i = βj with 1 6 i, j 6 m. Eventually we conclude that φ(∆j) = β∆j, j = 1, . . . , m. Let E11 = e1,n+m+1 − en+1,1 ∈ osp(n, m) and φ(E11) = y = ∆1 in (1) we have νk,lek,l. If we put x = E11 and 2m+n Pk,l=1 2m+n Pk,l=1 2m+n Pk,l=1 νk,lek,l = φ(E11) = 2φ(E11 ◦ ∆1) = 1 (νk,n+1ek,n+1 + νk,n+m+1ek,n+m+1)+ (νn+1,len+1,l + νn+m+1,len+m+1,l) + αE11!, 2m+n Pl=1 2 2m+n Pk=1 7 whence ν1,m+n+1 = νn+1,1 = α. Further, for x = E11 and y = ∆1 substituted in (1), we obtain 2m+n Pk,l=1 νk,lek,l = φ(E11) = 2φ((E11) ◦ ∆1) = 1 2(cid:18)2m+n Pl=1 ν1,le1,l + 2m+n Pk=1 νk,1ek,1 + βE11(cid:19) and ν1,m+n+1 = νn+1,1 = β. Thus α = β and φ(E) = αE. From (2), it follows that φ(y) = αy for any element y ∈ osp(n, m). The lemma is proved. LEMMA 3.4. Let φ be a 1 2 -derivation of P (n). Then φ(x) = αx, where α ∈ F . Proof. Let ∆i,j = ei,j + en+j,n+i, E = ∆i,i be unity in the superalgebra P (n), and φ(∆i,j) = 2n Pk,l=1 αi,j k,lek,l. If in (1) we put x = y = ∆i,i we arrive at 2n Pk,l=1 αi,i k,lek,l = φ(∆i,i) = φ(∆2 i,i) = 1 2 2n Pl=1 αi,i n+i,len+i,l + 2n Pk=1 αi,i k,n+iek,n+i + αi,i i,lei,l + 2n Pl=1 αi,i k,iek,i!. 2n Pk=1 The definition of P (n) implies αi,i αi,i n+i,ien+i,i. i,n+i = 0. Therefore, φ(∆i,i) = αi,i i,iei,i + αi,i n+i,n+ien+i,n+i + Put x = ∆i,i and y = ∆i,j in (1). Then n Pi=1 2 αi,i Pl=1 + 2n αi,j k,lek,l = φ(∆i,j) = 2φ(∆i,i ◦ ∆i,j) 2n Pk,l=1 = 1 i,iei,j + αi,i n+i,n+ien+j,n+i + αi,i n+i,ien+j,i + αi,i n+i,ien+i,j αi,j i,l ei,l + 2n Pk=1 αi,j k,iek,i + 2n Pl=1 αi,j n+i,len+i,l + αi,j k,n+iek,n+i!. 2n Pk=1 Pi=1 Thus αi,i i,i = αi,j n+i,i = αi,j Arguing similarly for ∆j,j and ∆i,j, we obtain αj,j n+j,n+i, and αi,i n+i,n+i = αi,i i,j, αi,i n+j,j = n+j,i. In view of the definition of P (n) and the relations above, we have φ(∆i,i) = α∆i,i+βen+i,i. n+j,n+i, and αj,j i,j, αj,j n+j,n+j = αi,i n+j,i. j,j = αi,j αi,j n The fact that the mapping φ is linear implies φ(E) = αE + β∆, ∆ = (en+i,i). Suppose that β 6= 0 and φ(x) = αx + β∆ ◦ x is a 1 P (n), where ψ(x) = ∆ ◦ x, likewise is a 1 Let bj,i = ej,n+i − ei,n+j. Then ψ(∆i,j ◦ bj,i) = ψ(0) = 0; but 1 1 2((∆i,j ◦ ∆) ◦ bj,i + ∆i,j ◦ (bj,i ◦ ∆)) = 1 en+i,n+j) ◦ (ei,j + en+j,n+i)) = 1 Therefore, β = 0 and φ(x) = αx. The lemma is proved. 2 -derivation. Then a mapping ψ : P (n) → 2-derivation. We argue to show that this is not so. 2(ψ(∆i,j) ◦ bj,i + ∆i,j ◦ ψ(bj,i)) = 4((en+j,i + en+i,j) ◦ (ej,n+i − ei,n+j) + (ej,i − ei,j − en+j,n+i + 6= 0 on the other hand. Hence ψ is not a 1 2-derivation. 8 ∆i,i We define the Jordan superalgebra J(V, f ). Let V = V0 + V1 be a Z2-graded vector space on which a non-degenerate superform f (. , .) : V × V → F is defined so that it is symmetric on V0 and is skew-symmetric on V1. Also f (V1, V0) = f (V0, V1) = 0. Consider a direct sum of vector spaces, J = F ⊕ V . Let e be unity in the field F . Define, then, multiplication by the formula (α + v)(β + w) = (αβ + f (v, w))e + (αw + βv). The given superalgebra has grading J0 = F + V0, J1 = V1. It is easy to see that e is unity in J(V, f ). LEMMA 3.5. Let φ be a 1 2 -derivation of J(V, f ). Then φ(x) = αx, where α ∈ F . 8 Proof. Let φ(e) = αe + v0 + v1, vi ∈ Vi. Putting x = zi, y = e, and zi ∈ Vi in (1), we obtain φ(zi) = 2φ(zie)−φ(zi) = φ(zi)e+ziφ(e)−φ(zi) = αzi+f (zi, vi)e, whence φ(zi) = αzi+f (zi, vi)e. 2(φ(z1)z0 + z1φ(z0)) = f (z1, v1)z0 + f (z0, v0)z1. By the definition of a superform f , we have v0 = 0 and v1 = 0, that is, φ(e) = αe. Using (2) yields φ(x) = αx, α ∈ F , for any x ∈ J(V, f ). The lemma is proved. If we put x = z0 and y = z1 in (1) we arrive at 0 = φ(z1z0) = 1 Consider the Grassmann algebra Γ with (odd) anticommutative generators e1, e2, . . . , en, . . . . In order to define new multiplication, we use the operation if j = ik, if j 6= il, l = 1, . . . , n. ∂ ∂ej (ei1ei2 . . . ein) =(cid:26) (−1)k−1ei1ei2 . . . eik−1eik+1 . . . ein For f, g ∈ Γ0S Γ1, Grassmann multiplication is defined thus: 0 ∞ {f, g} = (−1)p(f ) ∂f ∂ej ∂g ∂ej . Pj=1 Let Γ be an isomorphic copy of Γ under the isomorphic mapping x → x. Consider a direct sum of vector spaces, J(Γ) = Γ + Γ, and endow it with the structure of a Jordan superalgebra, setting A0 = Γ0 + Γ1 and A1 = Γ1 + Γ0, with multiplication •. We obtain a • b = ab, a • b = (−1)p(b)ab, a • b = ab, a • b = (−1)p(b){a, b}, where a, b ∈ Γ0S Γ1 and ab is the product in Γ. Let Γn be a subalgebra of Γ generated by elements e1, e2, . . . , en. By J(Γn) we denote the subsuperalgebra Γn + Γn of J(Γ). If n > 2 then J(Γn) is a simple Jordan superalgebra. LEMMA 3.6. Let φ be a 1 Proof. Let φ(1) = αγ + βν, where α, β ∈ F , γ ∈ Γ, and ν ∈ Γ. Put y = 1 in (1); then 2 -derivation of J(Γn). Then φ(x) = αx, where α ∈ F . φ(x) = 2φ(x • 1) − φ(x) = φ(x) + x • φ(1) − φ(x) = x • φ(1). (4) If in (1) we put x = ei, y = ei, i = 1, . . . , n, with (4) in mind, we arrive at φ(1) = φ(ei • ei) = 1 2(φ(ei) • ei + ei • φ(ei)) = φ(ei) • ei = ei • (ei • φ(1)). For any x of the form ei1ei2 . . . eik, obviously, we have ei • (ei • x) =(x if ∂x ei • (ei • x) =(x if ∂x 0 0 ∂ei = 0, otherwise; ∂ei 6= 0, otherwise. (5) (6) Let γ = γi+ + eiγi− and ν = νi+ + eiνi−, where γi−, γi+, νi−, νi+ do not contain ei. Since i is arbitrary, in view of (5) and (6), we have γ = 1 and ν = e1 . . . en. Thus φ(1) = α · 1 + βe1 . . . en. Relation (4) entails φ(e1) = e1 • φ(1) = e1 • (α · 1 + βe1 . . . en) = αe1, φ(e1) = e1 • φ(1) = e1 • (α · 1 + βe1 . . . en) = αe1 + βe2 . . . en. The relations above, combined with the condition in (1), imply 0 = φ(e1 • e1) = 1 2 (e1 • 2 e1 . . . en; that is, φ(1) = α · 1. From (2), we conclude that φ(x) = αx for φ(e1) + φ(e1) • e1) = β any element x ∈ J(Γn). The lemma is proved. 9 4. δ-DERIVATIONS FOR JORDAN SUPERALGEBRAS K3, Dt, K10 In this section, we confine ourselves to non-trivial δ-derivations of simple finite-dimensional Jordan superalgebras K3, K10, and Dt over an algebraically closed field of characteristic p not equal to 2. For the superalgebra K10, we require in addition that p 6= 3. In conclusion, we formulate a theorem on δ-derivations for simple finite-dimensional Jordan superalgebras over an algebraically closed field of characteristic 0. The three-dimensional Kaplansky superalgebra K3 is defined thus: (K3)0 = F e, (K3)1 = F z + F w, where e2 = e, ez = 1 2z, ew = 1 2w, and [z, w] = e. LEMMA 4.1. Let φ be a non-trivial δ-derivation of K3. Then δ = 1 2 and φ(x) = αx, where α ∈ F . Proof. Let φ(e) = αee + βez + γew, φ(z) = α1e + β1z + γ1w, and φ(w) = α2e + β2z + γ2w, where αe, α1, α2, βe, β1, β2, γe, γ1, γ2 ∈ F . If we put x = y = e in (1) we obtain αee + βez + γew = φ(e) = φ(e2) = δ(eφ(e) + φ(e)e) = δ(2αee + βez + γew). Thus it suffices to consider the following two cases: (1) δ = 1 2; (2) δ 6= 1 2, φ(e) = 0. In the former case, φ(e) = αe, where α = αe. Case (1), for x = e and y = z, entails α1e + β1z + γ1w = φ(z) = 2φ(ez) = 2 · 1 2(β1z + γ1w + αz), whence β1 = 1 2 γ1; that is, β1 = α and γ1 = 0. Similarly, substituting in (1) x = e and y = w, we obtain γ2 = α and β2 = 0. For x = z and y = w in (1), we have αe = φ(e) = φ([z, w]) = 1 2 α1w + αe), whence φ(e) = αe, φ(z) = αz, and φ(w) = αw, where α ∈ F . Consequently, φ(x) = αx for any x ∈ K3. 2(eφ(z) + φ(e)z) = α1e + 1 2 (β1 + α) and γ1 = 1 2 (zφ(w) + φ(z)w) = 1 2( 1 2α2z + αe + 1 We handle the second case. For x = e and y = z in (1), we have α1e + β1z + γ1w = φ(z) = 2φ(ez) = 2δ(eφ(z) + φ(e)z) = δ(2α1e + β1z + γ1w), which yields φ(z) = 0. Similarly, we arrive at φ(w) = 0. The fact that φ is linear implies φ = 0. The lemma is proved. At the moment, we define a one-parameter family of four-dimensional superalgebras Dt. For t ∈ F fixed, the given family is defined thus: Dt = (Dt)0 + (Dt)1, where (Dt)0 = F e1 + F e2, (Dt)1 = F x + F y, e2 e1 + te2, i = 1, 2. i = ei, e1e2 = 0, eix = 1 2x, eiy = 1 2y, [x, y] = LEMMA 4.2. Let φ be a non-trivial δ-derivation of Dt. Then δ = 1 2 and φ(x) = αx, where α ∈ F . Proof. Let φ(e1) = α1e1 + β1e2 + γ1z + λ1w, φ(e2) = α2e1 + β2e2 + γ2z + λ2w, φ(z) = αze1 + βze2 + γzz + λzw, φ(w) = αwe1 + βwe2 + γwz + λww, with coefficients in F . 10 Putting x = y = e1 and then x = y = e2 in (1), we obtain α1e1 + β1e2 + γ1z + λ1w = φ(e1) = 1) = 2δ(e1φ(e1)) = 2δα1e1 +δγ1z +δλ1w and α2e1 +β2e2 +γ2z +λ2w = 2δβ2e2 +δγ2z +δλ2w, φ(e2 whence α1 = 2δα1, β1 = 0, γ1 = δγ1, λ1 = δλ1, α2 = 0, β2 = 2δβ2, γ2 = δγ2, λ2 = δλ2. 2, β1 = α2 = γ1 = γ2 = λ1 = λ2 = 0; 2, α1 = α2 = β1 = β2 = γ1 = γ2 = λ1 = λ2 = 0. There are two cases to consider: (1) δ = 1 (2) δ 6= 1 In the former case, φ(e1) = α1e1 and φ(e2) = β2e2. Put x = e1 and y = z in condition (1); 2 (γzz+λzw+α1z), 2(e1φ(z)+φ(e1)z) = αze1+ 1 then αze1+βze2+γzz+λzw = φ(z) = 2φ(e1z) = 2· 1 which yields α1 = γz, βz = λz = 0. For x = e2 and y = z in (1), we have αze1 + γzz = φ(z) = 2φ(e2z) = 2 · 1 2 (e2φ(z) + φ(e2)z) = 1 2(γzz + β2z), whence γz + β2 = 2γz, αz = 0, α1 = β2, and φ(z) = αz, where α = α1. Similarly, we conclude that φ(w) = αw. The mapping φ is linear; so φ(x) = αx, α ∈ F , for any x ∈ Dt. We handle the second case. Put x = e1 and y = z in (1); then αze1 + βze2 + λzz + γzw = φ(z) = 2φ(e1z) = 2δ(e1φ(z) + φ(e1)z) = δ(2αze1 + λzz + γzw), which yields φ(z) = 0. Arguing similarly for w, we arrive at αwe1 + βwe2 + γwz + λww = δ(2αwe1 + γwz + λww). Consequently, φ(w) = 0. Ultimately, the linearity of φ implies φ = 0. The lemma is proved. The simple ten-dimensional Kac superalgebra K10 is defined thus: K10 = A ⊕ M, (K10)0 = A, (K10)1 = M, where A = A1 ⊕ A2, A1 = F e1 + F uz + F uw + F vz + F vw, A2 = F e2, M = F z + F w + F u + F v. Multiplication is specified by the following conditions: i = ei, e1 is unity in A1, eim = 1 e2 [u, z] = uz, [u, w] = uw, [v, z] = vz, [v, w] = vw, 2m for any m ∈ M, [z, w] = e1 − 3e2, [u, z]w = −u, [v, z]w = −v, [u, z][v, w] = 2e1; all other non-zero products are obtained from the above either by applying one of the skew- symmetries z ↔ w or u ↔ v or by substituting z ↔ u and w ↔ v simultaneously. LEMMA 4.3. Let φ be a non-trivial δ-derivation of K10. Then δ = 1 2 and φ(x) = αx, where α ∈ F . Proof. Let φ(e1) = α1e1 + α2e2 + α3z + α4w + α5u + α6v + α7uz + α8uw + α9vz + α10vw, φ(e2) = β1e1 + β2e2 + β3z + β4w + β5u + β6v + β7uz + β8uw + β9vz + β10vw, φ(z) = γz 10vw, 9 vz + γw φ(w) = γw φ(u) = γu 10vw, φ(v) = γv 10vw, 2e2 + γz 2 e2 + γw 2 e2 + γu 2 e2 + γv 1e1 + γz 1 e1 + γw 1 e1 + γu 1 e1 + γv 3z + γz 3 z + γw 3 z + γu 3 z + γv 4 w + γw 4 w + γu 4 w + γv 5 u + γu 5 u + γv 6 v + γu 6 v + γv 7 uz + γu 7 uz + γv 8 uw + γu 8 uw + γv 9 vz + γu 9 vz + γv 8 uw + γw 5 u + γw 6 v + γw 7 uz + γw 4w + γz 5u + γz 6v + γz 7uz + γz 8uw + γz 9vz + γz 10vw, where all coefficients are in F . For x = y = e1 in (1), we have α1e1 + α2e2 + α3z + α4w + α5u + α6v + α7uz + α8uw + α9vz + α10vw = 2δ(α1e1 + 1 2α3z + 1 φ(e1) = φ(e2 2α4w + 1 2α5u + 1 1) = δ(φ(e1)e1 + e1φ(e1)) = 2 α6v + α7uz + α8uw + α9vz + α10vw), 11 whence α1 = 2δα1, α2 = 0, α3 = δα3, α4 = δα4, α5 = δα5, α6 = δα6, α7 = 2δα7, α8 = 2δα8, α9 = 2δα9, α10 = 2δα10. Putting x = y = e2 in (1), we obtain β1e1 + β2e2 + β3z + β4w + β5u + β6v + β7uz + β8uw + β9vz + β10vw = φ(e2) = φ(e2 2) = δ(φ(e2)e2 + e2φ(e2)) = 2δe2φ(e2) = 2δ(β2e2 + 1 2β3z + 1 2β4w + 1 2β5u + 1 2β6v), which yields β1 = 0, β2 = 2δβ2, β3 = δβ3, β4 = δβ4, β5 = δβ5, β6 = δβ6, β7 = β8 = β9 = β10 = 0. Consequently, it suffices to consider the following two cases: (1) δ = 1 2; (2) δ 6= 1 2, φ(e1) = φ(e2) = 0. In the former case, φ(e1) = α1e1 + α7uz + α8uw + α9vz + α10vw and φ(e2) = αe2. Put x = e2 and y = z in (1); then 1e1 + γz γz 2e2 + γz 3z + γz 4w + γz 5u + γz 6v + γz 7uz + γz 8uw + γz 9vz + γz 10vw = φ(z) = 2φ(ze2) = φ(z)e2 + zφ(e2) = 6v + 1 4w + 1 5u + 1 3z + 1 2e2 + 1 γz 2 γz 2γz 2γz 2γz 2 αz, 2e2 + αz. If in (1) we put x = e1 and y = z we obtain γz and so φ(z) = γz 2e2 + αz = φ(z) = 2φ(ze1) = φ(z)e1 + zφ(e1) = (γz 2e2 + αz)e1 + z(α1e1 + +α7uz + α8uw + α9vz + α10vw), whence γz 2 = 0 and α = α1; that is, φ(z) = αz. Similarly, for w, u, and v, we have φ(u) = αu, φ(v) = αv, and φ(w) = αw. Hence φ(uz) = φ([u, z]) = 1 2 (α[u, z] + α[u, z]) = αuz. Analogously, we obtain φ(uw) = αuw, φ(vz) = αvz, and φ(vw) = αvw. 2(φ(u)z + uφ(z)) = 1 Let x = [u, z] and y = [v, w] in (1); then 2φ(e1) = φ([u, z][v, w]) = 1 2 (φ([u, z])[v, w] + [u, z]φ([v, w])) = α[u, z][v, w] = 2αe1. The fact that φ is linear implies φ(x) = αx, α ∈ F , for x ∈ K10 arbitrary. We handle the second case. Put x = z and y = e1 in (1). Then γz 1e1 + γz 2e2 + γz 4w + γz 3z + γz φ(z) = 2φ(ze1) = 2δ(φ(z)e1 + zφ(e1)) = 7uz + γz 5u + γz 6v + γz 8uw + γz 9vz + γz 10vw = 2δ(γz 1e1 + 1 2γz 3z + 1 2γz 4w + 1 2 γz 5u + 1 2γz 6v + γz 7uz + γz 8uw + γz 9vz + γz 10vw), which yields φ(z) = 0. Similarly, we arrive at φ(w) = φ(v) = φ(u) = 0. Since e1, e2, z, v, u, w generate K10, we have φ = 0. The lemma is proved. THEOREM 4.4. Let A be a simple finite-dimensional Jordan superalgebra over an algebraically closed field of characteristic 0, and let φ be a non-trivial δ-derivation of A. Then δ = 1 2 and φ(x) = αx for some α ∈ F and for any x ∈ A. The proof follows from Theorems 1.2, 2.1 and Lemmas 3.1-3.6, 4.1-4.3. Acknowledgments. I am grateful to A. P. Pozhidaev and V. N. Zhelyabin for their assistance. Список литературы [1] N. Jacobson, Lie Algebras, Wiley, New York (1962). 12 [2] I. N. Herstein, Jordan derivations of prime rings, Proc. Am. Math. Soc., 8, 1104-1110 (1958). [3] N. C. Hopkins, Generalized derivations of nonassociative algebras, Nova J. Math. Game Theory Alg., 5, No. 3, 215-224 (1996). [4] V. T. Filippov, On δ-derivations of prime Lie algebras, Sib. Mat. Zh., 40, No. 1, 201-213 (1999). [5] V. T. Filippov, δ-Derivations of prime alternative and Mal'tsev algebras, Algebra Logika, 39, No. 5, 618-625 (2000). [6] K. A. Zhevlakov, A. M. Slin'ko, I. P. Shestakov, and A. I. Shirshov, Jordan Algebras [in Russian], Novosibirsk State Univ., Novosibirsk (1978). [7] C. T. Wall, Graded Brauer groups, J. Reine Ang. Math., 213, 187-199 (1964). [8] I. L. Kantor, Jordan and Lie superalgebras defined by the Poisson algebra, in Algebra and Analysis [in Russian], Tomsk State Univ., Tomsk (1989), pp. 55-80. [9] V. G. Kac, Classification of simple Z-graded Lie superalgebras and simple Jordan superalgebras, Comm. Alg., 5, 1375-1400 (1977). [10] V. T. Filippov, On δ-derivations of Lie algebras, Sib. Mat. Zh., 39, No. 6, 1409-1422 (1998). 13
1605.00848	1	1605	2016-05-03T11:47:04	Solvable Leibniz algebras with naturally graded non-Lie $p$-filiform nilradicals	[ "math.RA" ]	In this paper solvable Leibniz algebras with naturally graded non-Lie $p$-filiform $(n-p\geq4)$ nilradical and with one-dimensional complemented space of nilradical are described. Moreover, solvable Leibniz algebras with abelian nilradical and extremal (minimal, maximal) dimensions of complemented space nilradical are studied. The rigidity of solvable Leibniz algebras with abelian nilradical and maximal dimension of its complemented space is proved.	math.RA	math	SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE p-FILIFORM NILRADICALS J. Q. ADASHEV1, M. LADRA2, B. A. OMIROV1 1Institute of Mathematics, National University of Uzbekistan, 100125, Tashkent, Uzbekistan, [email protected], [email protected] 2Department of Algebra, University of Santiago de Compostela, 15782 Santiago de Compostela, Spain, [email protected] Abstract. In this paper solvable Leibniz algebras with naturally graded non-Lie p-filiform (n − p ≥ 4) nilradical and with one-dimensional complemented space of nilradical are described. Moreover, solvable Leibniz algebras with abelian nilradical and extremal (minimal, maximal) dimensions of complemented space nilradical are studied. The rigidity of solvable Leibniz algebras with abelian nilradical and maximal dimension of its complemented space is proved. 1. Introduction Leibniz algebras are generalizations of Lie algebras and they have been introduced by Loday in [17] as a non-antisymmetric version of Lie algebras. These algebras preserve a unique property of Lie algebras - the right multiplication operators are derivations. Since the 1993 when Loday's work was published, many researchers have been attracted to Leibniz algebras, with remarkable activity during the last decades. From the classical theory of Lie algebras it is well known that the study of finite-dimensional Lie algebras was reduced to the nilpotent ones [19]. In the Leibniz algebra case we have an analogue of Levi's theorem [6]. Namely, the decomposition of a Leibniz algebra into a semidirect sum of its solvable radical and a semisimple Lie algebra is obtained. The semisimple part can be described from simple Lie ideals (see [6]) and therefore, the main problem is to study the solvable radical. Based on the work of [20], a new approach for the investigation of solvable Lie algebras by using their nilradicals is developed in the works [1 -- 3, 21, 23 -- 25] and others. This approach is based on the nil-independent derivations of nilradical. In fact, for a given solvable Leibniz algebra with a fixed nilradical the complemented space to nilradical has a basis with the condition that the restriction of operators of right multiplication on a basis element is nil-independent derivation of nilradical [12]. Since the description of finite-dimensional solvable Lie algebras is a boundless problem, lately geo- metric approaches are developing. Relevant tools of geometric approaches are Zariski topology and natural action of linear reductive group on varieties of algebras in a such way that orbits under the action consists of isomorphic algebras. It is a well-known result of algebraic geometry that any al- gebraic variety (evidently, algebras defined via identities form an algebraic variety) is a union of a finite number of irreducible components. The most important algebras are those whose orbits under the action are open sets in sense of Zariski topology. The algebras of a variety with open orbits are important since the closures of orbits of such algebras form irreducible components of the variety. At the same time there exist an irreducible component which is not closure of orbit of any algebra. This fact does not detract the importance of algebras with open orbits. This is a motivation of many works focused to discovering of algebras with open orbits and to description of sufficient properties of such algebras [7, 15, 16]. In this paper under the condition to a operator of right multiplication we classify solvable Leibniz algebras with abelian nilradical and with one-dimensional complemented space of nilradical. The clas- sification of solvable Leibniz algebras with abelian nilradical and maximal dimension of complemented 2010 Mathematics Subject Classification. 17A32, 17A36, 17B30, 17B56. Key words and phrases. Lie algebra, Leibniz algebra, solvable algebra, abelian algebra, nilradical, natural graduation, p-filiform algebra, group of cohomology, rigidity. 1 2 SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE P -FILIFORM NILRADICALS space is obtained, as well. The rigidity of such algebras is proved. Moreover, we present a description of solvable algebras with naturally graded non-Lie p-filiform nilradicals (n − p ≥ 4) under some condition to operator of right multiplication to an element of one-dimensional complemented space to nilradical. The classifications problems of obtained algebras of the descriptions are studied. Throughout the paper we consider finite-dimensional vector spaces and algebras over the field C. Moreover, in the multiplication table of an algebra omitted products are assumed to be zero and if it is not noticed we shall consider non-nilpotent solvable algebras. In this section we give necessary definitions and preliminary results. 2. Preliminaries Definition 2.1 ( [17]). A vector space with bilinear bracket (L, [−, −]) over a field F is called a Leibniz algebra if for any x, y, z ∈ L the so-called Leibniz identity holds, or equivalently,(cid:2)[x, y], z(cid:3) =(cid:2)[x, z], y(cid:3) +(cid:2)x, [y, z](cid:3). (cid:2)x, [y, z](cid:3) =(cid:2)[x, y], z(cid:3) −(cid:2)[x, z], y(cid:3) Here, we adopt the right Leibniz identity; since the bracket is not skew-symmetric, there exists the version corresponding to the left Leibniz identity, For examples of Leibniz algebras we refer to papers [17] and [18]. Further we will use the notation (cid:2)[x, y], z(cid:3) =(cid:2)x, [y, z](cid:3) −(cid:2)y, [x, z](cid:3) . L(x, y, z) = [x, [y, z]] − [[x, y], z] + [[x, z], y]. It is obvious that Leibniz algebras are determined by the identity L(x, y, z) = 0. From the Leibniz identity we conclude that the elements [x, x], [x, y] + [y, x] for any x, y ∈ L lie in Annr(L) = {x ∈ L [y, x] = 0, for all y ∈ L}, the right annihilator of the Leibniz algebra L. Moreover, it is easy to see that Annr(L) is a two-sided ideal of L. The two-sided ideal Center(L) = {x ∈ L [x, y] = 0 = [y, x], for all y ∈ L} is said to be the center of L. Definition 2.2. A linear map d : L → L of a Leibniz algebra (L, [−, −]) is said to be a derivation if for all x, y ∈ L the following condition holds: d([x, y]) = [d(x), y] + [x, d(y)]. (2.1) Note that the right multiplication operator Rx : L → L, Rx(y) = [y, x], y ∈ L, is a derivation (for a left Leibniz algebra L, the left multiplication operator Lx : L → L, Lx(y) = [x, y], y ∈ L, is also derivations). Definition 2.3. For a given Leibniz algebra (L, [−, −]) the sequences of two-sided ideals defined recursively as follows: L1 = L, Lk+1 = [Lk, L], k ≥ 1, L[1] = L, L[s+1] = [L[s], L[s]], s ≥ 1, are said to be the lower central and the derived series of L, respectively. Definition 2.4. A Leibniz algebra L is said to be nilpotent (respectively, solvable), if there exists n ∈ N (m ∈ N) such that Ln = 0 (respectively, L[m] = 0). The minimal number n (respectively, m) with such property is said to be the index of nilpotency (respectively, index of solvability) of the algebra L. Evidently, the index of nilpotency of an n-dimensional nilpotent algebra is not greater than n + 1. Definition 2.5. The maximal nilpotent ideal of a Leibniz algebra is said to be the nilradical of the algebra. Let R be a solvable Leibniz algebra with nilradical N . We denote by Q the complementary vector space of the nilradical N to the algebra R. Let us consider the restrictions to N of the right multipli- cation operator on an element x ∈ Q (denoted by RxN ). From [12] we know that for any x ∈ Q, the operator RxN is a non-nilpotent derivation of N . Let {x1, . . . , xm} be a basis of Q, then for any scalars {α1, . . . , αm} ∈ C \ {0}, the matrix α1Rx1N + is non-nilpotent, which means that the elements {x1, . . . , xm} are nil-independent [20]. · · · + αmRxmN SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE p-FILIFORM NILRADICALS 3 Therefore, we have that the dimension of Q is bounded by the maximal number of nil-independent derivations of the nilradical N (see [12, Theorem 3.2]). Moreover, similar to the case of Lie algebras, for a solvable Leibniz algebra R the inequality dim N ≥ dim R holds. Let L be a nilpotent Leibniz algebra and x ∈ L \ L2. Denote by C(x) = (n1, n2, . . . , nk) the decreasing sequence which consists of the dimensions of the Jordan blocks of the operator Rx. On the set of such sequences we consider the lexicographic order. 2 Definition 2.6. The sequence C(L) = max x∈L\L2 C(x) is called the characteristic sequence of the Leibniz algebra L. Below we define the notion p-filiform Leibniz algebra. Definition 2.7. A Leibniz algebra L is called p-filiform if C(L) = (n − p, 1, . . . , 1 ), where p ≥ 0. Note that above definition, when p > 0 agrees with the definition of p-filiform Lie algebras [8]. Since in the case of Lie algebras there is no singly-generated algebra, the notion of 0-filiform algebra for Lie algebras has no sense, while for the Leibniz algebras case in each dimension there exists up to isomorphism a unique null-filiform algebra [4]. p {z } Definition 2.8. Given an n-dimensional p-filiform Leibniz algebra L, put Li = Li/Li+1, 1 ≤ i ≤ n−p, and gr L = L1 ⊕ L2 ⊕ · · · ⊕ Ln−p. Then [Li, Lj] ⊆ Li+j and we obtain the graded algebra gr L. If gr L and L are isomorphic, gr L ∼= L, we say that L is naturally graded. Due to voluminous of the list of naturally graded p-filiform (1 ≤ p ≤ n − 1) Lie algebras we refer the reader to the work [9]. Since we shall consider naturally graded non-Lie p-filiform nilradical we give its classification. Theorem 2.9 ( [10]). An arbitrary n-dimensional naturally graded non-split p-filiform Leibniz algebra (n − p ≥ 4) is isomorphic to one of the following non-isomorphic algebras: p = 2k is even 1 ≤ i ≤ n − 2k − 1, 1 ≤ j ≤ k, 1 ≤ i ≤ n − 2k − 1, 2 ≤ i ≤ n − 2k − 1, 2 ≤ j ≤ k, µ2 : [e1, fj] = fk+j, [ei, e1] = ei+1, [e1, f1] = e2 + fk+1, [ei, f1] = ei+1, [e1, fj] = fk+j, µ1 :( [ei, e1] = ei+1,   µ3 :  p = 2k + 1 is odd 1 ≤ i ≤ n − 2k − 2, [ei, e1] = ei+1, [e1, fj] = fk+1+j, 1 ≤ j ≤ k, [ei, fk+1] = ei+1, 1 ≤ i ≤ n − 2k − 2, where {e1, e2, . . . , en−p, f1, f2, . . . , fp} is a basis of the algebra. In order to simplify our further calculations for the algebra µ3, by taking the change of basis in the following form: e′ 2 = e1 − fk+1, e′ 1 = e1, e′ i+1 = ei, 2 ≤ i ≤ n − 2k − 1, f ′ j = fj, f ′ k+j = fk+1+j, 1 ≤ j ≤ k, we obtain the table of multiplication of the algebra µ3, which we shall use throughout the paper: [e1, e1] = e3, [ei, e1] = ei+1, [e1, fj] = fk+j, 1 ≤ j ≤ k, [e2, fj] = fk+j, 1 ≤ j ≤ k. 2 ≤ i ≤ n − 2k − 1, µ3 :   For acquaintance with the definition of cohomology group of Leibniz algebras and its applications to the description of the variety of Leibniz algebras (similar to Lie algebras case) we refer the reader to 4 SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE P -FILIFORM NILRADICALS the papers [5, 13, 14, 17, 18, 22]. Here we just recall that the second cohomology group of a Leibniz algebra L with coefficients in a corepresentation M is the quotient space HL2(L, M ) = ZL2(L, M )/BL2(L, M ), where the 2-cocycles ϕ ∈ ZL2(L, M ) and the 2-coboundaries f ∈ BL2(L, M ) are defined as follows (d2ϕ)(a, b, c) = [a, ϕ(b, c)] − [ϕ(a, b), c] + [ϕ(a, c), b] + ϕ(a, [b, c]) − ϕ([a, b], c) + ϕ([a, c], b) = 0 (2.2) and (2.3) The linear reductive group GLn(F) acts on Leibn, the variety of n-dimensional Leibniz algebra f (a, b) = [d(a), b] + [a, d(b)] − d([a, b]) for some linear map d. structures, via change of basis, i.e., (g ∗ λ)(x, y) = g(cid:16)λ(cid:0)g−1(x), g−1(y)(cid:1)(cid:17), g ∈ GLn(F), λ ∈ Leibn . The orbits Orb(−) under this action are the isomorphism classes of algebras. Recall, Leibniz algebras with open orbits are called rigid. Note that solvable Leibniz algebras of the same dimension also form an invariant subvariety of the variety of Leibniz algebras under the mentioned action. Remark 2.10. Due to results of the paper [4] we have a sufficient condition for a Leibniz algebra being rigid algebra. Namely, if the second cohomology of a Leibniz algebra with coefficients in itself is trivial, then it is a rigid algebra. 3. Solvable Leibniz algebras with abelian nilradical and extremal dimensions of complemented space Q We denote by A(k) the k-dimensional abelian algebras. For solvable Leibniz algebras with nilradical A(k) and dimension of complemented space of nilradical to an algebra is equal to s, we shall use the notation R(A(k), s). This section is devoted to the classification of solvable Leibniz algebras with nilradical A(k) under the condition that the complemented space to the nilradical have extremal dimensions. The extremal dimensions of the complemented space means the minimal and maximal possible dimensions of the space. Evidently, the candidate for minimal dimension of complemented space is equal one. Firstly, we consider the case of solvable algebras with one-dimensional complemented space of A(k). Let {f1, f2, . . . , fk, x} be a basis of the algebra R(A(k), 1). Evidently, the space of derivations of the algebra A(k) coincided with the space of k × k matrices. Let us assume that the operator RxA(k) has Jordan block form, that is, RxA(k) = Jλ. Theorem 3.1. An arbitrary algebra of the family R(A(k), 1) is isomorphic to one of the following non-isomorphic algebras: R1 : ( [fi, x] = fi + fi+1, 1 ≤ i ≤ k − 1, [fk, x] = fk, R2 : 1 ≤ i ≤ k − 1, [fi, x] = fi + fi+1, [fk, x] = fk, [x, fi] = −fi − fi+1, 1 ≤ i ≤ k − 1, [x, fk] = −fk.   Proof. Since λ = 0 implies the nilpotency of the algebra R we get λ 6= 0. Moreover, by scaling the basis elements we can suppose λ = 1. Therefore, the multiplication table of the algebra R has the following form: x′ = λ−1x, i = λ1−ifi, 1 ≤ i ≤ k, f ′   [fi, x] = fi + fi+1, [fk, x] = fk, 1 ≤ i ≤ k − 1, αi,jfj, 1 ≤ i ≤ k − 1, k [x, fi] = [x, x] = k Pj=1 Pi=1 θifi. SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE p-FILIFORM NILRADICALS 5 From the equalities L(x, fi, x) = L(x, x, f1) = 0 with 1 ≤ i ≤ k, derive the restrictions: 1 ≤ j < i ≤ k, 1 ≤ i ≤ j ≤ k, (3.1) αi,j = 0, αi,j = α1,j−i+1, α2 1,1 = −α1,1, j   Pi=1 α1,iα1,j−i+1 = −α1,j−1 − α1,j, 2 ≤ j ≤ k. Consider the possible cases. Case 1. Let α1,1 = 0. Then from the fourth equation of restrictions (3.1) we get α1,i = 0, 2 ≤ i ≤ k. By taking the change in the following form: x′ = x + k Xi=1 i Xj=1 (−1)i+j−1θjfi we obtain the algebra R1. Case 2. Let α1,1 = −1. Then from restrictions (3.1) we obtain α1,2 = −1 and α1,i = 0, 3 ≤ i ≤ k. (cid:3) The equality L(x, x, x) = 0 implies θi = 0, 1 ≤ i ≤ k. Thus, we obtain the algebra R2 Now we shall classify the opposite case to one-dimensional complemented space Q, that is, we consider the maximal dimension of Q. Theorem 3.2. The maximal possible dimension of algebras of the family R(A(k), s) is equal to 2k, that is, s = k. Moreover, an arbitrary algebra of the family R(A(k), k) is decomposed into a direct sum of copies of two-dimensional non-trivial solvable Leibniz algebras. Proof. Let {x1, x2, . . . , xs} be a basis of the complemented space of A(k) to the algebra R(A(k), s). Then from Leibniz identity 0 = [fi, [xj, xt]] = [[fi, xj], xt] − [[fi, xt], xj] we conclude that the operators RxiA(k) RxtA(k) simultaneously transformed to their Jordan forms by a basis transformation. for any 1 ≤ j, t ≤ s. This implies that all operators RxiA(k) , 1 ≤ i ≤ s, commute pairwise, that is, Rxj A(k) ◦ Rxj A(k) = , 1 ≤ i ≤ s could be ◦ RxtA(k) 1 , λ(i) Let {λ(i) Consider the vectors αi = (λ(i) 2 , . . . , λ(i) k } be the eigenvalues of the operators corresponding to RxiA(k) , 1 ≤ i ≤ s. k ), 1 ≤ i ≤ s, of the k-dimensional vector space Ck. Since Der(A(k)) ∼= Mk,k and nil-independent derivations are RxiA(k) , 1 ≤ i ≤ s. We deduce that the maximal number of nil-independent among vectors αi, 1 ≤ i ≤ s is equal to k. This means that the maximal dimension of the complemented space is equal to k, i.e. s = k. 1 , λ(i) 2 , . . . , λ(i) Without loss of generality we can assume that αi = (0, . . . , 0, correspond to RxiA(k) , 1 ≤ i ≤ k. Thus, we obtain the products in the algebras R(A(k), k): , 0, . . . , 0), λi 6= 0, 1 ≤ i ≤ k, i-th place λi {z} 1 ≤ i ≤ k − 1, 1 ≤ j ≤ k, i 6= j, [fi, xj] = ai,jfi+1, [fi, xi] = λifi + ai,ifi+1, 1 ≤ i ≤ k − 1, [fk, xk] = λkfk, [fk, xj] = 0, 1 ≤ j ≤ k − 1.   By scaling basis elements x′ The equalities i = 1 λi , 1 ≤ i ≤ k, we obtain λi = 1. derive ai,j = 0, 1 ≤ i ≤ k − 1, 1 ≤ j ≤ k. Let us introduce the notations L(fi, xi, xj ) = L(fi, xi, xi+1) = 0 [xi, fj] = αt i,jft, [xi, xj ] = k Xt=1 From L(xi, fj, xj) = 0 we have [xi, fj] = αi,j fj. γt i,jft, 1 ≤ i, j ≤ k. k Xt=1 6 SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE P -FILIFORM NILRADICALS Moreover, the equalities L(xi, xj, fj) = L(xi, xj , fi) = 0 imply the restrictions: ( αi,j = 0, α2 i,i = −αi,i, 1 ≤ i ≤ k. 1 ≤ i, j ≤ k, i 6= j, Consider the chain of equalities [xi, [xj , xi]] = [[xi, xj], xi] − [[xi, xi], xj] = γi i,jfi − γj i,ifj. On the other hand, we have k Xt=1 γt j,ifti = αi,iγi j,ifi. i,j, 1 ≤ i, j ≤ k, i 6= j, [xi, [xj , xi]] =hxi,   αi,iγi αi,iγi γj i,i = 0, j,i = γi i,i = 0, By comparing the coefficients at the basis elements we obtain 1 ≤ i ≤ k, i = j, 1 ≤ i, j ≤ k, i 6= j. (3.2) We can assume that γi restrictions we have γi again obtain γi i,i = 0. Therefore, γi Consider the chain of equalities i,i = 0 for 1 ≤ i ≤ k. Indeed, if αi,i 6= 0 for some i, then from the above i,ifi, we i,i = 0. For those i such that αi,i = 0 by taking the change x′ i,i = 0, 1 ≤ i ≤ k, that is, [xi, xi] = 0, 1 ≤ i ≤ k. i = xi − γi [xi, [xj , xt]] = [[xi, xj ], xt] − [[xi, xt], xj] = γt i,jft − γj i,tfj. On the other hand, we have [xi, [xj, xt]] =hxi, k Xl=1 From these we obtain γl j,tfli = αi,iγi j,tfi. γt i,j = 0, 1 ≤ i, j, t ≤ k, i 6= j, i 6= t, j 6= t. By taking the change of basis element: x′ i = xi − k Xt=1, t6=i γt i,tft, 1 ≤ i ≤ k, and by taking into account restrictions (3.2) we can conclude that [xi, xj ] = 0, 1 ≤ i, j ≤ k. Thus, we have the multiplication table of the family of algebras R(A(k), k): ( [fi, xi] = fi, 1 ≤ i ≤ k, [xi, fi] = αi,ifi, 1 ≤ i ≤ k, where α2 i,i = −αi,i, 1 ≤ i ≤ k. (cid:3) Remark 3.3. The number of non-isomorphic algebras in the family R(A(k), k) is equal to k + 1. It should be noted that in the work [11] the algebras R(A(2), 2) were already classified. Let l2 : [e, x] = e and r2 : [e, x] = e, [x, e] = −e be two-dimensional non-Lie Leibniz and Lie algebras. Consider the algebra Lt = l2 ⊕ l2 ⊕ · · · ⊕ l2 ⊕ r2 ⊕ r2 ⊕ · · · ⊕ r2, where t (0 ≤ t ≤ k) is the number of entries of l2 in Lt. Then there exists a basis {e1, e2, . . . , et, x1, x2, . . . , xt, y1, y2, . . . , yk−t, f1, f2, . . . fk−t} of Lt such that the multiplication table has the form: [ei, xi] = ei, 1 ≤ i ≤ t, [fj, yj] = −[yj, fj] = fj, 1 ≤ j ≤ k − t. Let us present the general form of a derivation of the algebra Lt. Proposition 3.4. Any derivation d of the algebra from Lt has the following form: d(ei) = aiei, 1 ≤ i ≤ t, d(fj) = bjfj, d(yj) = cjfj, 1 ≤ j ≤ k − t. Proof. The proof is carrying out by straightforward verification of derivation property (2.1). (cid:3) Now we could to easily calculate the dimensions of the spaces Der(Lt). SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE p-FILIFORM NILRADICALS 7 Corollary 3.5. dim Der(Lt) = 2k − t, dim BL2(Lt, Lt) = 4k2 − 2k + t. In order to prove the triviality of the second group of cohomology for the algebra Lt with coefficients in itself we need to describe the space of 2-cocycles. Proposition 3.6. Any element of ϕ ∈ ZL2(Lt, Lt) has the following form: t Pm=1 β2 i,mem + β3 i,mym + k−t Pm=1 k−t Pm=1 1 ≤ i, j ≤ t, β4 i,mfm, 1 ≤ i ≤ t, , 1 ≤ i, j ≤ t, i 6= j, 1 ≤ i ≤ t, 1 ≤ j ≤ k − t, 1 ≤ i < j ≤ k − t, j,ifi + δ2 j,ifj, ϕ(fi, yi) = ξ1 i,mxm + ξ2 i,mem + t Pm=1 i,jfi + ηj,ifj, k−t Pm=1 ϕ(xi, xj) = αi,j ej, ϕ(ei, xi) = β1 i,mxm + i,jei + νj,iej, ϕ(ej , xi) = −β2 ϕ(yj , xi) = γ1 j,ifj, ϕ(yj , yi) = −ϕ(yi, yj) = δ1 j,iei + γ2 t Pm=1 t Pm=1 ϕ(fj , yi) = −ϕ(yi, fj) = −ξ4 ϕ(fj , xi) = −ξ2 i,jei + θj,ifj, ϕ(ei, yj) = −β4 i,jfj + τj,iej, ϕ(xi, yj) = −γ2 i,jfj, ϕ(ei, ej) = −β1 j,iei, ϕ(fj , ei) = −β3 i,jfj, ϕ(yj , ei) = β4 ϕ(fi, fj) = −ϕ(fj, fi) = −ξ3 ϕ(ei, fj) = −ξ1 i,jfj, ϕ(xi, fj) = −θj,ifj, j,iei + β3 j,ifj, ξ3 i,mym + k−t Pm=1 ξ4 i,mfm, 1 ≤ i ≤ k − t, 1 ≤ i, j ≤ k − t, i 6= j, 1 ≤ i ≤ t, 1 ≤ j ≤ k − t, 1 ≤ i ≤ t, 1 ≤ j ≤ k − t, 1 ≤ i ≤ t, 1 ≤ j ≤ k − t, 1 ≤ i, j ≤ t, 1 ≤ i ≤ t, 1 ≤ j ≤ k − t, 1 ≤ i ≤ t, 1 ≤ j ≤ k − t, 1 ≤ j < i ≤ k − t, 1 ≤ i ≤ t, 1 ≤ j ≤ k − t, 1 ≤ i ≤ t, 1 ≤ j ≤ k − t. i,jfi + ξ3 j,ifj, Proof. The proof is carrying out by straightforward calculations of equations (2.2) on the basis elements of the algebra Lt. (cid:3) As consequence from Proposition 3.6 we have the following corollary. Corollary 3.7. dim ZL2(Lt, Lt) = 4k2 − 2k + t, dim HL2(Lt, Lt) = 0. Now we give the main result regarding the rigidity of algebras Lt. Theorem 3.8. The algebra Lt is rigid algebra for any values of t (0 ≤ t ≤ k). Proof. The proof of the theorem completes the argumentation of Remark 2.10 and Corollary 3.7. (cid:3) 4. Solvable n + 1-dimensional Leibniz algebras with n-dimensional naturally graded non-Lie p-filiform nilradicals. In this section we describe solvable Leibniz algebra with naturally graded non-Lie p-filiform nilrad- icals under the condition dim Q = 1. We focus in the non-split p-filiform non-Lie Leibniz algebras case. 4.1. Derivations of algebras µi, i = 1, 2, 3. In order to start the description we need to know the derivations of naturally graded non-Lie p- filiform Leibniz algebras. Proposition 4.1. Any derivation of the algebra µ1 has the following matrix form: where A = n−2k Xi=1 ia1ei,i + n−2k−1 Xi=1 D =(cid:18)A B C D(cid:19) , Xj=i+1 n−2k aj−i+1ei,j, B = bie1,i + 2k Xi=1 k Xi=1 bie2,k+i, 8 SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE P -FILIFORM NILRADICALS C = k Xi=1 ciei,n−2k, D =(cid:18)D1 0 D2 a1E + D1(cid:19) , A ∈ Mn−2k,n−2k, B ∈ Mn−2k,2k, C ∈ M2k,n−2k, D1, D2, E ∈ Mk,k and matrix units ei,j. Proof. Let {e1, f1, f2, . . . , fk} be a generator basis elements of the algebra µ1. We put d(e1) = n−2k Xi=1 aiei + 2k Xi=1 bifi, d(fi) = n−2k Xj=1 ci,jej + 2k Xj=1 di,jfj, 1 ≤ i ≤ k. From the derivation property (2.2) we have d(e2) = d([e1, e1]) = [d(e1), e1] + [e1, d(e1)] = 2a1e2 + ai−1ei + n−2k Xi=3 bifk+i. k Xi=1 Buy applying the induction and the derivation property (2.2) we derive d(ei) = ia1ei + n−2k Xt=i+1 at−i+1et, 3 ≤ i ≤ n − 2k. 0 = d([fi, e1]) = [d(fi), e1] + [fi, d(e1)] = n−2k−1 Xj=1 ci,jej+1, 1 ≤ i ≤ k. Consider Consequently, Similarly, from d(fk+i) = d([e1, fi]), 1 ≤ i ≤ k, we deduce ci,j = 0, 1 ≤ i ≤ k, 1 ≤ j ≤ n − 2k − 1. d(fk+i) = a1fk+i + k Xj=1 di,jfk+j, 1 ≤ i ≤ k. Proposition 4.2. Any derivation of the algebra µ2 has the following matrix form: (cid:3) where A = n−2k Xi=1 Xi=1 k B = 2k Xi=1 D1 = bie1,i + k Xi=1 k Xj=2 D =(cid:18)A B C D(cid:19) , Xi=1 n−2k−1 (ia1 + (i − 1)b1)ei,i + bie2,k+i, C = ciei,n−2k, k Xi=1 n−2k Xj=i+1 aj−i+1ei,j, di,jei,j + (a1 + b1)e1,1, D3 = D1 + a1E − 0 D3(cid:19) , D =(cid:18)D1 D2 Xj=1 bje1,j, k with A ∈ Mn−2k,n−2k, B ∈ Mn−2k,2k, C ∈ M2k,n−2k, D1, D2, D3, E ∈ Mk,k and matrix units ei,j. Proof. The proof is carrying out by straightforward calculation of the derivation property of the algebra µ2. (cid:3) Proposition 4.3. Any derivation of the algebra µ3 has the following matrix form: D =(cid:18)A B C D(cid:19) , where A = n−2k Xi=1 ((i − 1)a1 + a2)ei,i + n−2k Xi=2 aie1,i + n−2k−1 Xi=3 aie2,i + βe2,n−2k + n−2k−1 Xi=3 n−2k Xj=i+1 aj−i+2ei,j, SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE p-FILIFORM NILRADICALS 9 B = 2k Xi=1 b1,ie1,i + k Xi=1 b2,ie2,k+i + k Xi=1 b1,ie3,k+i, C = k Xi=1 ciei,n−2k, D =(cid:18)D1 0 D2 (a1 + a2)E + D1(cid:19) with A ∈ Mn−2k,n−2k, B ∈ Mn−2k,2k, C ∈ M2k,n−2k, D1, D2, E ∈ Mk,k and matrix units ei,j. Proof. The proof is carrying out by straightforward calculation of the derivation property of the algebra µ3. (cid:3) 4.2. Descriptions of algebras R(µi, 1), i = 1, 2, 3. Let us consider the solvable algebra R(µi, 1) = µi ⊕ Q, i = 1, 2, 3, and a basis {e1, e2, . . . , en−p, f1, f2, . . . , fp}. We set F := {f1, f2, . . . , fk} and consider the projection of the operator RxF to the space F (denoted ε(RxF )). Let us suppose that there exists a basis of the space F such that the Jordan form of the operator ε(RxF ) can be transformed into a Jordan block Jλ with λ 6= 0. Theorem 4.4. An arbitrary algebra of the family R(µ1, 1) admits a basis such that its multiplication table has the following form: R(µ1, 1)(a2, . . . , an−2k+1) : [ei, x] = aj−i+1ej, 1 ≤ i ≤ n − 2k, 1 ≤ i ≤ n − 2k − 1, 1 ≤ i ≤ k, 1 ≤ i ≤ k − 1, k + 1 ≤ i ≤ 2k − 1, 1 ≤ i ≤ k − 1, [ei, e1] = ei+1, [e1, fi] = fk+i, n−2k Pj=i+1 [fi, x] = fi + fi+1, [fk, x] = fk, [fi, x] = fi + fi+1, [f2k, x] = f2k, [x, fi] = −fi − fi+1, [x, fk] = −fk, [x, x] = an−2k+1en−2k.   Proof. From Proposition 4.1 we have the products in the algebra R(µ1, 1): [ei, e1] = ei+1, [e1, fi] = fk+i, n−2k [e1, x] = aiei + 1 ≤ i ≤ n − 2k − 1, 1 ≤ i ≤ k, bifi, [e2, x] = 2a1e2 + ai−1ei + [ei, x] = ia1ei + aj−i+1ej, [fi, x] = cien−2k + di,jfj, k Pi=1 bifk+i, 3 ≤ i ≤ n − 2k, 1 ≤ i ≤ k, Pi=1 2k n−2k n−2k Pi=1 Pi=3 Pj=i+1 Pj=1 Pj=1 2k k [fk+i, x] = a1fk+i + di,jfk+j, 1 ≤ i ≤ k.   Let us introduce the notations: [x, e1] = n−2k Xi=1 βiei+ 2k Xi=1 βn−2k+ifi, [x, fi] = n−2k Xj=1 γi,jej+ 2k Xj=1 ϕi,j fj, 1 ≤ i ≤ k, [x, x] = n−2k Xi=1 δiei+ θifi. 2k Xi=1 Since the space F forms an abelian algebra, we are in the conditions of Theorem 3.1. Moreover, the products [e1, e1] = e2, [e1, fi] = fk+i ensure that e1, fi /∈ Annr(R(µ1, 1), 1 ≤ i ≤ k. In particular, we 10 SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE P -FILIFORM NILRADICALS are in the conditions of the algebra R2. This implies the existence of a basis of F such that 2k Pj=k+1 di,j fj, 1 ≤ i ≤ k − 1, 2k Pj=k+1 dk,j fj, ϕi,jfj, 1 ≤ i ≤ k − 1, 2k Pj=k+1 ϕk,j fj, [fi, x] = cien−2k + fi + fi+1 + [fk, x] = cken−2k + fk + [x, fi] = [x, fk] = n−2k n−2k Pj=1 Pj=1 Pi=1 n−2k γi,j ej − fi − fi+1 + 2k Pj=k+1 θifi. γk,jej − fk + δiei + 2k Pi=k+1 L(x, fi, e1) = L(e1, x, e1) = L(x, e1, fi) = 0 [x, x] = The equalities imply γi,j = β1 = 0, α1 = 0, βn−2k+i = −bi, 1 ≤ i ≤ k, 1 ≤ i ≤ k. [x, fk+i] = 0, 1 ≤ i ≤ k, 1 ≤ j ≤ n − 2k − 1, Since {e2, e3, . . . , en−2k, fk+1, . . . , f2k} ⊆ Annr(R(µ1, 1)) and [x, x] ∈ Annr(R(µ1, 1)) while e1 /∈ Annr(R(µ1, 1)), we conclude δ1 = 0. By setting f ′ i = fi − χi,n−2ken−2k − 2k Xj=k+1 ψi,jfj, 1 ≤ i ≤ k, where χk,n−2k = γk,n−2k, ψk,j = ϕk,j and parameters χi,n−2k, ψi,j for 1 ≤ i ≤ k − 1, can be recursively obtained from the products [x, f ′ i ] + f ′ i = −f ′ i+1, we can assume that The equalities L(x, fi, x) = 0, 1 ≤ i ≤ k we derive ci = di,j = 0, 1 ≤ i ≤ k, k + 1 ≤ j ≤ 2k, that [x, fi] = −fi − fi+1, 1 ≤ i ≤ k − 1, [x, fk] = −fk. is, we obtain By taking the change of basis element: [fi, x] = fi + fi+1, 1 ≤ i ≤ k − 1, [fk, x] = fk. x′ = x − n−2k Xi=2 βiei−1, βn−k+ifk+i, that is, we can assume that βi = 0, 2 ≤ i ≤ n − 2k. we obtain [x′, e1] = − bifi + The equality L(x, e1, x) = 0 implies k Pi=1 k Pi=1 δi = 0, 2 ≤ i ≤ n − 2k − 1, βn−k+i = 0, 1 ≤ i ≤ k. By putting e′ 1 = e1 + (−1)i+j−1bjfi + k Xi=1 (−1)i+j−1bk+jfk+i, i Xj=1 k i Xj=1 Xi=1 Xj=1 Xi=1 k i x′ = x + we deduce (−1)i+j−1θk+j fk+i, Thus, we obtain the multiplication tables of the algebras R(µ1)(a2, . . . , an−2k+1) of the assertion of the theorem. (cid:3) bi = bk+i = θk+i = 0, 1 ≤ i ≤ k. SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE p-FILIFORM NILRADICALS 11 In the next proposition a necessary and sufficient condition for the existence of an isomorphism between two algebras of the family R(µ1, 1)(a2, . . . , an−2k+1) is established. Proposition 4.5. Two algebras R(µ1, 1)′(a′ phic if and only if there exists A ∈ C such that 2, . . . , a′ n−2k+1) and R(µ1, 1)(a2, . . . , an−2k+1) are isomor- a′ i = ai Ai−1 , 2 ≤ i ≤ n − 2k + 1. Proof. Let us consider the general change of generator basis elements of the algebra R(µ1, 1): Aiei + Bifi, Ci,j ej + Di,jfj, 1 ≤ i ≤ k, e′ 1 = f ′ i = n−2k n−2k Pi=1 Pj=1 2k 2k Pi=1 Pj=1 x′ = Hx + Eiei + n−2k Pi=1 Fifi. 2k Pi=1 From the products [e′ i, e′ 1] = e′ i+1, 1 ≤ i ≤ n − 2k − 1, of the algebra R(µ1, 1)′ we derive e′ 2 = A1 n−2k Xi=2 Ai−1ei + A1 k Xi=1 Bifk+i, i = Ai−1 e′ 1 n−2k Xj=i Aj−i+1ej, 3 ≤ i ≤ n − 2k. By considering we deduce [x′, e′ 1] = [f ′ i , e′ 1] = [f ′ i , f ′ i ] = 0, 1 ≤ i ≤ k, Ci,j = Bi = Ej = 0, 1 ≤ i ≤ k, 1 ≤ j ≤ n − 2k − 1. Similarly, from the products [x′, f ′ i ] = −f ′ i − f ′ i+1, 1 ≤ i ≤ k − 1, and [x′, f ′ k] = −f ′ k we conclude H = 1, Di,j = 0, Di,j = D1,j−i+1, Ci,n−2k = Di,k+j = 0, 1 ≤ i ≤ k, 1 ≤ j ≤ k. 1 ≤ j < i ≤ k, 1 ≤ i ≤ j ≤ k, The relations between parameters {a′ i} and {ai} follow from the products [e′ 1, x′] = ie′ a′ i, n−2k Xi=2 Namely, we have [x′, x′] = a′ n−2k+1e′ n−2k. a′ i = ai Ai−1 1 , 2 ≤ i ≤ n − 2k + 1. (cid:3) Below, we present an analogue of Theorem 4.4 for the family R(µ2, 1). 12 SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE P -FILIFORM NILRADICALS Theorem 4.6. An arbitrary algebra of the family R(µ2, 1) admits a basis such that its multiplication table has the following form: R(µ2, 1)(α, β, γ) : µ2, [e1, x] = f1 + αfk+1, [e2, x] = e2 + fk+1, [ei, x] = (i − 1)ei, [fi, x] = −[x, fi] = fi + fi+1, [fk, x] = −[x, fk] = fk, [fk+1, x] = fk+2, [fi, x] = fi + fi+1, [f2k, x] = f2k, [x, e1] = −f1 + β [x, x] = γfk+1. k Pi=1 (−1)i−1fk+i, 3 ≤ i ≤ n − 2k, 1 ≤ i ≤ k − 1, k + 2 ≤ i ≤ 2k − 1,   The classification of the family of algebras R(µ2, 1)(α, β, γ) is presented in the following proposition. Proposition 4.7. Any algebra of the family R(µ2, 1)(α, β, γ) is isomorphic to one of the following pairwise non-isomorphic algebras: R(µ2, 1)(0, 0, 0), R(µ2, 1)(0, 0, 1), R(µ2, 1)(0, 1, γ), R(µ2, 1)(1, β, γ), β, γ ∈ C. Proof. In a similar way as in the proof of Proposition 4.5, we consider the general transformation of generator basis elements: e′ 1 = f ′ i = x′ = Then from the products n−2k n−2k Pi=1 Pj=1 Pi=1 n−2k Aiei + Bifi, Ci,j ej + Di,jfj, 1 ≤ i ≤ k, Eiei + Fifi + Hx. 2k 2k Pi=1 Pj=1 Pi=1 2k [e′ i, e′ 1] = e′ i+1, 1 ≤ i ≤ n − 2k − 1, f ′ k+1 = [e′ 1, f ′ 1] − e′ 2, f ′ k+i = [e′ 1, f ′ i ], 2 ≤ i ≤ k, we derive the rest of new basis elements e′ 2 = (A1 + B1) Ai−1ei + A1 e′ i = (A1 + B1)i−1 Aj−i+1ej, n−2k n−2k Pi=2 Pj=i Di,jfk+j, (D1,i − Bi))fk+i, k Pi=1 Bifk+i, 3 ≤ i ≤ n − 2k, A1 6= 0 2 ≤ i ≤ k. f ′ k+1 = A1 f ′ k+i = A1 k k Pi=1 Pj=2 By verifying the multiplications of R(µ2, 1)(α′, β′, γ′) in the new basis we obtain the relations be- tween the parameters {α′, β′, γ′} and {α, β, γ} : α′ = α A1 , β′ = β A1 , γ′ = γ A2 1 . (cid:3) Analogously, we have the description of solvable algebras.   [e2, x] = (a1 + a2)e2 + [e3, x] = (2a1 + a2)e3 + aiei + βen−2k, ai−1ei + b1fk+1, n−2k−1 n−2k Pi=4 Pi=5 n−2k Pj=i+2 R(µ3, 1)(I) : [ei, x] = ((i − 1)a1 + a2)ei + aj−i+2ej, 4 ≤ i ≤ n − 2k, 1 ≤ i ≤ k − 1, 1 ≤ i ≤ k − 1, [fi, x] = −[x, fi] = fi + fi+1, [fk, x] = −[x, fk] = fk, [fk+i, x] = (a1 + a2 + 1)fk+i + fk+i+1, [f2k, x] = (a1 + a2 + 1)f2k, k [x, e1] = −a1e1 − b1f1 + θn−k+ifk+i, [x, e2] = γen−2k, [x, x] = δ1en−2k−1 + δ2en−2k + δ3fk+1, Pi=1 SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE p-FILIFORM NILRADICALS 13 Theorem 4.8. An arbitrary algebra of the family R(µ3, 1) admits a basis such that its multiplication table has the following form: [ei, e1] = ei+1, [e2, fi] = fk+i, [e1, x] = a1e1 + an−2ken−2k + b1f1 + b2fk+1, 2 ≤ i ≤ n − 2k − 1, 1 ≤ i ≤ k, where I = {ai, b1, b2, β, γ, δ1, δ2, δ3, θn−k+j } with 1 ≤ i ≤ n − 2k, i 6= 3, 1 ≤ j ≤ k, and for these parameters the following equalities hold true (a1 + a2)γ = 0, δ1 = −a1an−2k, θn−i = (−1)i(a2 + 1)iθn, 0 ≤ i ≤ k − 1. (n − 2k − 2)a1γ = 0, a1b2 = (−1)k−1(a2 + 1)kθn, (4.1) We set J := I \ {ai, b1, b2, β, γ, δ1, δ2, δ3, θn−k+j } with 4 ≤ i ≤ n − 2k, 1 ≤ j ≤ k. Lemma 4.9. Let algebras R(µ3, 1)(a1, a2, J) and R(µ3, 1)(a′ 1, a′ 2, J ′) are isomorphic. Then Proof. Let us take the general transformation of generators basis elements of R(µ3)(a1, a2, J) in the following form: the algebra a′ 1 = a1, a′ 2 = a2. e′ 1 = f ′ i = n−2k n−2k Pi=1 Pj=1 A1,iei + Ci,j ej + 2k 2k Pi=1 Pj=1 B1,ifi, e′ 2 = A2,iei + B2,ifi, Di,jfj, 1 ≤ i ≤ k, x′ = Hx + Eiei + n−2k Pi=1 2k Pi=1 n−2k Pi=1 Fifi. 2k Pi=1 Then from the products [e′ i, e′ 1] = e′ i+1, 2 ≤ i ≤ n − 2k − 1, we obtain e′ 3 = A1,1 i = Ai−2 e′ 1,1 Moreover, the equalities n−2k n−2k Xi=3 Xj=i A2,i−1ei + A2,2 B1,ifk+i, k Xi=1 A2,j−i+2ej, 4 ≤ i ≤ n − 2k. [e′ 1, e′ 1] = [e′ 2, e′ 2] = [f ′ i , e′ 1] = [e′ 3, f ′ i ] = 0, 1 ≤ i ≤ k, imply A1,i = 0, A2,1 = 0 = B2,i = Ci,j = 0, 1 ≤ i ≤ k, 1 ≤ j ≤ n − 2k − 1. 2 ≤ i ≤ n − 2k − 1, 14 SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE P -FILIFORM NILRADICALS The rest basis elements f ′ k+i, 1 ≤ i ≤ k, can be obtained from the products [e′ 2, f ′ i ] = f ′ k+i, 1 ≤ i ≤ k. Namely, we have f ′ k+i = A2,2 Consider [x′, f ′ k] = −f ′ k. Then we derive Di,jfk+j, 1 ≤ i ≤ k. k Xj=1 (Dk,i + Dk,i+1)H = Dk,i+1, 1 ≤ i ≤ k − 1, Dk,1H = Dk,1, Dk,iE2 = −Dk,k+i, Ck,n−2k = 0. 1 ≤ i ≤ k, If H 6= 1, then Dk,i = 0, 1 ≤ i ≤ k, which implies f ′ 2k = 0, this is a contradiction. Therefore, H = 1 and Dk,i = Dk,k+i = 0, 1 ≤ i ≤ k − 1. Now from [x′, f ′ i+1 with 1 ≤ i ≤ k − 1, we get restrictions: i − f ′ i ] = −f ′ Di,j = 0, Di,j = D1,j−i+1, Ci,n−2k = 0, Di,jE2 = −Di,k+j − Di+1,k+j , 1, x′] and [e′ By considering the products [e′ 2, x′], we deduce a′ 2 = a2. a′ 1 = a1, 1 ≤ j < i ≤ k, 1 ≤ i ≤ j ≤ k, 1 ≤ i ≤ k − 1, 1 ≤ i ≤ k − 1, 1 ≤ j ≤ k. Let us introduce the notations I1 = {ai, β, γ, δ2}, I2 = {a2 = −1, ai, b2, β, δ3, θn}, I3 = {ai, β}, I4 = {a1, a2, an−2k, b1, b2, δ1, δ2, δ3, θn−k+i}, 1 ≤ i ≤ k, 4 ≤ i ≤ n − 2k, 4 ≤ i ≤ n − 2k, 2 ≤ i ≤ n − 2k, i 6= 3, a2 /∈ {−1, 0}, where for parameters from the set I4 we have the relations (cid:3) Proposition 4.10. δ1 = −a1an−2k, b2 = (−1)k−1(a2 + 1)kθn a1 , θn−k+i = (−1)k−i(a2 + 1)k−iθn, 1 ≤ i ≤ k − 1. R(µ3, 1)(I) = R(µ3, 1)(Ij ). 4 [j=1 Proof. Thanks to Lemma 4.9 for analysis of equalities (4.1) it is sufficient to consider the following cases. Case 1. Let a1 = 0. Then we get δ1 = θn−k+i = 0, 1 ≤ i ≤ k − 1 and a2γ = (a2 + 1)θn = 0. By applying the change e′ 1 = e1 + (−1)ib1fi, we obtain b1 = 0. Case 1.1. Let a2 = 0. Then θn = 0. By taking the change of basis elements e′ 1 = e1 + (−1)ib2fk+i, x′ = x + (−1)iδ3fk+i, k Xi=1 we can assume b2 = δ3 = 0. Thus, the subfamily of algebras R(µ3, 1)(I1) is obtained. Case 1.2. Let a2 6= 0. Then we have γ = (a2 + 1)θn = 0. By setting x′ = x − δ2 a2 δ2 = 0. en−2k, we get k Pi=1 k Xi=1 SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE p-FILIFORM NILRADICALS 15 Case 1.2.1. Let a2 = −1. Then we obtain the subfamily R(µ3, 1)(I2). Case 1.2.2. Let a2 6= −1. Then θn = 0. By putting e′ 1 = e1 + we get b2 = δ3 = 0. (−1)i b2 (a2 + 1)i fk+i, x′ = x + k Xi=1 (−1)i δ3 (a2 + 1)i fk+i, k Xi=1 Thus, we have the class of algebras R(µ3, 1)(I3). Case 2. Let a1 6= 0. Then γ = 0. By applying the change e′ 1 = e1, e′ i = ei + 2 ≤ i ≤ n − 2k, n−2k 1 Aj−i+2ej, Xj=i+1 (i − 2)a1 (cid:16) i−1 Xj=3 (n − 2k − 2)a1(cid:16) n−2k−1 Xj=3 Ajai+2−j + ai(cid:17), 4 ≤ i ≤ n − 2k − 1, Ajan−2k+2−j + β(cid:17), with A3 = 0, a3 = 0, Ai = − An−2k = − 1 we derive β = ai = 0, 4 ≤ i ≤ n − 2k − 1. Therefore, we obtain the class of algebras R(µ3, 1)(I4). (cid:3) Remark 4.11. It should be noted that algebras from different subfamilies R(µ3, 1)(Ij ), 1 ≤ j ≤ 4, are not isomorphic. Acknowledgements This work was partially supported by Ministerio de Economía y Competitividad (Spain), grant MTM2013-43687-P (European FEDER support included), by Xunta de Galicia, grant GRC2013-045 (European FEDER support included) and by Kazakhstan Ministry of Education and Science, grant 0828/GF4: "Algebras, close to Lie: cohomologies, identities and deformations". The last named author was partially supported by a grant from the Simons Foundation. References [1] J. M. Ancochea Bermúdez, R. Campoamor-Stursberg, L. García Vergnolle Indecomposable Lie algebras with non- trivial Levi decomposition cannot have filiform radical. Int. Math. Forum, vol. 1(5-8), 2006, p. 309 -- 316. [2] J. M. Ancochea Bermúdez, R. Campoamor-Stursberg, L. García Vergnolle Solvable Lie algebras with naturally graded nilradicals and their invariants. J. Phys. A, vol. 39(6), 2006, p. 1339 -- 1355. [3] J. M. Ancochea Bermúdez, R. Campoamor-Stursberg, L. García Vergnolle Classification of Lie algebras with natu- rally graded quasi-filiform nilradicals. J. Geom. Phys., vol. 61(11), 2011, p. 2168 -- 2186. [4] Sh.A. Ayupov, B.A. Omirov On some classes of nilpotent Leibniz algebras. Siberian Math. J., vol. 42(1), 2001, p. 15 -- 24. [5] D. Balavoine Déformations et rigidité géométrique des algèbres de Leibniz. Comm. Algebra, vol. 24, 1996, p. 1017 -- 1034. [6] D.W. Barnes On Levi's theorem for Leibniz algebras. Bull. Aust. Math. Soc., vol. 86(2), 2012, p. 184 -- 185. [7] D. Burde Degenerations of 7-dimensional nilpotent Lie algebras. Comm. Algebra, vol. 33(4), 2005, p. 1259 -- 1277. [8] J.M. Cabezas, J.R. Gómez, A. Jimenez-Merchán Family of p-filiform Lie algebras. Algebra and operator theory (Tashkent, 1997), Kluwer Acad. Publ., Dordrecht, 1998, p. 93 -- 102. [9] J.M. Cabezas, E. Pastor Naturally graded p-filiform Lie algebras in arbitrary finite dimension. J. Lie Theory, vol. 15(2), 2005, p. 379 -- 391. [10] L.M. Camacho, J.R. Gómez, A.J. González, B.A. Omirov The classification of naturally graded p-filiform Leibniz algebras. Comm. Algebra, vol. 39(1), 2011, p. 153 -- 168. [11] E.M. Cañete, A. Kh. Khudoyberdiyev The classification of 4-dimensional Leibniz algebras. Lin. Alg. Appl., vol. 439(1), 2013, p. 273 -- 288. [12] J.M. Casas, M. Ladra, B.A. Omirov, I.A. Karimjanov Classification of solvable Leibniz algebras with null-filiform nilradical. Lin. Multilin. Algebra, vol. 61(6), 2013, p. 758 -- 774. [13] M. Goze, Yu. Khakhimdjanov, Nilpotent Lie algebras. Kluwer Academic Publishers, Dordrecht, vol. 361, 1996, 336 pp. [14] M. Gerstenhaber On the deformation of rings and algebras, I, III. Ann. of Math. (2), vol. 79, 1964, p. 59 -- 103. [15] F. Grunewald, J. O'Halloran A characterization of orbit closure and applications. J. Algebra, vol. 116(1), 1988, p. 163 -- 175. 16 SOLVABLE LEIBNIZ ALGEBRAS WITH NATURALLY GRADED NON-LIE P -FILIFORM NILRADICALS [16] F. Grunewald, J. O'Halloran Varieties of nilpotent Lie algebras of dimension less than six. J. Algebra, vol. 112(2), 1988, p. 315 -- 325. [17] J.-L. Loday Une version non commutative des algèbres de Lie: les algèbres de Leibniz. Enseign. Math. (2), vol. 39(3-4), 1993, p. 269 -- 293. [18] J.-L. Loday, T. Pirashvili Universal enveloping algebras of Leibniz algebras and (co)homology. Math. Ann., vol. 296, 1993, p. 139 -- 158. [19] A.I. Malcev Solvable Lie algebras. Amer. Math. Soc. Translation, vol. 36(27), 1950. [20] G.M. Mubarakzjanov On solvable Lie algebras (Russian). Izv. Vysš. Učehn. Zaved. Matematika, vol. 32(1), 1963, p. 114 -- 123. [21] J.C. Ndogmo, P. Winternitz Solvable Lie algebras with abelian nilradicals. J. Phys. A, vol. 27(2), 1994, p. 405 -- 423. [22] A. Nijenhuis, R.W. Richardson, Cohomology and deformations in graded Lie algebras. Bull. Amer. Math. Soc., vol. 72, 1966, p. 1 -- 29. [23] L. Šnobl, P. Winternitz A class of solvable Lie algebras and their Casimir invariants. J. Phys. A, vol. 38(12), 2005, p. 2687 -- 2700. [24] S. Tremblay, P. Winternitz Solvable Lie algebras with triangular nilradicals. J. Phys. A, vol. 31(2), 1998, p. 789 -- 806. [25] Y. Wang, J. Lin, Sh. Deng Solvable Lie algebras with quasifiliform nilradicals. Comm. Algebra, vol. 36(11), 2008, p. 4052 -- 4067.
1507.04466	1	1507	2015-07-16T07:03:56	Natural dualities through product representations: bilattices and beyond	[ "math.RA" ]	This paper focuses on natural dualities for varieties of bilattice-based algebras.Such varieties have been widely studied as semantic models in situations where information is incomplete or inconsistent. The most popular tool for studying bilattices-based algebras is product representation. The authors recently set up a widely applicable algebraic framework which enabled product representations over a base variety to be derived in a uniform and categorical manner. By combining this methodology with that of natural duality theory, we demonstrate how to build a natural duality for any bilattice-based variety which has a suitable product representation over a dualisable base variety. This procedure allows us systematically to present economical natural dualities for many bilattice-based varieties, for most of which no dual representation has previously been given. Among our results we highlight that for bilattices with a generalised conflation operation (not assumed to be an involution or commute with negation). Here both the associated product representation and the duality are new. Finally we outline analogous procedures for pre-bilattice-based algebras (so negation is absent).	math.RA	math	NATURAL DUALITIES THROUGH PRODUCT REPRESENTATIONS: BILATTICES AND BEYOND L. M. CABRER AND H. A. PRIESTLEY Abstract. This paper focuses on natural dualities for varieties of bilattice- based algebras. Such varieties have been widely studied as semantic models in situations where information is incomplete or inconsistent. The most popular tool for studying bilattices-based algebras is product representation. The au- thors recently set up a widely applicable algebraic framework which enabled product representations over a base variety to be derived in a uniform and cat- egorical manner. By combining this methodology with that of natural duality theory, we demonstrate how to build a natural duality for any bilattice-based variety which has a suitable product representation over a dualisable base va- riety. This procedure allows us systematically to present economical natural dualities for many bilattice-based varieties, for most of which no dual rep- resentation has previously been given. Among our results we highlight that for bilattices with a generalised conflation operation (not assumed to be an involution or commute with negation). Here both the associated product rep- resentation and the duality are new. Finally we outline analogous procedures for pre-bilattice-based algebras (so negation is absent). Keywords: product representation, natural duality, bilattice, conflation, double Ockham algebra. 1. Introduction Bilattices, with and without additional operations, have been identified by re- searchers in artificial intelligence and in philosophical logic as of value for analysing scenarios in which information may be incomplete or inconsistent. Over twenty years, a bewildering array of different mathematical models has been developed which employ bilattice-based algebras in such situations; [19, 23, 15, 26] give just a sample of the literature. Within a logical context, bilattices have been used to interpret truth values of formal systems. The range of possibilities is illustrated by [2, 1, 17, 18, 16, 5, 27, 25]. To date, the structure theory of bilattices has had two main strands: product representations (see in particular [4, 11, 9] and references therein) and topological duality theory [24, 22, 8]. In this paper we entwine these two strands, demonstrating how a dual representation and a product representation can be expected to fit together and to operate in a symbiotic way. Our work on distributive bilattices in [8] provides a prototype. Crucially, as in [8], we exploit the theory of natural dualities; see Section 3. In [9] we set up a uniform framework for product representation. We introduced a formal definition of duplication of a base variety of algebras which gives rise to a new variety with additional operations built by combining suitable algebraic terms in the base language and coordinate manipulation (details are recalled in Section 2). This construction led to a very general categorical theorem on product representation [9, Theorem 3.2] which makes overt the intrinsic structure of such representations. The examples we present below all involve bilattice-based varieties, but we stress that 2010 Mathematics Subject Classification. primary: 08C20, secondary 03G10, O3G25, 06B10, 06D50. 1 2 L. M. CABRER AND H. A. PRIESTLEY the scope of the theorem is not confined to such varieties. Our Duality Transfer Theorem (Theorem 3.1) demonstrates how a natural duality for a given base class immediately yields a natural duality for any duplicate of that class. Moreover, the dualities for duplicated varieties mirror those for the base varieties, as regards both advantageous properties and complexity (note the concluding remarks in Section 4). By combining the Duality Transfer Theorem with product representation we can set up dualities for assorted bilattice-based varieties (see Section 4, Table 1). In almost all cases the dualities are new. The varieties in question arise as duplicates of B (Boolean algebras), D (bounded distributive lattices) K (Kleene algebras), DM (De Morgan algebras), and DB (bounded distributive bilattices), all of which have amenable natural dualities (see [10] and also [8]). Variants are available when lattice bounds are omitted. We contrast key features of our natural duality approach with earlier work on dualities for bilattice-based algebras. We stress that our methods lead directly to dual representations which are categorical: morphisms do not have to be treated case-by-case as an overlay to an object representation (as is done in [24, 22]). Oth- ers' work on dualities in the context of distributive bilattices has sought instead, for a chosen class of algebras, a dual category which is an enrichment of a subcat- egory of Priestley spaces, that is, they start from Priestley duality, applied to the distributive lattice reducts of their algebras, and then superimpose extra structure to capture the suppressed operations. This strategy has been successfully applied to very many classes of distributive-lattice-based algebras, but it has drawbacks. Although the underlying Priestley duality is natural, the enriched Priestley space representation rarely is. Accordingly one cannot expect the rewards a natural du- ality offers, such as instant access to free algebras. Section 5 focuses on the variety DB´ of (bounded) distributive bilattices with a conflation operation ´ which is not assumed to be an involution or to commute with the negation. This variety has not been investigated before and would not have susceptible to earlier methods. We realise DB´ as a duplicate of the variety DO of double Ockham algebras and set up a natural duality for DO, whence we obtain a duality for DB´. Both results are new. This example is also a novelty within bilattice theory since it takes us outside the realm of finitely generated varieties without losing the benefits of having a natural duality. In Section 6 we consider the negation-free setting of pre-bilattice-based algebras, and link the ideas of [9, Section 9] with dual representations. Again, a very general theorem enables us to transfer a known duality from a base variety to a suitably constructed duplicate. Here multisorted duality theory is needed. Nonetheless the ideas and the categorical arguments are simple, and the proof of Theorem 3.1 is easily adapted. 2. The general product representation theorem recalled We shall assume that readers are familiar with the basic notions concerning bilattices. A summary can be found, for example, in [4] and a bare minimum in [9, Section 2]. Here we simply draw attention to some salient points concerning notation and terminology since usage in the literature varies. Except in Section 6 we assume that a negation operator is present. A (unbounded) bilattice is an algebra A " pA; _t, ^t, _k, ^k, q, where the reducts At :" pA; _t, ^tq and Ak :" pA; _k, ^kq are lattices (respectively the truth lattice and knowledge lattice). The operation , capturing negation, is an endomorphism of Ak and a dual endomorphism of At. NATURAL DUALITIES THROUGH PRODUCT REPRESENTATIONS 3 Bilattice models come in two flavours: with and without bounds. Which flavour is preferred (or appropriate) may depend on an intended application, or on math- ematical considerations. We refer to [8, Section 1] for the formal definition of the terms bounded and unbounded. Here we merely issue a reminder that when uni- versal bounds for the lattice order are not included in the algebraic language for a class of lattice-based algebras then the algebras involved may, but need not, have bounds; when bounds do exist these do not have to be preserved by homomorph- isms. A subscript u on the symbol denoting a category will indicate that we are working in the unbounded setting. So, for example, D denotes the category of bounded distributive lattices and Du the category of all distributive lattices. All the bilattices considered in this paper are distributive, meaning that each of the four lattice operations distributes over each of the other three. The weaker condition of interlacing is necessary and sufficient for a bilattice to have a product representation. However varieties of interlaced bilattice-based algebras seldom come within the scope of natural duality theory. Our investigations involve classes of algebras, viewed both algebraically and cat- egorically. We draw, lightly, on some of the basic formalism and theory of universal algebra, specifically regarding varieties (alias equational classes) and prevarieties; a standard reference for this material is [6]. A class of algebras over a common language will be regarded as a category in the usual way: the morphisms are all the homomorphisms. The variety generated by a family M of algebras of common type is denoted VpMq. Equivalently VpMq is the class HSPpMq of homomorphic images of subalgebras of products of algebras in M. The prevariety generated by M is the class ISPpMq whose members are isomorphic images of subalgebras of products of members of M. Usually the algebras in M will be finite. We now recall our general product representation framework [9, Section 3]. We fix an arbitrary algebraic language Σ and let N be a family of Σ-algebras. Let Γ be a set of pairs of Σ-terms such that, for pt1, t2q P Γ, the terms t1 and t2 have common even arity, denoted 2npt1,t2q. We view Γ as an algebraic language for a family of algebras PΓpNq (N P N ), where the arity of pt1, t2q P Γ is npt1,t2q. We write rt1, t2s when the pair pt1, t2q is regarded as belonging to Γ, qua language. For A P VpN q we define a Γ-algebra PΓpAq " pA A; trt1, t2sPΓpAq pt1, t2q P Γuq, in which the operation rt1, t2sPΓpAq : pA Aqn Ñ A A is given by rt1, t2sPΓpAqppa1, b1q, . . . , pan, bnqq " ptA 1 pa1, b1, . . . , an, bnq, tA 2 pa1, b1, . . . , an, bnqq, where n " npt1,t2q and pa1, b1q, . . . , pan, bnq P A A. It is easy to check that the assignment A ÞÑ PΓpAq (on objects) and h ÞÑ h h (on morphisms) defines a functor PΓ : VpN q Ñ VpPΓpN qq. We shall also need the following notation. Given a set X the map δX : X Ñ X X is given by δX pxq " px, xq and πX 2 : X X Ñ X denote the projection maps. 1 , πX We are ready to recall a key definition from [9, Section 3], where further details can be found. We say that Γ duplicates N and that A " VpPΓpN qq is a duplicate of B if the following conditions on N and Γ are satisfied: (L) for each n-ary operation symbol f P Σ and each i P t1, 2u there exists an i tPΓpNq pδN qn " f N n-ary Γ-term t (depending on f and i) such that πN for each N P N ; (M) there exists a binary Γ-term v such that vPΓpNqppa, bq, pc, dqq " pa, dq for N P N and a, b P N ; (P) there exists a unary Γ-term s such that sPΓpNqpa, bq " pb, aq for N P N and a, b P N . 4 L. M. CABRER AND H. A. PRIESTLEY We now present the Product Representation Theorem [9, Theorem 3.2]. Theorem 2.1. Assume that Γ duplicates a class of algebras N and let B " VpN q. Then the functor PΓ : B Ñ A sets up a categorical equivalence between B and its duplicate A " VpPΓpN qq. The classes of algebras arising in this section have prinicipally been varieties. In the next section we concentrate on singly-generated prevarieties. The following corollary tells us how the class operators HSP and ISP behave with respect to du- plication. It is an almost immediate consequence of the fact that PΓ is a categorical equivalence; assertion (c) follows directly from (a) and (b). Corollary 2.2. Assume that Γ duplicates a class of algebras M. The following statements hold for each A P VpMq: (a) HSPpPΓpAqq is categorically equivalent to HSPpAq. (b) ISPpPΓpAqq is categorically equivalent to ISPpAq. (c) If VpAq " ISPpAq then VpPΓpAqq " ISPpPΓpAqq. 3. Natural duality and product representation It is appropriate to recall only in brief the theory of natural dualities as we shall employ it. A textbook treatment is given in [10] and a summary geared to applications to distributive bilattices in [8, Sections 3 and 5]. Our object of study in this section will be a prevariety A generated by an algebra M, so that A " ISPpMq. (Only in Section 6 will we replace the single algebra M by a family of algebras M. We shall then need to bring multisorted duality theory into play.) Traditionally (and in [10] in particular) M is assumed to be finite. This suffices for our applications in Section 4. However our application to bilattices with gener- alised conflation will depend on the more general theory presented in [12]. Therefore we shall assume that M can be equipped with a compact Hausdorff topology T with respect to which it becomes a topological algebra. When M is finite T is necessarily discrete. Our aim is to find a second category X whose objects are topological structures of common type and which is dually equivalent to A via functors D : A Ñ X and E : X Ñ A. Moreover -- and this is a key feature of a natural duality -- we want each algebra A in A to be concretely representable as an algebra of continuous structure-preserving maps from DpAq (the dual space of A) into M„ , where M„ P X has the same underlying set M as does M. For this to succeed, some compatibility between the structures M and M„ will be necessary. We consider a topological structure M„ " pM ; G, R, Tq where ‚ T is a topology on M (as demanded above); ‚ G is a set of operations on M , meaning that, for g P G of arity n ě 1, the map g : Mn Ñ M is a continuous homomorphism (any nullary operation in G will be identified with a constant in the type of M); ‚ R is a set of relations on M such that if r P R is n-ary (n ě 1) then r is the universe of a topologically closed subalgebra r of Mn. We refer to such a topological structure M„ as an alter ego for M and say that M„ and M are compatible. Of course. the topological conditions imposed on G and R are trivially satisfied if M is finite. (The general theory in [10] allows an alter ego also to include partial operations, but they do not arise in our intended applications.) We use M„ to build a new category X. We first consider structures of the same type as M„ . These have the form X " pX; GX, RX, TXq where TX is a compact Hausdorff topology and GX and RX are sets of operations and relations NATURAL DUALITIES THROUGH PRODUCT REPRESENTATIONS 5 on X in bijective correspondence with those in G and R, with matching arities. Isomorphisms between such structures are defined in the obvious way. For any non-empty set S we give M S the product topology and lift the elements of G and R pointwise to M S. The topological prevariety generated by M„ is X :" IScP`pM„ q, the class of isomorphic copies of closed substructures of non-empty powers of M„ , with ` indicating that the empty structure is included. We make X into a category by taking all continuous structure-preserving maps as the morphisms. As a consequence of the compatibility of M„ and M, and the topological con- ditions imposed, the following assertions are true. Let A P A and X P X. Then A and XpX, M„ q as a sub- ApA, Mq may be seen as a closed substrucructure of M„ algebra of MX. We can set up well-defined contravariant hom-functors D : A Ñ X and E : XT Ñ A; on objects: on morphisms: and on objects: on morphisms: D : A ÞÑ ApA, Mq, D : x ÞÑ ´ x, E : X ÞÑ XpX, M„ q, E : φ ÞÑ ´ φ, The following assertions are part of the standard framework of natural duality theory. Details can be found in [10, Chapter 2]; see also [12, Section 2]. Given A P A and X P X, we have natural evaluation maps eA : a ÞÑ ´a and εX : x ÞÑ ´x, with eA : A Ñ EDpAq and εX : X Ñ DEpXq. Moreover pD, E, e, εq is a dual adjunction. Each of the maps eA and εX is an embedding. We say that M„ yields a duality on A, or simply that M„ dualises M, if each eA is surjective, so that it is an isomorphism eA : A -- EDpAq. A dualising alter ego M„ plays a special role in the duality it sets up: it is the dual space of the free algebra on one generator in A. This fact is a consequence of compatibility. More generally, the free algebra generated by a non-empty set S has dual space M„ Assume that M„ yields a duality on A and in addition that each εX is surjective and so an isomorphism. Then we say M„ fully dualises M or that the duality yielded by M„ is full. In this case A and X are dually equivalent. Full dualities are particularly amenable if they are strong; this is the requirement that the alter ego be injective in the topological prevariety it generates. We do not need here to go deeply into the topic of strong dualities (see [10, Chapter 3] for a full discussion) but we do note in passing that each of the functors D and E in a strong duality interchanges embeddings and surjections -- a major virtue if a duality is to be used to transfer algebraic problems into a dual setting. S. We are ready to present our duality theorem for duplicated (pre)varieties. Our notation is chosen to match that in Theorem 2.1. Theorem 3.1 (Duality Transfer Theorem). Let N be an algebra and assume that If the topological structure N„ " pN ; G, R, Tq yields a duality Γ duplicates N. 2 yields a duality on on B " ISPpNq with dual category Y " IScP`pN„q, then N„ A " ISPpPΓpNqq, again with Y as the dual category. If the former duality is full, respectively strong, then the same is true of the latter. Proof. For the purposes of the proof we shall assume that N , and hence also M , is finite. It is routine to check that the topological conditions which come into play when N is infinite lift to the duplicated set-up. We claim that N„ 2 acts as a legitimate alter ego for M :" PΓpNq. Certainly these structures have the same universe, namely N N . It follows from the definition of the operations of PΓpNq that PΓprq, whose universe is r r, is a subalgebra of pPΓpNqqn whenever r P R is the universe of a subalgebra r of Nn. But RN„2 consists 6 L. M. CABRER AND H. A. PRIESTLEY of the relations r r, for r P R. Likewise, an n-ary operation g in G gives rise to 2. Hence g g is the same operation, viz. g g, of PΓpNq and in the structure N„ compatible with PΓpNq. We now set up the functors for the existing duality for ISPpNq and for the 2q " X too. duality sought for ISPpMq. Let X " IScP`pN„ Let DB : B Ñ Y and EB : Y Ñ B be the functors determined by N„ and DA : A Ñ X 2. Since Y " X, the functors DB and DA and EA : X Ñ A those determined by N„ have a common codomain. 2q. Then Y " IScP`pN„ Let A P A. By Corollary 2.2, we may assume that A " PΓpBq, for some B P B. By Theorem 2.1 and the definition of PΓ on morphisms, ApA, PΓpNqq " PΓpBpB, Nqq " t y y y P BpB, Nq u. i pαpy yqq for y P BpB, Nq. 2q. For i " 1, 2, define αi : DBpBq Ñ N„ by Let α P EADApAq " XpDApAq, N„ αipyq " πN It is straightforward to see that αi P EBDBpBq. Therefore, for i " 1, 2, there exists bi P B such that αipyq " ypbiq for y P DBpBq. We claim that αpxq " xpb1, b2q for all x P ApA, PΓpNqq. We can write x " y y where y P BpB, Nq. Then αpxq " αpy yq " pπN 1 pαpy yqq, πN 2 pαpy yqqq " pα1pyq, α2pyqq " pypb1q, ypb2qq " py yqpb1, b2q " xpb1, b2q. This proves that eA : A Ñ EADApAq is surjective for each A P A, so that we do indeed have a duality for A based on the alter ego M„ " N„ We now claim that if N„ fully dualises N then M„ fully dualises M. To do this we shall show that the bijection η : DBpBq Ñ DApAq, defined by ηpyq " y y for each y P DBpBq, is an isomorphism (of topological structures) from DBpBq onto DApAq, where, as before, A " PΓpBq, see [10, Lemma 3.1.1]. Let r be an n-ary relation in N„. For y1, . . . , yn P DBpBq, 2. py1, . . . ,ynq P rDBpBq ðñ @a P N ppy1paq, . . . , ynpaqq P rq ðñ @pa1, a2q P M pppy1pa1q, y1pa2qq, . . . , pynpa1q, ynpa2qqq P r rq ðñ py1 y1, . . . , yn ynq P pr rqDApAq. A similar argument applies to operations. The map η has compact codomain and Hausdorff domain and hence is a homeo- morphism provided η´1 is continuous. To prove this it will suffice to show that each map πb η´1 is continuous, where πb denotes the projection from DBpBq, regarded B, onto the b-coordinate, for b P B. The map πpb,bq is defined as a subspace of N„ likewise. Let U be open in N . For y y P DApAq, y y P pπb η´1q´1pU q ðñ πbpyq P U ðñ ypbq P U ðñ py yqpb, bq P U U ðñ πpb,bqpy yq P U U ðñ py yq P pπpb,bqq´1pU U q. This proves the continuity assertion. Finally, since N„ is injective in Y if and only if N„ 2 is, N„ yields a strong duality on B if and only if N„ 2 yields a strong duality on A, by [10, Theorem 3.2.4]. (cid:3) The proof of Theorem 3.1 is essentially routine, given the Product Representation Theorem. The theorem should not be disparaged because it is easy to derive. Rather the reverse: almost all the dualities given in Section 4 are new, and obtained at a stroke. NATURAL DUALITIES THROUGH PRODUCT REPRESENTATIONS 7 Of course, though, Theorem 3.1 is only useful when we have a (strong) duality to hand for the base class ISPpNq we wish to employ. Nothing we have said about natural dualities so far tells us how to find an alter ego N„ for N, or even whether a duality exists. Fortunately, simple and well-understood strong dualities exist for the base varieties ISPpNq which support the miscellany of logic-oriented examples presented in Section 4. In all cases considered there, N is a small finite algebra with a lattice reduct. Existence of such a reduct guarantees dualisability [10, Section 3.4]: a brute-force alter ego N„ " pN ; SpN2q, Tq is available. However this default choice is likely to yield a tractable duality only when N is very small. Otherwise the subalgebra lattice SpN2q is generally unwieldy. Methodology exists for slimming down a given dualising alter ego to yield a potentially more workable duality (see [10, Chapter 8]), but it is preferable to obtain an economical duality from the outset. This is often possible when N is a distributive lattice, not necessarily finite: in many such cases one can apply the piggyback method which originated with Davey and Werner (see [10, Chapter 7] and [12]). We shall demonstrate its use in Section 5, where we develop a duality for double Ockham algebras, our base variety for studying generalised conflation. Against this background we can appreciate the merits of Theorem 3.1. Suppose we have a class ISPpMq (with M finite) which is expressible as a duplicate of a dualisable base variety ISPpNq. Then M " N 2 and, on cardinality grounds alone, finding an amenable duality directly for ISPpMq could be challenging, whereas the chances are much higher that we have available, or are able to set up, a simple dualising alter ego N„ for N. And then, given N„ we can immediately obtain an alter ego M„ for M, with the same number of relations and operations in M„ as in N„. 4. Examples of natural dualities via duplication We now present a miscellany of examples. All involve bilattices but, as noted earlier, the scope of our methods is potentially wider. We derive (strong) dualities for certain (finitely generated) duplicated varieties given in [9] by calling on well- known (strong) dualities for their base varieties. A catalogue of base varieties and duplicates is assembled in [9, Appendix, Table 1], with references to where in the paper these examples are presented. Table 1 lists alter egos for dualities for base varieties. These dualities are discussed in [10], with their sources attributed. Natural dualities for the indicated duplicated varieties, also strong, can be read off from the table, using the Duality Transfer Theorem. When specifying a generator for each base variety, we adopt abbreviations for standard sets of operations: FL " t_, ^, 0, 1u, FB " FDM " FK " FL Y t„u; we have elected to denote negation in Boolean algebras, De Morgan algebras and Kleene algebras by „, to distinguish it from bilattice negation, . The top row of Table 1 should be treated as a prototype, both algebraically and dually. There the base variety is D, the variety of bounded distributive lattices. The duplicated variety in this case is the variety DB of distributive bilattices. It is generated (as a prevariety) by the four-element algebra in DB. Full details of the natural duality for DB and its relationship to Priestley duality for the base variety D appear in [8]. All the other examples in the table work in essentially the same way. The examples we list may be grouped into two types. In one type, the duplicator Γ includes the set of terms used to duplicate the variety of bounded lattices to create bounded bilattices, augmented with additional terms to capture other operations from terms in the base language; this applies to DB itself, to implicative bilattices, to distributive bilattices with conflation, to the varieties 8 L. M. CABRER AND H. A. PRIESTLEY base variety and its natural duality variety generator alter ego duplicate variety non-bilattice operation added bounded pt0, 1u; FLq DL's pt0, 1u; ď, Tq [10, §4.3.1] N/A Boolean algebras pt0, 1u; FBq pt0, 1u; Tq [10, §4.1.2] implication, Ą [1], [4, §2] Moore's epistemic operator, L [20] negation-by-failure, { [26, §3] De Morgan pt0, 1u2; FDMq pt0, 1u2; ď, g, Tq [10, §4.3.15] conflation, ´ (with bounds) [22] algebras Kleene algebras pt0, a, 1u; FKq see [10, §4.3.9] negation-by-failure, { [26, §4] Table 1. Examples of natural dualities (bounded case) carrying Moore's operator. In examples of the second type the base-level generator N is already equipped with a (distributive) bilattice structure and Γ includes all the terms used to create DB plus terms to create any extra operation present in N. This is the situation with negation-by-failure. For the natural dualities recorded in Table 1, we note that, apart from D, the base variety in each case is De Morgan algebras or a subvariety thereof. The alter ego includes a partial order ď known as the alternating order in [10, Theo- rem 4.3.16]; in the case of DM, the relation ď on universe t0, 1u2 of the four-element generator 4DM is the knowledge order. The map g is the involution swapping the coordinates. Only simple modifications are needed to handle the case when the language of a lattice-based variety does not include lattice bounds as nullary operations. It is an old result that Priestley duality for the variety Du can be set up in much the same way as that for D, with the dual category being pointed Priestley spaces, as described in [10, Section 1.2 and Subsection 4.3.1]. Natural dualities for duplicates of Du are derived from those for corresponding duplicates of D simply by adding to the alter ego nullary operations p0, 0q and p1, 1q. Compare with [8, Section 4], which provides a direct treatment of duality for DBu; here, even more than in the bounded case, we see the merit of the automatic process that Theorem 3.1 supplies. A duality for DMu (De Morgan lattices) is obtained by adding the top and bottom elements for the partial order ď to the alter ego for DM. Our transfer theorem then applies to unbounded distributive bilattices with conflation. 5. Bilattices with generalised conflation In this section we break new ground, both in relation to product representation and in relation to natural duality. The bilattice-based variety DB´ that we study -- (bounded) distributive bilat- tices with generalised conflation -- has not been considered before. Previous authors who have studied product representation when conflation is present have assumed NATURAL DUALITIES THROUGH PRODUCT REPRESENTATIONS 9 that this operation is an involution that commutes with negation (see [14, Theo- rem 8.3], [4] and our treatment in [9, Section 5]). We shall demonstrate that neither assumption is necessary for the existence of a product representation. Our focus in this paper is on developing theoretical tools. Nevertheless we should supply application-oriented reasons to justify investigating generalised conflation. We first note that it is often, but not always, natural to assume that conflation be an involution. On the other hand, the justification for the commutation condition is less clear cut. Indeed, both the original definition in [14] and that in [25] exclude commutation, and this is brought in only later. In [25, Section 3] the emphasis is on truth values. The authors' desired interpretation then leads them to consider a special algebra SIXTEEN3, in which the conflation operation does commute with negation. In [18, Section 2] conflation is used to study (knowledge) consistent and exact elements of a lattice. The investigations in both [25] and [18] are intrinsically connected to the product representation for bilattices with conflation. Our product representation would permit similar interpretations when commutation fails and/or conflation is not an involution. In a different setting, conflation has been used in [15] to present an algebraic model of the logic system of revisions in databases, knowledge bases, and belief sets introduced in [23]. In this model the coordinates of a pair in a product representation of a bilattice are interpreted as the degrees of confidence for including in a database an item of information and for excluding it. Conflation then models the transformation of information that reinterprets as evidence for inclusion whatever did not previously count as evidence against, and vice versa. That is, conflation comprises two processes: given the information against (for) a certain argument, these capture information for (against) the same argument. In [15] these two transformations coincide, and are mutually inverse. Our work on generalised conflation would allow these assumptions to be weakened so facilitating a wider range of models. The class DB´ consists of algebras of the form A " pA; _t, ^t, _k, ^k, , ´, 0, 1q, where the reduct of A obtained by suppressing ´ belongs to DB and ´ is an endomorphism of At and a dual endomorphism of Ak. Here we elect to include bounds. The variety DBC of (bounded) distributive bilattices with conflation (where by convention conflation and negation do commute) is a subvariety of DB´. However DB´ and DBC behave quite differently: even though is an involution, ´ is not. As a consequence the monoid these operations generate is not finite, as is the case in DBC. (We note that the unbounded case of generalised conflation could also be treated by making appropriate modifications to the above definition and throughout what follows.) Our product representation for DB´ uses as its base variety the class DO of double Ockham algebras. This is a new departure as regards representations of bilattice expansions. A double Ockham algebra is a D-based algebras equipped with two dual endomorphisms of the D-reducts. An Ockham algebra carries just one such operation. The variety O of Ockham algebras, which includes Boolean algebras, De Morgan algebras and Kleene algebras among its subvarieties, has been exhaustively studied, both algebraically and via duality methods, as indicated by the texts [3, 10] and many articles. The variety DO is much less well explored. The remainder of the section is accordingly organised as follows. Proposition 5.1 presents the product representation for DB´ over the base variety DO. We then set DB´ aside while we develop the theory of DO which we need if we are to apply our Duality Transfer Theorem to DB´. This requires us first to identify an algebra M such that DO " ISPpMq (Proposition 5.2). We then set up an alter ego M„ for M and call on [12, Theorem 4.4] to obtain a natural duality for DO 10 L. M. CABRER AND H. A. PRIESTLEY (Theorem 5.6). This is then combined with Theorem 3.1 to arrive at a natural duality for DB´ (Theorem 5.7). To motivate how we can realise DB´ as a duplicate of DO we briefly recall from [9, Section 5] how DBC arises as a duplicate of DM. We adopt the notation introduced in [9, Section 4]. Let Σ be a language and f be an n-ary function symbol in Σ. For m ě n and i1, . . . , in P t1, . . . , mu we denote by f m i1in the m- ary term f m i1...in px1, . . . , xmq " f pxi1 , . . . , xin q. We can capture the extra operation ´ on the generator 16DBC of DBC using the De Morgan negation „, combined with coordinate-flipping: the family of terms ΓDBC " ΓDB Y tp„2 1qu acts as a duplicator for DM with DBC as the duplicated variety; here ΓDB duplicates bounded lattices. (See [9, Section 5] for an explanation as to why the form of the operations in DBC dictates that DM should be used as the base variety.) 2, „2 We now present our duplication result linking DO and DB´. Proposition 5.1. The set ΓDB ´ " ΓDB Y tpf 2 2 , g2 1qu duplicates DO. Moreover, is identified with the language of DB´. DB´ " V`PΓDB ´ pDOq, where ΣΓDB ´ Proof. Certainly ΓDB´ duplicates DO because pf 2 duplicate for ΣD on D. 2 , g2 1q P ΓDB´ and ΓDB is a Now let A P DB´. By the product representation of DB over D, the bilattice reduct ADB -- PΓDBpLq, for some L P D. We identify A and L L and define f, g : L Ñ L by f paq " π1p´p0, aqq and gpaq " π2p´pa, 0qq, for a P L. For a, b P L, gpa _ bq " π2p´pa _ b, 0qq " π2p´ppa, 0q _k pb, 0qqq " π2p´pa, 0q ^k ´pb, 0qq " π2p´pa, 0q _k ´pb, 0qq " π2p´pa, 0qq ^ π2p´pb, 0qq " gpaq ^ gpbq, gpa ^ bq " π2p´pa ^ b, 0qq " π2p´ppa, 0q ^k pb, 0qqq " π2p´pa, 0qq _ π2p´pb, 0qq " gpaq _ gpbq, gp1q " π2p´p1, 0qq " π2p1, 0q " 0, gp0q " π2p´p0, 0qq " π2p1, 1q " 1, and similarly for f . Hence B " pL; _, ^, f, g, 0, 1q P DO. Observe that π1p´pa, 0qq " π1p´ppa, 0q _t p0, 0qqq " π1p´pa, 0q _t p1, 1qq " 1; π2p´p0, bqq " π2p´pp0, bq ^t p1, 1qqq " π1p´p0, bq ^t p0, 0qq " 0. Hence ´pa, bq " ´ppa, 0q _k p0, bqq " ´pa, 0q ^k ´p0, bq " pπ1p´pa, 0qq, π2p´pa, 0qqq ^k pπ1p´p0, bqq, π2p´p0, bqqq " p1, π2p´pa, 0qqq ^k pπ1p´p0, bqq, 0q " p1, gpaqq ^k pf pbq, 0q " pf pbq, gpaqq " rf 2 2 , g2 1spa, bq. Therefore A -- PΓDB ´ pBq. (cid:3) This theorem gives insight into the effect of reinstating the assumptions cus- tomarily imposed on conflation and which we removed in passing from DBC to DB´. From the product representation for DB´, it follows that ´ is involutive if and only if f and g are. The resulting subvariety of DB´ is a duplicate of double De Morgan algebras (that is, algebras in DO such that both unary operations are involutions). Similarly, ´ commutes with if and only if f " g. This time we obtain a subvariety of DB´ which duplicates O. We now want to identify an (infinite) algebra which generates our base variety DO as a prevariety. We take our cue from the variety O of Ockham algebras: O is generated as a prevariety by an algebra M whose universe is t0, 1uN0, where N0 " t0, 1, 2. . . .u; lattice operations and constants are obtained pointwise from the two-element bounded lattice and, identifying the elements as infinite binary strings, NATURAL DUALITIES THROUGH PRODUCT REPRESENTATIONS 11 negation is given by a left shift followed by pointwise Boolean complementation on t0, 1u. See for example [12, Section 4] for details. We may view the exponent N0 as the free monoid on one generator e, with 0 as identity and n acting as the n-fold composite of e. For DO, analogously, we first consider the free monoid E " te1, e2u on two generators e1 and e2 and identify it with the set of all finite words in the language with e1 and e2 as function symbols, with the empty word corresponding to the identity element 1; the monoid operation is given by concatenation. For s P E, we denote the length of s by s. For us, DO will serve as a base variety. Accordingly we align our notation with that in Theorem 3.1. We now consider the algebra N with universe t0, 1uE with lattice operations and constants given pointwise. The lattice t0, 1uE is in fact a Boolean lattice, whose complementation operation we denote by c. The dual endomorphisms f and g are given as follows. For a P t0, 1uE we have f paqpsq " cpaps e1qq and gpaq " cpaps e2qq for every s P E. This gives us an algebra N :" pt0, 1uE; _, ^, f, g, 0, 1q P DO. For future use we show how to assign to each word s P E a unary term ts in the language of DO, as follows. If s " 1 (the empty word) then ts is the identity map; if s " e1 s1 then ts " f ts1; and if s " e2 s1 then ts " g ts1. Structural induction shows that the term function tN s is given by ptN s paqqpeq "#aps eq 1 ´ aps eq if s is even, if s is odd, for every a P N and s P E. Proposition 5.2. Let N be defined as above. Then DO " ISPpNq. Proof. It will suffice to show that given any A P DO and any a ‰ b in A, there exists a DO-morphism h from A into N such that hpaq ‰ hpbq; see [10, Theorem 1.3.1]. By the Prime Ideal Theorem there exists a D-morphism x from (the D- reduct of) A into 2 with xpaq ‰ xpbq. Define ϕ : A Ñ N by ϕpcqpsq "#xptspcqq 1 ´ xptspcqq if s is even, if s is odd, for c P A and s P E. It is routine to check that ϕ is a D-morphism which preserves f and g. Finally, ϕpcqp1q " xpcq, whence ϕpaq ‰ ϕpbq. (cid:3) We now seek a natural duality for DO which parallels that which is already known for the category O of Ockham algebras. Our treatment follows the same lines as that given for O in [12, Section 4], whereby a powerful version of the piggyback method is deployed. (The duality for O was originally developed by Goldberg [21] and re-derived as an early example of a piggyback duality by Davey and Werner [13].) A general description of the piggybacking method and the ideas underlying it can be found in [12, Section 3]. We wish to apply to DO a special case of [12, Theorem 4.4]. We first make some comments and establish notation. We piggyback over Priestley duality between D " ISPp2q and P " IScP`p 2„q (where 2 and 2„ are the two-element objects in D and P with universe t0, 1u, defined in the usual way). We denote the hom-functors setting up the dual equivalence between D and P by H and K. The aim is to find an element ω P DpN5, 2q which, together with endomorphisms of N, captures enough information to build an alter ego N„ of N which yields a full duality, in fact, a strong duality. We now work towards showing that we can apply [12, Theorem 4.4] to DO " ISPpNq, where N is as defined above. We shall take ω : N Ñ 2 to be the projection map given by ωpaq " ap1q. We want to set up an alter ego N„ " pt0, 1uE; G, R, Tq 12 L. M. CABRER AND H. A. PRIESTLEY 5 such that ω P PpN„ 5, 2„q. so that in particular N„ has a Priestley space reduct N„ Moreover we need the structure N„ to be chosen in such a way that the conditions (1) -- (3) in [12, Theorem 4.4] are satisfied. We define T to be the product topology on N " t0, 1uE derived from the discrete topology on t0, 1u; this is compact and Hausdorff and makes N into a topological algebra. We now need to specify G and R. We would expect R to contain an order relation ď such that pt0, 1uE; ď, Tq P P. For Ockham algebras -- where one uses the free monoid on one generator as the exponent rather than E -- the corresponding order relation is the alternating order in which alternate coordinates are order-flipped; see [10, Section 7.5] (and recall the comment about De Morgan algebras, a subvariety of O, in Section 4). The key point is that a composition of an even (respectively odd) number of order-preserving self- maps on an ordered set is order-preserving (respectively order-reversing). Hence the definition of ď in Lemma 5.3 is entirely natural. Lemma 5.3. Let N be as above. Then ď, given by a ď b ðñ @z P E #apsq ď bpsq apsq ě bpsq if s is even, if s is odd, is an order relation making pt0, 1uE; ď, Tq a Priestley space. Moreover ď is the universe of a subalgebra of N2 and this subalgebra is the unique maximal subalgebra of pω, ωq´1pďq " t pa, bq P N 2 ωpaq ď ωpbq u. B (that is, 2„ with the order reversed) is a Proof. Each of 2„ and the structure 2„ Priestley space. It follows that the topological structure pt0, 1uE; ď, Tq is a product of Priestley spaces and so itself a Priestley space. Take a, b, c, d in N such that a ď b and c ď d and let s P E. Then pa ^ cqpsq " apsq ^ cpsq ď bpsq ^ dpsq " pb ^ dqpsq if s is even, pa ^ cqpsq " apsq ^ cpsq ě bpsq ^ dpsq " pb ^ dqpsq if s is odd. Hence a ^ c ď b ^ d. Similarly a _ c ď b _ d. Also 0 ď 0 and 1 ď 1. If s is even, f paqpsq " pc a e1qpsq " 1 ´ pape1 sqq ď 1 ´ pbpe1 sqq " f pbqpsq, since a ď b and e1 s is odd. Similarly, if s is odd then f paqpsq ě f pbqpsq. Therefore f paq ď f pbq. Likewise gpaq ď gpbq. Thus ď is indeed the universe of a subalgebra of N2. Now let r be the universe of a subalgebra of N2 maximal with respect to inclusion in pω, ωq´1pďq. Then, with ts as defined earlier for s P E, we have pa, bq P r ùñ p@s P Eq`ptspaq, tspbqq P r ùñ p@s P Eq`tspaq ď tspbq ùñ p@s P Eqp@e P Eq`tspaqpeq " 1 ùñ tspbqpeq " 1. But ptspaqqpeq "#aps eq 1 ´ aps eq if s is even, if s is odd. We deduce that r is a subset of ď. In addition a ď b implies ωpaq ď ωpbq: consider s " 1. Maximality of r implies that r equals ď. Consequently ď is the unique maximal subalgebra contained in pω, ωq´1pďq. (cid:3) We now introduce the operations we shall include in our alter ego N„. Let the map γi : E Ñ E be given by γipsq " s ei. Then we can define an endomorphism ui of N by uipaq " a γi, for i " 1, 2. These maps are continuous with respect to the topology T we have put on N . We define N„ :" pt0, 1uE; u1, u2, ď, Tq. Then N„ is compatible with N. We let Y :" IScP`pN„q be the topological prevariety generated by N„ and by 5 the forgetful functor from Y into P which suppresses the NATURAL DUALITIES THROUGH PRODUCT REPRESENTATIONS 13 operations u1 and u2. We note that now ω, as defined earlier, may be seen to 5, 2„q. The following two lemmas concern the interaction belong to DpN5, 2q X PpN„ of N, N„ and ω as regards separation properties. Lemma 5.4. Assume that N, N„ and ω are defined as above. Then, given a ‰ b in N , there exists a unary term u in the language of pN ; u1, u2q such that ωpupaqq ‰ ωpupbqq. Proof. Let a ‰ b P N. There exists s P E with s ‰ 1 such that apsq ‰ bpsq. Write s as a concatenation ei1 ein , where i1, . . . , in P t1, 2u. For each j " 1, . . . , n, there is an associated unary term uj such that, for all w P E, puij paqqpwq " pa γij qpwq " apw eij q. Write uin . . . ui1 as us. Then uspcqp1q " cpsq for all c P N and hence pω usqpaq " uspaqp1q " apsq ‰ bpsq " uspbqp1q " pω usqpbq. (cid:3) Lemma 5.5. If a ę b in N„ that ωptpaqq " 1 and ωptpbqq " 0. 5, then there exists a unary term function t of N such Proof. We have a ę b ðñ Ds P E#apsq " 1 & bpsq " 0 if s is even, apsq " 0 & bpsq " 1 if s is odd. When s is even, ωptspaqq " tspaqp1q " apsq " 1 and ωptspbqq " tspbqp1q " bpsq " 0. Similarly, if s is odd, ωptspaqq " tspaqp1q " c apsq " 1 ´ apsq " 1 and ωptspbqq " tspbqp1q " c bpsq " 1 ´ bpsq " 0. (cid:3) Theorem 5.6 (Strong Duality Theorem for Double Ockham Algebras). Let N " pt0, 1uE; _, ^, f, g, 0, 1q and N„ " pt0, 1uE; u1, u2, ď, Tq be as defined above. Let 5, 2„q be given by evaluation at 1, the identity of the monoid E. ω P DpN5, 2q X PpN„ Let D : DO Ñ Y and E : Y Ñ DO be the hom-functors: D :" DOp´, Nq and E :" Yp´, N„q. Then N„ strongly dualises N, that is, D and E establish a strong duality between DO and Y. Moreover DpAq5 -- HpA5q in P and EpYq5 -- KpY5q in D, for A P DO and Y P Y, where the isomorphisms are set up by ΦA x P DpAq, and ΨY ω : α ÞÑ ω α, for α P EpYq. ω : x ÞÑ ω x, for Proof. We simply need to confirm that the conditions of [12, Theorem 4.4] are satisfied. We have everything set up to ensure that all the functors work as the theorem requires. In addition Lemmas 5.3 -- 5.5 tell us that Conditions (1) -- (3) in the theorem are satisfied. (cid:3) Some remarks are in order here. We stress that it is critical that we could find a map ω which acts as a morphism both on the algebra side and on the dual side, and has the separation properties set out in Lemmas 5.4 and 5.5. We also observe that for our application of [12, Theorem 4.4], its Condition (3) is met in a simpler way than the theorem allows for: the special form of the f, g (viz. dual endomorphisms with respect to the bounded lattice operations) that forces pω, ωq´1pďq to contain just one maximal subalgebra. We should comment too on how our natural duality for DO relates to a Priestley- style duality for DO. The latter can be set up in just the same way as that for O originating in [28]. This duality is an enrichment of that between D and P, whereby f and g are captured on the dual side via a pair of order-reversing continu- ous maps p and q, and morphisms are required to preserve these maps. Theorem 5.6 tells us that, for any A P DO, there is an isomorphism between the Priestley space 14 L. M. CABRER AND H. A. PRIESTLEY reduct DpAq5 of the natural dual of A P DO and the Priestley dual HpA5q of the D-reduct of A. Both these Priestley spaces carry additional structure: u1 and u2 in the former case and p and q in the latter. When the reducts of the natural and Priestley-style dual spaces of the algebras are identified these pairs of maps coincide. Thus the two dualities for DO are essentially the same and one may toggle between them at will. We have a new example here of a 'best of both worlds' scenario, in which we have both the advantages of a natural duality and the benefits, pictorially, of a duality based on Priestley spaces. See [7, Section 3], [8, Section 6] and [12, Section 4] for earlier recognition of occurrences of this phenomenon: other varieties for which it arises are De Morgan algebras and Ockham algebras. In general it is not hereditary: it fails to occur for Kleene algebras, for example. Combining our results we arrive at our duality for the variety DB´. Theorem 5.7 (Strong Duality Theorem for Bounded Distributive Bilattices with Generalised Conflation). Let N„ " pt0, 1uE; u1, u2, ď, Tq be as in Theorem 5.6. Then N„ N„ yields a strong duality on DB´. Moreover the dual category for this duality is Y :" IScP`pN„q which may, in turn, be identified with the category PDO of double Ockham spaces. To illustrate the rewards derived from a natural duality for FDB´, we highlight for a non- the simple description of free objects that follows from Theorem 5.7: 2qS as its natural dual space. empty set S, the free algebra FDB´pSq on S has pN„ Hence FDB´pSq can be identified with the family of continuous structure-preserving maps from pN„ (Recall the remark on free algebras in Section 3.) 2, with the operations defined pointwise. 2qS into N„ 6. Dualities for pre-bilattice-based varieties In this final section we consider dualities for pre-bilattice-based varieties. Here we call on the adaptation of the product representation theorem given in [9, The- orem 9.1]. Hitherto in this paper we have worked with dualities for prevarieties of the form ISPpMq, thereby encompassing dualities for many classes of interest in the context of bilattices. However when we drop negation and so move from bilattices to pre-bilattices the situation changes and we encounter classes of the form ISPpMq, where M is a finite set of algebras over a common language. For example, for distributive pre-bilattices M consists of a pair of two-element alge- bras, one with truth and knowledge orders equal, the other with these as order duals. Fortunately a form of natural duality theory exists which is applicable to classes of the form ISPpMq; this makes use of multisorted structures on the dual side. So in this section we shall consider dualities for pre-bilattice-based varieties. As a starting point we have the treatment of distributive pre-bilattices given in [8, Sections 9 and 10]; a self-contained summary of the rudiments of multisorted duality theory can also be found there or see [10, Chapter 7]. We first recall how [9, Theorem 9.1] differs from Theorem 2.1. We start from a base class VpN q, where N is a class of algebras over a common language Σ. Let Γ and PΓpN q be as in Section 2. Negation in a product bilattice links the two factors, and condition (P) from the definition of duplication by Γ reflects this. In the absence of negation, (P) is dropped and the following condition is substituted: (D) for pt1, t2q P Γ with npt1,t2q " n, there exist n-ary Σ-terms r1 and r2 such that t1px1, . . . , x2nq " r1px1, x3, . . . , x2n´1q and t2px1, . . . , x2nq " r2px2, x4, . . . , x2nq. A product algebra associated with Γ now takes the form P dΓ Q " pP Q; trt1, t2sPdΓQ pt1, t2q P Γuq, NATURAL DUALITIES THROUGH PRODUCT REPRESENTATIONS 15 where P, Q belong to the base variety B " VpN q. This construction is used to define a functor dΓ : B B Ñ A as follows: on objects: on morphisms: pP, Qq ÞÑ P dΓ Q, dΓ ph1, h2qpa, bq " ph1paq, h2pbqq. Theorem 6.1. [9, Theorem 9.3] Let N be a class of Σ-algebras and let Γ a set of pairs of Σ-terms satisfying (L), (M) and (D). Let B " VpN q. Then the functor dΓ : BB Ñ A, sets up a categorical equivalence between BB and A " VptP dΓ Q P, Q P VpN quq. We move on to consider dualities for duplicated varieties. For simplicity we shall first assume that the base variety B " ISPpNq has a single-sorted duality with alter ego N„ " pN ; G, R, Tq. Our next task is to determine a set of generators for A as a prevariety. We denote the trivial algebra by T. For C P B let f C : C Ñ T be the unique homomorphism from C into T. Lemma 6.2. If B " ISPpNq " VpNq for some algebra N, then A " VptP dΓ Q P, Q P ISPpNquq " ISPpN dΓ T, T dΓ Nq. Proof. Let A P A and a ‰ b P A. By Theorem 6.1, we may assume that there exist B, C P B such that A " B dΓ C. Let a1, b1 P B and a2, b2 P C such that a " pa1, a2q and b " pb1, b2q. By simmetry we may assume that a1 ‰ b1. Then there exists a homomorphism h : B Ñ N such that hpa1q ‰ hpb1q. Now h dΓ f C : B dΓ C Ñ N dΓ T is such that ph dΓ f Cqpaq " phpa1q, f C pa2qq ‰ phpb1q, f Cpb2qq " ph dΓ f Cqpbq. (cid:3) Let M " tN dΓ T, T dΓ Nu. We now 'double up' N „ in the obvious way. Let „ Z N N „ " pN1 9YN2; G1, G2, R1, R2, Tq, based on disjointified universes N1 and N2, such that pNi; Gi, Ri, TaeNiq is isomorphic to N „ for i " 1, 2. Identify N1 with N T and N2 with T N and define M „ Z N „ . We now present our transfer theorem for natural dualities associated with Theo- rem 6.1 (the single-sorted case). Its proof is largely a diagram-chase with functors. Below, IdC denotes the identity functor on a category C and -- is used to denote natural isomorphism. „ " N Theorem 6.3. Let N be a Σ-algebra and assume that Γ satisfies (L), (M) and (D) relative to N. Assume that N„ " pN ; G, R, Tq yields a duality on B " ISPpNq " VpNq with dual category Y " IScP`pN„q. Let M and M „ be defined as above. Then M „ yields a multisorted duality for A " ISPpMq " VpP dΓ Q P, Q P VpN qq for which the dual category is X -- Y Y. If the duality for B is full, respectively strong, then the same is true of that for A. Proof. Let pX1, X2q P Y Y " IScP`pN„q IScP`pN„q. We identify this structure with X1 Z X2 " pX1 9YX2; G1, G2, R1, R2, Tq, where as before 9Y denotes disjoint union and the topology T is the union of T1 and T2. Morphisms in X are maps f : X1 9YX2 Ñ Y1 9YY2 that respect the structure and are such that f pxq P Yi when x P Xi and i P t1, 2u. Hence the assignment: on objects: on morphisms: pX1, X2q ÞÑ X1 Z X2, pf, gq ÞÑ f 9Yg sets up a categorical equivalence, Z. Let F : X Ñ Y Y denote its inverse. „ Z N Identify N dΓ T and T dΓ N with N1 and N2 respectively. One sees that „ " pN1 9YN2; G1, G2, R1, R2, Tq is a legitimate alter ego for M. Let „ :" N M DB : B Ñ Y and EB : Y Ñ B, and DA : A Ñ X and EA : X Ñ A be the hom- functors determined by N „ respectively. By Theorem 6.1, there exists a „ and M 16 L. M. CABRER AND H. A. PRIESTLEY DB DB B B Y Y B B EB EB C A DA Z X dΓ A EA Figure 1. Natural duality by duplication Y Y F X functor C : A Ñ BB that together with dΓ : BB Ñ A determines a categorical equivalence. Take A, B P B and let DApA dΓ Bq " pX1 9YX2; G1, G2, R1, R2, Tq. Again by Theorem 6.1, X1 " ApA dΓ B, N dΓ Tq " tphA, f Bq hA P BpA, Nqu " BpA, Nq tf Bu, and likewise X2 " tf Au BpB, Nq. i 2 i 1 if and only if hi " pf be the corresponding relation in RAdΓB if and only if hi " pgi, f For an n-ary relation r P R, let rAdΓB Ď i (i " t1, 2u). So ph1, . . . , hnq P rAdΓB X n Bq P BpA, Nq tf Bu for i P t1, . . . , nu and pg1, . . . , gnq P rA. Similarly, a tuple ph1, . . . , hnq belongs to rAdΓB Au BpB, Nq for i P t1, . . . , nu and pg1, . . . , gnq P rB. The same argument applied to G proves that pX1; G1, R1, TaeX1 q and pX2; G2, R2, TaeX2 q are isomorphic to DBpAq and DBpBq, respectively. Thus FpDApA dΓ Bqq is isomorphic to pDBpAq, DBpBqq in Y Y. Moreover, it is easy to see that the assignment FpDApA dΓ Bqq ÞÑ pDBpAq, DBpBqq determines a natural isomorphism between F DA dΓ and DB DB : B B Ñ X X. A, giq P tf Similarly, for each pX, Yq P X X, EApX Z Yq " pEBpXq dΓ Tq pT dΓ EBpYqq -- pEBpXq Tq dΓ pT EBpYqq -- EBpXq dΓ EBpYq. Moreover, the assignment EApX Z Yq ÞÑ EBpXq dΓ EBpYq is natural in X and Y, that is, EA Z -- pEB EBq dΓ. So (up to natural isomorphism) the diagrams in Figure 1 commute. A symbol- chase now confirms that M„ dualises M because N„ dualises N: EA DA -- dΓ pEB EBq F Z pDB DBq C " dΓ pEB EBq pDB DBq C -- dΓ pIdB IdBq C " dΓ C -- IdA. B B dΓ C A DB DB EB EB DA EA Y Y Z F X Figure 2. Full duality by duplication Assume that N„ yields a full duality. Then the diagram in Figure 2 commutes. „ yields a full duality. Moreover, We can easily prove that DA EA -- IdX, that is, M NATURAL DUALITIES THROUGH PRODUCT REPRESENTATIONS 17 if N„ is injective in Y then pN„, N„q is injective in Y Y, or equivalently M is injective in X. Hence M „ yields a strong duality if N does. „ " N„ Z N„ (cid:3) Theorem 6.3 applies to the variety pDBu of (unbounded) distributive pre-bilattices. Its members are algebras A " pA; _t, ^t, _k, ^kq for which pA; _t, ^tq P Du and pA; _k, ^kq P Du. The well-known product representation for pDBu comes from the observation that the set ΓpDBu " tp_4 24q, p^4 24q, p_4 24q, p^4 13, _4 13, ^4 13, _4 13, ^4 24qu satisfies (L), (M) and (D) [9, Section 9]. Since 2„u strongly dualises Du, the struc- ture 2„u Z 2„u determines a multisorted strong duality for pDBu. This was estab- lished by different techniques in [8, Theorem 10.2]. Theorem 6.3 also yields dualities for distributive trilattices. These are (to the best of our knowledge) new. As with pre-bilattices, we opt for the unbounded case. An unbounded distributive trilattice is an algebra pA; _t, ^t, _f , ^f , _i, ^iq such that pA; _t, ^tq, pA; _f , ^f q and pA; _i, ^iq are distributive lattices. Let DT u de- note the variety of (unbounded) distributive trilattices. An algebra pA; _t, ^t, _f , ^f , _i, ^i, ´tq is a distributive trilattice with t-involution if pA; _t, ^t, _f , ^f , _i, ^iq P DTu and ´t is an involution that preserves the f - and i-lattice operations and reverses _t and ^t. Let DT´t denote the variety of unbounded distributive trilattices with t-involution. Take DBu as the base variety and let ΓDT ´t " tpp^tq4 13, p^tq4 pp^kq4 24q, pp_tq4 13, p_kq4 13, p_tq4 24q, pp^kq4 24q, pp_kq4 13, p^kq4 13, p^kq4 24q, pp_kq4 24q, 13, p_kq4 24q, p 2 1, 2 2qu. ´t Then ΓDT satisfies (L), (M) and (D) over DBu (see [8, Example 9.4]). In Sec- tion 4, we used Theorem 3.1 to prove that p 2„uq2 yields a strong duality on DBu. Now Theorem 6.3 implies that p 2„uq2 Z p 2„uq2 determines a multisorted strong du- ality for unbounded distributive trilattices with t-involution. We can easily adapt our results to cater for a base variety which admits a multi- sorted duality rather than a single-sorted one. Predictably this leads to multisort- edness at the duplicate level. In the case of Theorem 3.1, one obtains the required alter ego by squaring the base level alter ego, sort by sort; as before, the base va- riety and its duplicate have the same dual category. The extension of Theorem 6.3 employs two disjoint copies of each sort of the base-level alter ego. The proofs of these results involve only minor modifications of those for the single-sorted case. As an example, the multisorted version of Theorem 6.3 combined with the results in [9, Example 9.4] leads to a strong duality for unbounded distributive trilattices which has four sorts, obtained from the two-sorted duality for pDBu. References [1] Arieli, O., Avron, A.: Reasoning with logical bilattices. Logic, Lang. Inform. 5 (1996), 25 -- 63 [2] Belnap, N.D.: A useful four-valued logic: How a computer should think. In: A.R. Anderson and N.D. Belnap, Entailment. The Logic of Relevance and Necessity, vol. II , pp. 506 -- 541, Princeton University Press (1992) [3] Blyth, T., Varlet, J.: Ockham Algebras. Oxford University Press (1994) [4] Bou, F., Jansana, R., Rivieccio, U.: Varieties of interlaced bilattices. Algebra Universalis 66 (2011), 115 -- 141 [5] Bou, F., Rivieccio, U.: The logic of distributive bilattices. Logic J. IGPL 19 (2011), 183 -- 216 [6] Burris, S.N., Sankappanavar, H.P.: A Course in Universal Algebra. Graduate Texts in Mathematics 78. Springer-Verlag (1981); Free download at http://www.math.waterloo.ca/ ~snburris [7] Cabrer, L.M., Priestley, H.A.: Coproducts of distributive lattice-based algebras. Algebra Universalis 72 (2014), 251 -- 286 [8] Cabrer, L.M., Priestley, H.A.: Distributive bilattices from the perspective of natural duality theory. Algebra Universalis 73 (2015), 103 -- 141 18 L. M. CABRER AND H. A. PRIESTLEY [9] Cabrer, L.M., Priestley, H.A.: A general framework for product representations: bilattices and beyond. Submitted (available at arXiv:1503.06921) [10] Clark, D.M., Davey, B.A.: Natural Dualities for the Working Algebraist. Cambridge Univer- sity Press (1998) [11] Davey, B.A.: The product representation theorem for interlaced pre-bilattices: some historical remarks. Algebra Universalis 70 (2013), 403 -- 409 [12] Davey, B.A. Haviar, M., Priestley, H.A.: Piggyback dualities revisited. Algebra Universalis (to appear), (available at arXiv:1501.02512v1) [13] Davey, B.A., Werner, H.: Piggyback-Dualitaten. Bull. Austral. Math. Soc. 32 (1985), 1 -- 32 [14] Fitting, M.: Kleene's three-valued logics and their children. Fund. Inform. 20 (1994), 113 -- 131 [15] Fitting, M.: Annotated revision specification programs. In proceedings LPNR'95 Lecture Notes in Comp. Sci. 928 (1995), 143 -- 155 [16] Fitting, M.: Bilattices are nice things. Self-reference, CSLI Lecture Notes 178, pp. 53 -- 77, CSLI Publ., Stanford, CA, (2006) [17] Font, J.M.: Belnap's four-valued logic and De Morgan lattices. Log. J. IGPL 5 (1997), 413 -- 440 [18] Gargov, G.: Knowledge, uncertainty and ignorance: bilattices and beyond. J. Appl. Non- classical Logics 9 (1999), 195 -- 283 [19] Ginsberg, M. L.: Multivalued logics: A uniform approach to inference in artificial intelligence. Comput. Intelligence 4 (1988), 265 -- 316 [20] Ginsberg, M. L.: Bilattices and modal operators. J. Logic Comput. 1 (1990), 41 -- 69 [21] Goldberg, M.S.: Topological duality for distributive Ockham algebras. Studia Logica 42 (1983), 23 -- 31 [22] Jung, A, Rivieccio, U.: Priestley duality for bilattices. Studia Logica 100 (2012), 223 -- 252 [23] Marek, V. W., Truszczy´nski, M.: Revision specifications by means of programs. In proceed- ings of JELIA'95, Lecture Notes in Comp. Sci. 838 (1994). 122 -- 136 [24] Mobasher, B., Pigozzi, D., Slutski, V., Voutsadakis, D.: A duality theory for bilattices. Algebra Universalis 43 (2000), 109 -- 125 [25] Odintsov, S.P., Wansing, H.: The logic of generalized truth values and the logic of bilattices. Studia Logia 103 (2015), 91 -- 112 [26] Ruet, P., Fages, F.: Combining explicit negation and negation by failure via Belnap's logic. Theoret, Comp. Sci. 171 (1997), 61 -- 75 [27] Shramko, Y., Wansing H.: Truth and Falsehood. An Inquiry into Generalized Logical Values. Springer (2011) [28] Urquhart, A.: Distributive lattices with a dual homomorphic operation. Studia Logica 38 (1979), 201 -- 209 (L. M. Cabrer) Institute of Computer Languages, Technische Universitat Wien, Fa- voritenstrasse 9-11, A-1040 Wien, Austria E-mail address: [email protected] (H. A. Priestley) Mathematical Institute, University of Oxford, Radcliffe Observa- tory Quarter, Oxford OX2 6GG, United Kingdom E-mail address: [email protected]
1607.05499	3	1607	2017-03-01T20:50:16	Applications of normal forms for weighted Leavitt path algebras: simple rings and domains	[ "math.RA", "math.KT" ]	Weighted Leavitt path algebras (wLpas) are a generalisation of Leavitt path algebras (with graphs of weight 1) and cover the algebras $L_K(n, n + k)$ constructed by Leavitt. Using Bergman's Diamond lemma, we give normal forms for elements of a weighted Leavitt path algebra. This allows us to produce a basis for a wLpa. Using the normal form we classify the wLpas which are domains, simple and graded simple rings. For a large class of weighted Leavitt path algebras we establish a local valuation and as a consequence we prove that these algebras are prime, semiprimitive and nonsingular but contrary to Leavitt path algebras, they are not graded von Neumann regular.	math.RA	math	APPLICATIONS OF NORMAL FORMS FOR WEIGHTED LEAVITT PATH ALGEBRAS: SIMPLE RINGS AND DOMAINS ROOZBEH HAZRAT AND RAIMUND PREUSSER Abstract. Weighted Leavitt path algebras (wLpas) are a generalisation of Leavitt path algebras (with graphs of weight 1) and cover the algebras LK (n, n + k) constructed by Leavitt. Using Bergman's diamond lemma, we give normal forms for elements of a weighted Leavitt path algebra. This allows us to produce a basis for a wLpa. Using the normal form we classify the wLpas which are domains, simple and graded simple rings. For a large class of weighted Leavitt path algebras we establish a local valuation and as a consequence we prove that these algebras are prime, semiprimitive and nonsingular but contrary to Leavitt path algebras, they are not graded von Neumann regular. 1. Introduction In a series of papers William Leavitt studied algebras that are now denoted by LK(n, n + k) and have been coined Leavitt algebras. Let X = (xij) and Y = (yji) be n × (n + k) and (n + k) × n matrices consisting of symbols xij and yji, respectively. Then for a field K, LK(n, n + k) is a K-algebra generated by all xij and yji subject to the relations XY = In+k and Y X = In. In [12, p.190] Leavitt studied these algebras for n = 2 and k = 1, in [13, p.322] for any n ≥ 2 and k = 1 and finally in [14, p.130] for arbitrary n and k. He established that these algebras are of type (n, k). He further showed that LK(1, k + 1) are simple rings and LK(n, n + k), n ≥ 2 are domains. Recall that a ring A is of type (n, k) if n and k are the least positive integers such that An ∼= An+k as left A-modules. He proved these statements by formulating a normal form for the elements of his algebras. This normal form was worked out more systematically by P.M. Cohn in [8] who showed that LK(n, n + k) is a domain using a trace method. The normal forms for algebras defined by generators and relations were streamlined by G. Bergman in his influential paper [6], called the diamond lemma, following the paper [17]. Leavitt path algebras were introduced a decade ago [1, 5], associating a K-algebra to a directed graph. For a graph with one vertex and k + 1 loops, it recovers the Leavitt algebra LK(1, k + 1). The definition and the development of the theory were inspired on the one hand by Leavitt's construction of LK(1, k + 1) and on the other hand by Cuntz algebras On [9] and Cuntz-Krieger algebras in C ∗-algebra theory [19]. The Cuntz algebras and later Cuntz-Krieger type C ∗-algebras revolutionised C ∗-theory, leading ultimately to the astounding Kirchberg-Phillips classification theorem [18]. In the last decade the Leavitt path algebras have created the same type of stir in the algebraic community. The development of Leavitt path algebras and its interaction with graph C ∗-algebras have been well-documented in several publications and we refer the reader to [2] and the references therein. Since their introductions, there have been several attempts to introduce a generalisation of Leavitt path algebras which would cover the algebras LK(n, n + k) for any n ≥ 1, as well. Ara and Goodearl's Leavitt path algebras of separated graphs were introduced in [4] which gives LK(n, n + k) as a corner ring of some separated graphs. The weighted Leavitt path algebras were introduced in [10] which gives LK(n, n + k) for a weighted graph with one vertex and n + k loops of weight n. If the weights of all the edges are 1 (i.e., the graph is unweighted), then the weighted Leavitt path algebras reduce to the usual 2000 Mathematics Subject Classification. 16S10, 16W10, 16W50, 16D70. Key words and phrases. Weighted Leavitt path algebra, diamond lemma, simple ring, prime ring, nonsingular ring. The first author would like to acknowledge Australian Research Council grants DP150101598 and DP160101481. A part of this work was done at the University of Munster, where the first author was a Humboldt Fellow. 1 2 ROOZBEH HAZRAT AND RAIMUND PREUSSER Leavitt path algebras. The structure of weighted Leavitt path algebras remained to be explored. In this paper we take a step in this direction (no one had looked at the topic systematically so far). In Section 2 we develop systematically a normal form for elements of weighted Leavitt path algebras by using Bergman's diamond machinery. This allows us to describe a basis for such algebras. In turn we can then characterise simple and graded simple weighted Leavitt path algebras (cf. Section 3). There are unexpected interesting cases. For example, for the weighted graphs E and F below with one edge of weight two and the rest of weight one, the weighted Leavitt path algebra L(E, ω) is simple, whereas L(F, w) is not (Z2-graded) simple. α1,α2 v β E : u F : u α1,α2 v . β γ In Theorem 34 we show that a simple weighted Leavitt path algebra is isomorphic to a Leavitt path algebra. In Section 4 we construct a local valuation for a large class of weighted Leavitt path algebras (so-called LV-algebras). Using the valuation we show these algebras are prime, semiprimitive and nonsingular but contrary to Leavitt path algebras, they are not graded von Neumann regular. Further we classify the weighted Leavitt path algebras which are domains (see Theorem 41). This allows us to obtain a much larger class of prime and nonsingular rings than Leavitt path algebras. We finish this introduction by mentioning that K. McClanahan [15, 16] studied U nc n,n+k-algebras (first considered by D.V. Voiculescu). These are C ∗-algebras generated by elements uij, 1 ≤ i ≤ n, 1 ≤ j ≤ n + k subject to the relations uu∗ = In and u∗u = In+k, where u = (uij)n×(n+k). Note that the Cuntz algebra On corresponds to U nc n,n+k corresponds to the Leavitt algebra LC(n, n + k). However, the concept of weighted graph C ∗-algebra which as a special case cover U nc n,n+k is yet to be defined and explored. 1,n. Clearly in the pure algebra setting, U nc 2. Normal forms for weighted Leavitt path algebras We begin this section by recalling the concept of weighted graphs and weighted Leavitt path algebras, first introduced in [10]. Throughout the semigroup of positive integers is denoted by N and the monoid of non-negative integers by N0. Definition 1 (Weighted graph). A weighted graph E = (E0, Est, E1, s, r, ω) consists of three countable sets, E0 called vertices, Est structured edges and E1 edges, maps s, r : Est → E0, and a weight map ω : Est → N such that E1 = G α∈Est {αi 1 ≤ i ≤ ω(α)}, i.e., for any α ∈ Est, with ω(α) = k, there are k distinct elements {α1, ..., αk}, and E1 is the disjoint union of all such sets for all α ∈ Est. Remark 2. We sometimes write (E, ω) to emphasise the graph is weighted. If s−1(v) is a finite set for every v ∈ E0, then the graph is called row-finite. In this note we will consider only row-finite graphs. In this setting, if the number of vertices, i.e., E0, is finite, then the number of edges, i.e., E1, is finite as well and we call E a finite graph. Definition 3 (Weighted Leavitt path algebra). Let (E, ω) denote a weighted graph and R a unital ring. Set X := E0 ∪ E1 ∪ (E1)∗, where (E1)∗ = {α∗ i αi ∈ E1}. The quotient RhXi/I of the free R-ring RhXi generated by X and the ideal I of RhXi generated by the relations (1) vw = δvwv for every v, w ∈ E0, (2) s(α)αi = αir(α) = αi and r(α)α∗ i s(α) = αi for all α ∈ Est and 1 ≤ i ≤ ω(α), i = α∗ [ [ W W [ [ NORMAL FORMS FOR WEIGHTED LEAVITT PATH ALGEBRAS 3 (3) P{α∈Est,s(α)=v} αiα∗ j = δijv for all v ∈ E0 and 1 ≤ i, j ≤ max{ω(α) α ∈ Est, s(α) = v}, (4) P 1≤i≤max{ω(α),ω(β)} i βi = δαβr(α), for all α, β ∈ Est α∗ is called weighted Leavitt path algebra of (E, ω) and is denoted by LR(E, ω) or just L(E, ω). In relations (3) and (4), we set αi and α∗ i zero whenever i > ω(α). When R is not commutative, then we consider LR(E, ω) as a left R-module. Weighted Leavitt path algebras are involutary graded rings with unit if E0 is finite and local units otherwise. In fact, the weighted Leavitt path algebra LR(E, ω) is a Zn-graded ring, where n = max{ω(α) α ∈ Est}. The grading defined as follows: for v ∈ E0 define deg(v) = 0 and for α ∈ Est, deg(αi) = ei and i ) = −ei, 1 ≤ i ≤ ω(α), where αi ∈ E1 (note that the grading depends on the ordering of edges E1). deg(α∗ Here ei denotes the element of Zn whose i − th component is 1 and whose other components are 0 Example 4. Let K be a field. Then the weighted Leavitt path algebra of a weighted graph consisting of one vertex and n + k loops of weight n is LK(n, n + k). To show this, let Est = {y1, . . . , yn+k} with ω(yi) = n, 1 ≤ i ≤ n+k. Denote the n edges corresponding to the structure edge yi ∈ Est by {y1i, . . . , yni}. We visualise this data as follows: y1,n+k,...,yn,n+k • y11,...,yn1 y13,...,yn3 y12,...,yn2 Set xsr = y∗ x12 x22 ... rs for 1 ≤ r ≤ n and 1 ≤ s ≤ n + k and arrange the y's and x's in the matrices   . . . x1n . . . x2n ... . . . . . . xn+k,n xn+k,1 xn+k,2   y11 y12 y21 y22 ... ... yn1 yn2   , x11 x21 ...   . . . . . . . . . . . . y1,n+k y2,n+k ... yn,n+k Y = X = Then condition (3) of Definition 1 precisely says that Y · X = In,n and condition (4) is equivalent to X · Y = In+k,n+k which are the generators of LK(n, n + k). Example 5. Let (E, ω) be a weighted graph where w : Est → N is the constant map ω(α) = 1 for all α ∈ Est. Then Est = E1 and L(E, ω) is isomorphic to the usual Leavitt path algebra L(E). Example 6. In Example 4, the map defined by y1i 7→ yi, xi1 7→ yi 2 ≤ i ≤ n and yij 7→ 0, xji 7→ 0 otherwise, induces a surjective ring homomorphism ∗, 1 ≤ i ≤ k + 1, yi,i+k 7→ 1, xi+k,i 7→ 1 LK(n, n + k) −→ LK (1, k + 1) showing the Leavitt algebra L(1, k + 1) is a quotient of L(n, n + k). In Theorem 34 we show that a simple weighted Leavitt path algebras has to be a simple Leavitt path algebra. Example 7. Consider a weighted graph with one vertex and one structured edge α of weight n, i.e., E1 = {α1, . . . , αn} and an unweighted graph F with one vertex and n edges {α1, . . . , αn}. Then the map (E, ω) −→ F, αi 7→ α∗ i , induces an isomorphism on the level of LPAs, namely α1,...,αn αn L(cid:0) • , ω(cid:1) ∼= L(cid:0) • α3 α1 (cid:1). α2 Note that this isomorphism is not graded as L(E, ω) is Zn-graded, whereas L(F ) is just Z-graded. The graph F is called the unweighted graph associated with E (see Definition 26). e e Q Q E E e e Q Q E E 4 ROOZBEH HAZRAT AND RAIMUND PREUSSER Until the end of this section R denotes a unital ring and (E, ω) a weighted graph. For any v ∈ E0 which is not a sink (i.e., s−1(v) 6= ∅) fix an αv ∈ Est such that s(αv) = v and ω(αv) = ω(v) (1) where ω(v) = max{ω(α) α ∈ Est, s(α) = v}. Definition 8 (Generalised path). Set s(v) := v, r(v) := v, s(αi) := s(α), r(αi) := r(α), s(α∗ i ) := r(α) and r(α∗ i ) := s(α) for any v ∈ E0, α ∈ Est and 1 ≤ i ≤ ω(α). Let hXi denote the set of all nonempty words over X := E0 ∪ E1 ∪ (E1)∗. A word p ∈ hXi is called a generalised path if either p = x1x2 . . . xn for some n ≥ 1 and x1, . . . , xn ∈ E1 ∪ (E1)∗ such that r(xi) = s(xi+1), 1 ≤ i ≤ n − 1 or p = x1 for some x1 ∈ E0. The length p of a generalised path p = x1 . . . xn is n if n ≥ 1 and x1, . . . , xn ∈ E1 ∪ (E1)∗ or 0 if n = 1 and x1 ∈ E0. p is called trivial if p = 0 and nontrivial if p ≥ 1. Further we set s(p) := s(x1), and r(p) := r(xn). j )∗ for Definition 9 (Normal element of RhXi). A word A ∈ hXi is called word of type I if A = αv i (αv some v ∈ E0 which is not a sink and some 1 ≤ i, j ≤ ω(αv). A is called word of type II if A = α∗ 1β1 for some α, β ∈ Est. A generalised path is called normal if it does not contain any subwords of type I or type II. An element of RhXi is called normal if it lies in the linear span RhXiN of all normal generalised paths. We will show that any element of LR(E, ω) has precisely one normal representative in RhXi. For this we need some definitions and results from [6] which we will recall below. Note that a weighted Leavitt path algebra is a quotient of a free R-ring where R is a not necessarily commutative ring while in [6] free associative, unital algebras over commutative rings are considered. Hence we have to make a few adaptations (see Remark 19). Here we recall the basics of Bergman's diamond machinery needed in the paper. Until the end of the proof of Theorem 15, R denotes a unital ring and X any set. By an R-ring we mean a (not necessarily unital) ring which is an R-bimodule such that the multiplication is left linear in the first argument and right linear in the second one. By an ideal of an R-ring A we mean an ideal of the ring A which is an R-subbimodule of A. Let hXi denote the semigroup (with juxtaposition) of all nonemtpy words over X and set hXi := hXi ∪ {empty word}. Further let RhXi denote the free R-ring generated by X, i.e. the free left R-module generated by hXi made an R-ring by the multiplication ( Px∈hXi rxsyxy. Definition 10 (Reduction system). Let S be a set of pairs of the form σ = (Wσ, fσ), where Wσ ∈ hXi and fσ ∈ RhXi such that all coefficients of fσ lie in the center of R. Then S is called a reduction system for RhXi. For any σ ∈ S and A, B ∈ hXi, let rAσB denote the R-bimodule endomorphism of RhXi that maps AWσB to AfσB and fixes all other elements of hXi. The maps rAσB : RhXi → RhXi are called reductions. syy) = Px,y∈hXi rxx)( Py∈hXi Until the end of the proof of Theorem 15, S denotes a reduction system for RhXi. Definition 11 (Irreducible element, final sequence of reduction). We shall say a reduction rAσB acts trivially on an element a ∈ RhXi if the coefficient of AWσB in a is zero, and we shall call a irreducible (under S) if every reduction is trivial on a, i.e., if a involves none of the monomials AWσB. The R-subbimodule of all irreducible elements of RhXi will be denoted RhXiirr. A finite sequence of reductions r1, . . . , rn will be said to be final on a ∈ RhXi if rn . . . r1(a) ∈ RhXiirr. Definition 12 (Reduction-finite element, reduction-unique element). An element a of RhXi will be called reduction-finite if for every infinite sequence r1, r2, . . . of reductions, ri acts trivially on ri−1 . . . r1(a) for all sufficiently large i. If a is reduction-finite, then any maximal sequence of reductions ri, such that each ri acts nontrivially on ri−l . . . r1(a), will be finite, and hence a final sequence. It follows from their definition that the reduction-finite elements form an R-subbimodule of RhXi. We shall call an element a ∈ RhXi reduction-unique if it is reduction-finite, and if its images under all final sequences of reductions are the same. This common value will be denoted rS(a). The set of reduction-unique elements NORMAL FORMS FOR WEIGHTED LEAVITT PATH ALGEBRAS 5 of RhXi forms an R-subbimodule, and rS is a bilinear map (i.e. an R-bimodule homomorphism) of this submodule into RhXiirr (see [6, proof of Lemma 1.1(i)]). Definition 13 (Ambiguity, resolvable ambiguity). A 5-tuple (σ, τ, A, B, C) with σ, τ ∈ S and A, B, C ∈ hXi, such that Wσ = AB and Wτ = BC is called an overlap ambiguity of S. We shall say the overlap ambiguity (σ, τ, A, B, C) is resolvable if there exist compositions of reductions, r and r′, such that r(fσC) = r′(Afτ ). Similarly, a 5-tuple (σ, τ, A, B, C) with σ 6= τ and A, B, C ∈ hXi will be called an inclusion ambiguity if Wσ = B, Wτ = ABC and such an ambiguity will be called resolvable if AfσC and fτ can be reduced to a common expression. Definition 14 (Semigroup partial ordering compatible with S). By a semigroup partial ordering on hXi we shall mean a partial order ≤ such that B < B′ ⇒ ABC < AB′C for any B, B′ ∈ hXi, A, C ∈ hXi. We call ≤ compatible with S if for all σ ∈ S, fσ is a linear combination of monomials < Wσ. We are in a position to state Bergman's diamond lemma [6, Theorem 1.2]. This theorem will be used to find a basis for the weighted Leavitt path algebras. Theorem 15. Let ≤ be a semigroup partial ordering on hXi compatible with S and having descending chain condition. Then the following conditions are equivalent: (1) All ambiguities of S are resolvable. (2) All elements of RhXi are reduction-unique under S. (3) RhXiirr is a set of representatives for the elements of the R-ring RhXi/I, where I is the ideal of RhXi generated by the elements Wσ − fσ (σ ∈ S). When these conditions hold, RhXi/I may be identified with the R-bimodule RhXiirr made an R-ring by the multiplication a · b = rS(ab). Now we can use the previous theorem in order to prove that any element of LR(E, ω) has precisely one normal representative in RhXi where X = E0 ∪ E1 ∪ (E1)∗. Theorem 16. Let R be a unital ring and (E, ω) a row-finite weighted graph. Then the weighted Leavitt path algebra LR(E, ω) has a basis consisting of normal generalised paths. Namely, the basis elements are of the form p = x1 . . . xn, xi ∈ E1 ∪ (E1)∗, r(xi) = s(xi+1), 1 ≤ i ≤ n − 1 or p = x1, x1 ∈ E0 such that 1β1 where α, β ∈ Est is none of the words αv a subword of p. j )∗ where v ∈ E0 is not a sink and 1 ≤ i, j ≤ ω(αv) and α∗ i (αv Proof. In order to be able to apply Theorem 15, we replace the relations (1)-(4) in Definition 3 by the relations (1') For any v, w ∈ E0, vw = δvwv, (2') For any v ∈ E0, α ∈ Est and 1 ≤ i ≤ ω(α), vαi = δvs(α)αi, αiv = δvr(α)αi, vα∗ i = δvr(α)α∗ α∗ i v = δvs(α)α∗ i i and 6 ROOZBEH HAZRAT AND RAIMUND PREUSSER (3') For any α, β ∈ Est, 1 ≤ i ≤ ω(α) and 1 ≤ j ≤ ω(β), αiβj = 0 if r(α) 6= s(β), α∗ i βj = 0 if s(α) 6= s(β), αiβ∗ α∗ i β∗ j = 0 if r(α) 6= r(β) and j = 0 if s(α) 6= r(β) (4') For all v ∈ E0 which are not sinks and 1 ≤ i, j ≤ ω(αv), αv i (αv j )∗ = δijv − X α∈Est,s(α)=v α6=αv αiα∗ j and (5') For all α, β ∈ Est such that s(α) = s(β), α∗ 1β1 = δαβr(α) − X 2≤i≤max{ω(α),ω(β)} α∗ i βi. In relations (4') and (5'), we set αi and α∗ i zero whenever i > ω(α). Clearly the relations (1')-(5') generate the same ideal I of RhXi as the relations (1)-(4). Denote by S the reduction system for RhXi defined by the relations (1')-(5') (i.e., S is the set of all pairs σ = (Wσ, fσ) where Wσ equals the left hand side of an equation in (1')-(5') and fσ the corresponding right hand side). For any A = x1 . . . xn ∈ hXi set l(A) := n and m(A) := (cid:12)(cid:12){i ∈ {1, . . . , n − 1}xixi+1 is of type I or II}(cid:12)(cid:12). Define a partial ordering ≤ on hXi by A ≤ B ⇔ (cid:2)A = B(cid:3) ∨ (cid:2)l(A) < l(B)(cid:3) ∨ (cid:2)l(A) = l(B) ∧ ∀C, D ∈ hXi : m(CAD) < m(CBD)(cid:3). Clearly ≤ is a semigroup partial ordering on hXi compatible with S and the descending chain condition is satisfied. It remains to show that all ambiguities of S are resolvable. In the table below we list all types of ambiguities which may occur. (1') uvw (1') (2') αivw, α∗ (3') (4') (5') - - - i vw vαiw, vα∗ i w, αivβj, etc. (2') vwαi,vwα∗ i αiβjv, α∗ i (αv αv α∗ 1β1v i βjv etc. j )∗w (3') - vαiβj, vαiβ∗ αiβjγk, αiβjγ∗ j )∗γk, αv i (αv i (αv αv α∗ 1β1γi, α∗ 1β1γ∗ j etc. k etc. j )∗γ∗ k i (4') (5') βkαv i (αv i (αv - i (αv wαv j )∗, β∗ j )∗ kαv - α∗ 1αv 1(αv j )∗ - vα∗ 1β1 kα∗ 1β1, γ∗ j )∗ γkα∗ 1)∗β1 i (αv αv - 1β1 Note that there are no inclusion ambiguities. The ((4')-(5') and (5')-(4')) ambiguities αv α∗ 1αv ones which are most difficult to resolve. We will show how to resolve the ambiguity αv 1)∗β1 and j )∗, where v ∈ E0 is not a sink, 1 ≤ i, j ≤ ω(αv) and α, β ∈ Est such that s(α) = s(β) = v are the 1)∗β1 and leave 1(αv i (αv i (αv NORMAL FORMS FOR WEIGHTED LEAVITT PATH ALGEBRAS 7 the other cases to the reader. αv i (αv 1)∗β1 (4′) ①①①①①①①①①①①①①①①① (5′) "❉❉❉❉❉❉❉❉❉❉❉❉❉❉❉❉ αiα∗ 1)β1 (δi1v − Ps(α)=v = δi1vβ1 − Ps(α)=v α6=αv α6=αv αiα∗ 1β1 = δαv βαv i r(αv) − αv i (δαv βr(αv) − j )∗βj) ω(αv ) (αv Pj=2 αv i (αv ω(αv ) Pj=2 j )∗βj (5′) (4′) = = δi1vβ1 αi(δαβ r(α) − ω(αv ) Pj=2 α∗ j βj) − Ps(α)=v α6=αv δi1vβ1 − Ps(α)=v α6=αv δαβαir(α) ω(αv ) αiα∗ j βj α6=αv Pj=2 + Ps(α)=v δi1vβ1 − δβ6=αv βir(β) + Ps(α)=v Pj=2 αiα∗ ω(αv ) j βj α6=αv ✸✸✸✸✸✸✸✸✸✸✸✸ (2′) − ω(αv ) Pj=2 i r(αv) δαv βαv (δijv − Ps(α)=v α6=αv αiα∗ j )βj = δαv βαv i r(αv) − ω(αv ) + Pj=2 Ps(α)=v α6=αv δijvβj ω(αv ) Pj=2 αiα∗ j βj = δαv ββir(β) − δi≥2vβi ω(αv ) + Pj=2 Ps(α)=v α6=αv αiα∗ j βj ☞☞☞☞☞☞☞☞☞☞☞☞ (2′) ω(αv ) δi1β1 − δβ6=αv βi + Ps(α)=v α6=αv ω(αv ) Pj=2 = δαv ββi − δi≥2βi + Pj=2 Ps(α)=v α6=αv αiα∗ j βj αiα∗ j βj. It follows from Theorem 15, that RhXiirr is a set of representatives for the elements of RhXi/I = LR(E, ω). But clearly RhXiirr = RhXiN . (cid:3) In [3] a basis for a Leavitt path algebras were described. Here we obtain this result as a corollary of Theorem 16. Corollary 17. Let E be a directed graph and LR(E) the associated Leavitt path algebra. Then the monomi- m 6= αv(αv)∗ als pq∗, where p = x1 . . . xn and q = y1 . . . ym are paths (of possibly length zero) such that xny∗ for any v ∈ E0 which is not a sink, form a basis for LR(E). " 8 ROOZBEH HAZRAT AND RAIMUND PREUSSER In Section 3 we will use Theorem 16 to determine when a weighted Leavitt path algebra is simple resp. graded simple. In Section 4 we will use it to determine when a weighted Leavitt path algebra is a domain. In order to do this we need the concept of normal forms. Definition 18 (Normal form of an element of LR(E, ω)). Let a ∈ LR(E, ω). Then the unique normal representative of a ∈ RhXi is called the normal form of a and is denoted by NF(a). It follows from [6, Lemma 1.1] that NF : LR(E, ω) −→ RhXiN a 7−→ NF(a) is an isomorphism of R-bimodules (note that NF = rS). NF(a) · NF(b) := NF(ab), then NF is an isomorphism of R-rings. If we make RhXiN an R-ring by defining Remark 19. As mentioned above, for the diamond lemma, Bergman's starting point is a unital free algebra. One can state and use the diamond lemma in the setting of non-unital free algebras as our treatment in this section. However one can also start with X = E0 ∪ E1 ∪ E1∗ and consider the unital free algebra on X subject to the weighted Leavitt path algebra relations. This gives the unitisation ring LR(E, ω) × R. One can then conduct the proof of Theorem 16 in this ring. It is then easy to obtain the normal forms for LR(E, ω) from this setting as well. 3. Classification of simple and graded simple weighted Leavitt path algebras In this section R denotes a ring and (E, ω) a weighted graph. As usual, we call an ideal J of LR(E, ω) proper if J 6= {0} and J 6= LR(E, ω). Note that an ideal of the ring LR(E, ω) is the same as an ideal of the R-ring LR(E, ω) since LR(E, ω) has local units. In Definition 23 we define reducible and irreducible weighted graphs. We will show that if (E, ω) is reducible, then LR(E, ω) is isomorphic to LR(F ) for some unweighted graph F . It is an open question if there are examples of irreducible graphs (E, ω) such that LR(E, ω) is isomorphic to LR(F ) for some unweighted graph F . However we will show that if (E, ω) is irreducible, then LR(E, ω) is not graded simple. The main idea to show that LR(E, ω) is not graded simple provided (E, ω) is irreducible is to find a nontrivial lr-normal generalised path (cf. Definition 29). These are normal generalised paths p which have the property that if o and q are nontrivial normal generalised paths such that r(o) = s(p) and s(q) = r(p), then opq again is a normal generalised path. It is easy to see that a nontrivial lr-normal generalised path generates a proper graded ideal (follows from the uniqueness of the normal form). Definition 20 (Path, tree, cycle). A generalised path x1 . . . xn is called a path if n = 1 and x1 ∈ E0 or n ≥ 1 and x1, . . . , xn ∈ E1. If u, v ∈ E0 and there is a path p such that s(p) = u and r(p) = v, then we write u ≥ v. Clearly ≥ is a preorder on E0. If u ∈ E0 then T (u) := {v ∈ E0 u ≥ v} is called tree of u. A nontrivial path p such that v = s(p) = r(p) is called a closed path based at v. If p = x1 . . . xn is a closed path based at v = s(p) and s(xi) 6= s(xj) for every i 6= j, then p is called a cycle. Definition 21 (Connected components, dual of a generalised path). If u, v ∈ E0 and there is a generalised path p such that s(p) = u and r(p) = v, then we write u ≥g v. Clearly ≥g is an equivalence relation on E0. The equivalence classes of ≥g are called connected components. (E, ω) is called connected if there is only one connected component. Set v∗ := v for any v ∈ E0 and (α∗ i )∗ := αi for any αi ∈ E1. If 1 is called dual of p. Note that p∗ is a generalised p = x1 . . . xn is a generalised path, then p∗ := x∗ path such that s(p∗) = r(p) and r(p∗) = s(p). n . . . x∗ Definition 22 (Circle graph, line graph, oriented line graph). A weighted graph (E, ω) is called cyclic if it contains a cycle and acyclic otherwise. (E, ω) is called a circle graph, if it is connected, cyclic and s−1(v), r−1(v) ≤ 1 for any v ∈ E0. (E, ω) is called a line graph if it is connected, acyclic and s−1(v) + r−1(v) ≤ 2 for any v ∈ E0. (E, ω) is called an oriented line graph if it is a line graph such that s−1(v), r−1(v) ≤ 1 for any v ∈ E0. NORMAL FORMS FOR WEIGHTED LEAVITT PATH ALGEBRAS 9 Definition 23 (Unweighted graph, weight forest, reducible graph, irreducible graph). A If (E, ω) is unweighted, we identify E1 and Est weighted graph (E, ω) is called unweighted if ω = 1. and write LR(E) instead of LR(E, ω) (see Example 5). An α ∈ Est is called weighted if ω(α) > 1 and ω . A v ∈ E0 is called unweighted otherwise. The set of all weighted elements of Est is denoted by Est weighted if ω(v) > 1 and unweighted otherwise. The set of all weighted elements of E0 is denoted by E0 ω. The set E0 ω 6= ∅ is called T (v) is called weight forest of (E, ω). A weighted graph (E, ω) with E0 ω := Sv∈E0 ω reducible if s−1(v), r−1(v) ∩ s−1(E0 ω) ≤ 1 for any v ∈ E0 ω and irreducible otherwise. Consider the weighted graph (E′, ω′) one gets by dropping all vertices which do not belong to the weight forest E0 ω and all structured edges α such that s(α) or r(α) does not belong to the weight forest. One checks easily that (E, ω) is reducible if and only if all connected components of (E′, ω′) are either circle graphs or oriented line graphs. Example 24. Let (E, ω) be the weighted graph below. Note that E0 ω = {u, v, w}. Then x γ y δ u α1,α2 / v / w β u α1,α2 / v / w β is the weighted graph (E′, ω′) one gets as described in the paragraph after Definition 23. Since the only connected component of (E′, ω′) is an oriented line graph, (E, ω) is reducible. Example 25. Let (E, ω) be the weighted graph below. Note that E0 ω = {u, v, w, y}. Then x γ y δ1,δ2 u α1,α2 / v / w β y δ1,δ2 u α1,α2 / v / w β is the weighted graph (E′, ω′) one gets as described in the paragraph after Definition 23. Since the only connected component of (E′, ω′) is neither a circle graph nor an oriented line graph, (E, ω) is irreducible. Definition 26 (The unweighted graph associated with (E, ω)). Let (E, ω) be a weighted graph. We construct a unweighted graph F as follows. Let F 0 = E0, F 1 = {eαi αi ∈ E1}, s(eαi) = s(α) and r(eαi) = r(α) if s(α) 6∈ E0 ω. The graph F is called the unweighted graph associated with E. ω and s(eαi) = r(α) and r(eαi) = s(α) if s(α) ∈ E0 Thus F has the same vertices as E, an edge αi ∈ E1 is kept as it is if s(α) 6∈ E0 ω and it is reversed if s(α) ∈ E0 ω. / / / / / / / / 10 ROOZBEH HAZRAT AND RAIMUND PREUSSER Example 27. The unweighted F associated with the reducible weighted graph (E, ω) from Example 24 is the graph x v eγ eβ y eδ w. u eα1 eα2 It follows from the next proposition that LR(E, ω) ∼= LR(F ). Proposition 28. If (E, ω) is reducible, then LR(E, ω) ∼= LR(F ), where F is the unweighted graph associ- ated with (E, ω). Proof. Let X := E0 ∪ E1 ∪ (E1)∗ and X ′ := F 0 ∪ F 1 ∪ (F 1)∗. Then the bijection X → X ′ mapping v 7→ v for any v ∈ E0, αi 7→ eαi and α∗ i 7→ eαi for any αi ∈ E1 such that s(α) ∈ E0 ω induces an isomorphism f : RhXi → RhX ′i. Let I and I ′ be ideals of RhXi and RhX ′i generated by the relations (1)-(4) in Definition 3, respectively (hence LR(E, ω) = RhXi/I and LR(F ) = RhX ′i/I ′, see Example 5). In order to show that LR(E, ω) ∼= LR(F ) it suffices to show that f (I) = I ′. Set αi for any αi ∈ E1 such that s(α) 6∈ E0 ω and αi 7→ e∗ αi and α∗ i 7→ e∗ A(2) := (cid:8)s(α)αi − αi, αir(α) − αi, r(α)α∗ and for any v ∈ E0 which is not a sink A(1) := (cid:8)vw − δvwv v, w ∈ E0(cid:9), i s(α) − α∗ i − α∗ i , α∗ i α ∈ Est, 1 ≤ i ≤ ω(α)(cid:9), A(3) v := n X α∈s−1(v) αiα∗ j − δijv 1 ≤ i, j ≤ ω(v)o and max{ω(α),ω(β)} A(4) v := n i βi − δαβr(α) α, β ∈ s−1(v)o. α∗ X i=1 Then I is generated by A(1), A(2), the A(3) v 's and the A(4) v ∈ F is not a sink, analogously. Then I ′ is generated by B(1), B(2), the B(3) f (A(1)) = B(1) and f (A(2)) = B(2). Let v ∈ E0 be not a sink. One checks easily that if v ∈ E0 \ E0 f (A(3) v . Now assume that v ∈ E0 v is not a sink and (E, ω) is reducible. Set w := r(α). Then v ∈ RhX ′i, where v 's. Clearly ω, then ω. Then s−1(v) = {α} for some α ∈ Est since v 's. Define B(1), B(2), B(3) v 's and the B(4) v ) = B(4) v ) = B(3) v and f (A(4) v , B(4) and A(3) v = (cid:8)αiα∗ j − δijv 1 ≤ i, j ≤ ω(α)(cid:9) v = n A(4) ω(α) X i=1 i αi − wo. α∗ It is easy to show that s−1(w) = {eα1 , . . . , eαω(α)} in F (note that all edges which w emits in E get reversed since clearly w ∈ E0 ω) = {α} in E since (E, ω) is reducible). Hence ω; further r−1(w) ∩ s−1(E0 and w = n B(3) ω(α) X i=1 eαie∗ αi − wo v ) = B(4) Clearly f (A(3) is not a sink in F , there is a v ∈ E0 LR(E, ω) ∼= LR(F ). w and f (A(4) w . It follows that f (I) = I ′ (note that for any w ∈ E0 ω which ω and an α ∈ Est such that s(α) = v and r(α) = w). Thus (cid:3) B(4) w = (cid:8)eα∗ w ) = B(3) i eαj − δijv 1 ≤ i, j ≤ ω(α)(cid:9). \ \ o o NORMAL FORMS FOR WEIGHTED LEAVITT PATH ALGEBRAS 11 One idea to show that LR(E, ω) is not graded simple provided (E, ω) is irreducible, is to find nontrivial lr-normal generalised paths. We define them below. Definition 29 (l-normal, r-normal, lr-normal generalised paths). An element x ∈ X = E0 ∪ E1 ∪ (E1)∗ is called l-normal, iff there is no y ∈ X such that yx is of type I or II (see Definition 9). x is called r-normal, iff there is no y ∈ X such that xy is of type I or II. x is called lr-normal, iff it is l-normal and r-normal. More generally a normal gen. path p = x1...xn is called l-normal if x1 is l-normal, r-normal if xn is r-normal and lr-normal if it is l-normal and r-normal. Let p be a nontrivial lr-normal gen. path and J the ideal of LR(E, ω) generated by (the image of) p. One checks easily that NF(J) is the linear span of the normal generalised paths containing p as a subword (note that if o and q are nontrivial normal generalised paths such that r(o) = s(p) and s(q) = r(p), then opq again is a normal generalised path). It follows from the uniqueness of the normal form (see Theorem 16) that J 6= LR(E, ω) (for instance J does not contain any vertex). Hence nontrivial lr-normal gen. paths generate proper graded ideals. Lemma 30. If there is a v ∈ E0 such that s−1(v) contains two distinct weighted structured edges α, β, then there is a nontrivial lr-normal gen. path. Hence LR(E, ω) is not graded simple. Proof. The normal form defined in Section 2 depends on the choice of elements αv ∈ s−1(v) (v not a sink) such that ω(αv) is maximal (see (1)). Without loss of generality assume that ω(α) ≥ ω(β). Then clearly one can choose αv 6= β. One checks easily that β2 is lr-normal. (cid:3) An example of an irreducible weighted graph satisfying the condition of the previous lemma is the following weighted graph: α1,α2 E : u v . β1,β2 Unfortunately there are irreducible weighted graphs without nontrivial lr-normal gen. paths, for example α1,α2 F : u v . β One can use the Diamond Lemma to show that the ideal of LR(F, ω) generated by α1 is proper, see the proof of the next lemma. Lemma 31. If there is a v ∈ E0 such that s−1(v) contains a weighted structured edge α and an unweighted structured edge β, then LR(E, ω) is not graded simple. Proof. By the previous lemma we can assume that α is the only weighted structured edge in s−1(v). Hence we can choose αv = α. Let J be the ideal of LR(E, ω) = RhXi/I generated by α1 + I. Using Bergman's machinery (see §2) we will show that LR(E, ω)/J is not trivial which implies that J 6= LR(E, ω). It is easy to show that LR(E, ω)/J is isomorphic to the quotient RhXi/I ′ where I ′ is the ideal of RhXi generated by the relations (1)-(4) in Definition 3 and the relation 1, β1β∗ (5) α1 = 0. We call the words α1, α∗ 2α2 ∈ hXi words of type III. Further we call a generalised path strongly normal if it is normal and does not contain a subword of type III. An element of RhXi is called strongly normal if it lies in the linear span RhXiSN of all strongly normal generalised paths. Using Theorem 15 we will show that RhXiSN is a set of representatives for the elements of RhXi/I ′. In order to be able to apply Theorem 15 we replace the relations (1)-(5) by the relations (1')-(5') in the proof of Theorem 16 and the relations 1, α∗ C C C C 12 ROOZBEH HAZRAT AND RAIMUND PREUSSER (6') α1 = 0, α∗ (7') β1β∗ 1 = 0, 1 = v − Pγ∈s−1(v) γ1γ∗ 1 and γ6=α,β (8') α∗ 2α2 = r(α) − ω(α) Pi=3 α∗ i αi. Clearly the relations (1')-(8') generate the same ideal J of RhXi as the relations (1)-(5). Denote by S the reduction system for RhXi defined by the relations (1')-(8') (i.e., S is the set of all pairs σ = (Wσ, fσ) where Wσ equals the left hand side of an equation in (1')-(8') and fσ the corresponding right hand side). For any A = x1 . . . xn ∈ hXi set l(A) := n, mI,II(A) := {i ∈ {1, . . . , n − 1} xixi+1 is of type I or II} and mIII(A) := {i ∈ {1, . . . , n} either xi is of type III or i ≤ n − 1 and xixi+1 is of type III}. Define a partial ordering ≤ on hXi by A ≤ B ⇔ [A = B] ∨ [l(A) < l(B)] ∨ [l(A) = l(B) ∧ ∀C, D ∈ hXi : mI,II(CAD) < mI,II(CBD)] ∨ [l(A) = l(B) ∧ ∀C, D ∈ hXi : mI,II(CAD) ≤ mI,II(CBD) ∧ ∀C, D ∈ hXi : mIII(CAD) < mIII(CBD)]. Clearly ≤ is a semigroup partial ordering on hXi compatible with S and the descending chain condition is satisfied. Further it is easy to show that all ambiguities of S are resolvable. For example 2α2α∗ α∗ 2 α∗ 2v (4′) ~⑤⑤⑤⑤⑤⑤⑤⑤⑤⑤⑤⑤⑤⑤ ❇❇❇❇❇❇❇❇❇❇❇❇❇❇ (2′) ω(α) Pi=3 i αi(cid:1)α∗ α∗ 2 (8′) $■■■■■■■■■■■■■ (cid:0)r(α) − z✉✉✉✉✉✉✉✉✉✉✉✉✉✉ (2′),(4′) α∗ 2. It follows from Theorem 15, that RhXiirr is a set of representatives for the elements of RhXi/I ′. But clearly RhXiirr = RhXiSN. It follows that RhXi/I ′ has more than one element since RhXiSN has more 1 ∈ RhXiSN). Since LR(E, ω)/J ∼= RhXi/I ′, it follows that than one element (for example v, α2, α∗ LR(E, ω)/J has more than one element and hence J 6= LR(E, ω). Thus J is a proper graded ideal. 2, β1, β∗ (cid:3) We can now use Lemma 30 and Lemma 31 to prove that if (E, ω) is irreducible, then LR(E, ω) is not graded simple (note that there are irreducible weighted graphs which neither satisfy the condition of Lemma 30 nor the condition of Lemma 31, e.g. the weighted graph (E, ω) from Example 25). Proposition 32. If (E, ω) is irreducible, then LR(E, ω) is not graded simple. Proof. Since (E, ω) is irreducible, there is a v ∈ E0 ω such that s−1(v) > 1 or r−1(v) ∩ s−1(E0 ω) > 1. $ ~ z NORMAL FORMS FOR WEIGHTED LEAVITT PATH ALGEBRAS 13 Case 1 Assume that s−1(v) > 1. ω, there is a u ∈ E0 Since v ∈ E0 ω and a path p such that s(p) = u and r(p) = v. By Lemma 30 and Lemma 31, we may assume that s−1(u) = {α} for some α ∈ Est ω . It follows that p is nontrivial and p = αip′ for some 1 ≤ i ≤ ω(α) and a path p′ (here we allow p′ to be the empty word). Choose a β ∈ s−1(v) such that β 6= βv (this is possible as s−1(v) > 1). One checks easily that α2p′β1 is lr-normal. Case 2 Assume r−1(v) ∩ s−1(E0 ω) > 1. Since r−1(v) ∩ s−1(E0 ω and distinct α, β ∈ Est such that s(α) = u1, ω) > 1, there are u1, u2 ∈ E0 ω, there are w1, w2 ∈ E0 s(β) = u2 and r(α) = r(β) = v. Since u1, u2 ∈ E0 ω and paths p1 and p2 such that s(p1) = w1, r(p1) = u1, s(p2) = w2 and r(p2) = u2. By Lemma 30 and Lemma 31, we may assume that s−1(w1) = {γ} and s−1(w1) = {ǫ} for some γ, ǫ ∈ Est ω . Assume that p1 and p2 2 for some 1 ≤ i ≤ ω(γ), 1 ≤ j ≤ ω(ǫ) and paths p′ are nontrivial. Then p1 = γip′ 1 and p′ 1 and p′ 2 is lr-normal. The case that p1 or p2 is trivial can be handled analogously. (cid:3) 1 and p2 = ǫjp′ 2 to be the empty word). One checks easily that γ2p′ 1 (here we allow p′ 1α1β∗ 1 (p′ 2)∗ǫ∗ Example 33. Let (E, ω) be the irreducible weighted graph from Example 25. One checks easily that α2β1δ∗ 2 is lr-normal. Hence LR(E, ω) is not graded simple. We are ready to classify the simple weighted Leavitt path algebras. Theorem 34 (Simplicity Theorem). The weighted Leavitt path algebra LR(E, ω) is simple if and only if (E, ω) is reducible and LR(F ) is simple where F is the unweighted graph associated with (E, ω). Proof. Follows from Proposition 28 and Proposition 32. (cid:3) Theorem 34 shows that although weighted Leavitt path algebras produce a wide range of algebras which are not covered by Leavitt path algebras (such as L(n, n + k), n ≥ 2), the class of simple weighted algebras doesn't produce new examples. We can prove a graded version of Theorem 34 in the case that R is a field. Note that LR(E, ω) is Zn-graded where n = max{ω(α) α ∈ Est} while LR(F ) is Z-graded. Theorem 35 (Graded Simplicity Theorem). If R is a field, then the weighted Leavitt path algebra LR(E, ω) is graded simple if and only if (E, ω) is reducible and LR(F ) is graded simple where F is the unweighted graph associated with (E, ω). Proof. (⇒) Assume that LR(E, ω) is graded simple. Then (E, ω) is reducible by Proposition 32 and hence LR(E, ω) ∼= LR(F ) by Proposition 28. Assume that LR(F ) contains a proper graded ideal J. Then J is generated by elements of F 0 by [2, Theorem 2.5.8] (namely J is generated by a hereditary and saturated subset of F 0). But the isomorphism between LR(E, ω) and LR(F ) established in Proposition 28 maps E0 onto F 0. Therefore the image of J in LR(E, ω) is generated by elements of E0 and therefore it is a proper graded ideal of LR(E, ω). But this contradicts the assumption that LR(E, ω) is graded simple. Thus LR(F ) is graded simple. (⇐) Assume that LR(F ) is graded simple and (E, ω) is reducible. Let φ : LR(E, ω) → LR(F ) be the isomorphism induced by the map f : RhXi → RhX ′i defined in Proposition 28. The only hereditary and saturated subsets of F 0 are ∅ and F 0 by [2, Theorem 2.5.8]. It follows from [2, Theorem 2.8.10] that every proper ideal J of LR(F ) is generated by terms of the form v +r1c+· · ·+rmcm where m ≥ 1, r1, . . . , rm ∈ R, rm 6= 0 and c is a cycle based at v without exit. But the elements f −1(v), f −1(r1c), . . . , f −1(rmcm) are homogeneous in LR(E, ω). Assume that f −1(J) is a graded ideal of LR(E, ω), then, by the definition of a graded ideal, all the elements f −1(v), f −1(r1c), . . . , f −1(rmcm) are contained in f −1(J) and hence all the elements v, r1c, . . . , rmcm are contained in J. But this implies that J is graded which is a contradiction. Thus LR(E, ω) is graded simple. (cid:3) 14 ROOZBEH HAZRAT AND RAIMUND PREUSSER 4. LV-algebras and classification of weighted Leavitt path algebras which are domains In [12] Leavitt proved that the algebra LZ(2, 3) is a domain. Namely he defined normal forms for the elements of LZ(2, 3) and showed that the map ν which associates to each x ∈ LZ(2, 3) the degree of NF(x) as a polynomial in the generators of LZ(2, 3) (with the convention ν(0) = −∞) is a valuation, i.e. ν(xy) = ν(x) + ν(y). It follows that if 0 = xy, then −∞ = ν(0) = ν(xy) = ν(x) + ν(y) and hence x or y must be 0. Later Cohn [8] proved, using the same method, that the Leavitt algebras LK(n, n + k), n ≥ 2 are domains if K is a field. Here we adapt Leavitt's approach to study certain weighted Leavitt path algebras. Clearly there is no valuation on LR(E, ω) if there is more than one vertex (since in this case there are zero divisors). But for a large class of weighted Leavitt path algebras, so-called LV-algebras (Definition 38), one can define a "local" valuation. Using the local valuation we prove that LV-algebras are prime, semiprimitive and non-singular, similar to the case of Leavitt path algebras. However they are not (graded) von Neumann regular. Thus we obtain a much larger class of prime and nonsingular rings than Leavitt path algebras. Definition 36 (Support of an element of LR(E, ω)). Let (E, ω) be a weighted graph and R a ring. If a ∈ LR(E, ω), then the set supp(a) of all normal generalised paths occurring in NF(a) with nonzero coefficient is called the support of a. Definition 37 (local valuation). Let (E, ω) be a weighted graph and R a ring. A local valuation on LR(E, ω) is a map ν : LR(E, ω) −→ N0 ∪ {−∞} such that (1) ν(a) = −∞ if and only if a = 0, (2) ν(a) = 0 if and only if a 6= 0 and supp(a) ⊆ E0, (3) ν(a + b) ≤ max{ν(a), ν(b)} for any a, b ∈ LR(E, ω) and (4) ν(ab) = ν(a) + ν(b) for any v ∈ E0, a ∈ LR(E, ω)v and b ∈ vLR(E, ω). We use the conventions −∞ ≤ x and x + (−∞) = (−∞) + x = −∞ for any x ∈ N0 ∪ {−∞}. For a certain type of weighted Leavitt path algebras, we can construct local valuations. Let (E, ω) be a weighted graph and set ν := deg ◦ NF (for a more formal definition of ν see Proposition 40). Assume that Est contains an unweighted structured edge α. Then ν(α∗ 1)+ν(α1) by relation (4) in Definition 3. Hence ν is not a local valuation. Assume now that there is a v ∈ E0 and an α ∈ s−1(v) such that ω(α) > ω(β) for any β ∈ s−1(v) \ {α}. Then ν(α∗ ω(α)) + ν(αω(α)) by relation (3) in Definition 3 and again ν is not a local valuation. This motivates the following definition. ω(α)αω(α)) = ν(v) = 0 6= 2 = ν(α∗ 1α1) = ν(r(α)) = 0 6= 2 = ν(α∗ Definition 38 (LV-graph, LV-rose, LV-algebra). A weighted graph (E, ω) is called an LV-graph if the condition ω(α) ≥ 2 ∀α ∈ Est and {α ∈ s−1(v) ω(α) = ω(v)} ≥ 2 ∀v ∈ E0, v not a sink (LV) is satisfied. Recall that ω(v) = max{ω(α) α ∈ s−1(v)} for any v ∈ E0 which is not a sink. In order to simplify the exposition, we additionally require that a LV-graph has edges (i.e., Est 6= ∅) and is connected (see Definition 21). An LV-graph (E, ω) such that E0 = 1 is called an LV-rose. The weighted Leavitt path algebras LR(E, ω) where R is a ring and (E, ω) is an LV-graph are called LV-algebras. Example 39. The weighted graph α1,α2,α3 E : δ1,δ2 • β1,β2,β3 γ1,γ2,γ3 is an LV-rose. We will see later that if K is a field, then LK(E, ω) is a domain which is not isomorphic to any of the algebras LK(n, n + k). e e E E % % NORMAL FORMS FOR WEIGHTED LEAVITT PATH ALGEBRAS 15 For a generalised path p, recall the length p from Definition 8. Proposition 40. If (E, ω) is an LV-graph and R a domain, then the map ν : LR(E, ω) −→ N0 ∪ {−∞} a 7−→ max{p p ∈ supp(a)}. is a local valuation on LR(E, ω). Here we use the convention max(∅) = −∞. It remains to show (4). Let v ∈ E0, a ∈ LR(E, ω)v and b ∈ Proof. Obviously (1), (2) and (3) hold. vLR(E, ω). If one of the terms ν(a) and ν(b) equals 0 or −∞, then clearly ν(ab) = ν(a) + ν(b). Suppose now ν(a), ν(b) ≥ 1. Clearly ν(ab) ≤ ν(a) + ν(b) since a reduction preserves or decreases the length of a generalised path. It remains to show that ν(ab) ≥ ν(a) + ν(b). Let be the elements of supp(a) with maximal length (namely ν(a)) and pk = xk 1 . . . xk ν(a) (1 ≤ k ≤ r) ql = yl 1 . . . yl ν(b) (1 ≤ l ≤ s) be the elements of supp(b) with maximal length (namely ν(b)). We assume that the pk's are pairwise distinct and also that the ql's are pairwise distinct. Since NF is a linear map, we have NF(pkql) =   pkql if xk ν(a)yl 1 is not of type I or II, 1 . . . xk ν(a)−1yl 2 . . . yl ν(a)−1αiα∗ ν(b)]) j yl 1 . . . xk xk 2 . . . yl ν(b) NF([δij xk − Pα∈s−1(u), α6=αu 1 . . . xk NF([δαβ xk P − 2≤i≤max{ω(α),ω(β)} 2 . . . yl ν(a)−1yl 1 . . . xk xk ν(b)]) ν(a)−1α∗ i βiyl 2 . . . yl ν(b) if xk ν(a)yl 1 is of type I, if xk ν(a)yl 1 is of type II. Case 1 Assume that xk ν(a)yl 1 is not of type I or II for any k, l. Then pkql ∈ supp(ab) for any k, l. It follows that ν(ab) ≥ pkql = ν(a) + ν(b). Case 2 Assume that there are k, l such that xk ν(a)yl 1 is of type I. Then there are a u ∈ E0 and 1 ≤ i, j ≤ ω(αu) such that xk weight ω(u). This is possible since (E, ω) is an LV-graph. ν(a)yl 1 = αu i (αu j )∗. Choose an α 6= αu of Case 2.1 Assume pk′ql′ 6= xk 1 . . . xk ν(a)−1αiα∗ j yl 2 . . . yl ν(b) for any k′, l′. Then xk 1 . . . xk ν(a)−1αiα∗ j yl 2 . . . yl ν(b) ∈ supp(ab) since it doesn't cancel with another term. It follows that ν(ab) ≥ ν(a) + ν(b). Case 2.2 Assume pk′ql′ = xk ν(a)−1αiα∗ j yl One checks easily that in this case 1 . . . xk 2 . . . yl ν(b) for some k′, l′. pkql′ = xk 1 . . . xk ν(a)−1αu i α∗ j yl 2 . . . yl ν(b) ∈ supp(ab). It follows that ν(ab) ≥ ν(a) + ν(b). Case 3 Assume that there are k, l such that xk Then there are α, β ∈ Est such that xk ν(a)yl ν(a)yl 1 is of type II. 1 = α∗ 1β1. Since (E, ω) is an LV-graph, ω(α), ω(β) ≥ 2. 16 ROOZBEH HAZRAT AND RAIMUND PREUSSER Case 3.1 Assume pk′ql′ 6= xk 1 . . . xk ν(a)−1α∗ 2β2yl 2 . . . yl ν(b) for any k′, l′. Then xk 1 . . . xk ν(a)−1α∗ 2β2yl 2 . . . yl ν(b) ∈ supp(ab) since it doesn't cancel with another term. It follows that ν(ab) ≥ ν(a) + ν(b). Case 3.2 Assume pk′ql′ = xk 1 . . . xk ν(a)−1α∗ 2β2yl 2 . . . yl ν(b) for some k′, l′. One checks easily that in this case 1 . . . xk pkql′ = xk ν(a)−1α∗ 1β2yl 2 . . . yl ν(b) ∈ supp(ab) It follows that ν(ab) ≥ ν(a) + ν(b). Hence (4) also holds and thus ν is a local valuation on LR(E, ω). (cid:3) Theorem 41. Let (E, ω) be a weighted graph and R a ring. Then LR(E, ω) is a domain if and only if R is a domain and (E, ω) is either an unweighted rose with not more than one petal or an LV-rose. Proof. One checks easily that LR(E, ω) is the zero ring or has zero divisors if R is not a domain or (E, ω) is neither an unweighted rose with not more than one petal nor an LV-rose. Suppose now that R is a domain. If (E, ω) is a rose with no petals, then LR(E, ω) ≃ R and if (E, ω) is an unweighted rose with one petal, then LR(E, ω) ≃ R[X, X −1]. Hence LR(E, ω) is a domain in these cases. If (E, ω) is an LV-rose, then there is a local valuation on LR(E, ω) by the previous proposition. It follows from (1) and (4) in Definition 37 that LR(E, ω) is a domain. (cid:3) We recover the theorem of Leavitt [14, footnote 6] and Cohn [8]. Corollary 42. For a domain R, the Leavitt algebras LR(n, n + k), n ≥ 2, are domains. In contrast to the fact that the class of weighted Leavitt path algebras does not contain any new examples of simple algebras, it contains new examples of domains. We use the dependence number, a ring-invariant introduced by Cohn in [8], in order to prove that there are weighted Leavitt path algebras which are domains but are not isomorphic to any of Leavitt's algebras. Definition 43 (Filtration, valuation). Let R be a ring. A (positive increasing) filtration on R is a map ν : R → N0 ∪ {−∞} such that ν(x) = −∞ ⇔ x = 0, ν(x − y) ≤ max{ν(x), ν(y)}, ν(xy) ≤ ν(x) + ν(y) ∀x, y ∈ R. (2) Let ν be a filtration on R and set Rn := {x ∈ R ν(x) ≤ n} for any n ∈ N0 ∪ {−∞}. Then each Rn is an additive subgroup of R and RmRn ⊆ Rm+n ∀m, n ∈ N0, [ n∈N0 Rn = R, {0} = R−∞ ⊆ R0 ⊆ R1 ⊆ . . . . (3) Conversely if {Rn n ∈ N0} is a family of additive subgroups of R such that (3) holds, then the map ν : R → N0 ∪ {−∞} defined by ν(x) = min{n ∈ N0 ∪ {−∞} x ∈ Rn}, is a filtration on R. Hence fixing a filtration on R is the same as fixing a family of additive subgroups of R such that (3) holds. Every ring has the trivial filtration ν defined by ν(0) = −∞ and ν(x) = 0 ∀x 6= 0. A filtration ν on R such that ν(xy) = ν(x) + ν(y) ∀x, y ∈ R is called a valuation. Definition 44 (Dependence number of a ring). Let R be a ring and ν a filtration on R. (1) A subset X of R is called R-dependent if X = {0} or if X = {x1, . . . , xr} and there exist a1, . . . , ar ∈ R such that . ν(x1) + ν(a1) = · · · = ν(xr) + ν(ar) > v( X i=1 xiai) r NORMAL FORMS FOR WEIGHTED LEAVITT PATH ALGEBRAS 17 (2) An element y ∈ R is called R-dependent on a subset X of R if y = 0 or if there exist x1, . . . , xr ∈ X and a1, . . . , ar ∈ R such that ν(y − r X i=1 xiai) < ν(y), ν(xi) + ν(ai) ≤ ν(y) (i = 1, . . . , r). Further, a subset X of R is called strongly R-dependent if it is R-dependent and any element of maximal value in X is R-dependent on the remaining elements of X. The dependence number of R relative to ν, λν(R), is the least integer n for which there exists an R-dependent set of n elements which is not strongly R-dependent. The supremum of the λν (R) for all filtrations ν on R is called the dependence number of R and is denoted by λ(R). Theorem 45. Let K be a field and (E, ω) an LV-rose such that the minimal weight is 2. Then the dependence number of LK(E, ω) equals 2. 1, α∗ 1 − α∗ 1β1 + α∗ 2β2 = 0. It follows that the set {α∗ 2} is not strongly L-dependent since ν(α∗ Proof. Set L := LK(E, ω). Let α ∈ Est be of weight 2 and choose a β ∈ Est such that β 6= α (possible since any LV-algebra contains at least 2 structured edges). Let ν be valuation on L defined in Proposition 1, α∗ 40. By relation (4) in Definition 3 we have α∗ 2} is L-dependent 2x) ≥ ν(α∗ with respect to ν. But {α∗ 1) for any x ∈ L. Hence λν(L) ≤ 2. On the other hand λν(L) > 1 by [8, Proposition 4.1]. Thus λν(L) = 2. Assume there is a filtration ν′ on L such that λν ′(L) ≥ 3. Then ν′ is a valuation by [8, Proposition 4.1]. Hence ν′(1) = ν′(1 · 1) = ν′(1) + ν′(1) and therefore ν′(1) = 0. It follows that ν′(x) = 0 for any right invertible element x. On the other hand if ν′(x) = 0, then the set {x, 1} is L-dependent with respect to ν′. Hence it is strongly so and we get that x is right invertible. But the right invertible elements of L are precisely the elements of K \ {0} (since ν is a valuation). Hence we have shown that ν′(x) = 0 if and only if x ∈ K \ {0}. W.l.o.g. assume that ν′(α∗ 2. By applying an analog of the Euclidean algorithm to r−1 and r0 we get elements q1, . . . , qn, r1, . . . , rn ∈ L, where n ≥ 1, such that ν′(q1) ≥ 0, ν′(q2), . . . , ν′(qn) > 0, ν′(r0) > ν′(r1) > · · · > ν′(rn) = −∞ and 1) ≥ ν′(α∗ 2). Set r−1 := α∗ 1 and r0 := α∗ r−1 = r0q1 + r1, r0 = r1q2 + r2, . . . , rn−3 = rn−2qn−1 + rn−1, rn−2 = rn−1qn + rn (see [7, pp. 340 -- 341]). We prove by induction on i that ν(ri) = 1 + i X j=1 ν(qj) for any i ∈ {0, . . . , n} (4) (which means that ν(ri) increases as i increases while ν′(ri) decreases). One checks easily that (4) holds for i = 0, 1. Let now 2 ≤ i ≤ n. We have ν(ri) = ν(ri−2 − qiri−1). By the induction hypothesis, ν(ri−2) = 1 + i−2 Pj=1 ν(qj) and ν(qiri−1) = ν(qi) + ν(ri−1) = 1 + i Pj=1 ν(qj). But ν′(qi) > 0 since i ≥ 2. Hence qi 6∈ K and therefore ν(qi) > 0. It follows that ν(qiri−1) > ν(ri−2) and hence ν(ri) = ν(qiri−1) = 1 + i Pj=1 ν(qj). Therefore (4) holds. It follows that −∞ = ν(0) = ν(rn) contradiction. Thus λ(L) = 2. (4) = 1 + n Pj=1 ν(qj) ≥ 1 and hence we have a (cid:3) Let K be a field and (E, ω) an LV-rose such that the minimal weight is 2, the maximal weight is l ≥ 3 and the number of structured edges is l + m for some m > 0. By [10, Theorem 5.21] and the previous theorem, LK(E, ω) has module type (l, m) (cf. [14]) and dependence number 2. Let n, k ≥ 1. By Example 4, [10, Theorem 5.21] and [8, Theorem 5.2], LK(n, n + k) has module type (n, k) and dependence number n. Hence LK (E, ω) cannot be isomorphic to one of Leavitt's algebras LK(n, n + k). In particular, if (E, ω) is the LV-rose from Example 39 and K a field, then the domain LK(E, ω) is not isomorphic to any of the algebras LK(n, n + k). 18 ROOZBEH HAZRAT AND RAIMUND PREUSSER In the next three theorems we show that LV-algebras over domains are prime, semi-primitive and nonsingular rings. We further show that contrary to the case of Leavitt path algebras, they are not graded von Neumann regular. Theorem 46. Let (E, ω) be an LV-graph and R a domain. Then LR(E, ω) is a prime ring. Proof. Let a, b ∈ LR(E, ω) \ {0}. Choose u, v ∈ E0 such that au, vb 6= 0. It follows from (1) in Definition 37 that ν(au), ν(vb) ≥ 0. Since (E, ω) is connected, there is a generalised path p such that s(p) = u and r(p) = v. Clearly ν(p) ≥ 0 since ν is a local valuation. It follows that ν(apb) =ν(a(upv)b) =ν((au)p(vb)) (4) = ν(au) + ν(p) + ν(vb) ≥ 0. It follows from (1) in Definition 37 that apb 6= 0 and thus LR(E, ω) is prime. (cid:3) Theorem 47. Let (E, ω) be an LV-graph and R a domain. Then LR(E, ω) is a nonsingular ring. Proof. Let a ∈ LR(E, ω)\{0}. Choose a v ∈ E0 such that av 6= 0. Consider the right ideal vLR(E, ω). Then annr(a) ∩ vLR(E, ω) = 0. For, if there is b ∈ vLR(E, ω) such that ab = 0, then condition (4) of Definition 37 implies b = 0 (as LV-algebras are "locally" domain). This shows that annr(a), a ∈ LR(E, ω) \ {0}, is not essential and thus LR(E, ω) is right nonsingular. The proof for left nonsingularity is similar. (cid:3) Recall that a ring A is called von Neumann regular if for any a ∈ A, there is b ∈ A such that aba = a. If A is a graded ring, then A is called graded von Neumann regular if the identity above holds for homogeneous elements. It is known that Leavitt path algebras are graded von Neumann regular rings [11, Corollary 1.6.17]. In contrast we have the following theorem. Theorem 48. Let (E, ω) be an LV-graph and R a domain. Then LR(E, ω) is not (graded) von Neumann regular. Proof. Choose an α ∈ Est. Assume that there is an a ∈ LR(E, ω) such that α1aα1 = α1. Set u := s(α) and v := r(α). Then clearly vau 6= 0 (otherwise α1aα1 = 0). It follows that ν(α1aα1) =ν((α1v)a(uα1)) =ν(α1(vau)α1)) (4) = ν(α1) + ν(vau) + ν(α1) (1),(2) > ν(α1). Since this is a contradiction, LR(E, ω) is not (graded) von Neumann regular. (cid:3) Lemma 49. Let (E, ω) be an LV-graph and R a domain. If J is a nonzero ideal of LR(E, ω), then for any n ∈ N and u, v ∈ E0 there is an a ∈ J ∩ uLR(E, ω)v such that ν(a) > n. Proof. Let J be a nonzero ideal of LR(E, ω), n ∈ N and u, v ∈ E0. Choose a nonzero element a′ ∈ J. Then there are z1, z2 ∈ E0 such that z1a′z2 6= 0. Now it easy to show that there are generalised paths p and q of length > n such that s(p) = u, r(p) = z1, s(q) = z2, r(q) = v (note that any vertex must emit or receive an structured edge since (E, ω) is an LV-graph). Clearly ν(p) = p, ν(q) = q > n. Set a := pa′q ∈ J ∩ uLR(E, ω)v. Then ν(a) = ν(pa′q) (4) = ν(p) + ν(z1a′z2) + ν(q) > n. (cid:3) Theorem 50. Let (E, ω) be an LV-graph and R a domain. Then the Jacobson radical of LR(E, ω) is zero. NORMAL FORMS FOR WEIGHTED LEAVITT PATH ALGEBRAS 19 Proof. Suppose the Jacobson radical J of LR(E, ω) is not zero. Choose a v ∈ E0. Then, by Lemma 49, there is an a ∈ J ∩ vLR(E, ω)v such that ν(a) > 0. Since a ∈ J, a is left quasi-regular, i.e., there is a b ∈ LR(E, ω) such that b+a = ba. By multiplying v from the right and from the left one gets vbv+a = vbva. Hence we may assume that b ∈ vLR(E, ω)v. It follows that max{ν(b), ν(a)} (3) ≥ ν(b + a) = ν(ba) (4) = ν(b) + ν(a). This implies that ν(b) = 0 and hence, by (2), b = λv for some λ ∈ R \{0}. It follows that λv = b = ba− a = (λv)a − a = λa − a = (λ − 1)a. But this is a contradiction since ν(λv) = 0 but either ν((λ − 1)a) = −∞, if λ = 1, or ν((λ − 1)a) = ν(a) > 0, if λ 6= 1. Thus the Jacobson radical of LR(E, ω) is zero. (cid:3) References [1] G. Abrams, G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293 (2005), no. 2, 319 -- 334. 1 [2] G. Abrams, P. Ara, M. Siles Molina, Leavitt path algebras, primer handbook, to appear. 1, 13 [3] A. Alahmadi, H. Alsulami, S. Jain, E. Zelmanov, Leavitt path algebras of finite Gelfand-Kirillov dimension, J. Algebra Appl. 11 (2012), no. 6, 1250 -- 225. 7 [4] P. Ara, K. Goodearl, Leavitt path algebras of separated graphs, J. reine angew. Math. 669 (2012), 165 -- 224. 1 [5] P. Ara, M.A. Moreno, E. Pardo, Nonstable K-theory for graph algebras, Algebr. Represent. Theory 10 (2007), no. 2, 157 -- 178. 1 [6] G. Bergman, The diamond lemma for ring theory, Adv. in Math. 29 (1978), no. 2, 178 -- 218. 1, 4, 5, 8 [7] P.M. Cohn, Rings with a weak algorithm, Trans. Amer. Math. Soc. 109 (1963), no. 2, 332 -- 356. 17 [8] P.M. Cohn, Some remarks on the invariant basis property, Topology 5 (1966), 215 -- 228. 1, 14, 16, 17 [9] J. Cuntz, Simple C ∗-algebras generated by isometries, Comm. Math. Phys. 57 (1977), no. 2, 173 -- 185. 1 [10] R. Hazrat, The graded structure of Leavitt path algebras, Israel J. Math. 195 (2013), no. 2, 833 -- 895. 1, 2, 17 [11] R. Hazrat, Graded rings and graded Grothendieck groups, London Math. Society Lecture Note Series, Cambridge Uni- versity Press, 2016. 18 [12] W.G. Leavitt, Modules over rings of words, Proc. Amer. Math. Soc. 7 (1956), 188 -- 193. 1, 14 [13] W.G. Leavitt, Modules without invariant basis number, Proc. Amer. Math. Soc. 8 (1957), 322 -- 328. 1 [14] W.G. Leavitt, The module type of a ring, Trans. Amer. Math. Soc. 103 (1962) 113 -- 130. 1, 16, 17 [15] K. McClanahan, C ∗-algebras generated by elements of a unitary matrix, J. Funct. Anal. 107 (1992) 439 -- 457. 2 [16] K. McClanahan, K-theory and Ext-theory for unitary C ∗-algebras, Rocky Mountain J. Math. 23 (1993) 1063 -- 1080. 2 [17] M.H.A. Newman, On theories with a combinatorial definition of "equivalence", Ann. of Math. 43 (1942) 223 -- 243. 1 [18] N.C. Phillips, A classification theorem for nuclear purely infinite simple C ∗-algebras, Doc. Math. 5 (2000), 49 -- 114. 1 [19] I. Raeburn, Graph algebras. CBMS Regional Conference Series in Mathematics, 103 American Mathematical Society, Providence, RI, 2005. 1 Centre for Research in Mathematics, Western Sydney University, Australia E-mail address: [email protected] Department of Mathematics, University of Brasilia, Brazil E-mail address: [email protected]
1009.4152	7	1009	2012-05-22T20:18:55	Skew polynomial rings, Groebner bases and the letterplace embedding of the free associative algebra	[ "math.RA" ]	In this paper we introduce an algebra embedding $\iota:K< X >\to S$ from the free associative algebra $K< X >$ generated by a finite or countable set $X$ into the skew monoid ring $S = P * \Sigma$ defined by the commutative polynomial ring $P = K[X\times N^*]$ and by the monoid $\Sigma = < \sigma >$ generated by a suitable endomorphism $\sigma:P\to P$. If $P = K[X]$ is any ring of polynomials in a countable set of commuting variables, we present also a general Gr\"obner bases theory for graded two-sided ideals of the graded algebra $S = \bigoplus_i S_i$ with $S_i = P \sigma^i$ and $\sigma:P \to P$ an abstract endomorphism satisfying compatibility conditions with ordering and divisibility of the monomials of $P$. Moreover, using a suitable grading for the algebra $P$ compatible with the action of $\Sigma$, we obtain a bijective correspondence, preserving Gr\"obner bases, between graded $\Sigma$-invariant ideals of $P$ and a class of graded two-sided ideals of $S$. By means of the embedding $\iota$ this results in the unification, in the graded case, of the Gr\"obner bases theories for commutative and non-commutative polynomial rings. Finally, since the ring of ordinary difference polynomials $P = K[X\times N]$ fits the proposed theory one obtains that, with respect to a suitable grading, the Gr\"obner bases of finitely generated graded ordinary difference ideals can be computed also in the operators ring $S$ and in a finite number of steps up to some fixed degree.	math.RA	math	. SKEW POLYNOMIAL RINGS, GR OBNER BASES AND THE LETTERPLACE EMBEDDING OF THE FREE ASSOCIATIVE ALGEBRA ROBERTO LA SCALA∗ AND VIKTOR LEVANDOVSKYY∗∗ Abstract. In this paper we introduce an algebra embedding ι : KhXi → S from the free associative algebra KhXi generated by a finite or countable set X into the skew monoid ring S = P ∗Σ defined by the commutative polynomial ring P = K[X × N∗] and by the monoid Σ = hσi generated by a suitable endo- morphism σ : P → P . If P = K[X] is any ring of polynomials in a countable set of commuting variables, we present also a general Grobner bases theory for graded two-sided ideals of the graded algebra S = Li Si with Si = P σi and σ : P → P an abstract endomorphism satisfying compatibility conditions with ordering and divisibility of the monomials of P . Moreover, using a suitable grading for the algebra P compatible with the action of Σ, we obtain a bijective correspondence, preserving Grobner bases, between graded Σ-invariant ideals of P and a class of graded two-sided ideals of S. By means of the embedding ι this results in the unification, in the graded case, of the Grobner bases the- ories for commutative and non-commutative polynomial rings. Finally, since the ring of ordinary difference polynomials P = K[X × N] fits the proposed theory one obtains that, with respect to a suitable grading, the Grobner bases of finitely generated graded ordinary difference ideals can be computed also in the operators ring S and in a finite number of steps up to some fixed degree. 1. Introduction Let P be a K-algebra and let Σ be a monoid of endomorphisms of P . If I is an ideal of P which is invariant under the maps in Σ then it is possible to codify the action of P and Σ over I as a single left module structure with respect to the skew monoid (or semigroup) ring S = P ∗ Σ. The study of some properties of I, as for instance its finite Σ-generation, can be reduced hence to that of general properties of the operators ring S as its Noetherianity (see [26, 21]). Ideals which are stable under the action of monoids of endomorphisms or groups of automorphisms are natural in many contexts as the representation theory (a classical reference is [6]), or in the study of PI-algebras [8, 13] where P is the free associative algebra and Σ the complete monoid of endomorphisms of P . Another context of relevant interest is the study of so-called "difference ideals" [23] which are ideals invariant under shift operators in applications to combinatorics, (nonlinear) differential and difference equations. For the viewpoint of computing in the ring of (differential-difference) operators an important contribution is [25]. To control in an effective way the structure of the left S-module P/I one generally needs to compute a K-basis of it. If P is a ring of polynomials in commutive or 2000 Mathematics Subject Classification. Primary 16Z05. Secondary 13P10, 68W30. Key words and phrases. Skew polynomial rings; Free algebras; Grobner bases. Partially supported by Universit`a di Bari, Ministero dell'Universit`a e della Ricerca, and by the DFG Graduiertenkolleg "Hierarchie und Symmetrie in mathematischen Modellen" at RWTH Aachen, Germany. 1 2 R. LA SCALA AND V. LEVANDOVSKYY non-commutative variables and one fixes a suitable ordering for the monomials of P , then a K-linear basis of monomials for P/I can be obtained by using the elements of a suitable generating set of I as rewriting rules. Such generating set is usually called a "Grobner basis" of I. Since I is a Σ-invariant ideal, it is natural to consider Σ-bases of I that is sets G ⊂ I such that I is the smallest Σ-ideal of P containing G. In other words, G is a generating set of I as left S-module. It follows that one has to harmonize the notion of Grobner basis with that of Σ-basis and attempts in this direction can be found for instance in [1, 3] and also in [9, 22]. If the elements of Σ are automorphisms, the main obstacle in the definition of a Grobner Σ-basis is that their action on P does not preserve the monomial ordering. Then, it has been shown in [3] and before in [22] that an appropriate setting to define Grobner Σ-bases is that of a commutative polynomial ring P = K[X] in an infinite number of variables and a monoid Σ of monomial monomorphisms of infinite order which are compatible with the ordering and divisibility of monomials of P . In this paper we propose a systematic study of the case when Σ is generated by a single map σ. In this case the skew monoid ring S coincides with the skew polynomial ring P [s; σ] which is an Ore extension where σ-derivation is zero. The approach we follow is to consider an abstract map σ satisfying compatibility condi- tions able to provide a "natural" Grobner bases theory. Note that this generalizes in particular the results contained in [30] where the map σ : xi 7→ xe i with e > 1 has been studied. We choose to consider a single endomorphism essentially be- cause a major application of our theory is the unification, in the graded case, of the Grobner bases theory for non-commutative polynomials introduced in [15, 27, 29] with the classical commutative theory based on the notion of S-polynomial (see for instance [17]). In Section 6 we show in fact that there exists an algebra embedding ι : KhXi → S where KhXi is the free associative algebra generated by the variables xi and S is the skew polynomial ring defined by the algebra P of commutative poly- nomials in double indexed variables xi(j) and the endomorphism σ : P → P such that xi(j) 7→ xi(j + 1), for all i, j. This algebra embedding is a significant improve- ment of the linear map ι′ : KhXi → P defined as xi1 · · · xid 7→ xi1 (1) · · · xid (d) and introduced by [11, 7] for the aims of physics and invariant/representation theory. In fact, the use of the map ι clarifies the phenomenon found in [22] of a bijective correspondence between all graded two-sided ideals of KhXi and some class of Σ- invariant ideals of P . Note that in the same paper, a competitive new algorithm for non-commutative homogeneous Grobner bases based on this correspondence has been implemented and experimented in Singular [5]. In Section 2 one finds a brief account of the equivalence between the notion of Σ-invariant P -module and that of left S-module, together with the description of some properties of the generating sets of graded two-sided ideals of S = Li Si with Si = P si. A Grobner basis theory for such ideals is introduced in Section 3 where we assume P = K[x0, x1, . . .], Σ = hσi and σ : P → P be a monomorphism of infinite order sending monomials into monomials. Additional assuptions for σ are that gcd(σ(xi), σ(xj )) = 1 for i 6= j and the monomial ordering of P is such that m ≺ n implies that σ(m) ≺ σ(n), for all monomials m, n. Such conditions are quite natural in many contexts as the shift operators defining difference ideals [23] or the maps used in [3]. Note that the algorithms we introduce for the computation of homogeneous Grobner bases in S are based on the free P -module structure of this ring and hence they appear as a variant of the classical module Buchberger SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 3 algorithm where the number of S-polynomials to be considered is reduced owing to the symmetry defined by Σ on the graded ideals of the ring S. In Section 5 we analyze the notion of Grobner Σ-basis for Σ-invariant ideals of P . When P can be endowed with a suitable grading compatible with the action of Σ, we describe a bijective correspondence between all graded Σ-invariant ideals of P and some class of graded two-sided ideals of S. Such correspondence preserves Grobner bases and gives rise to a "duality" between homogeneous algorithms in P and in S. Note that for finitely generated ideals all these procedures admit termination when truncated at some degree. As we said earlier, in Section 6 the algebra embedding ι : KhXi → S is introduced and a bijective correspondence between the ideals of KhXi and suitable ideals of S is hence obtained by extension. The Grobner bases are preserved by this correspondence and one obtains an alternative algorithm for computing non-commutative homogeneous Grobner bases of KhXi in the free P - module S. By means of the bijection of the Section 5, we reobtain in Section 7 the ideal correspondence and related algorithms introduced in [22] which provide the computation of non-commutative homogeneous Grobner bases directly in P . Therefore, the theory for such bases can be deduced by the classical Buchberger algorithm for commutative polynomial rings. In Section 8 we propose the explicit computation of a finite Grobner basis of an ideal of ordinary difference polynomials that can be obtained as a special case by the algorithms we introduced. Moreover, in this section we provide some timings obtained by an improvement of the library freegb.lib of Singular initially developed for [22]. Finally, in Section 9 we propose some conclusions and suggestions for future developments of the theory of Grobner Σ-bases and its methods. The preliminary full-size version of this paper has been accepted for oral presen- tation at MEGA 2011 conference in Stockholm. 2. Modules over skew monoid rings Fix K any field and let P be a commutative K-algebra. Let now Σ ⊂ EndK(P ) a submonoid of the monoid of K-algebra endomorphisms of P . Denote S = P ∗ Σ the skew monoid ring defined by Σ over P that is S is the free P -module with (left) basis Σ and the multiplication is defined by the identity σf = σ(f )σ, for all f ∈ P, σ ∈ Σ. If Σ 6= {id} then S is a non-commutative K-algebra where the ring P and the monoid Σ are embedded. Note that if Σ = hσi with σ : P → P a map of infinite order one has that S ≈ P [s; σ], the skew polynomial ring in the variable s and coefficients in P defined by the endomorphism σ. Moreover, if P is a domain and all maps in Σ are injective then S is also a domain. To simplify notations, we denote f σ = σ(f ) for any f ∈ P, σ ∈ Σ. Definition 2.1. Let M be a P -module. We call M a Σ-invariant module if there is a monoid homomorphism ρ : Σ → EndK(M ) such that ρ(σ)(f x) = f σρ(σ)(x), for all f ∈ P, x ∈ M and σ ∈ Σ. We denote as usual σ · x = ρ(σ)(x). If M, M ′ are Σ-invariant modules and ϕ : M → M ′ is a P -module homomorphism such that ϕ(σ · x) = σ · ϕ(x) for all x ∈ M, σ ∈ Σ, then the map ϕ is called a homomorphism of Σ-invariant modules. Proposition 2.2. The category of Σ-invariant P -modules is equal to the category of left S-modules. 4 R. LA SCALA AND V. LEVANDOVSKYY Proof. Let M be a left S-module. Then M is a P -module since P ⊂ S. By restric- tion to Σ ⊂ S, one has a monoid homomorphism ρ : Σ → EndK(M ). Moreover we have σ · (f x) = (σf ) · x = (f σσ) · x = f σ(σ · x), for all f ∈ P, x ∈ M and σ ∈ Σ. Let now M be a Σ-invariant P -module. We can define a left S-module structure by putting (Pi fiσi) · x = Pi fi(σi · x) with fi ∈ P, σi ∈ Σ and x ∈ M . Consider a homomorphism ϕ : M → M ′ of Σ-invariant modules. Since ϕ is P -linear, one has ϕ((Pi fiσi) · x) = Pi fiϕ(σi · x) = Pi fi(σi · ϕ(x)) = (Pi fiσi) · ϕ(x). (cid:3) Let M be a left S-module and let G = {gi} ⊂ M be a generating set of M . Note that x ∈ M if and only if x = Pi,σ fiσσ · gi with fiσ ∈ P that is M is generated by Σ · G = {σ · gi}i,σ as P -module. We want now to describe homogeneous bases for graded two-sided ideals of S. In fact, the algebra S has a natural grading over the monoid Σ that is S = Lσ∈Σ Sσ and SσSτ ⊂ Sστ where Sσ = P σ. Note that Sid = P , all Sσ are P -submodules of S and Sστ = Sστ . Proposition 2.3. Let J ⊂ S be a graded (two-sided) ideal and let G ⊂ J be a set of homogeneous elements. Then G is a generating set of J if and only if G Σ is a left basis of J that is Σ G Σ is a basis of J as P -module. Proof. Assume G = {giσi} with gi ∈ P, σi ∈ Σ, for all i. Let pi, qi ∈ S with qi = Pσ qiσσ and qiσ ∈ P . It is sufficient to note that Pi pigiσiqi = Pi,σ pigiσiqiσσ = Pi,σ piqσi (cid:3) iσ giσiσ. Corollary 2.4. Let f, g ∈ S and let g be a homogeneous element. Then, one has that f = pgq with p, q ∈ S if and only if f belongs to the (graded) left ideal generated by {gσ}σ∈Σ. 3. Monomial orderings and Grobner bases Denote N = {0, 1, . . .} the set of non-negative integers and let X = {x0, x1, . . .} be a countable set. From now on, we make the assumption that P = K[X] is a commutative polynomial ring in the variables set X. Starting from Section 6 we will assume in particular that this set has the form X × N. Moreover, we fix σ : P → P an algebra homomorphism of infinite order and define the monoid Σ = hσi ≈ N. Then, the skew monoid ring S = P ∗ Σ is isomorphic to the skew polynomial ring P [s; σ] and we identify Σ = {σi} with the monoid {si} of powers of the variable s. Note that S is a free P -module of infinite rank. We denote f si = σi(f ), for all f ∈ P, i ≥ 0. Moreover, a homogeneous element f ∈ Si = P si for some i is also called s-homogeneous and we put degs(f ) = i. Note finally that in the theory of difference ideals [23], the ring S is called ring of ordinary difference operators over P . = f σi Denote by Mon(P ) the set of all monomials of P (including 1). Clearly, Mon(P ) is a multiplicative K-basis of P that is mn ∈ Mon(P ) for all m, n ∈ Mon(P ). By definition of S, a K-basis of such algebra is given by the elements msi where m ∈ Mon(P ) and i ≥ 0 is an integer. We call such elements the monomials of S and we denote the set of them as Mon(S). Clearly Mon(P ) ⊂ Mon(S). Note that Mon(S) is in fact the "monomial basis" of S as a free P -module. In what follows, we assume also that the endomorphism σ : P → P is injective and monomial that is it stabilizes the set Mon(P ). In other words, {σ(xi)} is a set of algebraically independent monomials. Since P is a domain, it follows that S is also SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 5 a domain and the K-basis Mon(S) is multiplicative since msinsj = mnsisi+j 6= 0, for all m, n ∈ Mon(P ) and i, j ≥ 0. We want to study now some divisibility relations in Mon(S). Let f, g ∈ S. We say that f left-divides g if there is a ∈ S such that g = af . Clearly, left divisibility is a partial ordering (up to units). Since σ is a monomial injective map, one has that if f, g ∈ Mon(S) then also a ∈ Mon(S). Proposition 3.1. Let v = msi, w = nsj ∈ Mon(S) with m, n ∈ Mon(P ). Then v left-divides w if and only if i ≤ j and msj−i Proof. Let a = psk ∈ Mon(S) with p ∈ Mon(P ) such that nsj = pskmsi = pmsk (cid:3) sk+i. Then, we have that j − i = k ≥ 0 and msk n. n. Note that S has also a free P -module structure and so Mon(S) inherits another notion of divisibility. Precisely, let v, w ∈ Mon(S). We say that v P -divides w if degs(v) = degs(w) and there is a ∈ Mon(P ) such that w = av. Clearly P - divisibility is a partial ordering and one has the following result. Proposition 3.2. Let v, w ∈ Mon(S). Then v left-divides w if and only if skv P -divides w for some k ≥ 0. Note that left divisibility coincides with P -divisibility when the monomials have the same s-degree. If there are v, w, a, b ∈ Mon(S) such that w = avb we say that v (two-sided) divides w. It is easy to prove that such divisibility is also a partial ordering. Proposition 3.3. Let v, w ∈ Mon(S). Then w is a multiple of v if and only if there is j ≥ 0 such that w is a left multiple of vsj, that is w is a P -multiple of sivsj for some i, j ≥ 0. Proof. Since monomials are s-homogeneous elements of S, by applying Corollary 2.4 we obtain that w is a multiple of v if and only if w belongs to the (graded) left ideal generated by {vsj}j≥0. Clearly, this happens when w is a left multiple of vsj for some j. (cid:3) We start now considering monomial orderings. Definition 3.4. Let ≺ be a total ordering on the set Mon(S). We call ≺ a mono- mial ordering of S if it satisfies the following conditions (i) ≺ is a well-ordering, that is every non-empty set of Mon(S) has a minimal element; (ii) ≺ is compatible with multiplication, that is if v ≺ w then pvq ≺ pwq, for all v, w, p, q ∈ Mon(S). It follows immediately that 1 (cid:22) w for any w ∈ Mon(S) and if w = pvq with p 6= 1 or q 6= 1 then v ≺ w for all v, w, p, q ∈ Mon(S). Note that the above conditions agree with general definitions of orderings on K-bases of associative algebras that provide a Grobner basis theory (see for instance [16, 24]). The same conditions define monomial orderings of the free algebras KhXi and K[X]. Note that such algebras can be endowed with a monomial ordering even if the set of variables X is countable. This is provided by the Higman's lemma [19] which implies that any multiplicatively compatible total ordering of the monomials such that 1 ≺ x0 ≺ x1 ≺ . . . is a monomial ordering. Recall that f s stands for σ(f ) for any f ∈ P . 6 R. LA SCALA AND V. LEVANDOVSKYY Definition 3.5. Let ≺ be a monomial ordering on P . We call σ compatible with ≺ if σ is a strictly increasing map when restricted to Mon(P ), that is m ≺ n implies that ms ≺ ns for all m, n ∈ Mon(P ). The following result is based essentially on Remark 3.2 in [3]. Proposition 3.6. Assume σ be compatible with ≺. Then σ is not an automorphism and m (cid:22) ms, for all m ∈ Mon(P ). If m ≻ ms, by Proof. Since σ 6= id, there is m ∈ Mon(S) such that m 6= ms. compatibility of σ one gets an infinite descending chain m ≻ ms ≻ ms2 ≻ . . . which contradicts the condition that ≺ is a well-ordering. We conclude that m ≺ ms. Assume that σ has the inverse σ−1. By applying σ, from ms−1 it follows that m ≺ n. Since σ−1 is injective, we have therefore that m ≺ n implies that ms−1 which contradicts m ≺ ms. (cid:3) . Now, by compatibility of σ−1 we obtain m ≺ ms−1 ≺ ns−1 ≺ ns−1 There are many endomorphisms σ with are compatible with usual monomial orderings on P like lex, degrevlex, etc. For instance, we have the following maps. • σ(xi) = xf (i) for any i, where f : N → N is a strictly increasing map. Such maps have been considered in [3]. In particular, one may define the shift operator σ(xi) = xi+1 which is used in difference algebra. • σ(xi) = xe i for any i, with e > 1. This map has been considered in [30]. Proposition 3.7. Let ≺ be a monomial ordering on S. Then σ is compatible with the restriction of ≺ to Mon(P ). Moreover, for any m, n ∈ Mon(P ) and i, j ≥ 0 one has that msi ≺ nsj implies that m ≺ n or i < j. Proof. Suppose m ≺ n with m, n ∈ Mon(P ). Then sm ≺ sn that is mss ≺ nss. If ms (cid:23) ns then mss (cid:23) nss which is a contradiction. We conclude that ms ≺ ns. Now, assume that m (cid:23) n and i ≥ j. We have msi (cid:23) msj (cid:23) nsj. (cid:3) Assume now σ be compatible with a monomial ordering ≺ of P . We define a total ordering on Mon(S) by putting msi ≺′ nsj if and only if i < j, or i = j and m ≺ n, for all m, n ∈ Mon(P ) and i, j ≥ 0. Clearly, the restriction of ≺′ to Mon(P ) is ≺. Proposition 3.8. The ordering ≺′ is a monomial ordering on S that extends ≺. Proof. Clearly, an infinite descending sequence in Mon(S) implies an infinite de- scending sequence in Mon(P ) which contradicts the condition that ≺ is a well- ordering. Let msi, nsj ∈ Mon(S) and suppose msi ≺ nsj that is i < j, or i = j and m ≺ n. Let qsk ∈ Mon(S) and consider right multiplications msiqsk = mqsi si+k and nsjqsk = nqsj sj+k. If i < j then i + k < j + k. If i = j and m ≺ n then si+k ≺ nqsj . We conclude in both cases that mqsi mqsi = nqsj sj+k. For left multiplications qskmsi = qmsk sk+i and qsknsj = qnsk sk+j, note that m ≺ n implies that msk . Then, one may argue in a similar way as for right multi- plications. (cid:3) ≺ nqsi ≺ nsk Clearly, a byproduct of Proposition 3.7 and Proposition 3.8 is that there exist monomial orderings on the skew polynomial ring S if and only if σ is compatible with a monomial ordering of P . Note that ≺′ is well-known as module ordering when we consider S as a free P -module. Moreover, by Proposition 3.7 it follows SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 7 also that the monomial ordering of S is uniquely defined by the one of P when one compares monomials of the same s-degree. From now on, we assume S be endowed with a monomial ordering ≺. Definition 3.9. Let f ∈ S, f = Pi αimisi with mi ∈ Mon(P ), αi ∈ K ∗. Then, we denote lm(f ) = mksk = max≺{misi}, lc(f ) = αk and lt(f ) = lc(f )lm(f ). If G ⊂ S we put lm(G) = {lm(f ) f ∈ G, f 6= 0}. We denote as LM(G) and LMl(G) respectively the two-sided ideal and the left ideal of S generated by lm(G). Moreover, we denote by LMP (G) the P -submodule of S generated by lm(G). Proposition 3.10. Let J be an ideal (respectively left ideal) of S. Then, the set {w + J w ∈ Mon(S) \ LM(J)} (resp. {w + J w ∈ Mon(S) \ LMl(J)}) is a K-basis of the space S/J. If J ⊂ S is a P -submodule, in the same way one defines the K-basis {w + J w ∈ Mon(S) \ LMP (J)}. Proof. Let w ∈ Mon(S). By induction on the monomial ordering of S, we can assume that for any monomial v ∈ Mon(S) such that v ≺ w there is a polynomial f ∈ S belonging to the span of N = Mon(S) \ LM(J) such that v − f ∈ J. If w /∈ N then there is g ∈ J such that w = plm(g)q with p, q ∈ Mon(S). Therefore f = w − (1/lc(g))pgq is such that lm(f ) ≺ w and by induction f − f ′ ∈ J for some f ′ ∈ hN iK. We conclude that w − f ′ ∈ J. Finally if f ∈ N ∩ J then necessarily f = 0. Mutatis mutandis one proves the remaining assertions. (cid:3) Definition 3.11. Let J be an ideal (respectively left ideal) of S and G ⊂ J. We call G a Grobner basis (resp. left basis) of J if LM(G) = LM(J) (resp. LMl(G) = LMl(J)). As usual, if J is a P -submodule of S then G is a Grobner P -basis of J when LMP (G) = LMP (J). Proposition 3.12. Let J be an ideal (respectively left ideal) of S and G ⊂ J. The following conditions are equivalent: (i) G is a Grobner basis (resp. left basis) of J; (ii) for any f ∈ J, one has a Grobner representation of f with respect to G that is f = Pi figihi (resp. f = Pi figi) with lm(f ) (cid:23) lm(fi)lm(gi)lm(hi) (resp. lm(f ) (cid:23) lm(fi)lm(gi)) and fi, hi ∈ S, for all i. A similar characterization holds for Grobner P -bases. Proof. It follows immediately by the reduction process which is implicit in the proof of Proposition 3.10. (cid:3) Proposition 3.13. Let J be a graded ideal of S and G ⊂ J be a subset of s- homogeneous elements. The following conditions are equivalent: (i) G is a Grobner basis of J; (ii) G Σ is a Grobner left basis of J; (iii) Σ G Σ is a Grobner P -basis of J. Proof. Assume G = {gi} is a Grobner basis of J and put di = degs(gi). If f ∈ J then one has f = Pi figihi where fi, hi ∈ S and lm(f ) (cid:23) lm(fi)lm(gi)lm(hi), for all i. Decompose hi = Pj hij sj with hij ∈ P for any i, j. Then, we have lm(f ) (cid:23) lm(fi)lm(gi)lm(hij )sj, for all i, j. Since lm(gi) has s-degree di, one obtains lm(fi)lm(gi)lm(hij )sj = lm(fi)lm(hij)sdi lm(gisj). Moreover, as in Proposition 2.3, we have f = Pi,j figihij sj = Pi,j fihsdi ij gisj. From σ compatible with ≺ it follows 8 R. LA SCALA AND V. LEVANDOVSKYY that lm(hsdi ij ) = lm(hij)sdi and hence f has a left Grobner representation with respect to G Σ, that is this set is a left Grobner basis of J. The rest of the proof is straightforward. (cid:3) 4. Buchberger algorithm After Proposition 3.13, in order to obtain a homogeneous Grobner basis G of a (two-sided) graded ideal J ⊂ S one has to start with a homogeneous generating set H and consider the P -basis H ′ = Σ H Σ. Then, one should transform H ′ into a homogeneous Grobner P -basis G′ of J and finally reduce G′ as G′ = Σ G Σ with G ⊂ J. Apart with problems concerning termination of the module Buchberger algorithm (P is not Noetherian and S is a P -module of countable rank) that we will show solvable for the truncated algorithm up to some s-degree (see Proposition 4.7), it is more desirable to have a procedure able to compute G without actually considering all elements of G′. To obtain this, we need an additional requirement for the endomorphism σ. Note that, since σ : P → P is a ring homomorphism, such map is increasing with respect to the divisibility relation in P , that is f g implies that f s gs and in this case (g/f )s = gs/f s with f, g ∈ P . Proposition 4.1. The following conditions are equivalent: (a) gcd(xs (b) gcd(ms, ns) = gcd(m, n)s, for all m, n ∈ Mon(P ). j ) = 1, for all i 6= j; i , xs Moreover, in this case one has m n if and only if ms ns and lcm(ms, ns) = lcm(m, n)s with m, n ∈ P . In other words, σ is a lattice homomorphism with respect to the divisibility relation in Mon(P ). Proof. Assume (a) and let m, n ∈ Mon(P ) such that gcd(m, n) = 1. If m = i1 · · · xs xi1 · · · xik and n = xj1 · · · xjl then ms = xs with ik i , xs {i1, . . . , ik} ∩ {j1, . . . , jl} = ∅. Since gcd(xs j ) = 1 for all i 6= j, we conclude that gcd(ms, ns) = 1. Assume now gcd(m, n) = u and hence gcd(m/u, n/u) = 1. Then gcd(ms/us, ns/us) = gcd((m/u)s, (n/u)s) = 1 and therefore gcd(ms, ns) = us that is (b) holds. Suppose ms ns that is ms = gcd(ms, ns) = gcd(m, n)s. Since σ is injective we have that m = gcd(m, n) that is m n. Moreover, one ob- tains lcm(m, n)s = (mn/ gcd(m, n))s = (mn)s/ gcd(m, n)s = msns/ gcd(ms, ns) = lcm(ms, ns) for all m, n ∈ Mon(P ). (cid:3) and ns = xs j1 · · · xs jl Definition 4.2. We say that σ is compatible with divisibility in Mon(P ) if for all i , xs i 6= j, one has gcd(xs j are disjoint. j) = 1 that is the variables occuring in the monomials xs i , xs Note that if a monomial endomorphism of P is compatible with divisibility then it is automatically injective since the monomials xs i are algebraically independent. Let be the divisibility relation and ≺ a monomial ordering on Mon(P ). Throughout the rest of the paper, we make the assumption that the monomial endomorphism σ : P → P is compatible both with and with ≺. We recall now some basic results in the theory of module Grobner bases by applying them to the free P -module S whose (left) free basis is Σ = {si}i≥0. Consider f, g ∈ S \ {0} two elements whose leading monomials have the same s-degree (component), that is lm(f ) = msi, lm(g) = nsi with m, n ∈ Mon(P ) and i ≥ 0. If we put lc(f ) = α, lc(g) = β and l = lcm(m, n), one defines the SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 9 S-polynomial spoly(f, g) = (l/αm)f − (l/βn)g. Clearly spoly(f, g) = −spoly(g, f ) and spoly(f, f ) = 0. Proposition 4.3 (Buchberger criterion). Let G be a generating set of a P -submo- dule J ⊂ S. Then G is a Grobner basis of J if and only if for all f, g ∈ G \ {0} such that degs(lm(f )) = degs(lm(g)), the S-polynomial spoly(f, g) has a Grobner representation with respect to G. Usually the above result, see for instance [10, 17], is stated when P is a polyno- mial ring with a finite number of variables and S is a P -module of finite rank. In fact such assumptions are not needed since Noetherianity is not used in the proof, but only the existence of a monomial ordering for the ring P and the free module S. See also the comprehensive Bergman's paper [2] where the "Diamond Lemma" is proved without any restriction on the finiteness of the variable set. In the fol- lowing results we show how the Buchberger criterion, and hence the corresponding algorithm, can be reduced up to the symmetry defined by the monoid Σ on S. Lemma 4.4. Let f, g ∈ S \ {0} and let i ≤ j such that degs(lm(f )) + i = degs(lm(g)) + j. Then spoly(sif, sjg) = sispoly(f, sj−ig) and spoly(f si, gsj) = spoly(f, gsj−i)si. si+k, lt(sjg) = βnsj Proof. Let lt(f ) = αmsk, lt(g) = βnsl with α, β ∈ K ∗ and m, n ∈ Mon(P ). Then lt(sif ) = αmsi sj−i+l. By compati- bility of σ with divisibility in Mon(P ), if q = lcm(m, nsj−i ) = qsi ) then lcm(msi . Therefore h = spoly(f, sj−ig) = (q/αm)f − (q/βnsj−i )sj−ig and hence we have sih = (qsi )sjg = spoly(sif, sjg). sj+l and lt(sj−ig) = βnsj−i , nsj /αmsi )sif − (qsi /βnsj Note now that lt(f si) = αmsi+k, lt(gsj) = βnsj+l and lt(gsj−i) = βnsj−i+l. If q = lcm(m, n) and h = spoly(f, gsj−i) = (q/αm)f − (q/βn)gsj−i we have simply that hsi = (q/αm)f si − (q/βn)gsj = spoly(f si, gsj). (cid:3) Proposition 4.5 (Two-sided Σ-criterion). Let G be an s-homogeneous basis of a graded two-sided ideal J ⊂ S. Then G is a Grobner basis of J if and only if for all f, g ∈ G \ {0} and for any i, j ≥ 0, the S-polynomials spoly(f, sigsj) (degs(f ) = degs(g) + i + j) and spoly(f si, sjg) (degs(f ) + i = degs(g) + j) have a Grobner representation with respect to Σ G Σ. Proof. By Proposition 3.13 we have to prove that G′ = Σ G Σ is a Grobner ba- sis of J as P -module, that is G′ is P -basis of J and the S-polynomials h = spoly(sif sk, sjgsl) have a Grobner representation with respect to G′ for all f, g ∈ G \ {0} and for any i, j, k, l ≥ 0 such that degs(f ) + i + k = degs(g) + j + l. Since G is a homogeneous basis of J as two-sided ideal, from Proposition 2.3 it follows that G′ is a generating set of J ′ as P -module. Consider now all possibilities i ≤ j or i ≥ j and k ≤ l or k ≥ l and apply Lemma 4.4. If i ≤ j, k ≤ l one has h = sispoly(f, sj−igsl−k)sk, if i ≤ j, k ≥ l then h = sispoly(f sl−k, sj−ig)sl, and so on. Then, assume that a S-polynomial h = spoly(f, g), with f, g ∈ G′ \ {0}, has a Grobner representation with respect to G′ as P -basis of J, that is h = Pi figi with fi ∈ P, gi ∈ G′ and lm(h) ≥ lm(fi)lm(gi), for all i. We have to prove that skhsl has also a Grobner representation with respect to G′ for any k, l ≥ 0. One has that skhsl = Pi f sk i skgisl and lm(skhsl) = sklm(h)sl ≥ sklm(fi)lm(gi)sl = lm(fi)sk i )lm(skgisl). Since skgisl ∈ G′ = Σ G Σ, one obtains that skhsl has a Grobner representation with respect to G′. (cid:3) sklm(gi)sl = lm(f sk 10 R. LA SCALA AND V. LEVANDOVSKYY A criterion similar to Proposition 4.5 holds clearly for Grobner left bases of left ideals of S where no restrictions about the s-homogeneity of bases and ideals are needed. Proposition 4.6 (Left Σ-criterion). Let G be a basis of a left ideal J ⊂ S. Then G is a Grobner basis of J if and only if for all elements f, g ∈ G \ {0} such that i = degs(lm(f )) − degs(lm(g)) ≥ 0, the S-polynomial spoly(f, sig) has a Grobner representation with respect to Σ G. A standard procedure in the (module) Buchberger algorithm is the following. Algorithm 4.1 Reduce Input: f ∈ S and G ⊂ S. Output: h ∈ S such that f − h ∈ hGiP and h = 0 or lm(h) /∈ LMP (G). h := f ; while h 6= 0 and lm(h) ∈ LMP (G) do choose g ∈ G, g 6= 0 such that lm(g) P -divides lm(h); h := h − (lt(h)/lt(g))g; end while; return h. Note that if lt(g) = αmsi, lt(h) = βnsi with α, β ∈ K ∗ and m, n ∈ Mon(P ), by definition lt(h)/lt(g) = (αm)/(βn). Moreover, the termination of Reduce is provided since ≺ is a well-ordering on Mon(S). In particular, even if G is an infinite set, there are only a finite number of elements g ∈ G, g 6= 0 such that lm(g) P -divides lm(h) and hence lm(g) (cid:22) lm(h). It is well-known that if Reduce(f, G) = 0 then f has a Grobner representation with respect to G. Moreover, if Reduce(f, G) = h 6= 0 then clearly we have Reduce(f, G ∪ {h}) = 0. Therefore, from Proposition 4.5 it follows immediately the correctness of the following algorithm. SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 11 Algorithm 4.2 SkewGBasis Input: H, an s-homogeneous basis of a graded two-sided ideal J ⊂ S. Output: G, an s-homogeneous Grobner basis of J. G := H; B := {(f, g) f, g ∈ G}; while B 6= ∅ do choose (f, g) ∈ B; B := B \ {(f, g)}; for all i, j ≥ 0 such that i + j = degs(f ) − degs(g) do h := Reduce(spoly(f, sigsj), Σ G Σ); if h 6= 0 then B := B ∪ {(h, h), (h, k), (k, h) k ∈ G}; G := G ∪ {h}; end if ; end for; for all i, j ≥ 0 such that j − i = degs(f ) − degs(g) do h := Reduce(spoly(f si, sjg), Σ G Σ); if h 6= 0 then B := B ∪ {(h, h), (h, k), (k, h) k ∈ G}; G := G ∪ {h}; end if ; end for; end while; return G. Clearly, all well-known criteria (product criterion, chain criterion, etc) can be applied to SkewGBasis to shorten the number of S-polynomials to be considered. In fact, this algorithm can be understood as the usual (module) Buchberger proce- dure applied to the P -basis Σ H Σ, where an additional criterion to avoid "useless pairs" is provided by Proposition 4.5. Note that owing to Proposition 4.6, one has also a similar procedure for computing a Grobner left basis of any left ideal of S. Since the set Σ H Σ if infinite even if the basis H is eventually finite (S is a non-Noetherian ring) one has that SkewGBasis does not admit general termina- tion. In particular, the cycle "for all i, j ≥ 0 such that j − i = degs(f ) − degs(g) do" never stops unless one bounds the s-degree degs(f ) + i = degs(g) + j. As for other non-Noetherian structures like the free associative algebra that in fact can be embedded in S (see Section 6), the termination of homogeneous Grobner bases computations can be obtained only under truncation. Proposition 4.7. Let J ⊂ S be a graded two-sided ideal and fix d ≥ 0. Assume that J has a s-homogeneous basis H such that Hd = {f ∈ H degs(f ) ≤ d} is a finite set. Then, there exists an s-homogeneous Grobner basis G of J such that Gd is also finite. In other words, if we consider a selection strategy for the S- polynomials based on their s-degree, we obtain that the d-truncated version of the algorithm SkewGBasis terminates in a finite number of steps. Proof. Denote H ′ one has that H ′ occurring in the elements of H ′ d = {sif sj f ∈ Hd, i, j ≥ 0, i+j +degs(f ) ≤ d}. Since Hd is finite d is also finite. Then, consider Xd the finite set of variables of P d and define P (d) = K[Xd] and S(d) = Li≤d P (d)si. 12 R. LA SCALA AND V. LEVANDOVSKYY In fact, the d-truncated algorithm SkewGBasis computes a subset of a Grobner basis of the P (d)-submodule J (d) ⊂ S(d) generated by H ′ d. By Noetherianity of the ring P (d) and the free P (d)-module S(d) which has finite rank, we clearly obtain termination. (cid:3) Note that the above result implies algorithmic solution of the membership prob- lem for graded ideals of S which are finitely generated up to any degree. 5. Σ-invariant ideals of P In this section we define Grobner bases of Σ-invariant ideals I ⊂ P which gener- ates I up to the action of Σ. Moreover, if P can be endowed with a suitable grading, we show how such bases can be computed in the algebra S for a class of graded Σ-invariant ideals. As usual, we fix a monomial endomorphism σ : P → P which is compatible both with the divisibility and a monomial ordering on Mon(P ) and we extend this to an ordering on Mon(S). From Section 2 we know that Σ-invariant ideals of P are just left S-submodules of P . Since we make use of identification Σ = {si}, for all f ∈ P ⊂ S and for any i ≥ 0 one has that si · f = f si = σi(f ) and sif = (si · f )si. Definition 5.1. Let I ⊂ P be a Σ-invariant ideal and G ⊂ I. We say that G is a Σ-basis of I if G is a basis of I as left S-module. In other words, Σ · G is a basis of I as P -ideal. Proposition 5.2. Let G ⊂ P . Then lm(Σ · G) = Σ · lm(G). In particular, if I is a Σ-invariant P -ideal then LMP (I) is also Σ-invariant. Proof. Since σ is compatible with the monomial ordering of P , it is sufficient to note that lm(si · f ) = si · lm(f ) for any f ∈ P and i ≥ 0. (cid:3) Definition 5.3. Let I ⊂ P be a Σ-invariant ideal and G ⊂ I. We call G a Grobner Σ-basis of I if lm(G) is a basis of LMP (I) as left S-module. In other words, Σ · G is a Grobner basis of I as P -ideal. The computation of Grobner Σ-bases of Σ-invariant P -ideals is relevant, for in- stance, in applications to difference algebra (cf. [23], Chapter 3). Such computations appear also in other contexts, see for instance [9] and [3]. Note that in the latter paper Grobner Σ-bases are named "equivariant Grobner bases". In analogy with Proposition 4.5 and Proposition 4.6, we present here a Σ- criterion that allows to reduce the number of S-polynomials to be checked to provide that a Σ-basis is of Grobner type. Proposition 5.4 (Σ-criterion in P ). Let G be a Σ-basis of a Σ-invariant ideal I ⊂ P . Then G is a Grobner Σ-basis of I if and only if for all f, g ∈ G \ {0} and for any i ≥ 0, the S-polynomial spoly(f, si · g) has a Grobner representation with respect to Σ · G. Proof. Consider any pair of elements si · f, sj · g ∈ Σ · G (f, g ∈ G) and let i ≤ j. By compatibility of σ with divisibility in Mon(P ) (cf. Lemma 4.4), one has that spoly(si · f, sj · g) = si · spoly(f, sk · g) with k = j − i. Assume that spoly(f, sk · g) = h = Pl fl(sl · gl) (fl ∈ P, gl ∈ G) is a Grobner representation with respect to Σ · G. Since the endomorphism σ is compatible with the monomial ordering of P , we have also the Grobner representation spoly(si · f, sj · g) = si · h = Pl(si · fl)(si+l · gl). (cid:3) SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 13 Note that some version of this criterion can be found in [3], Theorem 2.5, where it is called "equivariant Buchberger criterion". Before than this, the same ideas have been used in [22] for the Proposition 3.11. From this criterion it follows immediately the correctness of the following algorithm. Algorithm 5.1 SigmaGBasis Input: H, a Σ-basis of a Σ-invariant ideal I ⊂ P . Output: G, a Grobner Σ-basis of I. G := H; B := {(f, g) f, g ∈ G}; while B 6= ∅ do choose (f, g) ∈ B; B := B \ {(f, g)}; for all i ≥ 0 do h := Reduce(spoly(f, si · g), Σ · G); if h 6= 0 then B := B ∪ {(h, h), (h, k), (k, h) k ∈ G}; G := G ∪ {h}; end if ; end for; end while; return G. 14 R. LA SCALA AND V. LEVANDOVSKYY As for the algorithm SkewGBasis, all criteria to avoid useless pairs can be added to SigmaGBasis. Note that termination of this algorithm is not provided in general (note the infinite cycle "for all i ≥ 0 do") and this is, in fact, one of the main problems in applications to differential/difference algebra. Nevertheless, in what follows we describe some class of Σ-invariant ideals of P where a truncated version of the algorithm SigmaGBasis stops in a finite number of steps. Such ideals are in bijective correspondence with a class of graded (two-sided) ideals of S which have truncated termination of SkewGBasis provided by Proposition 4.7. Consider now the P -module homomorphism π : S → P such that si 7→ 1, for all i. Clearly π is a left S-module epimorphism whose kernel is the left ideal of S generated by s − 1. Definition 5.5. Let J be a graded ideal of S and put J P = π(J). Clearly J P is a Σ-invariant ideal of P . Proposition 5.6. Let J ⊂ S be a graded ideal. If G is a homogeneous basis of J then GP = π(G) is a Σ-basis of J P . Proof. Since the map π is a left S-module homomorphism, it is sufficient to note that G Σ is a left basis of J and π(G Σ) = π(G) = GP . (cid:3) We introduce now a grading on the algebra P which is compatible with action of Σ. We start extending the structure (N, max, +) in the following way. Definition 5.7. Let −∞ be an element disjoint by N and put N = {−∞} ∪ N. Then, we define a commutative idempotent monoid ( N, max) with identity −∞ that extends (N, max) (with identity 0) by imposing that max(−∞, x) = x for any x ∈ N. Moreover, we define a commutative monoid ( N, +) with identity −∞ extending the monoid (N, +) by putting −∞ + x = −∞, for all x ∈ N. Since + clearly distributes over max, one has that ( N, max, +) is a commutative idempotent semiring, also known as commutative dioid or max-plus algebra [14]. Note that if σ−∞ ∈ EndK(P ) is the map such that xi 7→ 0 for any xi ∈ X, then Σ = {σ−∞} ∪ Σ is a commutative monoid isomorphic to ( N, +). Denote now M = Mon(P ) the set of monomials of the polynomial ring P . Definition 5.8. A mapping w : M → N such that for all m, n ∈ M and xi ∈ X one has (i) w(1) = −∞; (ii) w(mn) = max(w(m), w(n)); (iii) w(s · xi) = 1 + w(xi). is called a weight function of P endowed with σ. Note that if m = xi1 · · · xid 6= 1 with w(xi1 ) ≤ . . . ≤ w(xid ) then w(m) = w(xid ). Moreover, the condition (iii) implies that w(si · m) = i + w(m) for all i ∈ N, m ∈ M and hence si · m = m if and only if m = 1 or i = 0. We put Mi = {m ∈ M w(m) = i} for all i ∈ N and define Pi ⊂ P the subspace spanned by Mi. We have that P−∞ = K. Clearly P = Li∈N Pi is a grading of the algebra P defined by the monoid ( N, max) by means of the function w. Then, an element f ∈ Pi is said w-homogeneous of weight i. In what follows, we assume that P is endowed with a weight function. In fact, such functions are easily to define. Consider for instance the polynomial ring P = SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 15 K[X × N] and denote xi(j) each variable (xi, j) ∈ X × N. Let σ : P → P be the algebra monomorphism of infinite order such that σ(xi(j)) = xi(j + 1), for all i, j. Clearly σ is a monomial map compatible with divisibility in Mon(P ) and many usual monomial orderings on P , like lex, degrevlex, etc, are compatible with σ. For the algebra P endowed with the map σ we can clearly define the weight function w(xi(j)) = j. In Section 6 we show how to embed the free associative algebra KhXi into the skew polynomial ring defined by P and the monoid Σ = hσi. Moreover, if we put xi(j) = σj (ui) where xi(0) = ui = ui(t) is a set of (algebraically independent) univariate functions and σ is the shift operator ui(t) 7→ ui(t + h) then P = K[X × N] is by definition the ring of ordinary difference polynomials with constant coefficients in the field K (see [23]). Such algebra is used to study systems of (ordinary) difference equations for applications in combinatorics or discretization of systems of differential equations. Definition 5.9. Let I be an ideal of P . We call I w-graded if I = Pi Ii with Ii = I ∩ Pi for any i ∈ N. Define now the skew monoid ring S = P ∗ Σ extending S = P ∗ Σ and let π : S → P the left S-module epimorphism extending π that is si 7→ 1, for all i ∈ N. The existence of a weight function for P implies that one has also a mapping ξ : P → S such that πξ = id. Proposition 5.10. Define ξ : P → S the homogeneous injective map such that f 7→ Pi∈N fisi, for all f = Pi∈N fi ∈ P . Then ξ is a Σ-equivariant map. Proof. For all i, j ∈ N and f ∈ Pj one has that si · f ∈ Pi+j and therefore ξ(si · f ) = (si · f )si+j = sif sj = siξ(f ). (cid:3) Let I ⊂ P be a w-graded Σ-invariant ideal and consider ξ(I) ⊂ S. Note that if I 6= P then I−∞ = 0 and the set ξ(I) is in fact contained in S. Then, to get rid of the symbol −∞ we restrict ourselves to ideals not containing constants. Definition 5.11. Let I ( P be a w-graded Σ-invariant ideal of P . Denote by I S the graded (two-sided) ideal of S generated by ξ(I) ⊂ S. In other words, if we put G = ξ(Si≥0 Ii) = {f si f ∈ Ii, i ≥ 0} then I S is the left ideal generated by G Σ = {f sj f ∈ Ii, j ≥ i ≥ 0} or equivalently I S has the basis G Σ = Σ G Σ as P -submodule of S. We call I S the skew analogue of I. Proposition 5.12. Let I ( P be a w-graded Σ-invariant ideal. Then I SP = I, that is there is a bijective correspondence between all w-graded Σ-invariant ideals different from P and their skew analogues in S. Proof. Put J = I SP = π(I S). For any f ∈ Ii and j ≥ i we have clearly π(f sj) = f . Since the elements f sj are a left basis of I S, the ideal I is Σ-invariant and π is a left S-module homomorphism, we have that J ⊂ I. Moreover, the elements f ∈ Ii are a basis of I = Pi Ii and one has also that I ⊂ J. (cid:3) The next propositions need the following lemmas. Lemma 5.13. If sk · m divides n, with m, n ∈ M , then w(n) − k ≥ w(m). Proof. Since n = q(sk · m) with q ∈ M , we have w(n) ≥ w(sk · m) = k + w(m). (cid:3) Lemma 5.14. Let m, n ∈ M and put l = lcm(m, n). Then, one has that w(l) = max(w(m), w(n)). 16 R. LA SCALA AND V. LEVANDOVSKYY Proof. By property (ii) of Definition 5.8, it is sufficient to note that max is an idempotent operation and hence the weight of a monomial depends only on the variables occurring in the support. (cid:3) Proposition 5.15. Let I ( P be a w-graded Σ-invariant ideal, then I S is a graded ideal of S. Let G = Si Gi be a w-homogeneous Σ-basis of I that is Gi ⊂ Ii. Then GS = ξ(G) = {f si f ∈ Gi, i ≥ 0} is an s-homogeneous basis of I S. Proof. Consider the elements f sj with f ∈ Ii, j ≥ i which form a left basis of I S. Since G is a Σ-basis, one has f = Pk fk(sk · gk) with fk ∈ P, gk ∈ Gik . From w(f ) = i, by Lemma 5.13 we obtain that i − k ≥ ik. We have therefore that f sj = Pk fk(sk · gk)sksj−k = Pk fksk(gksj−k) with j − k ≥ i − k ≥ ik and hence gksj−k ∈ GS Σ, for all k. (cid:3) Note now that by Proposition 3.7 we have that msi ≺ nsi if and only if m ≺ n, for all m, n ∈ M and for any i ≥ 0. In other words, if f si (f ∈ P ) is an s-homogeneous element of S then lm(f si) = lm(f )si. If G = Si Ii, by Lemma 5.16. Let I ( P be a w-graded Σ-invariant ideal. definition I S is the graded ideal of S generated by GS = ξ(G). Then GS is an s-homogeneous Grobner basis of I S. Proof. Let f si, gsj ∈ GS Σ that is the w-homogeneous elements f, g ∈ G are such that i ≥ w(f ), j ≥ w(g). Assume i ≥ j and put k = i − j. By Proposition 4.6 we have to check for Grobner representations of the S-polynomial spoly(f si, skgsj) with respect to Σ GS Σ. Since G is clearly a Grobner Σ-basis of I, one has that the S-polynomial spoly(f, sk · g) has a Grobner representation with respect to Σ · G, say h = spoly(f, sk ·g) = Pl fl(sl ·gl) with fl ∈ P, gl ∈ G. Note that spoly(f si, skgsj) = hsi = Pl fl(sl · gl)si. We have to prove now that i ≥ l + w(gl) for any l, because in this case one has the Grobner representation hsi = Pl flsl(glsi−l). In fact, by Lemma 5.13 and Lemma 5.14 we have that max(w(f ), w(g)) = w(h) ≥ l + w(gl). Then, from i ≥ w(f ) and i ≥ j ≥ w(g) one obtains the claim. (cid:3) Proposition 5.17. Let G ⊂ Si≥0 Pi. Then lm(G)S = lm(GS). Moreover, if I ( P is a w-graded Σ-invariant ideal then LMP (I)S = LM(I S). Proof. If f ∈ Pi is a w-homogeneous element then w(lm(f )) = w(f ) = i and lm(f )si = lm(f si). We obtain that lm(G)S = lm(GS). Consider now G = Si Ii. By definition I S is the ideal of S generated by GS. Moreover, since I = Pi Ii one has that lm(G) = lm(I) and hence LMP (I)S is the ideal generated by lm(G)S = lm(GS). Finally, by Lemma 5.16 one has that LM(I S) is the ideal of S generated by lm(GS). (cid:3) Proposition 5.18. Let I ( P be a w-graded Σ-invariant ideal. Let G = Si Gi be a w-homogeneous Grobner Σ-basis of I. Then, GS = ξ(G) is an s-homogeneous Grobner basis of I S. Proof. By hypothesis lm(G) is a Σ-basis of LMP (I). Then lm(GS) = lm(G)S is a basis of LMP (I)S = LM(I S) that is GS is a Grobner basis of I S. (cid:3) Proposition 5.19. Let I ( P be a w-graded Σ-invariant ideal. homogeneous Grobner basis of I S then GP = π(G) is a Grobner Σ-basis of I. If G is an s- SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 17 Proof. Let f ∈ Il for some l ≥ 0 and consider the element f sl ∈ I S. Since G is an s-homogeneous Grobner basis of I S, there is gsk ∈ G (g ∈ P, k ≥ 0) such that lm(f sl) = qsilm(gsk)sj that is lm(f )sl = qsilm(g)sk+j = q(si · lm(g))sl with q ∈ M and i + j + k = l. It follows that lm(f ) = q(si · lm(g)) = qlm(si · g) with g = π(gsk) ∈ GP and we conclude that GP is a Grobner Σ-basis of I. (cid:3) Note that Proposition 5.18 and Proposition 5.19 explain that there is a complete equivalence between Grobner bases computations for w-graded Σ-invariant ideals I ( P and their skew analogues I S which are graded two-sided ideals of S. In particular, Grobner Σ-bases of I can be computed by the algorithm SkewGBasis when applied to I S. Precisely, if H = Si Hi is a w-homogeneous Σ-basis of I and G = SkewGBasis(H S) then GP = π(G) is a w-homogeneous Grobner Σ-basis of I. We may call this procedure SigmaGBasis2. The following result provides algorithmic solution of the membership problem for a class of Σ-invariant ideals. Note that such kind of results are quite rare, for instance, in the theory of difference ideals. Proposition 5.20. Let I ⊂ P be a w-graded Σ-invariant ideal and fix d ≥ 0. Assume that I has a w-homogeneous basis H such that Hd = {f ∈ H w(f ) ≤ d} is a finite set. Then, there is a w-homogeneous Grobner Σ-basis G of I such that Gd is also a finite set. In other words, if we consider for the algorithm SigmaGBasis a selection strategy of the S-polynomials based on their weights, we obtain that the d-truncated version of SigmaGBasis stops in a finite number of steps. Proof. First of all, note that the algorithm SigmaGBasis essentially computes a subset G of a Grobner basis Σ · G obtained by applying the Buchberger algorithm to the basis Σ · H of I. Moreover, by property (iii) of Definition 5.8 and Lemma 5.14 the elements of Σ · H and hence of Σ · G are all w-homogeneous. Denote H ′ d is also a finite set, consider Xd the finite set of variables of P occurring in the elements of H ′ d and define P (d) = K[Xd]. In fact, the d-truncated algorithm SigmaGBasis computes a subset of a Grobner basis of the ideal I (d) of P (d) generated by H ′ d. By Noetherianity of the ring P (d), we clearly obtain termination. (cid:3) d = {si · f i ≥ 0, f ∈ Hd, i + w(f ) ≤ d}. Since H ′ Note that the above result can be obtained also by Proposition 4.7. In fact, if I ( P is finitely Σ-generated up to weight d then I S is a graded ideal of S which is finitely generated up to s-degree d. Precisely, if H = Si Hi is a w-homogeneous Σ-basis of I and the set Si≤d Hi is finite for all d, then {f si f ∈ Hi, i ≤ d} is also a finite set that generates I S up to degree d. 6. The skew letterplace embedding Denote N∗ = {1, 2, . . .} the set of positive integers and let X = {x1, x2, . . .} be a finite or countable set of variables. We denote by xi(j) each element (xi, j) of the product set X × N∗ and define P = K[X × N∗] the polynomial ring in the commuting variables xi(j). Consider the algebra monomorphism of infinite order σ : P → P such that xi(j) 7→ xi(j + 1) for all i, j. Note that σ is a monomial map that is compatible with divisibility in Mon(P ). Then, put S = P [s; σ] the skew polynomial ring in the variable s defined by P and σ. Finally, let F = KhXi denote the free associative algebra generated by X. We consider F as a graded 18 R. LA SCALA AND V. LEVANDOVSKYY algebra with respect to the total degree. Recall that S = Li∈N Si is also a graded algebra with Si = P si. Definition 6.1. Let A ⊂ S be a K-subalgebra. If A is spanned by a submonoid M ⊂ Mon(S) then we call A a monomial subalgebra of S and we denote Mon(A) = M . In this case, a monomial ordering of S can be restricted to A. For instance, P is a monomial subalgebra of S. We have now a result about the possibility to embed the free associative algebra F into the skew polynomial ring S. Proposition 6.2. The graded algebra homomorphism ι : F → S, xi 7→ xi(1)s is injective. Then, the free associative algebra F is isomorphic to R = Im ι, a graded monomial subalgebra of S. Proof. It is sufficient to note that by the commutation rule of the variable s and the definition of the endomorphism σ, any word xi1 · · · xid ∈ Mon(F ) maps into xi1 (1) · · · xid (d)sd ∈ Mon(S). (cid:3) We call S the skew letterplace algebra and the algebra monomorphism ι the skew letterplace embedding. In Section 7 we will give motivation for such names. Fix now a monomial ordering ≺ on the algebra S that is σ is compatible with the restriction of ≺ to Mon(P ). It is easy to show that many usual monomial orderings on P (lex, degrevlex, etc) satisfy such condition. Recall that the existence of monomial orderings for P is provided by the Higman's lemma which implies the following result (see for instance [1], Corollary 2.3 and remarks at beginning of page 5175). Proposition 6.3. Let ≺ be a total ordering on the set Mon(P ) such that for all m, n, t ∈ Mon(P ) one has 1 (cid:22) m and if m ≺ n then tm ≺ tn. Then ≺ is also a well-ordering of Mon(P ) that is a monomial ordering of P if and only if the restriction of ≺ to the variables set X × N∗ is a well-ordering. Clearly, it is easy to assign well-orderings to the set X × N∗ which is in bijective correspondence to N2. Note that the algebra P has also a multigrading which is defined as follows. If m = xi1 (j1) · · · xid (jd) ∈ Mon(P ), then we denote ∂(m) = µ = (µk)k∈N∗ where µk = #{α jα = k}. If Pµ ⊂ P is the subspace spanned by all monomials of multidegree µ then P = Lµ Pµ is clearly a multigrading of the algebra P . If µ = (µk) is a multidegree, we denote i·µ = (µk−i)k∈N∗ where we put µk−i = 0 when k−i < 1. By definition of the map σ, if we denote Sµ,i = Pµsi one obtains that S = Lµ,i Sµ,i and Sµ,iSν,j ⊂ Sµ+(i·ν),i+j . The elements of each subspace Sµ,i ⊂ S are said multi-homogeneous. An ideal J ⊂ S is called multigraded if J = Pµ,i Jµ,i with Jµ,i = J ∩Sµ,i. In other words, the ideal J is generated by multi-homogeneous elements. For any integer i ≥ 0 we denote by 1i the multidegree µ = (µk)k∈N∗ such that µk = 1 if k ≤ i and µk = 0 otherwise. Clearly, a homogeneous element f si ∈ S (f ∈ P ) belongs to the graded subalgebra R if and only if f is multi-homogeneous and ∂(f ) = 1i. In other words, Ri = R ∩ Si = S1i,i = P1i si. Lemma 6.4. Let f sl ∈ S with f ∈ P a multi-homogeneous element and con- sider fijsi, gjsj, hjksk ∈ S where fij, gj, hjk ∈ P are multi-homogeneous elements such that f sl = Pi+j+k=l fijsigjsjhjksk. Then, from f sl ∈ R it follows that fijsi, gjsj, hjksk ∈ R, for all i, j, k. SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 19 jk . Denote µ = ∂(fij), ν = ∂(gsi Proof. Clearly we have f = Pi+j+k=l fijgsi j ) and ρ = ∂(hsi+j jk ) and put α = min{k νk > 0} and β = min{k ρk > 0}. By definition of the map σ, one has that α ≥ i + 1 and β ≥ i + j + 1. If we assume f sl ∈ R that is 1l = ∂(f ) = µ + ν + ρ, then necessarily µ = 1i, ν = i · 1j and ρ = (i + j) · 1k and hence ∂(fij) = 1i, ∂(gj) = 1j, ∂(hjk) = 1k. (cid:3) j hsi+j Proposition 6.5. Let I be a graded (two-sided) ideal of R ⊂ S and let J be the extension of I to S that is J is the (multigraded) ideal generated by I in S. If G is a multi-homogeneous basis of J then G ∩ R is a (homogeneous) basis of I. In particular, the contraction J ∩ R is equal to I, that is there is a bijective correspondence between all graded ideals of R and their extensions to S. Proof. Consider f sl ∈ I ⊂ R (f ∈ P ) a homogeneous element and let G = {gjsj} with gj ∈ P , gj multi-homogeneous. Since f is multi-homogeneous and G is a basis of J ⊃ I, one has f sl = Pi+j+k=l fij sigjsjhjksk with fij, hjk ∈ P , fij, hjk multi-homogeneous. From Lemma 6.4 it follows immediately that all elements fijsi, gjsj, hjksk ∈ R that is G ∩ R is a basis of I. (cid:3) Proposition 6.6. Let I ⊂ R be a graded ideal and let J ⊂ S be its extension. If G ⊂ J is a multi-homogeneous Grobner basis of J then G ∩ R is a homogeneous Grobner basis of I. Proof. If f sl = Pi+j+k=l fijsigjsjhjksk is a Grobner representation in S of a homogeneous element f sl ∈ I ⊂ J with respect to G = {gjsj}, then it is sufficient to use the same argument of Proposition 6.5 to obtain that f sl has a Grobner representation in R with respect to G ∩ R. (cid:3) We obtain finally an algorithm to compute Grobner bases of graded two-sided ideals of the subring R ⊂ S which is isomorphic to the free associative algebra F by the map ι. Note that the considered monomial orderings on F are obtained as the restriction of monomial orderings on S to the monomial subalgebra R. By applying Proposition 6.6, the computation of homogeneous Grobner bases in R is obtained as a slight modification of the algorithm SkewGBasis for the ideals of S. It is interesting to note that the latter procedure is in turn a variant of the Buchberger algorithm for modules over commutative polynomial rings. Thus, we may say that these computations in associative algebras are reduced to analogue ones over commutative rings via the notion of skew polynomial ring (see also Section 7). This reverses somehow the trivial fact that commutative algebras are just a subclass of the associative ones. 20 R. LA SCALA AND V. LEVANDOVSKYY Algorithm 6.1 FreeGBasis2 Input: H, a homogeneous basis of a graded two-sided ideal I ⊂ R. Output: G, a homogeneous Grobner basis of I. G := H; B := {(f, g) f, g ∈ G}; while B 6= ∅ do choose (f, g) ∈ B; B := B \ {(f, g)}; for all i, j ≥ 0, i + j = degs(f ) − degs(g) and spoly(f, sigsj) ∈ R do h := Reduce(spoly(f, sigsj), Σ G Σ); if h 6= 0 then B := B ∪ {(h, h), (h, k), (k, h) k ∈ G}; G := G ∪ {h}; end if ; end for; for all i, j ≥ 0, j − i = degs(f ) − degs(g) and spoly(f si, sjg) ∈ R do h := Reduce(spoly(f si, sjg), Σ G Σ); if h 6= 0 then B := B ∪ {(h, h), (h, k), (k, h) k ∈ G}; G := G ∪ {h}; end if ; end for; end while; return G. Note explicitely that conditions spoly(f, sigsj), spoly(f si, sjg) ∈ R are equiva- lent to ask that such multi-homogeneous elements of S have multidegrees of type (1d, d), for some d ≥ 0. Proposition 6.7. The algorithm FreeGBasis2 is correct. Proof. Since G is multi-homogeneous implies that Σ G Σ is also multi-homogeneous, the procedure Reduce clearly preserves multi-homogeneity. Moreover, any element f ∈ G (f /∈ H) is obtained by reduction of a S-polynomial, say h. Owing to Proposition 6.6 we are interested only in the elements f ∈ R and this holds if and only if h ∈ R. (cid:3) Assume now that the graded ideal I ⊂ R has a finite number of generators up to some degree d > 0. Note that the d-truncated algorithm FreeGBasis2 has termination provided by termination of SkewGBasis as stated in Proposition 4.7. This generalizes a well-known result about algorithmic solution of the word problem (membership problem) for finitely presented graded associative algebras. 7. Letterplace in P As in Section 5, consider the P -linear map π : S → P such that si 7→ 1, for all i. Note now that ι′ = πι : F → P is an injective K-linear map such that xi1 · · · xid ∈ Mon(F ) 7→ xi1 (1) · · · xid (d) ∈ Mon(P ). Recall that F = Li Fi is a graded algebra with respect to total degree. Moreover, consider the weight map w : Mon(P ) → N such that w(xi(j)) = j for all i, j ≥ 1 and the corresponding SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 21 grading P = Li∈N Pi defined by the monoid ( N, max). Then, we have that ι′ is a homogeneous map (note K = F0 = P−∞ and P0 = 0) and ι = ξι′ which is an algebra homomorphism. Definition 7.1. Let I ( F be a graded (two-sided) ideal. Denote by I ′ ( P the w-graded Σ-invariant ideal Σ-generated by ι′(I). In other words, if G = {ι′(f ) f ∈ Ii, i > 0} then I ′ is the ideal of P generated by Σ · G. We call I ′ the letterplace analogue of I. Proposition 7.2. Let I ( F be a graded ideal and I ′ ( P its letterplace analogue. Denote by J = I ′S the skew analogue of I ′ and call J the skew letterplace analogue of I. We have that J is the extension to S of the ideal ι(I) ⊂ R. Then, there is a bijective correspondence between all graded ideals of F and their (skew) letterplace analogues. Proof. Let J ′ be the extension of ι(I) to S. By definition J ′ is the ideal gener- ated by the elements ι(f ) = ι′(f )si, for all f ∈ Ii. Since I ′ is Σ-generated by the w-homogeneous elements ι′(f ) of weight i, we conclude that J = I ′S = J ′. Moreover, the bijective correspondence between graded two-sided ideals of F and their letterplace analogues in P is obtained by composing the bijections contained in Proposition 5.12 and Proposition 6.5. (cid:3) The bijection between graded ideals of F and their letterplace analogues has been introduced in [22] and called "letterplace correspondence". The motivation of such name is essentially historical since the linear map ι′ was first considered in [11, 7]. Note that in these articles the endomorphism σ and the algebra embedding ι were not introduced. The polynomial ring P was named there the "letterplace alge- bra" because in the monomial ι′(xi1 · · · xid ) = xi1 (1) · · · xid (d) the indices 1, . . . , d play the role of the "places" where the "letters" xi1 , . . . , xid occur in the word xi1 · · · xid ∈ Mon(F ). Fix now a monomial ordering ≺ on the algebra S that is σ is compatible with the restriction of ≺ to Mon(P ). By restricting ≺ to R one obtains a monomial ordering on F . Denote by V the image of the map ι′ that is V = Li Vi is a graded subspace of P where Vi = P1i ⊂ Pi. Note that V is a left R-module isomorphic to R ≈ F . In fact, V = π(R) and the restriction π : R → V has the restriction ξ : V → R as its inverse. In [22] one has the following result which is now a direct consequence of Proposition 5.18 and Proposition 6.6. Proposition 7.3. Let I ( F be a graded ideal and denote by J ( P its letterplace analogue. Then J is a multigraded (hence w-graded) Σ-invariant ideal of P . If G is a multi-homogeneous (hence w-homogeneous) Grobner Σ-basis of J then ι′−1(G∩V ) is a homogeneous Grobner basis of I. From this result and algorithm SigmaGBasis one obtains the correctness of the following procedure which also has been introduced in [22]. 22 R. LA SCALA AND V. LEVANDOVSKYY Algorithm 7.1 FreeGBasis Input: H, a homogeneous basis of a graded two-sided ideal I ( F . Output: ι′−1(G), a homogeneous Grobner basis of I. G := ι′(H); B := {(f, g) f, g ∈ G}; while B 6= ∅ do choose (f, g) ∈ B; B := B \ {(f, g)}; for all i ≥ 0 such that spoly(f, si · g) ∈ V do h := Reduce(spoly(f, si · g), Σ · G); if h 6= 0 then B := B ∪ {(h, h), (h, k), (k, h), k ∈ G}; G := G ∪ {h}; end if ; end for; end while; return ι′−1(G). Assume finally that the graded ideal I ( F has a finite number of generators up to some degree d > 0. Note that the d-truncated algorithm FreeGBasis has now termination provided by Proposition 5.20. 8. Examples and timings In this section we propose an explicit computation and some timings in order to provide some concrete experience with the algorithms we introduced. Let X = {x} and consider the ring of ordinary difference polynomials P = K[X × N] that is P is the polynomial ring in the variables x(j) which are the shifts of a single univariate function x = x(0). Moreover, let P be endowed with the lexicographic monomial ordering where x(0) < x(1) < . . .. Denote by J the difference ideal generated by the single difference polynomial g1 = x(2)x(0) − x(1). This ideal has been considered in [18] as an example of an ordinary difference equation with periodic solutions. A variant of this equation has been also considered in [12]. By the algorithm SigmaGBasis one can compute that J has a finite Grobner difference basis with elements g1 = x(2)x(0) − x(1), g2 = x(4)x(1) − x(3)x(0), g3 = x(3)2x(0) − x(3), g4 = x(4)x(3)x(0) − x(4), g5 = x(5) − x(4)x(0). If Σ = hσi Let us see how the algorithm SigmaGBasis is able to obtain this. where σ : P → P is the shift endomorphism x(i) 7→ x(i + 1), then the ideal J is by definition Σ-generated by {g1} that is it is generated by Σ · {g1} = {x(2)x(0) − x(1), x(3)x(1) − x(2), x(4)x(2) − x(3), . . .}. Up to the product criterion, to compute a Grobner basis of J one should consider all the S-polynomials spoly(σi ·g1, σi+2 ·g1) for any i ≥ 0. We are interested in fact in computing a Grobner Σ-basis of J and hence we can apply the Σ-criterion that kills all these S-polynomials except for spoly(g1, σ2 · g1). The reduction of this element with respect to Σ · {g1} leads SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 23 to g2 = x(4)x(1) − x(3)x(0). Now the current Σ-basis of J is {g1, g2}. The S- polynomials that survive to product and Σ-criterion are now spoly(g2, σ · g1), spoly(g2, σ2 · g1), spoly(g2, σ4 · g1), spoly(g1, σ · g2), spoly(g2, σ3 · g2). Then spoly(g2, σ2 · g1) → 0 and spoly(g2, σ · g1) reduces to g3 with respect to Σ · {g1, g2}. The list of new S-polynomials arising from g3 that pass product and Σ-criterion is spoly(g3, g1), spoly(g3, σ · g1), spoly(g3, σ3 · g1), spoly(g3, σ2 · g2), spoly(g2, σ · g3), spoly(g1, σ2 · g3), spoly(g3, σ3 · g3), spoly(g2, σ4 · g3). We have now that spoly(g3, σ · g1) → 0, spoly(g3, g1) → 0 and spoly(g2, σ · g3) → g4. Up to all criteria, including chain criterion, the list of S-polynomials has to be updated with the following ones spoly(g4, σ · g1), spoly(g4, σ2 · g1), spoly(g4, σ3 · g1), spoly(g4, σ4 · g1), spoly(g4, σ3 · g2), spoly(g2, σ · g4), spoly(g1, σ2 · g4), spoly(g3, σ3 · g4), spoly(g4, σ3 · g4), spoly(g2, σ4 · g4), spoly(g4, σ4 · g4). Now, one has the following reductions: spoly(g4, σ · g1) → 0, spoly(g4, σ2 · g1) → 0 and spoly(g1, σ · g2) → f = x(5)x(1) − x(3)x(0)2. We will show that the element f of the Grobner Σ-basis of J is in fact redundant because g5 is also in this basis. The new S-polynomials arising from f are spoly(f, σ · g1), spoly(f, σ5 · g1), spoly(f, g2), spoly(f, σ · g2), spoly(f, σ4 · g2), spoly(g1, σ · f ), spoly(g3, σ2 · f ), spoly(g4, σ2 · f ), spoly(g2, σ3 · f ), spoly(g4, σ3 · f ), spoly(f, σ4 · f ). Then, we start again with reductions: spoly(f, σ · g2) → 0, spoly(f, σ · g1) → 0, spoly(g3, σ3 · g1) → 0, spoly(g2, σ · g4) → g5 and therefore f is redundant. The last S-polynomials to be added are spoly(g5, σ3 · g1), spoly(g5, σ5 · g1), spoly(g5, σ · g2), spoly(g5, σ4 · g2), spoly(g5, f ), spoly(g5, σ4 · f ). If G = {g1, g2, g3, g4, g5} then we have that all remaining S-polynomials to be considered reduce to zero with respect to Σ · G, that is G is a Grobner Σ-basis (difference basis) of the Σ-ideal (difference ideal) J. We present now some timings obtained with an implementation of the algorithm FreeGBasis. This implementation, which is still under development, is an im- provement of the one we presented in [22]. We decided not to start implementing also FreeGBasis2 until FreeGBasis will evolute to some final form. We pro- pose here new comparisons with the system Magma that contains one of the most effective implementations of the classical algorithm [27, 15, 29] for computing non- commutative Grobner bases. Note that this implementation takes also advantage by the use of Faug`ere's F4 approach. The tests were performed on a PC with four Intel Core i7 CPU 940 2.93GHz processors with 12 GB RAM running Ubuntu Linux. We used Singular 3-1-3 with freegb.lib release 14203 and Magma version 2.17-8. We measured the time for real execution of the process (thus differently to the way we did comparisons in [22]) in "min:sec" format. The number of generators in the input and in the output are given as well. 24 R. LA SCALA AND V. LEVANDOVSKYY Example G3-5-6-2d12 G2-3-13-4d10 G3-8-13d8 serf-g2d8 cliff5d9 C41d6 C41Xd5 C41Yd5 C41Zd6 C41Wd6 0:10 0:05 0:05 0:05 0:08 0:05 0:08 0:05 0:08 0:05 1:15 0:01 0:04 0:01 0:12 0:10 0:04 0:03 0:10 0:01 Magma Singular #In #Out 5885 275 1490 11 10 18 17 41 6 6 6 6 6 6 168 50 44 44 44 35 This table shows essentially that the letterplace approach to the computation of non-commutative Grobner bases is comparable with the classical algorithms and hence it is feasible. From the viewpoint of implementations we record that Magma achieved significant improvements with respect to comparisons included in [22] and this stimulate us to further optimize our code. In fact, there is an ongoing work to enhance freegb.lib in Singular. We will make more extensive comparisons in future articles that will be concentrated on technical aspects of implementing the letterplace algorithms. Here is a brief description of the examples we considered for testing. In all the examples the last integer indicates the total degree that bounds the computa- tions. The examples G3-5-6-2, G2-3-13-4 refer to the class of presented groups G(l, m, n; q) = hr, s rl, sm, (rs)n, [r, s]qi, where [r, s] denotes the commutator. As for the example G3-8-13, this is one from the class of groups G(m, n, p) = ha, b, c am, bn, cp, (ab)2, (bc)2, (ca)2, (abc)2i. All these groups has been considered by Cox- eter [4] for the problem of determining their finiteness. For our computations we considered a homogenization of the ideal of the free associative algebra defining the group algebra of such groups. The example serf-g2 are modified full Serre relations built from the Cartan matrix G2. The following non-commutative polynomials are explicitly the generators we considered for homogenization. f1f2f2 − 2f2f1f2 + f2f2f1, e1e2e2 − 2e2e1e2 + e2e2e1, f1f1f1f1f2 − 4f1f1f1f2f1 + 6f1f1f2f1f1 − 4f1f2f1f1f1 + f2f1f1f1f1, e1e1e1e1e2 − 4e1e1e1e2e1 + 6e1e1e2e1e1 − 4e1e2e1e1e1 + e2e1e1e1e1, f2e1 − e1f2, f1e2 − e2f1, f1e1 − e1f1 + h1, f2e2 − e2f2 + h2, h1h2 − h2h1, h1e1 − e1h1 − 2e1, f1h1 − h1f1 − 2f1, h1e2 − e2h1 + e2, f2h1 − h1f2 + f2, h2e1 − e1h2 + 3e1, f1h2 − h2f1 + 3f1, h2e2 − e2h2 − 2e2, f2h2 − h2f2 − 2f2. Let F5 = Khx1, x2, x3, x4, x5i and define Γ ⊂ EndK(F5) the submonoid of all algebra endomorphisms sending variables into variables. The example cliff5 is the ideal I ⊂ F5 which is Γ-generated by the polynomials [x1 ◦ x2, x3] = (x1x2 + x2x1)x3 − x3(x1x2 + x2x1) and s5 = Pπ∈S5 sgn(π)xπ(1)xπ(2)xπ(3)xπ(4)xπ(5). The quotient ring F5/I is the generic Clifford algebra in 5 variables of a 4-dimensional vector space. Finally, the family C41 of examples originates from random linear substitutions into the ideal of 6 generators, defining the non-cancellative monoid C(4, 1) (see [20]) and includes also variations of those. For instance, C41W is given SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 25 by x4x4 − 25x4x2 − x1x4 − 6x1x3 − 9x1x2 + x1x1, x4x3 + 13x4x2 + 12x4x1 − 9x3x4 + 4x3x2 + 41x3x1 − 7x1x4 − x1x2, x3x3 − 9x3x2 + 2x1x4 + x1x1, 17x4x2 − 5x2x2 − 41x1x4, x2x2 − 13x2x1 − 4x1x3 + 2x1x2 − x1x1, x2x1 + 4x1x2 − 3x1x1. while C41 is given by 189x4x4 + 63x4x3 − 66x4x2 − 161x4x1 − 103x3x4 + 19x3x3 + 262x3x2 + 467x3x1− 360x2x4 − 144x2x3 + 24x2x2 + 136x2x1 + 175x1x4 + 35x1x3 − 160x1x2 − 315x1x1, 27x4x4 + 409x4x3 + 82x4x2 − 42x4x1 − 57x3x4 − 403x3x3 + 26x3x2 − 42x3x1− 50x2x4 − 434x2x3 − 12x2x2 − 14x2x1 + 45x1x4 + 435x1x3 + 30x1x2, 232x4x4 − 29x4x3 + 77x4x2 + 332x4x1 − 147x3x4 + 175x3x3 + 60x3x2 − 269x3x1− 107x2x4 + 184x2x3 + 83x2x2 − 217x2x1 + 28x1x4 − 217x1x3 − 139x1x2 + 120x1x1, 52x4x4 + 233x4x3 − 129x4x2 + 135x4x1 − 248x3x4 − 205x3x3 + 171x3x2 + 138x3x1+ 100x2x4 − 58x2x3 − 177x2x1 + 84x1x4 + 39x1x3 − 43x1x2 − 73x1x1, −225x4x4 − 150x4x3 − 179x4x2 − 262x4x1 + 91x3x4 − 94x3x3 + 225x3x2 + 74x3x1+ 214x2x4 + 224x2x3 + 90x2x2 + 266x2x1 − 175x1x4 − 50x1x3 − 205x1x2 − 190x1x1, 289x4x4 − 170x4x3 − 289x4x2 − 153x4x1 − 186x3x4 + 95x3x3 + 177x3x2 + 106x3x1− 231x2x4 + 35x2x3 + 168x2x2 + 175x2x1 + 241x1x4 + 60x1x3 − 115x1x2 − 233x1x1. 9. Conclusions and future directions From the previous sections we can conclude that, owing to the notion of Grobner Σ-basis and the skew letterplace embedding ι, the theory of non-commutative Grobner bases developed for the free associative algebra F = KhXi using the concepts of overlappings, tips or obstructions [15, 27, 29] can be deduced from, unified to the classical Buchberger theory for commutative polynomial rings based on S-polynomials, at least in the graded case. From a practical point of view, one obtains the alternative algorithms FreeGBasis and FreeGBasis2 which are im- plementable in any computer algebra system providing commutative Grobner bases. The feasibility of such methods has been already shown in [22] and confirmed by the new timings we have collected in Section 8. Moreover, the general theory developed in this paper can be applied to any context where a monoid of endomorphisms Σ acts on the polynomial algebra P = K[X] in a way which is compatible with Grobner bases theory. We propose not only an abstract definition of what this may mean contributing to a current research trend (see for instance [9, 1, 3]), but also a method to transfer the related algorithms from P to the skew monoid ring S = P ∗ Σ when a suitable grading is given for P . This theory applies in particular to the shift operators and hence a stimulating field of applications are the rings of difference polynomials. The simple calculation proposed in Section 8 gives some feeling of this. In particular, we aim to extend the Grobner Σ-bases theory to any finitely generated free commutative monoid Σ = hσ1, . . . , σri in order to cover partial difference ideals and to extend the letterplace method for F to the non-graded case by means of suitable (de)homogenization techniques. An effective implementation of all proposed algorithms will be clearly important to understand the actual performance of the methods. Acknowledgments First of all, we would like to thank Vladimir Gerdt for introducing us to the theory of difference polynomial rings and thus leading us to note that the letterplace algebra K[X × N] endowed with the endomorphism xi(j) 7→ xi(j + 1) is just an 26 R. LA SCALA AND V. LEVANDOVSKYY instance of them (ordinary case). We are grateful to Albert Heinle, Benjamin Schnitzler and Grischa Studzinski for adopting the SymbolicData project [28] for handling ideals in free algebras. The timings in this article were obtained with the help of the updated SymbolicData system. We like also to thank the anonymous reviewers for their valuable comments and suggestions. References [1] Aschenbrenner, M.; Hillar, C.J., Finite generation of symmetric ideals. Trans. Amer. Math. Soc., 359 (2007), no. 11, 5171 -- 5192. [2] Bergman, G. M., The diamond lemma for ring theory. Adv. in Math., 29 (1978), no. 2, 178 -- 218. [3] Brouwer, A.E.; Draisma, J., Equivariant Grobner bases and the Gaussian two-factor model. Math. Comp., 80 (2011), no. 274, 1123 -- 1133. [4] Coxeter, H.S.M., The abstract groups Gm,n,p, Trans. Amer. Math. Soc., 45, (1939), no. 1, 73 -- 150. [5] Decker, W., Greuel, G.-M., Pfister, G., Schonemann, H.: Singular 3-1-3 -- A computer algebra system for polynomial computations. http://www.singular.uni-kl.de (2011). [6] De Concini, C.; Eisenbud, D.; Procesi, C., Young diagrams and determinantal varieties. Invent. Math., 56, (1980), no. 2, 129 -- 165. [7] Doubilet, P.; Rota, G.-C.; Stein, J., On the foundations of combinatorial theory. IX. Combi- natorial methods in invariant theory. Studies in Appl. Math., 53 (1974), 185 -- 216. [8] Drensky, V., Free algebras and PI-algebras. Graduate course in algebra, Springer Singapore, Singapore, 2000. [9] Drensky, V.; La Scala, R., Grobner bases of ideals invariant under endomorphisms, J. Sym- bolic Comput., 41 (2006), no. 7, 835 -- 846. [10] Eisenbud, D., Commutative algebra with a view toward algebraic geometry. Graduate Texts in Mathematics, 150. Springer, New York, 1995. [11] Feynman, R.P., An operator calculus having applications in quantum electrodynamics. Phys- ical Rev. (2), 84, (1951), 108 -- 128. [12] Gerdt, V.P., Consistency Analysis of Finite Difference Approximations to PDE Systems. In: Proc. of Mathematical Modeling and Computational Physics. MMCP 2011, 28 -- 42, Lecture Notes in Comput. Sci., 7125. Springer, Berlin, 2012. [13] Giambruno, A.; Zaicev, M., Polynomial identities and asymptotic methods. Mathematical Surveys and Monographs, 122. American Mathematical Society, Providence, RI, 2005. [14] Gondran, M.; Minoux, M., Graphs, dioids and semirings. New models and algorithms. Oper- ations Research/Computer Science Interfaces Series, 41. Springer, New York, 2008. [15] Green, E.L., An introduction to noncommutative Grobner bases. Computational algebra (Fairfax, VA, 1993), 167 -- 190, Lecture Notes in Pure and Appl. Math., 151, Dekker, New York, 1994. [16] Green, E.L., Multiplicative bases, Grobner bases, and right Grobner bases. Symbolic com- putation in algebra, analysis, and geometry (Berkeley, CA, 1998). J. Symbolic Comput., 29 (2000), no. 4-5, 601 -- 623. [17] Greuel, G.-M.; Pfister, G., A Singular introduction to commutative algebra. Second edition. Springer, Berlin, 2008. [18] Grove, E.A.; Ladas, G., Periodicities in nonlinear difference equations. Advances in Discrete Mathematics and Applications, 4. Chapman & Hall/CRC, Boca Raton, FL, 2005. [19] Higman, G., Ordering by divisibility in abstract algebras. Proc. London Math. Soc. (3), 2, (1952). 326 -- 336. [20] Jespers, E.; Okni´nski, J., Noetherian semigroup algebras. Algebras and Applications, 7. Springer, Dordrecht, 2007. [21] Lam, T.Y., A first course in noncommutative rings. Graduate Texts in Mathematics, 131. Springer, New York, 2001. [22] La Scala, R.; Levandovskyy, V., Letterplace ideals and non-commutative Grobner bases. J. Symbolic Comput., 44 (2009), no. 10, 1374 -- 1393. [23] Levin, A., Difference algebra. Algebra and Applications, 8. Springer, New York, 2008. SKEW POLYNOMIAL RINGS, GR OBNER BASES . . . 27 [24] Li, H., Noncommutative Grobner bases and filtered-graded transfer. Lecture Notes in Math- ematics, 1795. Springer, Berlin, 2002. [25] Mansfield, E.L., Szanto, A., Elimination theory for differential difference polynomials. In: Sendra, J.R. (Ed.), Proc. of the 2003 International Symposium on Symbolic and Algebraic Computation, 191 -- 198, ACM Press, New York, 2003. [26] McConnell, J.C.; Robson, J.C., Noncommutative Noetherian rings. Graduate Studies in Mathematics, 30. American Mathematical Society, Providence, RI, 2001. [27] Mora, F., Grobner bases for noncommutative polynomial rings. Algebraic algorithms and er- ror correcting codes (Grenoble, 1985), 353 -- 362, Lecture Notes in Comput. Sci., 229, Springer, Berlin, 1986. [28] Symbolic Data Project, http://www.symbolicdata.org, (2012). [29] Ufnarovski, V.A., On the use of graphs for calculating the basis, growth and Hilbert series of associative algebras. (Russian) Mat. Sb. 180 (1989), no. 11, 1548 -- 1560, 1584; translation in Math. USSR-Sb., 68, (1991), no. 2, 417 -- 428. [30] Weispfenning, V., Finite Grobner bases in non-Noetherian skew polynomial rings. In: Wang, P.S. (Ed.), Proc. of the 1992 International Symposium on Symbolic and Algebraic Compu- tation, 329 -- 334, ACM Press, New York, 1992. ∗ Dipartimento di Matematica, via Orabona 4, 70125 Bari, Italia E-mail address: [email protected] ∗∗ RWTH Aachen, Templergraben 64, 52062 Aachen, Germany E-mail address: [email protected]
1609.05812	2	1609	2017-12-29T19:14:01	Wedderburn-Malcev decomposition of one-sided ideals of finite dimensional algebras	[ "math.RA" ]	Let $A$ be a finite dimensional associative algebra over a perfect field and let $R$ be the radical of $A$. We show that for every one-sided ideal $I$ of $A$ there exists a semisimple subalgebra $S$ of $A$ such that $I=I_{S}\oplus I_{R}$ where $I_{S}=I\cap S$. and $I_{R}=I\cap R$.	math.RA	math	WEDDERBURN-MALCEV DECOMPOSITION OF ONE-SIDED IDEALS OF FINITE DIMENSIONAL ALGEBRAS A.A. BARANOV, A. MUDROV, AND H.M. SHLAKA Abstract. Let A be a finite dimensional associative algebra over a perfect field and let R be the radical of A. We show that for every one-sided ideal I of A there exists a semisimple subalgebra S of A such that I = IS ⊕ IR where IS = I ∩ S and IR = I ∩ R. Wedderburn-Malcev Principal Theorem is one of the landmarks in the theory of associative algebras. Theorem 1 (Wedderburn, Malcev). Let A be a finite dimensional associative algebra and let R be the radical of A. Suppose that A/R is separable. Then the following hold. (1) There exists a semisimple subalgebra S of A such that A = S ⊕ R (Wedderburn). (2) If Q is a semisimple subalgebra of A then there exists r ∈ R such that Q ⊆ (1 + r)S(1 + r)−1 (Malcev). (3) If S1 and S2 are two subalgebras of A such that A = Si ⊕ R (i = 1, 2) then there exists r ∈ R such that S1 = (1 + r)S2(1 + r)−1 (Malcev). In this note we show that for every one-sided ideal I of A there exists a Wedderburn- Malcev decomposition S ⊕ R of A which splits I, i.e. I = IS ⊕ IR, where IS = I ∩ S and IR = I ∩ R. Note that the algebra A in Wedderburn-Malcev theorem is not required to be unital. If A doesn't contain 1 then the conjugation s 7→ (1 + r)s(1 + r)−1 should be rewritten in the obvious way: by expanding (1 + r)−1 = 1 − r + r2 − r3 + . . . (since r is nilpotent the sum is finite) we get (1 + r)s(1 + r)−1 = s + rs − sr − rsr + . . . . Although most of the books state this theorem for unital algebras only (e.g. [2]), one can find it in a non-unital context as well (as in the original papers of Wedderburn and Malcev), see, for example, [1]. The part (2) of the theorem is not normally mentioned in the books so we refer the reader to the original Malcev's paper for the proof [3]. Note that if A is a finite dimensional algebra over a perfect field then A/R is always separable, so the theorem holds in this case. For the sake of greater generality some of the results below are stated for the Artinian rings. Although every finite dimensional unital algebra is Artinian as a ring, this is not true for non-unital algebras (for example, any one dimensional algebra over Q with zero multiplication is not Artinian as a ring). However, as usual, it is easy to see that the results below hold for the finite dimensional algebras as well. 2010 Mathematics Subject Classification. 16P10, 16D20, 16D25. Supported by University of Leicester. Supported by University of Leicester. Supported by the Higher Committee for Educational Development in Iraq (HCED Iraq). 1 2 A.A. BARANOV, A. MUDROV, AND H.M. SHLAKA Throughout this paper, A is a (non-unital) associative left Artinian ring or a finite dimensional algebra. We denote by R its radical. If V is a subset of A we denote by ¯V its image in ¯A = A/R. Let I be a left ideal of A and let Q be a left ideal of ¯A. We say that I is Q-minimal (or simply bar-minimal ) if ¯I = Q and for every left ideal J of A with J ⊆ I and ¯J = Q one has J = I. It is well known that every non-nilpotent left ideal of a left Artinian ring contains an idempotent, see for example [2, Theorem 24.2]. We need a bit more precise version of this fact. Proposition 2. Let A be a left Artinian associative ring and let I be a left ideal of A. Suppose that I is bar-minimal. Then there is an idempotent e ∈ I such that I = Ae. Proof. We can assume that ¯I is non-zero (otherwise I = 0 is generated by the idempo- tent 0). Since ¯A is semisimple, there is a non-zero idempotent f ∈ ¯I such that ¯I = ¯Af . Fix any x ∈ I such that ¯x = f . Note that x2 6= 0 because ¯x2 = f 2 = f 6= 0. Consider the map ϕx : I → I defined by ϕx (y) = yx for all y ∈ I. Then ϕx is an endomorphism of the left A-module I. Note that the image J = ϕx(I) = Ix is a left ideal of A. We have ¯J = Ix = ¯I ¯x = ( ¯Af )f = ¯Af = ¯I. Since J ⊆ I and I is ¯I-minimal, we get I = J = Ix. Thus, ϕx is surjective. Since I is an Artinian module, ϕx is an automorphism. Hence, there is an element e ∈ I such that x = ϕx (e) = ex. We have so ϕx (cid:0)e2(cid:1) = e2x = e (ex) = ex = ϕx (e) , ϕx (cid:0)e2 − e(cid:1) = ϕx (cid:0)e2(cid:1) − ϕx (e) = 0 Since ϕx is injective, e2 − e = 0, so e is an idempotent. As ¯e ∈ ¯I = ¯Af , we have ¯e = ¯af for some ¯a ∈ ¯A. Thus, ¯ef = ¯af f = ¯af = ¯e. On the other hand, we have ex = x, so ¯ef = ¯e¯x = ¯x = f . Hence ¯e = f = ¯x. Therefore, Ae = ¯A¯e = ¯Af = ¯I. Since Ae ⊆ I is a left ideal of A with Ae = ¯I and I is ¯I-minimal, we must have I = Ae, as required. (cid:3) As a corollary we get the following simple fact, which is frequently mentioned in the textbooks. Corollary 3. Let A be a left Artinian ring and let I be a minimal non-nilpotent left ideal of A. Then I = Ae for some idempotent e ∈ I. Proof. Since I is non-nilpotent, I 6⊆ R, so ¯I is a non-zero left ideal of ¯A. Let J be a left ideal of A such that J ⊆ I and ¯J = ¯I. Then J is non-nilpotent (because ¯J = ¯I is non- nilpotent). Since I is minimal non-nilpotent, J = I. This implies that I is ¯I-minimal. Therefore, by Proposition 2, there is an idempotent e ∈ I such that I = Ae. (cid:3) The following theorem gives a complete characterisation of bar-minimal left ideals of left Artinian associative rings. WEDDERBURN-MALCEV DECOMPOSITION 3 Theorem 4. Let A be a left Artinian associative ring and let I be a left ideal of A. Then I is bar-minimal if and only if there is an idempotent e ∈ A such that I = Ae. Proof. The "only if" part is proved in Proposition 2. Suppose now that I = Ae, where e is an idempotent in A. Let J ⊆ I be an ¯I-minimal left ideal of A. We need to show that J = I. By Theorem 2, J = Ae1 for some idempotent e1 in A. Note that e1 = e1e1 ∈ Ae1 = J ⊆ I = Ae, so e1e = e1. Let e2 = ee1e = ee1. Then e2 is an idempotent. Indeed, e2 2 = ee1eee1e = ee1e1e = ee1e = e2. We have Ae2 = Aee1 ⊆ Ae1 ⊆ Ae. We claim that Ae2 = Ae1 = J. Since J is ¯I-minimal, it is enough to show that ¯A¯e2 = ¯A¯e = ¯I. As ¯A¯e1 = ¯I = ¯A¯e, we have ¯e = ¯e¯e ∈ ¯A¯e1, so ¯e¯e1 = ¯e. Therefore, ¯A¯e2 = ¯A¯e¯e1 = ¯A¯e, as required. We are going to show that e = e2. Since ¯e = ¯e¯e ∈ ¯A¯e = ¯A¯e2, we have ¯e¯e2 = ¯e. Recall that e2 = ee1, so ¯e = ¯e¯e2 = ¯e¯e¯e1 = ¯e¯e1 = ¯e2. Therefore, there is r ∈ R such that e2 = e + r. Since e2 = ee1e, we have e2 = ee2 = e2e. Hence e + r = e2 = ee2 = e(e + r) = e + er, so er = r. Similarly, re = r. Thus, e + r = e2 = e2 2 = (e + r)2 = e + 2r + r2. Therefore, r2 = −r and r2k e2 = e. Therefore, I = Ae = Ae2 = Ae1 = J, as required. = −r for all k ∈ N. Since R is nilpotent, we get r = 0, so (cid:3) Remark 5. Let e1 and e2 be idempotents of A. They are said to be right equivalent if e1e2 = e1 and e2e1 = e2. It is easy to see that Ae1 = Ae2 if and only if e1 and e2 are right equivalent. Thus, the bar-minimal left ideals of A are in bijective correspondence with the right equivalence classes of the idempotents of A. We are now ready to prove our main result. Theorem 6. Let A be a finite dimensional algebra and let I be a left ideal of A. Suppose that A/R is separable. Then there exists a semisimple subalgebra S of A such that A = S ⊕ R and I = IS ⊕ IR, where IS = I ∩ S and IR = I ∩ R. Proof. If I is nilpotent, then I ⊆ R, so I = IR as required. Suppose that I is non- nilpotent. Then ¯I is a non-zero left ideal of ¯A. Let J be a minimal left ideal of A such that J ⊆ I and ¯J = ¯I. Then by Proposition 2, J = Ae for some idempotent e ∈ J. By Wedderburn-Malcev Theorem, there is a semisimple subalgebra S of A such that e ∈ S and A = S ⊕ R. We have J = Ae = (S ⊕ R) e = Se ⊕ Re = JS ⊕ JR where JS = Se and JR = Re. Note that JS = J ∩ S (because e ∈ S) and JR = J ∩ R. Put IS = I ∩ S and IR = I ∩ R. Then the sum IS + IR ⊆ I is direct. We have JS ⊆ IS and ¯JS ⊆ ¯IS ⊆ ¯I = ¯J = ¯JS. 4 A.A. BARANOV, A. MUDROV, AND H.M. SHLAKA This implies I/IR ∼= ¯I = ¯IS ∼= IS, so I = IS ⊕ IR, as required. (cid:3) References [1] E.A. Behrens, Ring Theory, Academic Press, New York (1972). [2] C.W. Curtis, I. Reiner, Representation theory of finite groups and associative algebras. Interscience, New York, 1962. [3] A. Malcev. On the representation of an algebra as a direct sum of its radical and a semi-simple subalgebra. Dokl. Akad. Nauk 36 (1942), 42-46. Department of Mathematics, University of Leicester, Leicester, LE1 7RH, UK E-mail address: [email protected] Department of Mathematics, University of Leicester, Leicester, LE1 7RH, UK E-mail address: [email protected] Department of Mathematics, University of Leicester, Leicester, LE1 7RH, UK; De- partment of Material Engineering, University of Kufa, Al-Najaf, Iraq E-mail address: [email protected], [email protected]
1207.1733	1	1207	2012-07-06T21:04:34	Tolerances as images of congruences in varieties defined by linear identities	[ "math.RA" ]	An identity s=t is linear if each variable occurs at most once in each of the terms s and t. Let T be a tolerance relation of an algebra A in a variety defined by a set of linear identities. We prove that there exist an algebra B in the same variety and a congruence Theta of B such that a homomorphism from B onto A maps Theta onto T.	math.RA	math	Mailbox Tolerances as images of congruences in varieties defined by linear identities Ivan Chajda, G´abor Cz´edli, Radom´ır Halas, and Paolo Lipparini Abstract. An identity s = t is linear if each variable occurs at most once in each of the terms s and t. Let T be a tolerance relation of an algebra A in a variety defined by a set of linear identities. We prove that there exist an algebra B in the same variety and a congruence θ of B such that a homomorphism from B onto A maps θ onto T . An identity s = t is linear if each variable occurs at most once in each of the terms s and t, see, for example, M. N. Bleicher, H. Schneider and R. L. Wil- son [1, Theorem 4.19], W. Taylor [8], I. Bosnjak and R. Madar´asz [2], A. Pil- itowska [7], and their references. In the particular case where every variable occurs exactly twice, once in s and once in t, we speak of a balanced linear identity, see M. V. Lawson [6]. For example, the variety of semigroups and that of commutative semigroups are defined by balanced linear identities. Bi- nary reflexive, symmetric, and compatible relations are called tolerances; see I. Chajda [3]. If ϕ : B → A is a surjective homomorphism and θ is a congru- ence of the algebra B, then ϕ(θ) = {(ϕ(x), ϕ(y)) : (x, y) ∈ θ} is a tolerance of A. Each tolerance of A is obtained this way; this follows from our result below (applied for the variety defined by the empty set of linear identities). Sometimes, like in I. Chajda, G. Cz´edli, and R. Halas [4] or G. Cz´edli and G. Gratzer [5], we can choose an appropriate B from a given variety. We have the following additional result of this kind. Theorem. Assume that VVV is a variety defined by a set of linear identities, that A = (A, F ) ∈ VVV, and that T is a tolerance of A. Then there exist an algebra B ∈ VVV, a congruence θ of B, and a surjective homomorphism ϕ : B → A such that T = ϕ(θ). Proof. We generalize the idea of G. Cz´edli and G. Gratzer [5]. If D is an arbitrary algebra (not necessarily in VVV), then the complex algebra C of D, in other words the algebra of complexes of D, has the underlying set {X ⊆ D : X 6= ∅}, and for each basic operation f of D, the corresponding operation of C is defined by f (X1, . . . , Xn) = {f (x1, . . . , xn) : x1 ∈ X1, . . . , xn ∈ Xn}. 2010 Mathematics Subject Classification: Primary: 08A30. Secondary: 08B99, 20M07. Key words and phrases: Tolerance relation, homomorphic image of a congruence, linear identity, balanced identity. This research was supported the project Algebraic Methods in Quantum Logic, No.: CZ.1.07/2.3.00/20.0051 and by by the NFSR of Hungary (OTKA), grant numbers K77432 and K83219. 2 I. Chajda, G. Cz´edli, R. Halas, and P. Lipparini Algebra univers. If s is a linear term, which means that each variable occurs in s at most once, then it can be shown that s(X1, . . . , Xn) = {s(x1, . . . , xn) : x1 ∈ X1, . . . , xn ∈ Xn} holds for arbitrary Xi ∈ C (but this does not hold for arbitrary terms in general). This implies, as proved in [1] and [8], that if a variety is defined by linear identities, then it contains the complex algebra of each of its members. Next, let E denote the set {X ⊆ A : X 2 ⊆ T and X 6= ∅}. Since it is clearly a subalgebra of the complex algebra of A, the paragraph above implies that E = (E, F ) belongs to VVV. Let B = {(x, Y ) ∈ A × E : x ∈ Y }. Then B = (B, F ) also belongs to VVV since it is a subalgebra of A × E. Define θ = (cid:8)(cid:0)(x1, Y1), (x2, Y2)(cid:1) ∈ B2 : Y1 = Y2(cid:9). As the kernel of the second projection from B to E, it is a congruence of B. The first projection ϕ : B → A, (x, Y ) 7→ x, is a surjective homomorphism since, for every x ∈ A, x = ϕ(cid:0)x, {x}(cid:1). Clearly, if (cid:0)(x1, Y1), (x2, Y2)(cid:1) ∈ θ, then {x1, x2} ⊆ Y1 = Y2 ∈ E implies that (cid:0)ϕ(x1, Y1), ϕ(x2, Y2)(cid:1) = (x1, x2) ∈ T . Hence ϕ(θ) ⊆ T . Conversely, let (x1, x2) ∈ T . Then, with Y = {x1, x2}, we have that (x1, Y ), (x2, Y ) ∈ B, (cid:0)(x1, Y ), (x2, Y )(cid:1) ∈ θ, and xi = ϕ(xi, Y ). This implies that (x1, x2) ∈ ϕ(θ), and we conclude that T ⊆ ϕ(θ). (cid:3) References [1] Bleicher, M. N., Schneider, H., Wilson, R. L.: Permanence of identities on algebras. Algebra Universalis 3, 72 -- 93 (1973) [2] Bosnjak, I, Madar´asz R.: On power structures. Algebra and Discr. Math. 2, 14 -- 35 (2003) [3] Chajda, I.: Algebraic Theory of Tolerance Relations. Palack´y University Olomouc, Olomouc (1991) [4] Chajda, I., Cz´edli, G., Halas, R.: Independent joins of tolerance factorable varieties. Algebra Universalis (submitted) [5] Cz´edli, G., Gratzer, G.: Lattice tolerances and congruences. Algebra Universalis 66, 5 -- 6 (2011) [6] Lawson, M. V.: A correspondence between balanced varieties and inverse monoids. Int. J. Algebra Comput. 16, 887-924 (2006) [7] Pilitowska, A.: Linear identities in graph algebras. Comment. Math. Univ. Carol. 50, 11 -- 24 (2009) [8] Taylor, W.: Equational logic. Houston J. Math., Survey 1979. Abridged version in: G. Gratzer, Universal Algebra, 2nd ed., Appendix 4, Springer-Verlag (1979) Ivan Chajda Palack´y University Olomouc, Department of Algebra and Geometry, 17. listopadu 12, 771 46 Olomouc, Czech Republic e-mail : [email protected] G´abor Cz´edli University of Szeged, Bolyai Institute, Szeged, Aradi v´ertan´uk tere 1, Hungary 6720 e-mail : [email protected] URL: http://www.math.u-szeged.hu/~czedli/ Radom´ır Halas Vol. 00, XX Tolerances as images of congruences 3 Palack´y University Olomouc, Department of Algebra and Geometry, 17. listopadu 12, 771 46 Olomouc, Czech Republic e-mail : [email protected] Paolo Lipparini Department of Mathematics, Tor Vergata University of Rome, I-00133 Rome, Italy e-mail : [email protected]
1604.04561	1	1604	2016-04-15T16:37:38	Commutation principles in Euclidean Jordan algebras and normal decomposition systems	[ "math.RA", "math.OC" ]	The commutation principle of Ramirez, Seeger, and Sossa \cite{ramirez-seeger-sossa} proved in the setting of Euclidean Jordan algebras says that when the sum of a Fr\'{e}chet differentiable function $\Theta(x)$ and a spectral function $F(x)$ is minimized over a spectral set $\Omega$, any local minimizer $a$ operator commutes with the Fr\'{e}chet derivative $\Theta^{\prime}(a)$. In this paper, we extend this result to sets and functions which are (just) invariant under algebra automorphisms. We also consider a similar principle in the setting of normal decomposition systems.	math.RA	math	Commutation principles in Euclidean Jordan algebras and normal decomposition systems M. Seetharama Gowda Department of Mathematics and Statistics University of Maryland, Baltimore County Baltimore, Maryland 21250, USA [email protected] and Juyoung Jeong Department of Mathematics and Statistics University of Maryland, Baltimore County Baltimore, Maryland 21250, USA [email protected] September 17, 2018 Abstract The commutation principle of Ramirez, Seeger, and Sossa [13] proved in the setting of Euclidean Jordan algebras says that when the sum of a Fr´echet differentiable function Θ(x) and a spectral function F (x) is minimized over a spectral set Ω, any local minimizer a operator commutes with the Fr´echet derivative Θ′(a). In this paper, we extend this result to sets and functions which are (just) invariant under algebra automorphisms. We also consider a similar principle in the setting of normal decomposition systems. Key Words: Euclidean Jordan algebra, (weakly) spectral sets/functions, automorphisms, commu- tation principle, normal decomposition system, variational inequality problem, cone complemen- tarity problem. AMS Subject Classification: 17C20, 17C30, 52A41, 90C25. 1 1 Introduction Let V be a Euclidean Jordan algebra of rank n [3] and λ : V → Rn denote the eigenvalue map (which takes x to λ(x), the vector of eigenvalues of x with entries written in the decreasing order). A set Ω in V is said to be a spectral set [1] if it is of the form Ω = λ−1(Q), where Q is a permutation invariant set in Rn. A function F : V → R is said to be a spectral function [1] if it is of the form F = f ◦ λ, where f : Rn → R is a permutation invariant function. Extending an earlier result of Iusem and Seeger [7] for real symmetric matrices, Ramirez, Seeger, and Sossa [13] prove the following commutation principle. Theorem 1. Let V be a Euclidean Jordan algebra, Ω be a spectral set in V, and F : V → R be a spectral function. Let Θ : V → R be Fr´echet differentiable. If a is a local minimizer of Ω then a and Θ′(a) operator commute in V. min Θ(x) + F (x), (1) A number of important and interesting applications are mentioned in [13]. The proof of the above result in [13] is somewhat intricate, deep, and long. In our paper we extend the above result by assuming only the automorphism invariance of Ω and F , and at the same time provide (perhaps) a simpler and shorter proof. To elaborate, recall that an (algebra) automorphism on V is an invertible linear transformation on V that preserves the Jordan product. It is known (see [8], Theorem 2) that spectral sets and functions are invariant under automorphisms, but the converse may fail unless the algebra is either Rn or simple. By defining weakly spectral sets/functions as those having this automorphism invariance property, we extend the above result of Ramirez, Seeger, and Sossa as follows. Theorem 2. Let V be a Euclidean Jordan algebra and suppose that Ω in V and F : V → R are weakly spectral. Let Θ : V → R be Fr´echet differentiable. If a is a local minimizer of Ω then a and Θ′(a) operator commute in V. min Θ(x) + F (x). (2) By noting that an Euclidean Jordan algebra is an inner product space and the corresponding automorphism group is a subgroup of the orthogonal group (at least when the algebra is equipped with the canonical inner product), we state a similar result in the setting of a normal decomposition system. Such a system was introduced by Lewis [10] to unify various results in convex analysis on matrices. A normal decomposition system is a triple (X, G, γ) where X is a real inner product space, G is a (closed) subgroup of the orthogonal group of X, and γ : X → X is a mapping that 2 has properties similar to those of λ(x), see Section 4. Our commutation principle on such a system is as follows. Theorem 3. Let (X, G, γ) be a normal decomposition system. Let Ω be a convex G-invariant set in X, F : X → R be a convex G-invariant function, and Θ : X → R be Fr´echet differentiable. Suppose that a is a local minimizer of Ω Then −Θ′(a) and a commute in (X, G, γ). min Θ(x) + F (x). (3) The organization of our paper is as follows. We cover some preliminary material in Section 2. In Section 3, we define weakly spectral sets and present a proof of Theorem 2. In Section 4, we describe normal decomposition systems and present a proof of Theorem 3. In the Appendix, we state a structure theorem for the automorphism group of a Euclidean Jordan algebra and show that weakly spectral sets and spectral sets coincide only in an essentially simple algebra. 2 Preliminaries 2.1 Euclidean Jordan algebras Throughout this paper, V denotes a Euclidean Jordan algebra [3]. For x, y ∈ V, we denote their inner product by hx, yi and Jordan product by x ◦ y. We let e denote the unit element in V and V+ := {x ◦ x : x ∈ V} denote the corresponding symmetric cone. If V1 and V2 are two Euclidean Jordan algebras, then, V1 × V2 becomes a Euclidean Jordan algebra under the Jordan and inner products, defined, respectively by (x1, x2) ◦ (y1, y2) =(cid:16)x1 ◦ y1, x2 ◦ y2(cid:17) and h(x1, x2), (y1, y2)i = hx1, y1i + hx2, y2i. A similar definition is made for a product of several Euclidean Jordan algebras. Recall that a Euclidean Jordan algebra V is simple if it is not a direct sum/product of nonzero Euclidean Jordan algebras (or equivalently, if it does not contain any non-trivial ideal). It is known, see [3], that any nonzero Euclidean Jordan algebra is, in a unique way, a direct sum/product of simple Euclidean Jordan algebras. Moreover, every simple algebra is isomorphic to one of the following five algebras: (i) the algebra S n of n × n real symmetric matrices, (ii) the algebra Hn of n × n complex Hermitian matrices, (iii) the algebra Qn of n × n quaternion Hermitian matrices, (iv) the algebra O3 of 3 × 3 octonian Hermitian matrices, (v) the Jordan spin algebra Ln for n ≥ 3. 3 We say that V is essentially simple if it is either Rn or simple. An element c ∈ V is an idempotent if c2 = c; it is a primitive idempotent if it is nonzero and cannot be written as a sum of two nonzero idempotents. We say a finite set {e1, e2, . . . , en} of primitive idempotents in V is a Jordan frame if n ei ◦ ej = 0 if i 6= j and ei = e. Xi=1 It turns out that the number of elements in any Jordan frame is the same; this common number is called the rank of V. Proposition 1 (Spectral decomposition theorem [3]). Suppose V is a Euclidean Jordan algebra of rank n. Then, for every x ∈ V, there exist uniquely determined real numbers λ1(x), . . . , λn(x) (called the eigenvalues of x) and a Jordan frame {e1, . . . , en} such that x = λ1(x)e1 + · · · + λn(x)en. Conversely, given any Jordan frame {e1, . . . , en} and real numbers λ1, λ2, . . . , λn, the sum λ1e1 + λ2e2 + · · · + λnen defines an element of V whose eigenvalues are λ1, λ2, . . . , λn. We define the eigenvalue map λ : V → Rn by where λ1(x) ≥ λ2(x) ≥ . . . ≥ λn(x). This is well-defined and continuous [1]. λ(x) =(cid:16)λ1(x), λ2(x), . . . , λn(x)(cid:17), We define the trace of an element x ∈ V by tr(x) := λ1(x) + · · · + λn(x). Correspondingly, the canonical (or trace) inner product on V is defined by hx, yitr := tr(x ◦ y). This defines an inner product on V that is compatible with the given Jordan structure. With respect to this inner product, the norm of any primitive element is one. Throughout this paper, for a linear transformation A : V → V and x ∈ V, we use, depending on the context, both the function notation A(x) as well as the operator notation Ax. Given a ∈ V, we define the corresponding transformation La : V → V by La(x) = a◦ x. We say that two elements a and b operator commute in V if La Lb = Lb La. We remark that a and b operator commute if and only if there exist a Jordan frame {e1, e2, . . . , en} and real numbers a1, a2, . . . , an, b1, b2, . . . , bn such that a = a1e1 + a2e2 + · · · + anen and b = b1e1 + b2e2 + · · · + bnen, see [3], Lemma X.2.2. (Note that the ais and bis need not be in the decreasing order.) In particular, 4 in S n or Hn, operator commutativity reduces to the ordinary (matrix) commutativity. A linear transformation between two Euclidean Jordan algebras is a (Jordan algebra) homomor- phism if it preserves Jordan products. If it is also one-to-one and onto, then it is an isomorphism. If the algebras are the same, we call such an isomorphism an automorphism. Thus, a linear trans- formation A : V → V is an automorphism of V if it is invertible and A(x ◦ y) = Ax ◦ Ay ∀ x, y ∈ V. The set of all automorphisms of V is denoted by Aut(V). When V is a product, say, V = V1 × V2, for φi ∈ Aut(Vi), it is easy to see that φ defined by belongs to Aut(V). Thus, φ(x) :=(cid:16)φ1(x1), φ2(x2)(cid:17), x = (x1, x2) ∈ V1 × V2 Aut(V1) × Aut(V2) ⊆ Aut(V1 × V2). A similar statement can be made when V is a product of several factors. When V carries the canonical inner product, every automorphism is inner product preserving and so, Aut(V) is a closed subgroup of the orthogonal group of V. A linear transformation D : V → V is a derivation if D(x ◦ y) = D(x) ◦ y + x ◦ D(y) ∀ x, y ∈ V. We recall the following result from [8] (essentially, [3], Theorem IV.2.5). Proposition 2. Let V be essentially simple. If {e1, . . . , en} and {e′ frames in V, then there exists φ ∈ Aut(V) such that φ(ei) = e′ i for all i = 1, . . . , n. 1, . . . , e′ n} are any two Jordan 3 Weakly spectral sets and functions Definition 1. We say that a set E in V is weakly spectral if A function F : V → R is said to be weakly spectral if A(E) ⊆ E for all A ∈ Aut(V). F (Ax) = F (x) for all x ∈ V, A ∈ Aut(V). Remarks (1) Suppose E is a spectral set, that is, E = λ−1(Q) for some permutation invariant set Q in Rn. Then, x ∈ E, y ∈ V, λ(y) = λ(x) ⇒ y ∈ E. (4) Now, let x ∈ E and A ∈ Aut(V). As A maps Jordan frames to Jordan frames, λ(Ax) = λ(x). From (4), Ax ∈ E. This proves that E is weakly spectral. Hence, every spectral set is weakly spectral. 5 Now suppose F : V → R is a spectral function so that for some permutation invariant function f : Rn → R, F = f ◦ λ. It follows that F (Ax) = f (λ(Ax)) = f (λ(x)) = F (x) for any A ∈ Aut(V). Thus, F is weakly spectral. This proves that every spectral function is weakly spectral. (2) It has been observed in [8], Theorem 2, that in any essentially simple algebra, every weakly spectral set is spectral. The following example shows that weakly spectral sets/functions can be different from spectral sets/functions in general algebras. In the product algebra V = R × S 2, let Ω = R+ × S 2, and x =(cid:16)1, (cid:2) −1 0 0 2(cid:3)(cid:17). 0 2(cid:3)(cid:17), y =(cid:16) − 1, (cid:2) 1 0 Since x ∈ Ω, y 6∈ Ω, and λ(x) = (2, 1, −1)T = λ(y), we see that Ω cannot be of the form λ−1(Q) for any (permutation invariant) set Q in R3. Thus, Ω is not a spectral set in V. Now, identity transformation is the only automorphism of R and any automorphism of S 2 is of the form X 7→ U XU T for some orthogonal matrix U . As R and S 2 are non-isomorphic Euclidean Jordan algebras, we see (from Proposition 1 in [4] or Corollary 3 in the Appendix) that automorphisms of V are of the form (t, X) 7→ (t, U XU T), for some orthogonal matrix U . It follows that Ω is weakly spectral. The characteristic function of Ω is an example of a weakly spectral function that is not spectral. (3) As a consequence of Corollary 3 in the Appendix, one can show the following: Suppose V = V1 × V2 × · · · × Vm where V1, V2, . . . , Vm are non-isomorphic simple algebras. Let Ei be a spectral set in Vi, i = 1, 2, . . . , m. Then, E1 × E2 × · · · × Em is weakly spectral in V. Not every weakly spectral set in V arises this way: Referring to example given in Remark 2, {(t, X) ∈ Ω : t + tr(X) = 0} is weakly spectral but not a product of two spectral sets. We also note, as a consequence of Theorem 6 that a product of (weakly) spectral sets need not be (weakly) spectral. The set {0} × S 2 in S 2 × S 2 is one such example. (4) It will be shown in Theorem 7 that in any algebra that is not essentially simple, the class of weakly spectral sets is strictly larger than the class of spectral sets. This shows that Theorem 2 is applicable to a wider class of sets/functions than Theorem 1. Proof of Theorem 2. As a is a local minimizer of the problem (2), we have Θ(a) + F (a) ≤ Θ(x) + F (x) for all x ∈ Na ∩ Ω, where Na denotes some open ball around a. Let u and v be arbitrary (but fixed) elements of V. Let D := LuLv − LvLu, where Lu(x) := u ◦ x, etc. Then, Proposition II.4.1. in [3] shows that D is a derivation on V; hence, as observed in [3], p. 36, etD is an automorphism of V for all t ∈ R. Therefore, by the continuity of t 7→ etDa and the automorphism invariance of Ω, x := etDa ∈ Na ∩Ω 6 for all t close to zero in R. Then, Θ(a) + F (a) ≤ Θ(etDa) + F (etDa) for all t close to 0. As F (etDa) = F (a) by the automorphism invariance of F , we see that Θ(a) ≤ Θ(etDa) for all t near 0. This implies that the derivative of Θ(etDa) at t = 0 is zero. Thus, we have hΘ′(a), Dai = 0. Putting b := Θ′(a) and recalling D = LuLv − LvLu, we get hb, (LuLv − LvLu)(a)i = 0. Since Lu and Lv are self-adjoint, the above expression can be rewritten as hv ◦ a, u ◦ bi − hu ◦ a, v ◦ bi = 0. This, upon rearrangement, leads to h(LbLa − LaLb)u, vi = 0. As this equation holds for all u and v, we see that LbLa = LaLb, proving the operator commutativity of a and b (= Θ′(a)) in V. An immediate special case of Theorem 2 is obtained by taking F = 0: If Ω is weakly spectral and Θ is Fr´echet differentiable, then any local minimizer a of minΩ Θ(x) operator commutes with Θ′(a). This can further be specialized by assuming that Θ is linear, that is, of the form Θ(x) = hb, xi. A number of applications mentioned in [13] have analogs for weakly spectral sets and functions. We mention one application that is especially important. Theorem 4. Suppose Ω ⊆ V and F : V → R are weakly spectral. Let G : V → V be arbitrary. Consider the variational inequality problem VI(G, Ω, F ): Find x∗ ∈ Ω such that hG(x∗), x − x∗i + F (x) − F (x∗) ≥ 0 for all x ∈ Ω. If a solves VI(G, Ω, F ), then a operator commutes with G(a). Proof. The proof is similar to the one given in [13], Proposition 1.9. If a solves VI(G, Ω, F ), then hG(a), x − ai + F (x) − F (a) ≥ 0 for all x ∈ Ω. This implies hG(a), xi + F (x) ≥ hG(a), ai + F (a) for all x ∈ Ω. So, a minimizes hG(a), xi + F (x) over Ω. By Theorem 2 applied to Θ(x) := hG(a), xi, we see that a commutes with G(a). As an illustration of the above result, let K be a closed convex cone in V and G : V → V be arbitrary. Consider the cone complementarity problem CP(G, K) of finding an x∗ ∈ K such that x∗ ∈ K, G(x∗) ∈ K ∗, and hx∗, G(x∗)i = 0 7 where K ∗ is the dual of K defined by K ∗ = {y ∈ V : hy, xi ≥ 0 for all x ∈ K}. Corollary 1. Suppose K is weakly spectral. If a solves the cone complementarity problem CP(G, K), then a commutes with G(a). Remark (5) The above corollary yields the following: Suppose K is a closed convex cone in V that is weakly spectral. If a ∈ K and b ∈ K ∗ satisfy ha, bi = 0, then a and b operator commute. Such a result for K = V+ (the symmetric cone of V) is well-known, see Proposition 6 in [5]. 4 Normal decomposition systems Before giving a proof of Theorem 3, we briefly recall the definition of a normal decomposition system and mention relevant properties. Definition 2. Let X be a real inner product space, G be a closed subgroup of the orthogonal group of X, and γ : X → X be a mapping satisfying the following properties: (a) γ is G-invariant, that is, γ(Ax) = γ(x) for all x ∈ X, A ∈ G. (b) For each x ∈ X, there exists A ∈ G such that x = Aγ(x), and (c) For all x, w ∈ X, we have hx, wi ≤ hγ(x), γ(w)i. Then, (X, G, γ) is called a normal decomposition system [10]. In such a system, a set Ω ⊂ X is said to be G-invariant if A(Ω) ⊆ Ω for all A ∈ G; a function F : X → R is said to be G-invariant if F (Ax) = F (x) for all x ∈ X and A ∈ G. In [10], various results on normal decomposition systems are given. In particular, the following is proved: Proposition 3 ([10], Proposition 2.3). In a normal decomposition system, for any two elements x and w, we have max A∈G hAx, wi = hγ(x), γ(w)i . Also, hx, wi = hγ(x), γ(w)i holds for two elements x and w if and only if there exists an A ∈ G such that x = Aγ(x) and w = Aγ(w). Motivated by the above proposition, we say that x and w commute in (X, G, γ) if there exists an A ∈ G such that x = Aγ(x) and w = Aγ(w). Now consider an essentially simple Euclidean Jordan algebra V. We assume that V carries the 8 canonical inner product and let G = Aut(V). Let {e1, . . . , en} be a fixed Jordan frame in V (with specified order). Define for any x ∈ V, n γ(x) := λi(x)ei, (5) Xi=1 where λi(x) are components of λ(x). As eigenvalues are preserved under automorphisms, we see that γ satisfies condition (a) in the definition of normal decomposition system. Since V is essentially simple, any Jordan frame can be mapped onto any other by an element of G (by Proposition 2). 1 λi(x)fi, we can find A ∈ G such that A(ei) = fi for all i. Then, Thus, given any x ∈ V with its spectral decomposition x = Pn λi(x)ei! = Aγ(x). x = A n Xi=1 This verifies condition (b) in the definition of normal decomposition system. Finally, for all x, w ∈ X, we have the so-called Theobald- von Neumann inequality hx, wi ≤ hγ(x), γ(w)i, see for example, [12], [1], or [6]. Putting all these together, we have the following result: Proposition 4. Every essentially simple Euclidean Jordan algebra V is a normal decomposition system with X = V, G = Aut(V), and γ : V → V defined as in (5). In this setting, two elements x, y ∈ X commute if and only if there exists a Jordan frame {f1, f2, . . . , fn} such that n n x = λi(x)fi and y = λi(y)fi. (6) X1 X1 We note that this is stronger than the operator commutativity of x and y. For example, in V = R2, x = (1, 0)T and y = (0, 1)T operator commute but does not commute in the above sense. Remark (6) Lim, Kim, and Faybusovich ([12], Corollary 4), show that when V is a simple Eu- clidean Jordan algebra, (V, K, γ) is a normal decomposition system, where K is the connected component of identity in Aut(V) and γ is defined as in (5). In [10], Lewis provides numerous examples of normal decomposition systems. In particular, the algebras S n and Hn (see Section 2) are normal decomposition systems where G is the corresponding automorphism group, γ(X) is the diagonal matrix with λ(X) as the diagonal. Another example is the space Mm,n of all real m × n matrices with inner product hX, Y i := tr(X T Y ), with G consisting of transformations of the form X 7→ U XV T , where U and V are orthogonal matrices, and γ(X) is an m × n matrix with diagonal consisting of singular values of X written in the decreasing order and zeros elsewhere. 9 Proof of Theorem 3. Since a is a local minimizer of the problem (3), we have Θ(a) + F (a) ≤ Θ(x) + F (x) for all x ∈ Na ∩ Ω, where Na denotes (some) neighborhood of a. Let A be an arbitrary element of G. As Ω is convex and G-invariant, we have, for all positive t near zero, (1 − t)a + tAa ∈ Na ∩ Ω. Thus, for all positive t near 0. As F is convex and G-invariant, Θ(a) + F (a) ≤ Θ(cid:16)(1 − t)a + tAa(cid:17) + F(cid:16)(1 − t)a + tAa(cid:17), F(cid:16)(1 − t)a + tAa(cid:17) ≤ (1 − t)F (a) + tF (Aa) = (1 − t)F (a) + tF (a) = F (a); hence, for all positive t near 0. This implies that hΘ′(a), Aa − ai ≥ 0, that is, hΘ′(a), Aai ≥ hΘ′(a), ai. Now let b := −Θ′(a) so that hb, Aai ≤ hb, ai. Then, as A ∈ G is arbitrary, we have Θ(a) ≤ Θ(cid:16)(1 − t)a + tAa(cid:17) max A∈G hb, Aai ≤ hb, ai . Using Proposition 3, we see that hγ(b), γ(a)i ≤ hb, ai. Since the reverse inequality always holds in a normal decomposition system, the above inequality turns into an equality. By Proposition 3, a and b commute in (X, G, γ). Remark (7) One might ask if the commutativity of a and −Θ′(a) in the above theorem could be replaced by that of a and Θ′(a). To answer this, we consider X = R2 with the usual inner product, let G be the set of all 2 × 2 signed permutation matrices (having exactly one nonzero entry, either 1 or −1, in each row/column), and γ(x) = x↓ (which is the vector of absolute values of entries of x written in the decreasing order). Let a = (−1, 1)T and b = (3, −2)T, Ω := convex-hull{Aa : A ∈ G}, Θ(x) = hb, xi, and F = 0. Then, it is easy to see that a minimizes Θ over Ω and commutes with −b (which is −Θ′(a)), but does not commute with b. We now state analogs of Theorem 4 and Corollary 1 in normal decomposition systems. Theorem 5. Suppose Ω ⊆ X and F : X → R are convex and G-invariant. Let G : X → X be arbitrary. Consider the variational inequality problem VI(G, Ω, F ) on X: Find x∗ ∈ Ω such that hG(x∗), x − x∗i + F (x) − F (x∗) ≥ 0 for all x ∈ Ω. If a solves VI(G, Ω, F ), then a operator commutes with −G(a). When Ω = K is a closed convex cone and F = 0, we write CP(G, K) for VI(G, Ω, F ). 10 Corollary 2. Suppose K is closed convex cone in X that is G-invariant and G : V → V be arbitrary. If a solves the cone complementarity problem CP(G, K), then a commutes with −G(a). Remarks (8) The above corollary yields the following: Suppose K is a closed convex cone in X that is G-invariant. If a ∈ K and b ∈ K ∗ satisfy ha, bi = 0, then a and −b commute. (9) We specialize the above remark to essentially simple algebras. Let V be such an algebra and let K be a spectral cone (which is a closed convex cone that is spectral) in V. If a ∈ K and b ∈ K ∗ satisfy ha, bi = 0, then there exists a Jordan frame {f1, f2, . . . , fn} such that n n n λn+1−i(b)fi, and λi(a) λn+1−i(b) = 0. X1 a = λi(a)fi, b = X1 X1 This comes from (6) by noting −λi(−b) = λn+1−i(b) and ha, bi = 0. (10) Another consequence of Remark (8) is the following: Suppose (X, G, γ) is a normal decom- position system where hγ(x), γ(y)i = 0 ⇒ x = 0 or y = 0. (We note that Mm,n and the system considered in Remark (7) are such systems.) closed convex cone in X that is G-invariant, then K = {0} or X. This can be seen as follows. If K is a Suppose K is different from {0} and X. Let a be a nonzero element in the boundary of K. By an application of the supporting hyperplane theorem, we can find a nonzero b ∈ K ∗ such that ha, bi = 0. By Remark (8), a and −b commute, hence, a = Aγ(a), −b = Aγ(−b) for some A ∈ G. Then, hγ(a), γ(−b)i = ha, −bi = 0. It follows that a = 0 or b = 0 leading to a contradiction. 5 Appendix Here, we describe a result on the automorphism group of a Euclidean Jordan algebra which is written as a product of simple algebras. While this result can be deduced from Theorem VI.18 in [9], for completeness, we present a direct (perhaps, elementary) proof. Using this result, we show that a Euclidean Jordan algebra V is essentially simple if and only if every weakly spectral set in V is spectral. Consider a (general) Euclidean Jordan algebra V. We assume that V is product of distinct non- isomorphic simple algebras V1, V2, . . . , Vm (with possible repetitions). Regrouping the factors, we assume that V =(cid:16)V1 × V1 × · · · × V1(cid:17) ×(cid:16)V2 × V2 × · · · × V2(cid:17) × · · · ×(cid:16)Vm × Vm × · · · × Vm(cid:17). By letting Wi := Vi × Vi × · · · × Vi, we write V = W1 × W2 × · · · × Wm. 11 (7) (8) Theorem 6. Aut(V) = Aut(W1) × Aut(W2) × · · · × Aut(Wm). Moreover, any automorphism φ of Aut(Wi) has the following form: where k is the number of factors in Wi, φj ∈ Aut(Vi), j = 1, 2, . . . , k and σ is a k × k permutation matrix. φ =(cid:16)φ1, φ2, . . . , φk(cid:17) ◦ σ, Note: The explicit form of the automorphism φ written with a permutation σ is: φ(x) =(cid:16)φ1(xσ(1)), φ2(xσ(2)), . . . , φk(xσ(k))(cid:17) for all x = (x1, x2, . . . , xk) ∈ Vi × Vi × · · · × Vi. Before giving a proof, we present two lemmas. In what follows, we write dim(X) for the dimension of a space X. Lemma 1. Suppose that V and W are simple Euclidean Jordan algebras and A : V → W is a non-zero Jordan homomorphism. Then: (i) dim(V) ≤ dim(W). (ii) If dim(V) = dim(W), A is an isomorphism. (iii) If dim(V) < dim(W), then zero is the only homomorphism from W to V. Proof. (i) Since the kernel of a homomorphism is an ideal of V and V is simple, we see that either A is zero or one-to-one. Since our A is nonzero, its kernel is {0}; hence it is one-to-one and so dim(V) ≤ dim(W). (ii) When dim(V) = dim(W), this A is also onto; hence it is an isomorphism. (iii) Assume dim(V) < dim(W). If there is a nonzero Jordan homomorphism from W to V, by (i), dim(W) ≤ dim(V). This is a contradiction. Lemma 2. Consider a product Euclidean Jordan algebra Z = Z1 × Z2 · · · × Zm. Let A : Z → Z be a linear transformation written in the matrix form A = [ Aij], where Aij : Zj → Zi is linear. If A is a Jordan homomorphism, then so is Aij for any i, j. Furthermore, AT ik Ail = 0 for all i and k 6= l. Proof. For x, y ∈ Z, we have A(x ◦ y) = Ax ◦ Ay. Taking x = (0, . . . , 0, xj , 0, . . . , 0)T and y = (0, . . . , 0, yj , 0, . . . , 0)T, we get, for any i, j, Aij(xj ◦ yj) = Aij xj ◦ Aij yj. This proves that is a homomorphism. Now suppose k 6= l and let x = (0, . . . , 0, xk, 0, . . . , 0)T and y = Aij (0, . . . , 0, yl, 0, . . . , 0)T. Then A(x ◦ y) = Ax ◦ Ay yields, 0 = Aikxk ◦ Ailyl. This leads to hAikxk, Ailyli = 0 and to hxk, AT ik Ail yli = 0. As xk and yl are arbitrary, we get AT ik Ail = 0. 12 Proof of Theorem 6. We may assume without loss of generality that all algebras involved carry canonical inner products. Since Aut(W1)×Aut(W2)×· · ·×Aut(Wm) ⊆ Aut(V), it is enough to prove the reverse inclusion. As the result is obvious for m = 1, we assume that m ≥ 2. Let A ∈ Aut(V). Since we are given V by (7) and (8), we think of A as a matrix of linear transformations A = [Aij], where each Aij is a linear transformation from some Vk to Vl. By partitioning this matrix, we write A = [Bkl], where (each block) Bkl : Wl → Wk is a linear transformation. The main part of our proof consists in showing B1j = 0 for all j ≥ 2. (9) Once we establish this, the same argument can then be used for AT (which is the inverse of A as we are using the canonical inner product). This results in Bj1 = 0 for all j ≥ 2. It then follows that B11 ∈ Aut(W1) and A could be viewed as an element of Aut(W1) × Aut(W2 × W3 × · · · × Wm). We then invoke the induction principle to see that A ∈ Aut(W1) × Aut(W2) × · · · × Aut(Wm). Now towards proving (9), we make the following claims: Claim 1: (a) If for some k 6= l, (the off-diagonal block) Bkl has a nonzero entry, then, dim(Vl) < dim(Vk) and Blk = 0. (b) If Aij is a nonzero entry in (a diagonal block) Bkk, then all other entries in the row/column of A containing Aij are zero, that is, Ail = 0 and Ali = 0 for all l 6= j. To see (a), suppose Aij is a nonzero entry in Bkl. Then, Aij from Vl to Vk is a nonzero homo- morphism (by Lemma 2). As Vl and Vk are simple and non-isomorphic, by Lemma 1, dim(Vl) < dim(Vk). Lemma 1 also shows that there cannot be a nonzero homomorphism from Vk to Vl. Thus, every entry in Blk is zero. To see (b), suppose that Aij is a nonzero entry in a diagonal block Bkk. Then, by Lemma 2, Aij : Vk → Vk is an isomorphism. From the same lemma, for l 6= j, we have AT ij Ail = 0 and so Ail = 0. Thus, in the row containing Aij, all other entries are zero. By working with the transpose of A, we see that the column containing Aij is zero except for the Aijth entry. This proves the claim. Claim 2: Suppose for some l with 1 ≤ l ≤ m−1, B12, B23, . . . , Bl l+1 are nonzero. Then, l < m−1 and there exists j > l + 1 such that Bl+1 j is nonzero. If this were not true, then either l = m − 1 or l < m − 1 and Bl+1 j = 0 for all j > l + 1. From Claim 1(a), dim(Vl+1) < dim(Vl) < · · · < dim(V1). From Lemma 1(iii), Bl+1 1, Bl+1 2, . . . , Bl+1 l are all zero. This means that in the matrix with entries Bij, in the l + 1 row, all entries except possibly Bl+1 l+1, are zero. As A is invertible, this lone entry Bl+1 l+1 cannot be zero. In fact, no row in the matrix Bl+1 l+1 can be zero. By Claim 1(b), each row of Bl+1 l+1 contains exactly one nonzero entry. This implies that in the square matrix Bl+1 l+1, each column will also contain exactly one 13 nonzero entry. By Claim 1(b), all columns of Bl l+1 will be zero. This contradicts our assumption that Bl l+1 is nonzero. This proves our claim. Now suppose, if possible, (9) is false so that B1j 6= 0 for some j ≥ 2. We may assume, by relabeling, that j = 2, so B12 is nonzero. By Claim 2 (with l = 1), 2 < m and there exists j > 2 such that B2j is nonzero. Relabeling, we may assume that j = 3 so that B23 is nonzero. We can use Claim 2 again, to see that B34 is nonzero, etc. Claim 2 allows us to repeat this process; however, as m is finite, this cannot continue forever. Thus, we reach a contradiction. Hence, (9) holds and as discussed before, leads to the completion of the proof of the first part of the theorem. We now come to the second part of the theorem. For simplicity, we let i = 1. We need to describe the matrix A which is now B11. As A is invertible, each row of B11 is nonzero. By Claim 1(b), each row of B11 contains exactly one nonzero entry which, by Lemma 1, is an automorphism of V1. (This means that each column of B11 also has the same property.) Thus, B11 can be regarded as a permutation of a diagonal matrix of transformations where each diagonal entry is an automorphism of V1. This gives the stated assertion. The following is immediate. Corollary 3. Suppose V = V1×V2×· · ·×Vm, where V1, . . . , Vm are non-isomorphic simple algebras. Then, Aut(V) = Aut(V1) × Aut(V2) × · · · × Aut(Vm). As an application of the above results, we prove the following. Theorem 7. V is essentially simple if and only if every weakly spectral set in V is spectral. Proof. The 'only if' part has been observed in [8], Theorem 2. We prove the 'if' part. Suppose, if possible, V is not essentially simple; let V be given by (7) and (8). We consider two cases: Case 1: V = W1 × W2 × · · · × Wm, m ≥ 2. For i = 1, 2, . . . , m, let rank(Wi) = ni, Pi denote the set of all primitive idempotents in Wi, and 0 denote the zero element in any Wi. Since automorphisms map primitive idempotents to primitive idempotents, P1 is invariant under automorphisms of W1, and so, by Theorem 6, Ω := P1 × {0} × {0} . . . × {0} is weakly spectral in V. Let c1 ∈ P1, c2 ∈ P2, x =(cid:16)c1, 0, 0, . . . , 0(cid:17) and y =(cid:16)0, c2, 0, 0 . . . , 0(cid:17). As both x and y have eigenvalues 1 (appearing once) and 0 (appearing n1 + n2 + . . . + nm − 1 times), we see that λ(x) = λ(y). However, x ∈ Ω while y 6∈ Ω. Thus, Ω cannot be of the form λ−1(Q) for any (permutation invariant) set Q. Case 2: V = W1 = V1 × V1 × · · · × V1, where V1 is a simple algebra of rank at least 2 and the number of factors in this product, say, m, is more than one. Let n = rank(V1). In Rn, let si denote the coordinate vector containing 1 in its ith slot and zeros elsewhere; let Q = {s1, s2, . . . , sn}. As 14 Q is permutation invariant, the set P1 := λ−1(Q) (which equals the set of all primitive elements in V1) is a spectral set in V1, where λ : V1 → Rn is the eigenvalue map. As P1 is invariant under automorphisms of V1, an application of Theorem 6 shows that the set Ω := P1 × P1 × · · · × P1 is weakly spectral in V. We now claim that Ω is not a spectral set in V. Let e denote the unit vector in V1 and {e1, e2 . . . , en} be a Jordan frame in V1. Then, λ(e1) = (1, 0, 0, . . . , 0)T and so the vector x := (e1, e1, . . . , e1) in Ω has eigenvalues 1 (repeated m times) and 0 (repeated m(n − 1) λ(y) = λ(x). On the other hand, when n < m, we write m = nk + l with 0 ≤ l < n and define times). When m ≤ n, let y := (cid:16)e1 + e2 + · · · + em, 0, 0, . . . , 0(cid:17) ∈ V. We see that y 6∈ Ω while y :=(cid:16)e, e, . . . , e, e1 + e2 + · · · + el, 0, 0 . . . , 0(cid:17), where e is repeated k times. We see that y 6∈ Ω while λ(y) = λ(x). Thus, Ω is not a spectral set. This completes the proof. References [1] M. Baes, Convexity and differentiability of spectral functions on Euclidean Jordan algebra, Lin- ear Alg. Appl., 422 (2007) 664-700. [2] R. Bhatia, Matrix Analysis, Springer Graduate Texts in Mathematics, Springer-Verlag, New York, 1997. [3] J. Faraut and A. Kor´anyi, Analysis on Symmetric Cones, Clarendon Press, Oxford, 1994. [4] M.S. Gowda, Positive and doubly stochastic maps, and majorization in Euclidean Jordan alge- bras, To appear in Linear Alg. Appl., DOI 10.1016/j.laa.2016.02.024. [5] M.S. Gowda, R. Sznajder, and J. Tao, Some P-properties for linear transformations on Eu- clidean Jordan algebras, Linear Alg. Appl., 393 (2004) 203-232. [6] M.S.Gowda and J. Tao, The Cauchy interlacing theorem in simple Euclidean Jordan algebras and some consequences, Linear Multi. Alg., 59 (2011) 65-86. [7] A. Iusem and A. Seeger, Angular analysis of two classes of non-polyhedral convex cones: the point of view of optimization theory, Comp. Appl. Math., 26 (2007) 191-214. [8] J. Jeong and M.S. Gowda, Spectral sets and functions in Euclidean Jordan alge- bras, Research Report, Department of Mathematics and Statistics, University of Mary- land, Baltimore County, Baltimore, Maryland 21250, USA, January 2016, available at www.math.umbc.edu/∼gowda/tech-reports/index.html. [9] M. Koecher, The Minnesota Notes on Jordan Algebras and Their Applications, Lecture Notes in Mathematics 1710, Springer, Berlin 1999. 15 [10] A.S. Lewis, Group invariance and convex matrix analysis, SIAM J. Matrix Anal., 17 (1996) 927-949. [11] A. S. Lewis, Convex analysis on the Hermitian matrices, SIAM J. Optim., 6 (1996) 164-177. [12] Y. Lim, J. Kim, and L. Faybusovich, Simultaneous diagonalization on simple Euclidean Jordan algebras and its applications. Forum Math., 15 (2003) 639-644. [13] H. Ram´ırez, A. Seeger, and D. Sossa, Commutation principle for variational Problems on Euclidean Jordan algebras, SIAM J. Optim., 23 (2013) 687-694. [14] A. Seeger, Convex analysis of spectrally defined matrix functions, SIAM J. Optim., 7 (1997) 679-696. [15] D. Sossa, Euclidean Jordan algebras and variational problems under conic constraints, PhD Thesis, University of Chile, 2014. 16
1508.03798	1	1508	2015-08-16T07:55:07	Criteria for a ring to have a left Noetherian left quotient ring	[ "math.RA" ]	Two criteria are given for a ring to have a left Noetherian left quotient ring (this was an open problem since 70's). It is proved that each such ring has only finitely many maximal left denominator sets.	math.RA	math	Criteria for a ring to have a left Noetherian left quotient ring V. V. Bavula Abstract Two criteria are given for a ring to have a left Noetherian left quotient ring (this was an open problem since 70's). It is proved that each such ring has only finitely many maximal left denominator sets. Key Words: Goldie's Theorem, the left quotient ring of a ring, the largest left quotient ring of a ring, a maximal left denominator set, the prime radical. Mathematics subject classification 2010: 15P50, 16P60, 16P20, 16U20. 1. Introduction. Contents 2. Properties of rings with a left Noetherian left quotient ring. 3. Proof of Theorem 1.2. and the Second Criterion (via the associated graded ring). 4. Finiteness of max.Denl(R) when Q(R) is a left Noetherian ring. 1 Introduction In this paper, module means a left module, and the following notation is fixed: • R is a ring with 1; • C = CR is the set of regular elements of the ring R (i.e. C is the set of non-zero-divisors of the ring R); • Q = Q(R) := Ql,cl(R) := C−1R is the left quotient ring (the classical left ring of fractions) of the ring R (if it exists) and Q∗ is the group of units of Q; • n = nR is the prime radical of R, ν ∈ N ∪ {∞} is its nilpotency degree (nν 6= 0 but nν+1 = 0) and Ni := ni/ni+1 for i ∈ N; • R := R/n and π : R → R, r 7→ r = r + n; • C := CR is the set of regular elements of the ring R and Q := C • eC := π(C), eQ := eC−1R and C† := C eQ is the set of regular elements of the ring eQ; • Sl = Sl(R) is the largest left Ore of R that consists of regular elements and Ql = Ql(R) := R is its left quotient ring; −1 Sl(R)−1R is the largest left quotient ring of R [2, Theorem 2.1]; • Orel(R) := {S S is a left Ore set in R}; • Denl(R) := {S S is a left denominator set in R}. Criteria for a ring to have a left Noetherian left quotient ring. The aim of the paper is to give two criteria for a ring R to have a left Noetherian left quotient ring (Theorem 1.2 and Theorem 1.3). The case when R is a semiprime ring is a very easy special case. Theorem 1.1 Let R be a semiprime ring. Then the following statements are equivalent. 1 1. Q(R) is a left Noetherian ring. 2. Q(R) is a semisimple ring. 3. R is a semiprime left Goldie ring. 4. Ql(R) is a left Noetherian ring. 5. Ql(R) is a semisimple ring. If one of the equivalent conditions holds then Sl(R) = CR and Q = Ql(R). In particular, if the left quotient ring Q (resp. Ql(R)) is not a semisimple ring then the ring Q (resp. Ql(R)) is not left Noetherian. dx ,R i of polynomial integro-differential operators over a field K Example, [1]. The ring I1 := Khx, d of characteristic zero is a semiprime ring but not left Goldie (as it contains infinite direct sums of non-zero left ideals). Therefore, the largest left quotient ring Ql(I1) is not a left Noetherian ring (moreover, the left quotient ring Q(I1) does not exists). The ring Ql(I1) and Sl(I1) were described explicitly in [1]. The first criterion for a ring to have a left Noetherian left quotient ring is below, its proof is given in Section 3. Theorem 1.2 Let R be a ring. The following statements are equivalent. 1. The left quotient ring Q(R) of R is a left Noetherian ring. 2. (a) eC ⊆ C. (b) eC ∈ Orel(R). (c) eQ = eC−1R is a left Noetherian ring. (d) n is a nilpotent ideal of the ring R. degree of n and Ni := ni/ni+1). (e) The eQ-modules eC−1Ni, i = 1, . . . , ν, are finitely generated (where ν is the nilpotency (f ) For each element c ∈ eC, the left R-module Ni/Nic is eC-torsion for i = 1, . . . , ν. Remark. The conditions (a) and (b) above imply that the set eC is a left denominator set of the ring R, and so the ring eQ exists. The powers of the prime radical n, {ni}i≥0, form a descending filtration on R. Let gr R = R ⊕ n/n2 ⊕ · · · be the associated graded ring. The second criterion is given in terms of properties of the ring gr R, its proof is given in Section 3. Theorem 1.3 Let R be a ring. The following statements are equivalent. 1. The ring R has a left Noetherian left quotient ring Q. 2. The set eC is a left denominator set of the ring gr R, eC ⊆ C, eC−1gr R is a left Noetherian ring and n is a nilpotent ideal. If one of the equivalent conditions holds then gr Q ≃ eC−1gr R where gr Q := eQ ⊕ nQ/n2 Q ⊕ · · · is the associated graded ring with respect to the prime radical filtration. In particular, the ring gr Q is a left Noetherian ring. Finiteness of the set of maximal left denominators for a ring with a left Noetherian left quotient ring. For a ring R, the set max.Denl(R) of maximal left denominator sets (with respect to ⊆) is a non-empty set, [2]. It was proved that the set max.Denl(R) is a finite set if the left quotient ring Q(R) of R is a semisimple ring, [3], or a left Artinian ring, [5]. The next theorem extends this result for a larger class of rings that includes the class of left Noetherian ring for which the left quotient ring exists. 2 Theorem 1.4 Let R be a ring such that its left quotient ring Q(R) is a left Noetherian ring. Then i=1 Qi and Qi max.Denl(R) < ∞. Moreover, max.Denl(R) ≤ s = max.Denl(R) where Q ≃ Qs are simple Artinian rings (see Theorem 2.4.(3)). The next corollary follows at once from Theorem 1.4. Corollary 1.5 If R is a left Noetherian ring such that its left quotient ring Q(R) exists (i.e. C ∈ Orel(R)) then max.Denl(R) < ∞. The next corollary is an explicit description of the set max.Denl(R) for a ring R with a left Noetherian left quotient ring Q(R). Corollary 1.6 Let R be a ring such that its left quotient ring Q(R) is a left Noetherian ring. For ∗ each i = 1, . . . , s, let pi : R → Qi be the natural projection (see (8) and Theorem 1.4), Q i be the group of units of the simple Artinian ring Qi, S′ i be the largest element (w.r.t. ⊆) of the set ∗ Di = {S′ ∈ Denl(R) pi(S′) ⊆ Q i }. Then 1. max.Denl(R) is the set of maximal elements (w.r.t. ⊆) of the set {S′ 1, . . . , S′ s}. 2. For all i = 1, . . . , s, C ⊆ S′ i. 3. The rings S′−1 i R are left Noetherian where i = 1, . . . , s. A criterion for a left Noetherian ring to have a left Noetherian left quotient ring. In [7], Goldie posed a problem of deciding when a Noetherian ring possesses a Noetherian quotient ring. In [13], Stafford obtained a criterion to determine when a Noetherian ring is its own quotient ring. In [6], Chatters and Hajarnavis obtained necessary and sufficient conditions for a Noetherian ring which is a finite module over its centre to have a quotient ring. The next corollary follows from Theorem 1.2, Theorem 1.3 and the fact that the prime radical of a left Noetherian ring is a nilpotent ideal. Corollary 1.7 Let R be a left Noetherian ring. The following statements are equivalent. 1. The set C = CR is a left Ore set (i.e. the left quotient ring Q(R) = C−1R of R exists). 2. eC ∈ Denl(R, 0) and for each element c ∈ eC the left R-module Ni/Nic is eC-torsion for i = 3. eC ⊆ C and eC ∈ Den(gr R). 1, . . . , ν. Proof. (1 ⇔ 2) and (1 ⇔ 3) follow from Theorem 1.2 and Theorem 1.3, respectively. (cid:3) • Corollary 2.5 is a criterion for a ring to have a left Noetherian left quotient ring Q such that the factor ring Q/nQ is a semisimple ring (or eQ is a semisimple ring; or eC = C; or C = π−1(C)). • Theorem 4.2 is a criterion for a ring R to have a left Noetherian ring such that max.Denl(R) = max.Denl(R) (recall that, in general, max.Denl(R) ≤ max.Denl(R), Theorem 1.4). The set of maximal denominator sets for a commutative ring. The next theorem describes the set of maximal denominator sets (i.e. maximal Ore sets) for a commutative ring. Theorem 1.8 Let R be a commutative ring and Min(R) be the set of minimal prime ideals of the ring R. Then max.Den(R) = {Sp := R\p p ∈ Min(R)}. The paper is organized as follows. In Section 2, many properties of a ring R with a left Noetherian left quotient ring are proven (Theorem 2.4). In particular, the implication (1 ⇒ 2) of Theorem 2.4 is proven. In Section 3, proofs of Theorem 1.2 and Theorem 1.3 are given. In Section 4, Theorem 1.4 is proven. 3 2 Properties of rings with a left Noetherian left quotient ring In this section, we establish many properties of rings with left Noetherian left quotient ring (The- orem 2.4). In particular, the implication (1 ⇒ 2) of Theorem 1.2 is proven (Theorem 2.4.(2)). A criterion is given (Corollary 2.5) for the ring eQ to be a semisimple ring or a left Artinian ring. At the beginning of the section, we collect necessary results that are used in the proofs of this paper. More results on localizations of rings (and some of the missed standard definitions) the reader can find in [8], [14] and [9]. In this paper the following notation will remain fixed: • Sa = Sa(R) = Sl,a(R) is the largest element of the poset (Denl(R, a), ⊆) and Qa(R) := a R is the largest left quotient ring associated with a, Sa exists (Theorem 2.1, Ql,a(R) := S−1 [2]); • In particular, S0 = S0(R) = Sl,0(R) is the largest element of the poset (Denl(R, 0), ⊆) and Ql(R) := S−1 0 R is the largest left quotient ring of R, [2]; The largest regular left Ore set and the largest left quotient ring of a ring. Let R be a ring. A multiplicatively closed subset S of R or a multiplicative subset of R (i.e. a multiplicative sub-semigroup of (R, ·) such that 1 ∈ S and 0 6∈ S) is said to be a left Ore set if it satisfies the left Ore condition: for each r ∈ R and s ∈ S, SrT Rs 6= ∅. Let Orel(R) be the set of all left Ore sets of R. For S ∈ Orel(R), ass(S) := {r ∈ R sr = 0 for some s ∈ S} is an ideal of the ring R. A left Ore set S is called a left denominator set of the ring R if rs = 0 for some elements r ∈ R and s ∈ S implies tr = 0 for some element t ∈ S, i.e. r ∈ ass(S). Let Denl(R) be the set of all left denominator sets of R. For S ∈ Denl(R), let S−1R = {s−1r s ∈ S, r ∈ R} be the left localization of the ring R at S (the left quotient ring of R at S). In Ore's method of localization one can localize precisely at left denominator sets. In general, the set C of regular elements of a ring R is neither left nor right Ore set of the ring R and as a result neither left nor right classical quotient ring (Ql,cl(R) := C−1R and Qr,cl(R) := RC−1) exists. There exists the largest (w.r.t. ⊆) regular left Ore set S0 = Sl,0 = Sl,0(R), [2]. This means that the set Sl,0(R) is an Ore set of the ring R that consists of regular elements (i.e., Sl,0(R) ⊆ C) and contains all the left Ore sets in R that consist of regular elements. Also, there exists the largest regular right (resp. left and right) Ore set Sr,0(R) (resp. Sl,r,0(R)) of In general, all the sets C, Sl,0(R), Sr,0(R) and Sl,r,0(R) are distinct, for example, the ring R. when R = I1 = Khx, ∂,R i is the ring of polynomial integro-differential operators over a field K of characteristic zero, [1]. In [1], these four sets are found for R = I1. Definition, [1], [2]. The ring Ql(R) := Sl,0(R)−1R (respectively, Qr(R) := RSr,0(R)−1 and Q(R) := Sl,r,0(R)−1R ≃ RSl,r,0(R)−1) is called the largest left (respectively, right and two-sided) quotient ring of the ring R. In general, the rings Ql(R), Qr(R) and Q(R) are not isomorphic, for example, when R = I1, [1]. Small and Stafford [12] have shown that any (left and right) Noetherian ring R possesses a uniquely determined set of prime ideals P1, . . . , Pn such that CR = ∩n i=1C(Pi), an irreducible intersection, where C(Pi) := {r ∈ R r + Pi ∈ CR/Pi }. Michler and Muller [10] mentioned that the ring R contains a unique maximal (left and right) Ore set of regular elements Sl,r,0(R) and called the ring Q(R) the total quotient ring of R. For certain Noetherian rings, they described the set Sl,r,0(R) and the ring Q(R). For the class of affine Noetherian PI-rings, further generalizations were given by Muller in [11]. The next theorem gives various properties of the ring Ql(R). In particular, it describes its group of units. 4 Theorem 2.1 ([2].) 1. S0(Ql(R)) = Ql(R)∗ and S0(Ql(R)) ∩ R = S0(R). 2. Ql(R)∗ = hS0(R), S0(R)−1i, i.e. the group of units of the ring Ql(R) is generated by the sets S0(R) and S0(R)−1 := {s−1 s ∈ S0(R)}. 3. Ql(R)∗ = {s−1t s, t ∈ S0(R)}. 4. Ql(Ql(R)) = Ql(R). The maximal denominator sets and the maximal left localizations of a ring. The set (Denl(R), ⊆) is a poset (partially ordered set). In [2], it is proved that the set max.Denl(R) of its maximal elements is a non-empty set. Definition, [2]. An element S of the set max.Denl(R) is called a maximal left denominator set of the ring R and the ring S−1R is called a maximal left quotient ring of the ring R or a maximal left localization ring of the ring R. The intersection lR := l.lrad(R) := \ ass(S) S∈max.Denl(R) is called the left localization radical of the ring R, [2]. For a ring R, there is a canonical exact sequence 0 → lR → R σ→ Y S−1R, σ := Y σS, S∈max.Denl(R) S∈max.Denl(R) where σS : R → S−1R, r 7→ r 1 . (1) (2) Properties of the maximal left quotient rings of a ring. The next theorem describes various properties of the maximal left quotient rings of a ring, in particular, their groups of units and their largest left quotient rings. Theorem 2.2 ([2].) Let S ∈ max.Denl(R), A = S−1R, A∗ be the group of units of the ring A; a := ass(S), πa : R → R/a, a 7→ a + a, and σa : R → A, r 7→ r 1 . Then 1. S = Sa(R), S = π−1 a (S0(R/a)), πa(S) = S0(R/a) and A = S0(R/a)−1R/a = Ql(R/a). 2. S0(A) = A∗ and S0(A) ∩ (R/a) = S0(R/a). 3. S = σ−1 a (A∗). 4. A∗ = hπa(S), πa(S)−1i, i.e. the group of units of the ring A is generated by the sets πa(S) and π−1 a (S) := {πa(s)−1 s ∈ S}. 5. A∗ = {πa(s)−1πa(t) s, t ∈ S}. 6. Ql(A) = A and Assl(A) = {0}. In particular, if T ∈ Denl(A, 0) then T ⊆ A∗. A bijection between max.Denl(R) and max.Denl(Ql(R)). The next theorem shows that there is a canonical bijection between the maximal left denominator sets of a ring R and its largest left quotient ring Ql(R). Proposition 2.3 [3] Let R be a ring, Sl be the largest regular left Ore set of the ring R, Ql := S−1 l R be the largest left quotient ring of the ring R, and C be the set of regular elements of the ring R. Then 5 1. Sl ⊆ S for all S ∈ max.Denl(R). In particular, C ⊆ S for all S ∈ max.Denl(R) provided C is a left Ore set. 2. Either max.Denl(R) = {C} or, otherwise, C 6∈ max.Denl(R). 3. The map max.Denl(R) → max.Denl(Ql), S 7→ SQ∗ l = {c−1s c ∈ Sl, s ∈ S}, is a bijection with the inverse T 7→ σ−1(T ) where σ : R → Ql, r 7→ r sub-semigroup of (Ql, ·) generated by the set S and the group Q∗ S−1R = (SQ∗ is the l of units of the ring Ql, and 1 , and SQ∗ l l )−1Ql. 4. If C is a left Ore set then the map max.Denl(R) → max.Denl(Q), S 7→ SQ∗ = {c−1s c ∈ C, s ∈ S}, is a bijection with the inverse T 7→ σ−1(T ) where σ : R → Q, r 7→ r 1 , and SQ∗ is the sub-semigroup of (Q, ·) generated by the set S and the group Q∗ of units of the ring Q, and S−1R = (SQ∗)−1Q. A ring R is called a left Goldie ring if it satisfies ACC (the ascending chain condition) for left annihilators and contains no infinite direct sums of left ideals. Theorem 2.4 Let R be a ring such that its left quotient ring Q is a left Noetherian ring. Let σ : R → Q, r 7→ r 1 , (Q/nQ)∗ be the group of units of the ring Q/nQ, and σ′ : Q/nQ → Q(Q/nQ), q + nQ 7→ q 1 + nQ. Then 1. n = R ∩ nQ, C−1n = nQ, (C−1n)i = C−1ni for all i ≥ 1, and ν = νQ < ∞ where ν and νQ are the nilpotency degrees of the prime radicals n and nQ, respectively. 2. (a) C + n ⊆ C. (b) eC ∈ Denl(R, 0). In particular, eC ⊆ C. (c) eQ := eC−1R ≃ Q/nQ is a semiprime left Noetherian ring. (d) n is a nilpotent ideal of the ring R. tency degree of n). (e) The eQ-modules eC−1(ni/ni+1), i = 1, . . . , ν, are finitely generated (where ν is the nilpo- (f ) For each elements c ∈ eC, the left R-module Ni/Nic is eC-torsion where Ni := ni/ni+1. 3. The ring R is a semiprime left Goldie ring and its left quotient ring Q := Q(R) ≃ Q(Q/nQ) ≃ Q(eQ) is a semisimple ring. 4. 1 → 1 + nQ → Q∗ π∗ Q→ (Q/nQ)∗ → 1 is a short exact sequence of group homomorphisms where πQ : Q → Q/nQ, q 7→ q + nQ and π∗ Q := πQQ∗ . see (4). 5. C = σ−1(Q∗) = (πQσ)−1((Q/nQ)∗) = π−1(eσ−1((Q/nQ)∗)) where eσ : R → Q/nQ, r 7→ r 6. Let C† := C eQ, i.e. C† = CQ/nQ when we identify the rings eQ and Q/nQ via the isomorphism in statement 2(c). Then C† = σ′−1(Q(Q/nQ)∗) and C = eσ−1(C†) = (σ′eσ)−1(Q(Q/nQ)∗). 1 +nQ, Proof. 1. The prime radical nQ is a nilpotent ideal since the ring Q is a left Noetherian ring. Then the intersection R ∩ nQ is a nilpotent ideal of the ring R, hence R ∩ nQ ⊆ n. To establish the equality R ∩ nQ = n it suffices to show that the factor ring R/R ∩ nQ has no nonzero nilpotent ideals. Suppose that I is a nilpotent ideal of the ring R/R ∩ nQ then its preimage I under the epimorphism R → R/R ∩ nQ, r 7→ r + R ∩ nQ, is a nilpotent ideal of the ring R since the ideal 6 R ∩ nQ is a nilpotent ideal. We have to show that I = 0. The left ideal C−1I of Q is an ideal of the ring Q since the ring Q is a left Noetherian ring. Then IC−1 := {ic−1 i ∈ I, c ∈ C} ⊆ C−1I, and so (C−1I)i ⊆ C−1I i, i ≥ 1. (3) So, C−1I is a nilpotent ideal of Q, and so C−1I ⊆ nQ and I ⊆ R ∩ C−1I ⊆ R ∩ nQ. Therefore, I = 0, as required. Clearly, C−1n = C−1(R ∩ nQ) = C−1R ∩ C−1nQ = Q ∩ nQ = nQ, and (C−1n)i = C−1ni for i ≥ 1. In particular, ν = νQ < ∞. 2(a). Let c ∈ C and n ∈ n. Then the element c−1n ∈ C−1n = nQ (statement 1) is a nilpotent element of the ring Q and so the element 1 + c−1n is a unit of the ring Q. Now, 2(b,c) Since n = R ∩ nQ (statement 1), there is a commutative diagram of ring homomorphisms c + n = c(1 + c−1n) ∈ C. Q σ R πQ / Q/nQ eσ R π (4) where the horizontal maps are natural epimorphisms and the vertical maps are natural monomor- phisms (where eσ(r) := r 1 = r 1 + nQ). Since Q/nQ = {π(c)−1r c ∈ C, r ∈ R}, we see that eC = π(C) ∈ Denl(R, 0) and Q/nQ ≃ eC−1R is a semiprime left Noetherian ring. 2(d) Statement 1. 2(e) By the statement (c), eQ ≃ Q/nQ is a left Noetherian ring. Hence, the Q/nQ-modules Q/ni+1 ni Q ≃ (C−1n)i/(C−1n)i+1 st.1 ≃ C−1ni/C−1ni+1 ≃ eC−1(ni/ni+1) are finitely generated where i = 1, . . . , ν. 2(f) For each i = 1, . . . , ν, the left Q-module/eQ-module Ni/Nic is eC-torsion since eC−1(Ni/Nic) = eC−1(ni/(nic + ni+1)) = C−1(ni/(nic + ni+1)) = C−1ni/(C−1nic + C−1ni+1) = C−1ni/C−1ni = 0. 4. Statement 4 follows from the fact that nQ is a nilpotent ideal of the ring Q. 3. By statement 2(c), the ring eQ ≃ Q/nQ is a semiprime left Noetherian ring. In particular, it is a semiprime left Goldie ring and, by Goldie's Theorem, its left quotient ring Q(eQ) ≃ Q(Q/nQ) is a semisimple ring. Since R ⊆ Q/nQ ≃ eC−1R (statement 2(c)) we have Q(R) = Q(eC−1R) is a semisimple ring. By Goldie's Theorem, the ring R is a semiprime left Goldie ring. So, we can extend the commutative diagram (4) to the commutative diagram (which is used in the proof of statement 6) Q σ R πQ / Q/nQ eσ R π σ′ σ / Q(Q/nQ) ≃ / Q (5) where the maps σ′ and σ are monomorphisms, σ′(q + nQ) = q+nQ 1 and σ(r) = r 1 . 7 / / / O O O O / / / / O O O O / 5. By Theorem 2.1.(1), C = σ−1(Q∗). By statement 4, Q∗ = π−1 Q ((Q/nQ)∗). Then, in view of the commutative diagram (4), C = σ−1π−1 Q ((Q/nQ)∗) = (πQσ)−1((Q/nQ)∗) = (eσπ)−1((Q/nQ)∗) = π−1(eσ−1((Q/nQ)∗))). 6. By Theorem 2.1.(1), C† = σ′−1(Q(Q/nQ)∗) and C = σ−1(Q ∗ ) = σ−1(Q(Q/nQ)∗) (statement 3). Thus, the commutativity of the second square in the diagram (5) yields, C = (σ′eσ)−1(Q(Q/nQ)∗) = eσ−1(C†). (cid:3) The next corollary is a criterion for a ring to have a left Noetherian left quotient ring Q such that the factor ring Q/nQ is a semisimple ring (or eQ is a semisimple ring; or eC = C; or C = π−1(C)). Corollary 2.5 Let R be a ring such that its left quotient ring Q is a left Noetherian ring, we keep the notation of Theorem 2.4 and its proof. The following statements are equivalent (recall that eQ = Q/nQ, Theorem 2.4.(2c)). 1. eQ is a semisimple ring. 2. eQ = Q(eQ). 3. C = eσ−1(eQ∗). 4. C = π−1(C). 5. eC = C. 6. eQ is a left Artinian ring. Proof. (1 ⇒ 2) Trivial. (2 ⇒ 3) If eQ = Q(eQ), i.e. the map σ′ in (5) is an isomorphism, then the rings eQ ≃ Q/nQ and ) = eσ−1(eQ∗) where the Q are isomorphic, see the commutative diagram (5). Now, C = σ−1(Q first equality holds by Theorem 2.1.(1). ∗ (3 ⇒ 4) By Theorem 2.4.(5). C = π−1(eσ−1(eQ∗)) = π−1(C). (4 ⇒ 5) eC = π(C) = π(π−1(C)) = C since the map π is an epimorphism. (5 ⇒ 1) If eC = C then by (5), eQ ≃ Q is a semisimple ring, by Theorem 2.4.(3). (1 ⇒ 6) Trivial. (6 ⇒ 1) This implication follows from Theorem 2.4.(2c). (cid:3) 3 Proof of Theorem 1.2 and a Second Criterion (via the associated graded ring) The aim of this section is to give proofs of Theorem 1.2 and Theorem 1.3 which are criteria for a ring R to have a left Noetherian left quotient ring. Proof of Theorem 1.1. The implications (1 ⇐ 2) and (4 ⇐ 5) are obvious and the equiva- lence (2 ⇔ 3) is Goldie's Theorem. (1 ⇒ 2) Since R is a semiprime ring then so is the ring Q = Q(R). In particular, the ring Q is a semiprime left Goldie ring, hence, by Goldie's Theorem, Q(Q(R)) = Q(R) is a semisimple ring. (2 ⇔ 5) [2, Theorem 2.9]. (4 ⇒ 2) Since R is a semiprime ring then so is the ring Ql(R) (since the ring Ql(R) is left Noetherian). The ring Ql(R) is a left Noetherian ring, hence Ql(R) is a semiprime left Goldie ring. By Goldie's Theorem, Q(Ql(R)) is a semisimple ring. Then, by [2, Theorem 2.9], Q(Ql(R)) = Ql(Ql(R)) is a semisimple ring. By [2, Theorem 2.8.(4)], Ql(Ql(R)) = Ql(R), and so 8 Ql(R) is a semisimple ring. Then, by [2, Theorem 2.9], Q(R) = Ql(R), i.e. Q(R) is a semisimple ring. (cid:3) Proof of Theorem 1.2. (1 ⇒ 2) Theorem 2.4.(1,2). (1 ⇐ 2) (i) C ∈ Orel(R): We have to show that for given elements c ∈ C and r ∈ R there are elements c′ ∈ C and r′ ∈ R such that c′r = r′c. We can assume that r 6= 0 since otherwise take c′ = 1 and r′ = 0. To prove this fact we use a downward induction on the degree of the element r 6= 0: deg(r) := max{i r ∈ ni where 0 ≤ i ≤ ν}. Suppose that deg(r) = ν. By the condition (f), the R-module nν/nνc is eC-torsion, hence c′r = r′c for some elements c′ = c + n ∈ eC, c ∈ C and r′ ∈ nν. Then c′r = r′c. Suppose that i = deg(r) < ν, and the result is true for all elements c ∈ C and r ∈ R with deg(r) > i. Suppose that i = 0, i.e. r ∈ R\n. Since eC ∈ Denl(R, 0) (by the conditions (a) and (b)), c1r = r1c for some elements c1 ∈ C and r1 ∈ R. The difference a := c1r − r1c belongs to the ideal n, and so deg(a) > 0. By induction, c2a = bc for some elements c2 ∈ C and b ∈ R. Then c2c1r = (c2r1 + b)c, and it suffices to take c′ = c2c1 and r′ = c2r1 + b. Suppose that i > 0. By the condition (f), the left R-module Ni/Nic = ni/(nic + ni+1) is eC- torsion. Therefore, sr = xc + y for some elements s ∈ C, x ∈ ni and y ∈ ni+1. Since deg(y) ≥ i + 1, by induction, there are elements t ∈ C and z ∈ R such that ty = zc. Therefore, tsr = (tx + z)c and it suffices to take c′ = ts and r′ = tx + z. (ii) C−1n is an ideal of the ring Q such that Q/C−1n ≃ eQ: By (i), the left quotient ring Q = C−1R exists. Let σ : R → Q, r 7→ r 1 . By the universal property of left localization, there is a ring homomorphism πQ : Q → eQ, c−1r 7→ c−1r, where c = c + n and r = r + n, and we have the commutative diagram of ring homomorphisms Q σ πQ / eQ eσ R π R (6) where eσ : r 7→ r Applying the exact functor C−1(−) to the short exact sequence of R-modules 0 → n → R we obtain the short exact sequence of Q-modules 1 is a monomorphism and πQ is an epimorphism (by the very definition of πQ). π→ R → 0 0 → C−1n → Q πQ→ C−1R = eC−1R = eQ → 0. Therefore, ker(πQ) = C−1n is an ideal of Q (since πQ is a ring homomorphism) such that Q/C−1n ≃ eQ. (iii) (C−1n)i = C−1ni for i ≥ 1: This follows from (ii). (iv) The ring Q is a left Noetherian ring: By localizing the descending chain of ideals of the ring R: R ⊃ n ⊃ n2 ⊃ · · · ⊃ ni ⊃ · · · ⊃ nν ⊃ nν+1 = 0 we obtain the descending chain of ideals (by (iii)) of the ring Q: Q ⊃ C−1n ⊃ C−1n2 ⊃ · · · ⊃ C −1ni ⊃ · · · ⊃ C −1nν ⊃ C−1nν+1 = 0. (7) 9 / / / O O O O By (ii), eQ ≃ Q/C−1n is a left Noetherian Q-module since the ring eQ is a left Noetherian ring, by the condition (c). For each i = 1, . . . , ν, the left Q-module, C −1ni/C−1ni+1 ≃ (C−1n)i/(C−1n)i+1 is a Q/C−1n = eQ-module, by (ii). The eQ-modules C−1ni/C−1ni+1 ≃ C−1(ni/ni+1) ≃ eC−1(ni/ni+1) = eC−1Ni are finitely generated, by the condition (e), hence Noetherian since the ring eQ is a left Noetherian. Since all the factors of the finite filtration (7) are Noetherian eQ-modules/Q-modules, the ring Q is a left Noetherian ring. (cid:3) The minimal primes of the rings R, Q(R), Q and eQ. For a ring R such that Q(R) is a left Noetherian ring, the next corollary shows that the localizations of R at the maximal left denominator sets are left Noetherian rings and there are natural bijections between the sets of minimal primes of the rings R, Q(R), Q and eQ. Corollary 3.1 Let R be a ring such that Q(R) is a left Noetherian ring. Then 1. For every S ∈ max.Denl(R), the ring S−1R is a left Noetherian ring. 2. (a) The map Min(R) → Min(Q), p 7→ C −1 (p/n), is a bijection with the inverse q 7→ (σπ)−1(q) where the maps σ and π are defined in (8). (b) The map Min(R) → min(eQ), p 7→ eC−1(p/n), is a bijection with the inverse q 7→ τ −1(q) (c) The map Min(R) → Min(Q), p 7→ C −1p, is a bijection with the inverse q 7→ q ∩ R. where τ : R → eQ, r 7→ r+p 1 . Proof. 1. By Proposition 2.3.(1), C = CR ⊆ S for all S ∈ max.Denl(R). The ring Q := C−1R is a left Noetherian ring and the ring S−1R ≃ (SQ∗)−1Q is a left localization of the ring Q (by Proposition 2.3.(3), the map max.Denl(R) → max.Denl(Q), S 7→ SQ∗, is a bijection with S−1R ≃ (SQ∗)−1Q). Therefore, the ring S−1R is a left Noetherian ring since the ring Q is so. 2(a). The map Min(R) → Min(R), p 7→ p/n, is a bijection with the inverse p′ 7→ π−1(p′) where π : R → R, r 7→ r + p. The ring R is a semiprime left Goldie ring such that Q ≃ Q(Q/nQ) is a semisimple ring. Hence, the map min(R) → Min(Q), p′ 7→ C p′, is a bijection with the inverse q 7→ σ−1(q) where σ : R → Q, r 7→ r 1 . Now, the statement (a) follows. −1 bijection. Now, the statement (c) follows from the statement (b). (cid:3) (b) The ring eQ is a semiprime left Goldie ring (by Theorem 2.4.(2c)) and Q(eQ) ≃ eQ is a semisimple ring (Theorem 2.4.(3)). So, the map Min(eQ) → Min(Q), P 7→ Q ⊗ eQ P , is a bijection with the inverse P ′ 7→ P ′ ∩ eQ. Now, the statement (b) follows from the statement (a). (c) By Theorem 2.4.(2c), eQ ≃ Q/nQ and the map Min(Q) → Min(Q/nQ), P 7→ P/nQ, is a Lemma 3.2 Let G be a monoid and e ∈ G be its neutral element, A = Lg∈G Ag be a G-graded ring, 1 ∈ Ae, S ∈ Denl(A) and a = ass(S). If S ⊆ Ae then the ring S−1A = Lg∈G(S−1A)g is a G-graded ring where (S−1A)g = S−1Ag := {s−1ag s ∈ S, ag ∈ Ag}, S ∈ Denl(Ae) and a = Lg∈G ag is a G-graded ideal of the ring A, i.e. ag = a ∩ Ag for all g ∈ G. Proof. Straightforward. (cid:3) Suppose that a ring R has a left Noetherian left quotient ring Q. By Theorem 2.4, the associated graded ring gr Q := Q/nQ ⊕ nQ/n2 Q ⊕ · · · is equal to gr Q = eQ ⊕ eC−1(n/n2) ⊕ · · · ⊕ eC−1(nν /nν+1). Proof of Theorem 1.3. (1 ⇒ 2) Suppose that the ring Q is a left Noetherian ring, and so the conditions of Theorem 1.2 and Theorem 2.4 hold. In particular, eC ⊆ C and n is a nilpotent ideal. 10 (i) eC ∈ Orel(gr R): It suffices to show that for given elements c = c+n ∈ eC and r+ni+1 ∈ ni/ni+1 where c ∈ C, r ∈ ni and i = 0, 1, . . . , ν, there are elements c′ = c′ + n ∈ eC and r′ + ni+1 ∈ ni/ni+1 where c′ ∈ C, r′ ∈ ni such that c′(r + ni+1) = (r′ + ni+1)c. The case i = 0 is obvious, by Theorem 1.2.(2b). So, we can assume that i ≥ 1. By Theorem 2.4.(1), niC−1 := {nc−1 n ∈ ni, c ∈ C} ⊆ C−1ni. So, rc−1 = c′−1r′ for some elements c′ ∈ C and r′ ∈ ni, hence c′r = r′c. This equality implies the required one. (ii) eC ∈ Denl(gr R): We have to show that if rc = 0 for some elements r = r + ni+1 ∈ ni/ni+1 and c = c + n where r ∈ ni, c ∈ C and i = 0, 1, . . . , ν, then c1r = 0 for some element c1 = c1 + n where c1 ∈ C. The case i = 0 follows from Theorem 2.4.(2b). Suppose that i ≥ 1. Then n := rc ∈ ni+1, and so r = nc−1 = c−1 1 n1 for some elements c1 ∈ C and n1 ∈ ni+1, by Theorem 2.4.(2c). Hence, c1r = 0, as required. (iii) gr Q ≃ eC−1gr R: By Theorem 2.4.(1), nQ = C−1n. By Theorem 2.4.(2c), Q/nQ ≃ eQ. Now, using Theorem 2.4.(1,2), we have gr Q = eQ ⊕ · · · ⊕ ni Q/ni+1 Q ⊕ · · · = eQ ⊕ · · · ⊕ C −1ni/C−1ni+1 ⊕ · · · = eQ ⊕ · · · ⊕ C −1(ni/ni+1) ⊕ · · · = eQ ⊕ · · · ⊕ eC−1(ni/ni+1) ⊕ · · · ≃ eC−1gr R. (iv) eC−1gr R is a left Noetherian ring: The ring eQ is a left Noetherian ring (Theorem 2.4.(2c)) and the left eQ-modules eC−1(ni/ni+1) ≃ (C−1n)i/(C−1n)i+1 are finitely generated where i = 1, . . . , ν (since Q is a left Noetherian ring). Therefore, the left eQ-module gr Q = eQ ⊕ · · · ⊕ eC−1(ni/ni+1) ⊕ · · · ⊕ eC−1(nν/nν+1) is finitely generated, hence Noetherian. Since eQ ⊆ gr Q, the ring gr Q is a left Noetherian ring. (1 ⇐ 2) It suffices to show that the conditions (a)-(f) of Theorem 1.2.(2) hold. The conditions the condition (e) holds. ring, i.e. the condition (c) holds. (a) and (d) are given. The set eC is a left denominator set of the N-graded ring gr R such that eC ⊆ R. By Lemma 3.2, the ring eC−1gr R = eQ ⊕ · · · ⊕ eC−1Ni ⊕ · · · is an N-graded ring and eC ∈ Denl(R, 0), and so the condition (b) holds. The ring eQ is a factor ring of the left Noetherian ring eC−1gr R, hence eQ is a left Noetherian The ring eC−1gr R is left Noetherian, hence the eQ-modules eC−1Ni are finitely generated, i.e. The ring eC−1gr R is an N-graded ring. In particular, eC−1Ni eQ ⊆ eC−1Ni for all i. Therefore, NieC−1 = {nc−1 n ∈ Ni, c ∈ eC} ⊆ eC−1Ni, i.e. the condition (f) holds. (cid:3) Corollary 3.3 Let R be a ring with a left Noetherian left quotient ring Q and ea := assgr R(eC). Then the largest left quotient ring Ql(gr R/ea) of the ring gr R/ea is a left Noetherian ring. Proof. By Theorem 1.3.(2), the ring eC−1gr R is a left Noetherian ring. The ring Ql(gr R/ea) is a left localization of the ring eC−1gr R, hence is a left Noetherian ring. (cid:3) Proposition 3.4 Let R be a ring and S ∈ Denl(R). Then the ring R is a left Noetherian ring iff the ring S−1R is a left Noetherian ring and for each left ideal I of R the left R-module torS(R/I) is left Noetherian. Proof. (⇒) Trivial. (⇐) Let I be a left ideal of the ring R. We have to show that I is a finitely generated left R- module. The ring S−1R is a left Noetherian ring. Then its left ideal S−1I is a finitely generated S−1R-module. We can find elements u1, . . . , un ∈ I such that S−1I = Pn i=1 S−1Rui/1. Let I ′ := Pn i=1 Rui. Then I ′ ⊆ I and I/I ′ is a submodule of the left Noetherian R-module torS(R/I ′) and as a result the left R-module I/I ′ is finitely generated. This fact implies that the left ideal I is finitely generated. (cid:3) 11 Let R be a ring, S ∈ Orel(R) and M be a left R-module. Let ker(S, M ) = {ker(sM ) s ∈ S} where sM : M → M , m 7→ sm. The set (ker(S, M ), ⊆) is a poset. Let max ker(S, M ) be the set of its maximal elements and let max(S, M ) := {s ∈ S ker(sM ) ∈ max ker(S, M )}. Proposition 3.5 Let R be a ring, S ∈ Denl(R), M be a left R-module such that max ker(S, M ) 6= ∅. Then max ker(S, M ) = {torS(M )}, i.e. ker(sM ) = torS(M ) for all elements s ∈ max(S, M ). In particular, a := annR(torS(M )) 6= 0 and ∅ 6= max(S, M ) ⊆ S ∩ a. Proof. Suppose that ker(sM ) 6= torS(M ) for some s ∈ max(S, M ), we seek a contradiction. Fix m ∈ torS(M )\ker(sM ) and t ∈ S with tm = 0. Since S ∈ Orel(R), we have St ∩ Rs 6= ∅, i.e. s′ := s1t ∈ St ∩ Rs for some s1 ∈ S. Clearly, s′ ∈ S and ker(s′ M ) ⊇ ker(tM ) + ker(sM ) ' ker(sM ), a contradiction. (cid:3) Proposition 3.6 Let R be a ring, p1, . . . , pn be prime ideals of the ring R such that the rings R/pi (i = 1, . . . , n) are left Goldie rings. Let M be a left R-module such that M = Mn ⊃ Mn−1 ⊃ · · · ⊃ M1 ⊃ M0 = 0 is a chain of its submodules such that pi is maximal among left annihilators of non-zero submodules of M/Mi−1, Mi = annM/Mi−1 (pi) := {m ∈ M/Mi−1 pim = 0}, and max ker(CR/pi , R/Mi−1) 6= ∅ for i = 1, . . . , n. Then Tn i=1 C(pi) ⊆ ′CM (0) := {r ∈ R rM is an injection}. Proof. Let c ∈ Tn i=1 C(pi). We have to show that ker(cM ) = 0. Suppose that this is not true, i.e. cm = 0 for some 0 6= m ∈ M , we seek a contraction. Then m ∈ Mi\Mi−1 for some i. Then 0 6= m := m + Mi−1 ∈ τi := torCR/pi (Mi/Mi−1) since c := c + pi ∈ CR/pi. By Proposition 3.5, sτi = 0 for some s ∈ CR/pi (since max ker(CR/pi, R/Mi−1) 6= ∅), i.e. sτi = 0 for some s ∈ C(pi). This means that annR(τi) ' pi, a contradiction. (cid:3) 4 Finiteness of max.Denl(R) when Q(R) is a left Noetherian ring The aim of this section is to give a proof of Theorem 1.4. Let R be a ring. Let S and T be submonoids of the multiplicative monoid (R, ·). We denote by ST the submonoid of (R, ·) generated by S and T . This notation should not be confused with the product of two sets which is not used in this paper. The next result is a criterion for the set ST to be a left Ore (denominator) set, it is used at the final stage in the proof of Theorem 1.4. Lemma 4.1 ([4].) 1. Let S, T ∈ Orel(R). If 0 6∈ ST then ST ∈ Orel(R). 2. Let S, T ∈ Denl(R). If 0 6∈ ST then ST ∈ Denl(R). Proof of Theorem 1.4. Let S ∈ max.Denl(R) and π : R → R, r 7→ r := r + n. (i) S := π(S) ∈ Orel(R): This obvious since S ∩ n = ∅. (ii) C ⊆ S (Proposition 2.3.(1)). (iii) S−1R is a left Noetherian ring: By (ii), S−1R is a left localization of the left Noetherian ring Q, hence S−1R is a left Noetherian ring. (iv) S−1n is an ideal of S−1R such that (S−1n)i = S−1ni for all i ≥ 1: By (iii), S−1n is an ideal of the ring S−1R. In particular, nS−1 ⊆ S−1n, hence (S−1n)i = S−1ni for all i ≥ 1. (v) S ∈ Denl(R): In view of (i), we have to show that if rs = 0 for some r ∈ R and s ∈ S then tr = 0 for some element t ∈ S. The equality rs = 0 means that n := rs ∈ n. Then S−1R ∋ r 1 n1 for some s1 ∈ S and n1 ∈ n, (by (iv)). Hence, s2s1r = s2n1 ∈ n for some s2 ∈ S, and so tr = 0 where t = s2s1 ∈ S. 1 = ns−1 = s−1 The ring R is a semiprime left Goldie ring (Theorem 2.4.(3)), Q = Qs i=1 Qi and Qi are simple Artinian rings. By [3, Theorem 4.1], max.Denl(R) = s. Let max.Denl(R) = {T1, . . . , Ts}. 12 (vi) max.Denl(R) ≤ s: By (v), S ⊆ Ti for some i. Then S is the largest element with respect to inclusion of the set {S′ ∈ Denl(R) π(S′) ⊆ Ti} (as 0 ∈ SS1 for all distinct S, S1 ∈ max.Denl(R), by Lemma 4.1.(2), where SS1 is the multiplicative submonoid of R generated by S and S1). Hence, max.Denl(R) ≤ s. (cid:3) Let R be a ring such that its left quotient ring Q(R) is a left Noetherian ring. Define the ring homomorphism pi : R π→ R σ→ Q = sY j=1 Qi pi→ Qi. Let σi := piσ : R → Qi. By [3, Theorem 4.1], max.Denl(R) = {T1, . . . , Ts} where Ti := σ−1 {r ∈ R σi(r) ∈ Q i ∗ i }. (8) (Q ∗ i ) = i := SS ′∈Di Proof of Corollary 1.6. 1. The set Di is a non-empty set since R∗ ⊆ Di. By Lemma 4.1.(2), In the proof of Theorem 1.4, we 1, . . . , S′ s} the set S′ proved that max.Denl(R) ⊆ M where M is the set of maximal elements of the set {S′ (see the statement (vi)). The reverse inclusion is obvious. S′ is the largest element of the set Di. 2. By Theorem 2.4.(2b), eC ∈ Denl(R, 0) and eC ⊆ C. By Theorem 2.4.(3), the ring R is a i for all i = 1, . . . , s. semiprime left Goldie ring and Q is a semisimple ring. By Theorem 2.1.(1), C = R ∩ Q C ⊆ S′ . Hence, ∗ −1R is a localization of the left Noetherian 3. Statement 3 follows from statement 2: The ring S′ i ring C−1R (see statement 2). Hence it is a left Noetherian ring. (cid:3) Criterion for max.Denl(R) = max.Denl(R). Let R be a ring such that its left quotient ring Q(R) is a left Noetherian ring. In general, max.Denl(R) ≤ max.Denl(R), Theorem 1.4. The next theorem is a criterion for max.Denl(R) = max.Denl(R). Theorem 4.2 Let R be a ring such that its left quotient ring Q(R) is a left Noetherian ring. Then max.Denl(R) = max.Denl(R) iff for each pair of indices i 6= j where 1 ≤ i, j ≤ s = max.Denl(R) there exist Si, Sj ∈ Denl(R) such that 0 ∈ SiSj (where SiSj is the multiplica- ∗ tive submonoid of R generated by Si and Sj), pi(Si) ⊆ Q i and pj(Sj) ⊆ Q In this case, i is the largest element in the set {S′ ∈ Denl(R) pi(S′) ⊆ max.Denl(R) = {S′ Q s} where S′ 1, . . . , S′ ∗ j . ∗ i }. Proof. By [3, Theorem 4.1], max.Denl(R) = {T1, . . . , Ts} where Ti = (piσ)−1(Q ∗ i ) for i = 1, . . . , s. (⇒) It suffices to take Si = Ti, i = 1, . . . , s since 0 ∈ SiSj for all i 6= j, by Lemma 4.1.(2). (⇐) Suppose that S1, . . . , Ss are as in the theorem. For each i = 1, . . . , s, let S′ element with respect to inclusion of the set {S′ ∈ Denl(R) π(S′) ⊆ Ti}. Since Si ⊆ S′ Sj ⊆ S′ 4.1.(2)) and incomparable (i.e. S′ (vi), we have seen that max.Denl(R) ⊆ {S′ the sets S′ i be the largest i and s are distinct (Lemma j for all i 6= j). In the proof of Theorem 1.4, the statement s} since j for each distinct pair i 6= j and 0 ∈ SiSj, the elements S′ s}. Hence, max.Denl(R) = {S′ 1, . . . , S′ 1, . . . , S′ 1, . . . , S′ s are incomparable. (cid:3) 1, . . . , S′ i 6⊆ S′ The next theorem gives sufficient conditions for finiteness of the set of max.Denl(R). Theorem 4.3 Let R be a ring and I be an ideal of the ring Ql(R) such that S +I ⊆ S and S−1I is an ideal of S−1Ql(R) for all S ∈ max.Denl(Ql(R)). Then max.Denl(R) ≤ max.Denl(Ql(R)/I). In particular, if max.Denl(Ql(R)/I) < ∞ then max.Denl(R) < ∞. Proof. By Proposition 2.3.(4), there is a bijection between the sets max.Denl(R) and max.Denl(Ql(R)). Let Ql := Ql(R) and f : Ql → Ql := Ql/I, q 7→ q := q + I. (i) For all S ∈ max.Denl(Ql), S := f (S) ∈ Denl(Ql): Since S + I ⊆ S, S ∩ I = ∅ and S ∈ Orel(R). It remains to show that if qs = 0 for some elements q ∈ Ql and s ∈ S then tq = 0 13 for some element t ∈ S. The equality qs = 0 means that i := qs ∈ I. Then S−1Ql ∋ q 1 = is−1 = s−1 1 i1 for some elements s1 ∈ S and i1 ∈ I (since S−1R is an ideal of the ring S−1Ql). Hence, s2s1q = s2i ∈ I for some s2 ∈ S, and so tq = 0 where t = s2s1 ∈ S. (ii) max.Denl(R) ≤ max.Denl(Ql(R)/I): Let S ∈ max.Denl(Ql). By (i), S ⊆ T for some T = T (S) ∈ max.Denl(Ql). Then S is the largest element (w.r.t. ⊆) of the set {S′ ∈ Denl(R) f (S′) ⊆ T } (as 0 ∈ SS1 for all distinct S, S1 ∈ max.Denl(R), by Lemma 4.1.(2)), and the statement (ii) follows. (cid:3) Corollary 4.4 Let R be a ring and I be an ideal of Ql(R) such that S−1I is an ideal of the ring S−1Ql(R) with S−1I ⊆ rad(S−1Ql(R)). Then max.Denl(R) ≤ max.Denl(Ql(R)/I). Proof. In view of Theorem 4.3 and its proof, it suffices to show that S + I ⊆ S for all S ∈ max.Denl(Ql) where Ql = Ql(R). By the assumption, S−1I ⊆ rad(S−1Ql). Hence, 1+S−1I ⊆ (S−1Ql)∗. Now, for all s ∈ S and i ∈ I, s + i = s(1 + s−1i) ∈ (S−1Ql)∗. Let σS : Ql → S−1Ql, q 7→ q S ((S−1Ql)∗), and so s + i ∈ S, as required. (cid:3) 1 . By Theorem 2.1.(1), S = σ−1 Proof of Theorem 1.8 (description of the set max.Denl(R) for a commutative ring R). For each minimal prime ideal p of R, Sp ∈ max.Den(R) since the ring Rp := S−1 p R is a local ring such that its maximal ideal S−1p (which is also the prime radical of Rp) is a nil ideal, so every element of Rp is either a unit or a nilpotent element. Conversely, let S ∈ max.Den(R). We have to show that S = Sp for some p ∈ Min(R). In the ring S−1R, every element is either a unit or a nilpotent element. Hence, the ring S−1R is a local ring (S−1R, m) where the maximal ideal m is a nil ideal. Let σ : R → S−1R, r 7→ r 1 . Then S = σ−1((S−1R)∗) = Sp, by Theorem 2.2.(3). (cid:3) Proposition 4.5 Let R be a commutative ring, I be a nil ideal of R and πI : R → R/I, r 7→ r = r + I. Then 1. The map Den(R) → Den(R/I), S 7→ πI (S), is a surjection that respects inclusions. More- over, the map Den(R, I) := {S ∈ Den(R) S + I ⊆ S} → Den(R/I), S 7→ πI (S), is a bijection with the inverse T 7→ π−1 I (T ). 2. The map max.Den(R) → max.Den(R/I), S 7→ πI (S), is a bijection with the inverse T 7→ π−1 I (T ). Proof. 1. Straightforward. 2. Statement 2 follows statement 1 since if S ∈ max.Den(R) then S + I ⊆ S. (cid:3) Let R be a commutative ring such that its quotient ring Q = Q(R) is a Noetherian ring. We keep the notation of Theorem 3.5 and Theorem 1.2. For each i = 1, . . . , s, let pi : R → Qi be as in (8). The next theorem shows that the second inequality of Theorem 1.4 is an equality, it also provides another characterization of the set max.Den(R). Theorem 4.6 Let R be a commutative ring such that its quotient ring Q is a Noetherian ring. ∗ Then max.Den(R) = {Si i = 1, . . . , s} where Si := p−1 i ). In particular, max.Den(R) = s = max.Den(R). (Q i Proof. Notice that max.Den(Q) = {p−1 ∗ i ) i = 1, . . . , s}. By Proposition 2.3.(4), the map max.Den(Q) → max.Den(R), S 7→ σ−1(S), is a bijection. By Proposition 4.5.(2), the map max.Den(R) → max.Den(R), T 7→ π−1(T ), is a bijection. Therefore, max.Den(R) = {Si i = 1, . . . , s}. (cid:3) (Q i Proposition 4.7 Let R be a ring, S ∈ Denl(R, a), I be an ideal of R such that I ∩ S = ∅ and S−1I is an ideal of S−1R, π : R → R/I, r 7→ r := r + I, and S := π(S). Then 1. S ∈ Denl(R/I, b) for some ideal b of R/I such that a + I ⊆ b. 14 2. Let b := π−1(b). Then the map φ : S−1R/S−1I → S −1 (R/I), s−1r + S−1I 7→ s−1r, is an epimorphism with ker(φ) = S−1b/S−1I. Proof. 1. Since I ∩ S = ∅, S ∈ Orel(R/I). It remains to show that if rs = 0 for some elements r ∈ R/I and s ∈ S then tr = 0 for some t ∈ S. The equality rs = 0 means that i := rs ∈ I. Hence, S−1R ∋ r 1 i1 for some elements s1 ∈ S and i1 ∈ I (since S−1I is an ideal of S−1R). Therefore, s2s1r = s2i1 for some element s2 ∈ S, and so tr = 0 where t = s2s1 ∈ S. Clearly, a + I ⊆ b. 1 = is−1 = s−1 2. It is obvious that φ is an epimorphism. Since S ker(φ) = S−1b/S−1I. (cid:3) −1 (R/I) ≃ S−1(R/I) and I ⊆ b, we have Proposition 4.8 Let R be a ring, S ∈ Denl(R, 0), Q′ := S−1R and Q′∗ be the group of units of Q′. Suppose that S = R ∩ Q′∗ and one of the following statements holds: 1. Every left invertible element of Q′ is invertible. 2. Every right invertible element of Q′ is invertible. 3. The ring Q′ is a left Noetherian ring. 4. The ring Q′ is a right Noetherian ring. 5. The ring Q′ satisfies ACC on left annihilators. 6. The ring Q′ satisfies ACC on right annihilators. 7. The ring Q′ does not contain infinite direct sums of nonzero left ideals. 8. The ring Q′ does not contain infinite direct sums of nonzero right ideals. If yz ∈ S for some elements y, z ∈ R then y, z ∈ S. Proof. If q := yz ∈ S then q ∈ Q′∗. In particular, y · zq−1 = 1 and q−1y · z = 1. If condition 1 (resp. 2) holds then z (resp. y) is a unit of Q′∗. So, if any conditions 1 or 2 holds then y, z ∈ Q′∗, and so y, z ∈ R ∩ Q′∗ = S. Clearly, (3 ⇒ 5) and (4 ⇒ 6). So, it suffices to consider the case when one of the conditions 5-8 holds. Let x := zq−1. Then yx = 1 in Q′. The idempotents ei := xiyi − xi+1yi+1 = xie0yi, i = 0, 1, . . ., are orthogonal idempotents. Since one of conditions 5-8 holds we must have ei = 0 for some i then 0 = yieixi = yixie0yixi = e0 = 1 − xy, i.e. 1 = xy = yx. Hence, x, y ∈ Q′∗ and x, y ∈ R ∩ Q′∗ = S. (cid:3) Corollary 4.9 1. Let R be a ring such that its largest left quotient ring Q′ := Ql(R) satisfies one of conditions 1-8 of Proposition 4.8. If yz ∈ Sl(R) for some elements y, z ∈ R then y, z ∈ Sl(R). 2. Let R be a ring such that its left quotient ring Q′ := Q(R) satisfies one of conditions 1-8 of Proposition 4.8. If yz ∈ CR for some elements y, z ∈ R then y, z ∈ CR. Proof. 1. Statement 1 follows from Proposition 4.8 since Sl(R) = R ∩ Ql(R)∗ (Theorem 2.1.(1)). 2. Statement 2 is a particular case of statement 1. (cid:3) 15 References [1] V. V. Bavula, The algebra of integro-differential operators on an affine line and its modules, J. Pure Appl. Algebra, 217 (2013) 495-529. (Arxiv:math.RA: 1011.2997). [2] V. V. Bavula, The largest left quotient ring of a ring, Comm. Alg. (2015), to appear. (Arxiv:math.RA:1101.5107). [3] V. V. Bavula, New criteria for a ring to have a semisimple left quotient ring, Journal of Alg. and its Appl. 6 (2015) no. 6, [28 pages] DOI: 10.1142/S0219498815500905. (Arxiv:math.RA:1303.0859). [4] V. V. Bavula, The largest strong left quotient ring of a ring. J. Algebra (2015) [32 pages] http://dx.doi.org/10.1016/j.algebra.2015.04.037. (Arxiv:math.RA:1310.1077). [5] V. V. Bavula, Left localizations of left Artinian rings, Arxiv:math.RA:1405.0214. [6] A. W. Chatters and C. R. Hajarnavis, Quotient rings of Noetherian module finite rings, Proc. AMS 121 (1994) no. 2, 335 -- 341. [7] A. W. Goldie, Some aspects of ring theory, Bull. London Math. Soc. 1 (1969) 129-154. [8] A. V. Jategaonkar, Localization in Noetherian Rings, Londom Math. Soc. LMS 98, Cambridge Univ. Press, 1986. [9] J. C. McConnell and J. C. Robson, Noncommutative Noetherian rings. With the cooperation of L. W. Small. Revised edition. Graduate Studies in Mathematics, 30. American Mathematical Society, Providence, RI, 2001. 636 pp. [10] G. Michler and B. Muller, The maximal regular Ore set of a Noetherian ring, Arch. Math. 43 (1984) 218 -- 223. [11] B. Muller, Affine Noetherian PI-Rings Have Enough Clans, J. Algebra 97 (1985) 116 -- 129. [12] L. W. Small and J. T. Stafford, Regularity of zero divisors, Proc. London Math. Soc. (3) 44 (1982) 405-419. [13] J. T. Stafford, Noetherian full quotient rings, Proc. London Math. Soc. (3) , 44 (1982) 385-404. [14] B. Stenstrom, Rings of Quotients, Springer-Verlag, Berlin, Heidelberg, New York, 1975. Department of Pure Mathematics University of Sheffield Hicks Building Sheffield S3 7RH UK email: [email protected] 16
1509.09039	1	1509	2015-09-30T07:37:54	Hochschild homology and trivial extensions	[ "math.RA" ]	We prove that if an algebra is either selfinjective, local or graded, then the Hochschild homology dimension of its trivial extension is infinite.	math.RA	math	HOCHSCHILD HOMOLOGY AND TRIVIAL EXTENSIONS PETTER ANDREAS BERGH AND DAG OSKAR MADSEN Abstract. We prove that if an algebra is either selfinjective, local or graded, then the Hochschild homology dimension of its trivial extension is infinite. 1. Introduction Let A be a finite dimensional algebra over an algebraically closed field. It is well known that if its global dimension is finite, then its Hochschild cohomology and homology groups HHn(A) and HHn(A) vanish for all sufficiently large n. In his classical paper [Hap] on the cohomology of path algebras, Happel remarked that the converse was not known for Hochschild cohomology. As shown in [AvI], it does hold for commutative algebras. However, noncommutative counterexamples were given in [BGMS]: there exist algebras of infinite global dimension for which the Hochschild cohomology groups vanish in high degrees. The Hochschild homology groups of the algebras studied in [BGMS] do not be- have as the cohomology groups; they do not vanish in high degrees. This led Han to conjecture in [Han] that if the Hochschild homology groups of an alge- bra vanish in high degrees, then the algebra is necessarily of finite global dimen- sion. Han proved that this holds for monomial algebras, and just as for cohomol- ogy it also holds for commutative algebras, by [AV-P]. In the subsequent papers [BHM, BM1, BM2, SV-P], the conjecture has been shown to hold for several classes of algebras, including Koszul algebras, cellular algebras and local graded algebras. In this paper, we study the Hochschild homology groups of trivial extensions of algebras. The trivial extension of any algebra is a non-semisimple symmetric algebra, and therefore it always has infinite global dimension. For symmetric al- gebras, the vector space dimensions of the Hochschild homology groups equal the dimensions of the cohomology groups. Thus, for trivial extension algebras, Han's conjecture states that the Hochschild homology and cohomology groups do not all vanish in high degrees. We prove that this holds for trivial extensions of selfinjective algebras, local algebras and graded algebras. 2. Trivial extension algebras Throughout this paper, let k be an algebraically closed field, and A a finite dimensional k-algebra. Any such algebra is Morita equivalent to a basic algebra, and these again are isomorphic to quotients of path algebras by admissible ideals. Thus we may without loss of generality assume that our algebra A is of the form kQ/I for some finite quiver Q and admissible ideal I ⊆ kQ. 2010 Mathematics Subject Classification. 16E40, 16S70. Key words and phrases. Hochschild homology, trivial extensions. 1 2 PETTER ANDREAS BERGH AND DAG OSKAR MADSEN Denote by DA the k-dual Homk(A, k) of A, considered as a bimodule. The trivial extension of A by DA, denoted T (A) = A ⋉ DA, is the algebra with underlying vector space A ⊕ DA, and multiplication given by (a, f ) · (b, g) = (ab, ag + f b) for all a, b ∈ A and f, g ∈ DA. For any finite dimensional algebra, the trivial extension is symmetric. Moreover, there is a close relationship between the quiver eQ of T (A) and the defining quiver Q of A; we recall here this relationship as described in [FeP]. The radical of T (A) is r ⊕ DA, where r is the radical of A. Consequently, the quotient T (A)/ rad T (A) is isomorphic to A/r, from which it follows that the sizes of the complete sets of primitive orthogonal idempotents of A and T (A) are the same. This means Q and eQ have the same number of vertices. Next, consider the square of the radical of T (A). It is given by rad2 T (A) = r 2 ⊕ (rDA + DAr), and so rad T (A)/ rad2 T (A) = r/r 2 ⊕ DA/(rDA + DAr). The quotient DA/(rDA + DAr) is isomorphic to D(socAe A), where Ae denotes the enveloping algebra A ⊗k Aop of A, and socAe A denotes the socle of the A-bimodule A. Consequently, there is an isomorphism rad T (A)/ rad2 T (A) ≃ r/r 2 ⊕ D(socAe A) of A-bimodules. The presence of the summand r/r2 shows that the arrows of Q to a vertex j is equal to dimk ej(D(socAe A))ei, where ei denotes the primitive can be regarded as arrows also of eQ, whereas the summand D(socAe A) represents additional arrows. The number of such additional arrows in eQ from a vertex i idempotent in A corresponding to the vertex i in the quiver Q. Let eQ+ denote Q1 denotes the set of arrows of Q and eQ1 denotes the set of arrows of eQ, then eQ1 = Q1 ∪ eQ+. Denote by ξ the surjection DA → D(socAe A). For every arrow β ∈ eQ+ from i to j, choose an element xβ ∈ ej(DA)ei such that the set the collection of all new arrows; this set is non-empty since D(socAe A) 6= 0. If forms a k-basis for ej(D(socAe A))ei. Now define a surjective ring homomorphism {ξ(xβ) β ∈ eQ+, β : i → j} φ : keQ → T (A) as follows. For a primitive idempotent ei we set φ(ei) = (ei, 0), while φ(α) = (α, 0) for all α ∈ Q1. For all β ∈ eQ+ we set φ(β) = (0, xβ). Then the kernel eI of φ is an admissible ideal in keQ, and T (A) ≃ keQ/eI. While the quiver eQ is an invariant of the algebra T (A), the ideal eI may depend on the choices made; the algebra might admit several presentations as a bounded path algebra. Nevertheless, the following always holds. Lemma 2.1. Suppose β1 and β2 are arrows in eQ+. Then β2β1 ∈ eI. Proof. Using the ring homomorphism φ : keQ → T (A), we obtain where xβ1, xβ2 ∈ DA. Hence β2β1 ∈ ker φ = eI. φ(β2β1) = φ(β2) · φ(β1) = (0, xβ2) · (0, xβ1 ) = 0, (cid:3) HOCHSCHILD HOMOLOGY AND TRIVIAL EXTENSIONS 3 3. Hochschild homology dimension In this main section we prove that the Hochschild homology groups of the trivial extension algebra T (A) of A do not all vanish in high degrees, provided the algebra A is either local, selfinjective or graded. We define the Hochschild homology dimension of an algebra B as HHdim B def= sup{n HHn(B) 6= 0}, and so what we shall prove is that HHdim T (A) = ∞ when A is either local, selfinjective or graded. We recall first a key result from [BHM], which shows that the existence of a certain kind of cycle in the quiver of an algebra forces the Hochschild homology dimension to be unbounded. Let Γ be a finite quiver, J an admissible ideal in kΓ, and consider the algebra kΓ/J. A cycle in Γ is a path p of length at least one, starting and ending at the same vertex. Let p = αn . . . α2α1 be a cycle in Γ with αi ∈ Γ1 for all 1 ≤ i ≤ n. We say that p is a 2-truncated cycle in the algebra kΓ/J if αi+1αi ∈ J for all 1 ≤ i ≤ n − 1 and also α1αn ∈ J. Theorem 3.1 ([BHM, Theorem 3.1]). Let Γ be a finite quiver and J an admissible ideal in kΓ. If kΓ/J admits a 2-truncated cycle, then HHdim kΓ/J = ∞. In [BHM], this result was actually proved in a more general setting. Namely, the coefficients need not be taken from a field, they can be taken from any commutative ring. Moreover, the quiver need not be finite, it only has to have a finite number of vertices. We are now ready to prove our three results, proving that HHdim T (A) = ∞ whenever A is local, selfinjective or graded. We divide these three cases into sep- arate subsections. Recall that we may without loss of generality assume that A is the quotient kQ/I of a path algebra, where Q is a finite quiver an I ⊆ kQ an admissible ideal. We keep the notation from the previous section, in particular eQ denotes the quiver of T (A). 3.1. The local case. When the algebra A is local, the proof is short since the underlying quiver contains only a single vertex. Theorem 3.2. If A is a local finite dimensional k-algebra, then HHdim T (A) = ∞. Proof. The quiver Q, and therefore also eQ, has only one vertex. Let β be an arrow in eQ+. By Lemma 2.1, the path p = β is a 2-truncated cycle in keQ/eI, and this algebra is isomorphic to T (A). Hence HHdim T (A) = ∞ by Theorem 3.1. (cid:3) 3.2. The selfinjective case. Next, we treat the case when the algebra A is self- injective. Before the proof, we establish the following result, showing that every vertex in the quiver eQ of T (A) is the target of at least one new arrow. That is, given any vertex, there exists one arrow in eQ which does not correspond to an arrow in the quiver Q of A, and having the vertex as target. We denote by r the number of vertices in the quiver Q. Proposition 3.3. Suppose that A is a selfinjective k-algebra. Then for every 1 ≤ i ≤ r, there is an arrow in eQ+ ending at vertex i. Proof. Since A is selfinjective, there exists a permutation π : {1, . . . , r} → {1, . . . , r} 4 PETTER ANDREAS BERGH AND DAG OSKAR MADSEN of the vertices such that Aei ≃ D(eπ(i)A) for all 1 ≤ i ≤ r. So for any given i, the Loewy length of the left A-module Aei is equal to the Loewy length of the right A-module eπ(i)A. Denote this common length by Li, and choose a nonzero element x ∈ rLi−1ei. Now rLi−1ei equals socA Aei, which in turn is isomorphic to D(topAop eπ(i)A), where topAop eπ(i)A denotes the top of the right A-module eπ(i)A. Therefore x = eπ(i)xei, giving x ∈ eπ(i)r Li−1 = socAop eπ(i)A. Let α be any arrow in Q1. Then αx = 0 since x belongs to the socle of the left A-module Aei, and xα = 0 since x belongs to the socle of the right A-module eπ(i)A. Thus x is an element of socAe A, showing that eπ(i)(socAe A)ei is nonzero. But then ei (D(socAe A)) eπ(i) must also be nonzero, and so there exists an arrow (cid:3) in eQ+ from vertex π(i) to i. Remark 3.4. As can be seen from the proof of this proposition, the result does not require the algebra A to be selfinjective. Namely, the result holds under the weaker assumption that socA A ⊆ socAe A, that is, when the socle of A as a left module is contained in its bimodule socle. Proposition 3.3 guarantees that the quiver of the trivial extension of a selfinjec- tive algebra contains "enough" arrows. This is what we need in order to prove the result on the Hochschild homology dimension for such algebras. Theorem 3.5. If A is a selfinjective k-algebra, then HHdim T (A) = ∞. Proof. From Proposition 3.3, we know that for any vertex in eQ there is at least one arrow in eQ+ ending at that vertex. cycle p = βt . . . β2β1 in eQ consisting entirely of arrows in eQ+. By Lemma 2.1, the composition of any two such arrows is zero in T (A), hence p is a 2-truncated cycle in the trivial extension algebra. Theorem 3.1 now gives HHdim T (A) = ∞. (cid:3) It follows from this that there exists a Since the trivial extension of any algebra is symmetric, we obtain the following. Corollary 3.6. If A is any finite dimensional k-algebra, then HHdim T (T (A)) = ∞. 3.3. The graded case. In this final subsection, we treat the case when the algebra A is positively graded, so that A = A0 ⊕ A1 ⊕ · · · ⊕ As. Many of the algebras one normally studies are gradable, and they are therefore covered by the result. However, our proof requires the characteristic of the ground field k to be zero, and the degree zero part A0 of A to be isomorphic to a product k × · · · × k = k×r as a k-algebra. Let 1A = e1 + · · · + er be the corresponding decomposition of the identity. Example 3.7. We do not require the generators of A to be in degree 1. For example, let A be the path algebra A = kQ/I, where Q is the quiver Q : •1 α ;①①①①①①① ✹✹✹✹✹ γ •4 •2 δ β #❋❋❋❋❋❋❋ E✡✡✡✡✡ / •5 •3 ε # ; / E HOCHSCHILD HOMOLOGY AND TRIVIAL EXTENSIONS 5 and I is the ideal in kQ generated by the single relation εδγ − βα. This ideal is not homogeneous with the path length grading, where every arrow is assigned the degree one. However, if we let deg α = deg β = 3 and deg γ = deg δ = deg ε = 2, then A is a positively graded algebra. There are finite dimensional algebras that do not admit a positive grading with semisimple degree zero part, see [BBFS] for examples. Returning now to the general case, for 1 ≤ l ≤ s, let C l A be the r × r-matrix with A)i,j = dimk ejAlei. The graded Cartan matrix of A is defined to be the entries (C l r × r-matrix CA(x) = C0 A + C1 Ax + C2 Ax2 + · · · + C s Axs with entries in Z[x]. Its determinant det CA(x) is the graded Cartan determinant of A. The following result from [BM1] establishes a connection between this deter- minant and the Hochschild homology dimension of A. Theorem 3.8 ([BM1, Corollary 3.5]). If det CA(x) 6= 1, then HHdim A = ∞. Now we show how to give the trivial extension T (A) of A a positive grading, based on the grading of A. By definition, there is a vector space decomposition T (A) = A ⊕ DA. We keep the original grading of A, so that and then we give D(A) the following grading: deg Al = l; 1 ≤ l ≤ s, deg D(Al) = s + 1 − l; 1 ≤ l ≤ s. In this way, T (A) becomes a positively graded k-algebra with top degree s + 1. Next, we analyze the graded Cartan matrix CT (A)(x) = C0 T (A) + C1 T (A)x2 + · · · + C s T (A)xs + C s+1 T (A)xs+1 T (A)x + C2 T (A) and C s+1 of T (A). The matrices C0 T (A) must both be identity matrices, since dimk ej(T (A)0)ei = dimk ejA0ei = δij and dimk ej(T (A)s+1)ei = dimk ejD(A0)ei = δij, where δij denotes the Kronecker delta. It follows from this that the graded Cartan matrix of T (A) has the shape CT (A)(x) =   1 + p1,1(x) + xs+1 p2,1(x) ... pr,1(x) p1,2(x) 1 + p2,2(x) + xs+1 ... pr,2(x) · · · · · · . . . · · · p1,r(x) p2,r(x) ... 1 + pr,r(x) + xs+1   , where the entries pi,j(x), 1 ≤ i, j ≤ r, are integer polynomials of degree at most s and with zero constant term. This is the key ingredient when we now prove that the Hochschild homology dimension of T (A) is infinite. Theorem 3.9. Suppose that the characteristic of k is zero, and let A = A0 ⊕ A1 ⊕ · · · ⊕ As be a positively graded finite dimensional k-algebra, where A0 is a product of copies of k. Then HHdim T (A) = ∞. 6 PETTER ANDREAS BERGH AND DAG OSKAR MADSEN Proof. The product of the diagonal entries in the graded Cartan matrix CT (A)(x) is a monic polynomial of degree r(s + 1), and with constant term 1. All other products in the expression for the determinant involve off-diagonal entries, so they have degrees less than r(s + 1) and zero constant term. Therefore the determinant is of the form det CT (A)(x) = 1 + · · · + xr(s+1). Then HHdim T (A) = ∞ by Theorem 3.8. (cid:3) References [AvI] L.L. Avramov, S. Iyengar, Gaps in Hochschild cohomology imply smoothness for com- mutative algebras, Math. Res. Lett. 12 (2005), no. 5 -- 6, 789 -- 804. [AV-P] L.L. Avramov, M. Vigu´e-Poirrier, Hochschild homology criteria for smoothness, Internat. Math. Res. Notices (1992), no. 1, 17 -- 25. [BBFS] T. Belzner, W.D. Burgess, K.R. Fuller, R. Schulz, Examples of ungradable algebras, Proc. Amer. Math. Soc. 114 (1992), no. 1, 1 -- 4. [BHM] P.A. Bergh, Y. Han, D. Madsen, Hochschild homology and truncated cycles, Proc. Amer. Math. Soc. 140 (2012), no. 4, 1133 -- 1139. [BM1] P.A. Bergh, D. Madsen, Hochschild homology and global dimension, Bull. London Math. Soc. 41 (2009), no. 3, 473 -- 482. [BM2] P.A. Bergh, D. Madsen, Hochschild homology and split pairs, Bull. Sci. Math. 134 (2010), 665 -- 676. [BGMS] R.-O. Buchweitz, E.L. Green, D. Madsen, Ø. Solberg, Finite Hochschild cohomology [FeP] [Han] [Hap] without finite global dimension, Math. Res. Lett. 12 (2005), no. 5-6, 805 -- 816. E.A. Fern´andez, M.I. Platzeck, Presentations of trivial extensions of finite dimensional algebras and a theorem of Sheila Brenner, J. Algebra 249 (2002), no. 2, 326 -- 344. Y. Han, Hochschild (co)homology dimension, J. London Math. Soc. (2) 73 (2006), no. 3, 657 -- 668. D. Happel, Hochschild cohomology of finite-dimensional algebras, S´eminaire d'Alg`ebre Paul Dubreil et Marie-Paul Malliavin, 39`eme Ann´ee (Paris, 1987/1988), 108 -- 126, Lecture Notes in Math., 1404, Springer, Berlin, 1989. [SV-P] A. Solotar, M. Vigu´e-Poirrier, Two classes of algebras with infinite Hochschild homology, Proc. Amer. Math. Soc. 138 (2010), no. 3, 861 -- 869. Department of Mathematical Sciences, NTNU, NO-7491 Trondheim, Norway E-mail address: [email protected] Faculty of Professional Studies, University of Nordland, NO-8049 Bodø, Norway E-mail address: [email protected]
1105.5475	1	1105	2011-05-27T04:06:56	Jordan Triple Disystems	[ "math.RA" ]	We take an algorithmic and computational approach to a basic problem in abstract algebra: determining the correct generalization to dialgebras of a given variety of nonassociative algebras. We give a simplified statement of the KP algorithm introduced by Kolesnikov and Pozhidaev for extending polynomial identities for algebras to corresponding identities for dialgebras. We apply the KP algorithm to the defining identities for Jordan triple systems to obtain a new variety of nonassociative triple systems, called Jordan triple disystems. We give a generalized statement of the BSO algorithm introduced by Bremner and Sanchez-Ortega for extending multilinear operations in an associative algebra to corresponding operations in an associative dialgebra. We apply the BSO algorithm to the Jordan triple product and use computer algebra to verify that the polynomial identities satisfied by the resulting operations coincide with the results of the KP algorithm; this provides a large class of examples of Jordan triple disystems. We formulate a general conjecture expressed by a commutative diagram relating the output of the KP and BSO algorithms. We conclude by generalizing the Jordan triple product in a Jordan algebra to operations in a Jordan dialgebra; we use computer algebra to verify that resulting structures provide further examples of Jordan triple disystems. For this last result, we also provide an independent theoretical proof using Jordan structure theory.	math.RA	math	JORDAN TRIPLE DISYSTEMS MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA Abstract. We take an algorithmic and computational approach to a basic problem in abstract algebra: determining the correct generalization to dial- gebras of a given variety of nonassociative algebras. We give a simplified statement of the KP algorithm introduced by Kolesnikov and Pozhidaev for extending polynomial identities for algebras to corresponding identities for di- algebras. We apply the KP algorithm to the defining identities for Jordan triple systems to obtain a new variety of nonassociative triple systems, called Jordan triple disystems. We give a generalized statement of the BSO algo- rithm introduced by Bremner and S´anchez-Ortega for extending multilinear operations in an associative algebra to corresponding operations in an asso- ciative dialgebra. We apply the BSO algorithm to the Jordan triple product and use computer algebra to verify that the polynomial identities satisfied by the resulting operations coincide with the results of the KP algorithm; this provides a large class of examples of Jordan triple disystems. We formulate a general conjecture expressed by a commutative diagram relating the output of the KP and BSO algorithms. We conclude by generalizing the Jordan triple product in a Jordan algebra to operations in a Jordan dialgebra; we use com- puter algebra to verify that resulting structures provide further examples of Jordan triple disystems. For this last result, we also provide an independent theoretical proof using Jordan structure theory. 1. Introduction The theory of Jordan algebras, originally motivated by potential applications to quantum physics, was initiated by Jordan, von Neumann and Wigner [12], and the theory of Jordan triple systems was initiated by Jacobson [10]. Standard references on these topics are Braun and Koecher [1], Jacobson [11], McCrimmon [18], Neher [19] and Loos [17]; for applications to geometry and analysis see Faraut et al. [7]. The concept of an associative dialgebra was introduced by Loday [15]; the gener- alization of the Lie bracket produces Lie dialgebras (also called Leibniz algebras) which were first introduced by Cuvier [6] and Loday [14]. Numerous authors have considered other varieties of nonassociative dialgebras; in particular, the general- ization of the Jordan product produces Jordan dialgebras (also called quasi-Jordan algebras), which have been studied by Vel´asquez and Felipe [24, 25], Kolesnikov [13], Pozhidaev [22, 23], Bremner [2], Bremner and Peresi [4], and Voronin [27]. The purpose of the present paper is to introduce a new variety of triple systems of Jordan type, which we call Jordan triple disystems. The relation between Jordan triple disystems and associative dialgebras is analogous to the relation between Jordan triple systems and associative algebras. Section 2 recalls basic definitions for associative dialgebras. Section 3 presents a simplified statement of the general Kolesnikov-Pozhidaev (KP) algorithm for con- verting an arbitrary variety of multioperator algebras into a variety of dialgebras. We recall how this algorithm can be applied to the defining identities for Jordan 1 2 MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA algebras to obtain the variety of Jordan dialgebras. In Section 4 we apply the KP algorithm to the defining identities for Jordan triple systems; we obtain a system of polynomial identities which define our new variety of Jordan triple disystems. Section 5 presents a generalized statement of the Bremner-S´anchez-Ortega (BSO) algorithm for extending multilinear operations in an associative algebra to an as- sociative dialgebra. We apply the BSO algorithm to the Jordan triple product and use computer algebra to verify that the resulting polynomial identities coincide with the results of the KP algorithm; we therefore obtain a large class of examples of Jordan triple disystems. In Section 6 we formulate a general conjecture expressed by a commutative diagram relating the output of the KP and BSO algorithms. Section 7 generalizes the Jordan triple product in a Jordan algebra to operations in Jordan dialgebras; we use computer algebra to verify that resulting structures provide another large class of examples of Jordan triple disystems. We also provide an independent theoretical proof of this last result using Jordan structure theory. 2. Dialgebras 2.1. Dialgebras and Leibniz algebras. Dialgebras were introduced by Loday [14, 15, 16] to provide a natural setting for Leibniz algebras, a "noncommutative" generalization of Lie algebras. Definition 2.1. (Cuvier [6], Loday [14]) A Leibniz algebra is a vector space L together with a bilinear map L × L → L, denoted (a, b) 7→ [a, b] and called the Leibniz bracket, satisfying the Leibniz identity: [[a, b], c] ≡ [[a, c], b] + [a, [b, c]]. If [a, a] ≡ 0 then the Leibniz identity is the Jacobi identity and L is a Lie algebra. Every associative algebra becomes a Lie algebra if the associative product is replaced by the Lie bracket. Loday introduced the notion of dialgebra which gives, by a similar procedure, a Leibniz algebra: one replaces ab and ba by two distinct operations, so that the resulting bracket is not necessarily skew-symmetric. Definition 2.2. An associative dialgebra is a vector space D with two bilinear operations ⊣ : D × D → D and ⊢ : D × D → D, the left and right products, satisfying the left and right bar identities, and left, right and inner associativity: (a ⊣ b) ⊢ c ≡ (a ⊢ b) ⊢ c, a ⊣ (b ⊣ c) ≡ a ⊣ (b ⊢ c), (a ⊣ b) ⊣ c ≡ a ⊣ (b ⊣ c), (a ⊢ b) ⊢ c ≡ a ⊢ (b ⊢ c), (a ⊢ b) ⊣ c ≡ a ⊢ (b ⊣ c). From a dialgebra we construct a Leibniz bracket by a ⊣ b − b ⊢ a. 2.2. Free dialgebras. Loday has determined a basis for the free dialgebra. Definition 2.3. A dialgebra monomial on a set X is a product x = a1a2 · · · an where a1, . . . , an ∈ X with some placement of parentheses and some choice of operations. The center of x is defined inductively: if n = 1 then c(x) = x; if n ≥ 2 then x = y ⊣ z or x = y ⊢ z and we set c(y ⊣ z) = c(y) or c(y ⊢ z) = c(z). Lemma 2.4. (Loday [16]) If x = a1a2 · · · an is a monomial with c(x) = ai then x is determined by the order of its factors and the position of its center: x = (a1 ⊢ · · · ⊢ ai−1) ⊢ ai ⊣ (ai+1 ⊣ · · · ⊣ an). JORDAN TRIPLE DISYSTEMS 3 Definition 2.5. The right side of the last equation is the normal form of x and is abbreviated by the hat notation a1 · · · ai−1baiai+1 · · · an. Lemma 2.6. (Loday [16]) The set of monomials a1 · · · ai−1baiai+1 · · · an in normal form with 1 ≤ i ≤ n and a1, . . . , an ∈ X forms a basis of the free dialgebra on X. 3. The Kolesnikov-Pozhidaev algorithm This algorithm, introduced by Kolesnikov [13] and Pozhidaev [23], converts a multilinear polynomial identity of degree d for an n-ary operation into d multilinear identities of degree d for n new n-ary operations. Definition 3.1. KP Algorithm. Part 1: We consider a multilinear n-ary operation, denoted by the symbol (1) {−, −, . . . , −} (n arguments). Given a multilinear polynomial identity of degree d in this operation, we describe the application of the algorithm to one monomial in the identity, and from this the application to the complete identity follows by linearity. Let a1a2 . . . ad be a multilinear monomial of degree d, where the bar denotes some placement of n-ary operation symbols. We introduce n new n-ary operations, denoted by the same symbol but distinguished by subscripts: (2) {−, −, . . . , −}1, {−, −, . . . , −}2, . . . , {−, −, . . . , −}n. For each i ∈ {1, 2, . . . , d} we convert the monomial a1a2 . . . ad in the original n- ary operation (1) into a new monomial of the same degree d in the n new n-ary operations (2), according to the following rule which is based on the position of ai. For each occurrence of the original operation symbol in the monomial, either ai occurs within one of the n arguments or not, and we have the following cases: • If ai occurs within the j-th argument then we convert the original operation symbol {. . . } to the j-th new operation symbol {. . . }j. • If ai does not occur within any of the n arguments, then either -- ai occurs to the left of the original operation symbol, in which case we convert {. . . } to the first new operation symbol {. . . }1, or -- ai occurs to the right of the original operation symbol, in which case we convert {. . . } to the last new operation symbol {. . . }n. In this process, we call ai the central argument of the monomial. Part 2: In addition to the identities constructed in Part 1, we also include the following identities for all i, j ∈ {1, 2, . . . , n} with i 6= j and all k, ℓ ∈ {1, 2, . . . , n}: {a1, . . . , ai−1, {b1, · · · , bn}k, ai+1, . . . , an}j ≡ {a1, . . . , ai−1, {b1, · · · , bn}ℓ, ai+1, . . . , an}j. This identity says that the n new operations are interchangeable in the i-th argu- ment of the j-th new operation when i 6= j. Example 3.2. The defining identities for associative dialgebras can be obtained by applying the KP algorithm to the associativity identity, which we write in the form {{a, b}, c} ≡ {a, {b, c}}. The original operation produces two new operations {−, −}1 and {−, −}2. Since associativity has degree 3, Part 1 produces three new identities of degree 3 by making a, b, c in turn the central argument: {{a, b}1, c}1≡{a, {b, c}1}1, {{a, b}2, c}1≡{a, {b, c}1}2, {{a, b}2, c}2≡{a, {b, c}2}2, 4 MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA and Part 2 produces these two identities: {a, {b, c}1}1 ≡ {a, {b, c}2}1, {{a, b}1, c}2 ≡ {{a, b}2, c}2. If we revert to the standard notation by writing a ⊣ b = {a, b}1 and a ⊢ b = {a, b}2, then these five identities are the defining identities for associative dialgebras. Definition 3.3. A (linear) Jordan algebra is a vector space J over a field of characteristic 6= 2 with a bilinear operation J × J → J, denoted a ◦ b, satisfying commutativity and the Jordan identity for all a, b ∈ J: a ◦ b ≡ b ◦ a, ((a ◦ a) ◦ b) ◦ a ≡ (a ◦ a) ◦ (b ◦ a). Example 3.4. To apply the KP algorithm to Jordan algebras, we must start with multilinear identities, so we linearize the Jordan identity to obtain ((a ◦ c) ◦ b) ◦ d + ((a ◦ d) ◦ b) ◦ c + ((c ◦ d) ◦ b) ◦ a ≡ (a ◦ c) ◦ (b ◦ d) + (a ◦ d) ◦ (b ◦ c) + (c ◦ d) ◦ (b ◦ a). (For a general discussion of linearization, see Zhevlakov et al. [28], Chapter 1.) We rewrite commutativity and the linearized Jordan identity using the symbol {−, −}: {a, b} − {b, a} ≡ 0, {{{a, c}, b}, d} + {{{a, d}, b}, c} + {{{c, d}, b}, a} − {{a, c}, {b, d}} − {{a, d}, {b, c}} − {{c, d}, {b, a}} ≡ 0. The KP algorithm tells us to introduce two new operations {−, −}1 and {−, −}2. Part 1: Since commutativity has degree 2, we obtain two identities of degree 2 relating the two new operations: {a, b}1 − {b, a}2 ≡ 0, {a, b}2 − {b, a}1 ≡ 0, These identities are both equivalent to {a, b}2 ≡ {b, a}1: the second operation is the opposite of the first. Hence we can replace every occurrence of {−, −}2 by an occurrence of {−, −}1. Since the linearized Jordan identity has degree 4, we obtain four identities of degree 4 relating the two new operations: {{{a, c}1, b}1, d}1 + {{{a, d}1, b}1, c}1 + {{{c, d}2, b}2, a}2 − {{a, c}1, {b, d}1}1 − {{a, d}1, {b, c}1}1 − {{c, d}2, {b, a}2}2 ≡ 0, {{{a, c}2, b}2, d}1 + {{{a, d}2, b}2, c}1 + {{{c, d}2, b}2, a}1 − {{a, c}2, {b, d}1}2 − {{a, d}2, {b, c}1}2 − {{c, d}2, {b, a}1}2 ≡ 0, {{{a, c}2, b}1, d}1 + {{{a, d}2, b}2, c}2 + {{{c, d}1, b}1, a}1 − {{a, c}2, {b, d}1}1 − {{a, d}2, {b, c}2}2 − {{c, d}1, {b, a}1}1 ≡ 0, {{{a, c}2, b}2, d}2 + {{{a, d}2, b}1, c}1 + {{{c, d}2, b}1, a}1 − {{a, c}2, {b, d}2}2 − {{a, d}2, {b, c}1}1 − {{c, d}2, {b, a}1}1 ≡ 0. (3) (4) (5) (6) In identities (3) -- (6), we replace every instance of the second operation by the opposite of the first operation: {{{a, c}1, b}1, d}1 + {{{a, d}1, b}1, c}1 + {a, {b, {d, c}1}1}1 (7) (8) − {{a, c}1, {b, d}1}1 − {{a, d}1, {b, c}1}1 − {{a, b}1, {d, c}1}1 ≡ 0, {{b, {c, a}1}1, d}1 + {{b, {d, a}1}1, c}1 + {{b, {d, c}1}1, a}1 − {{b, d}1, {c, a}1}1 − {{b, c}1, {d, a}1}1 − {{b, a}1, {d, c}1}1 ≡ 0, JORDAN TRIPLE DISYSTEMS 5 {{{c, a}1, b}1, d}1 + {c, {b, {d, a}1}1}1 + {{{c, d}1, b}1, a}1 (9) − {{c, a}1, {b, d}1}1 − {{c, b}1, {d, a}1}1 − {{c, d}1, {b, a}1}1 ≡ 0, {d, {b, {c, a}1}1}1 + {{{d, a}1, b}1, c}1 + {{{d, c}1, b}1, a}1 (10) − {{d, b}1, {c, a}1}1 − {{d, a}1, {b, c}1}1 − {{d, c}1, {b, a}1}1 ≡ 0. Since we now have only one operation, we revert to a simpler notation, and write {a, b}1 simply as ab. Identities (7) -- (10) take the following form: (11) (12) (13) (14) ((ac)b)d + ((ad)b)c + a(b(dc)) − (ac)(bd) − (ad)(bc) − (ab)(dc) ≡ 0, (b(ca))d + (b(da))c + (b(dc))a − (bd)(ca) − (bc)(da) − (ba)(dc) ≡ 0, ((ca)b)d + c(b(da)) + ((cd)b)a − (ca)(bd) − (cb)(da) − (cd)(ba) ≡ 0, d(b(ca)) + ((da)b)c + ((dc)b)a − (db)(ca) − (da)(bc) − (dc)(ba) ≡ 0. Clearly (11) becomes (13) after the transposition ac, and (11) becomes (14) after the cyclic permutation adc. We discard (13) and (14) and retain (11) and (12). Part 2: We include identities which state that in the first (resp. second) argument of the second (resp. first) new operation, the two operations are interchangeable: {a, {b, c}1}1 ≡ {a, {b, c}2}1, {{a, b}1, c}2 ≡ {{a, b}2, c}2. Rewriting these in terms of the first operation gives {a, {b, c}1}1 ≡ {a, {c, b}1}1, {c, {a, b}1}1 ≡ {c, {b, a}1}1. These identities are both equivalent to right-commutativity a(bc) ≡ a(cb). Rearranging the terms in (11) and applying right-commutativity gives ((ac)b)d − (ac)(bd) + ((ad)b)c − (ad)(bc) − (ab)(cd) + a(b(cd)) ≡ 0. This can be reformulated in terms of associators as follows: (15) (ac, b, d) + (ad, b, c) − (a, b, cd) ≡ 0. If we assume characteristic not 2, then identity (15) is equivalent to (16) (a, b, c2) ≡ 2(ac, b, c). Apply right-commutativity to (12) gives (b(ac))d + (b(ad))c + (b(cd))a − (bd)(ac) − (bc)(ad) − (ba)(cd) ≡ 0. Setting a = c = d (and dividing by 3) gives (17) (ba2)a ≡ (ba)a2. If we assume characteristic not 3, then identities (12) and (17) are equivalent. Definition 3.5. Over a field of characteristic not 2, 3, a (right) Jordan dialgebra is a vector space D with a bilinear operation D × D → D, denoted ab, satisfying right commutativity, the Osborn identity, and the right Jordan identity: a(bc) ≡ a(cb), (a, b, c2) ≡ 2(ac, b, c), (ba2)a ≡ (ba)a2. Remark 3.6. The second identity in Definition 3.5 first appeared in Osborn [20]; see also Petersson [21]. In a Jordan dialgebra, the Osborn identity is equivalent to (18) (b, a2, c) = 2(b, a, c)a. 6 MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA The linearized forms of the Osborn identity, the right Jordan identity and (18) are O1(a, b, c, d) = ((ba)c)d + ((bd)c)a − (b(ac))d − (b(cd))a − (b(ad))c + b((ad)c), RJ(a, b, c, d) = (b(ac))d + (b(ad))c + (b(cd))a − (bd)(ac) − (bc)(ad) − (ba)(cd), O2(a, b, c, d) = ((ac)b)d + ((ad)b)c − (ab)(cd) − (ac)(bd) − (ad)(bc) + a((cd)b). Using right commutativity it is easy to verify that O2(a, b, c, d) = O1(c, a, b, d) + RJ(b, a, c, d). Remark 3.7. The notion of a Jordan dialgebra was discovered independently by various authors during the last few years. Kolesnikov [13] introduced a functorial approach to varieties of associative and nonassociative dialgebras and obtained the defining identities for left Jordan dialgebras (the opposite identities to those of Definition 3.5). Vel´asquez and Felipe [24] introduced the product a ⊣ b + b ⊢ a in an associative dialgebra, showed that it satisfies the first and third identities of Definition 3.5, and defined a quasi-Jordan algebra to be a vector space satisfying these identities. Bremner [2] used computer algebra to verify that a ⊣ b + b ⊢ a also satisfies the second identity of Definition 3.5. 4. Jordan triple disystems In this section we apply the KP algorithm to the defining identities of Jordan triple systems. We obtain a new variety of triple systems, which we call Jordan triple disystems, with two trilinear operations. Definition 4.1. A (linear) Jordan triple system (JTS) is a vector space T over a field of characteristic not 2 with a trilinear operation T × T × T → T , denoted {−, −, −}, satisfying these polynomial identities: {a, b, c} ≡ {c, b, a}, {a, b, {c, d, e}} ≡ {{a, b, c}, d, e} − {c, {b, a, d}, e} + {c, d, {a, b, e}}. Theorem 4.2. Applying the KP algorithm to Definition 4.1 produces two trilinear operations {−, −, −}1 and {−, −, −}2 satisfying these identities: {a, b, c}2 ≡ {c, b, a}2, {a, {b, c, d}1, e}1 ≡ {a, {b, c, d}2, e}1 ≡ {a, {d, c, b}1, e}1, {a, b, {c, d, e}1}1 ≡ {a, b, {c, d, e}2}1 ≡ {a, b, {e, d, c}1}1, {{a, b, c}1, d, e}2 ≡ {{a, b, c}2, d, e}2 ≡ {{c, b, a}1, d, e}2, {{e, d, c}1, b, a}1 ≡ {{e, b, a}1, d, c}1 − {e, {d, a, b}1, c}1 + {e, d, {c, b, a}1}1, {{e, d, c}2, b, a}1 ≡ {{e, b, a}1, d, c}2 − {e, {d, a, b}1, c}2 + {e, d, {c, b, a}1}2, {a, b, {c, d, e}1}1 ≡ {{a, b, c}1, d, e}1 − {c, {b, a, d}2, e}2 + {{a, b, e}1, d, c}1, {a, b, {c, d, e}1}2 ≡ {{a, b, c}2, d, e}1 − {c, {b, a, d}1, e}2 + {{a, b, e}2, d, c}1. Proof. Part 1: First, we consider the identity of degree 3: {a, b, c} − {c, b, a} ≡ 0. If we make a, b, c in turn the central argument we obtain three identities: {a, b, c}1 − {c, b, a}3 ≡ 0, {a, b, c}2 − {c, b, a}2 ≡ 0, {a, b, c}3 − {c, b, a}1 ≡ 0. The first and third identities are both equivalent to {a, b, c}3 ≡ {c, b, a}1: the third operation is the opposite of the first, and can be eliminated. The second identity JORDAN TRIPLE DISYSTEMS 7 says that the second operation is symmetric in its first and third arguments: (19) {a, b, c}2 ≡ {c, b, a}2. Second, we consider the identity of degree 5, {a, b, {c, d, e}} − {{a, b, c}, d, e} + {c, {b, a, d}, e} − {c, d, {a, b, e}} ≡ 0. If we make a, b, c, d, e in turn the central argument we obtain five identities; {a, b, {c, d, e}1}1 − {{a, b, c}1, d, e}1 + {c, {b, a, d}2, e}2 − {c, d, {a, b, e}1}3 ≡ 0, {a, b, {c, d, e}1}2 − {{a, b, c}2, d, e}1 + {c, {b, a, d}1, e}2 − {c, d, {a, b, e}2}3 ≡ 0, {a, b, {c, d, e}1}3 − {{a, b, c}3, d, e}1 + {c, {b, a, d}1, e}1 − {c, d, {a, b, e}1}1 ≡ 0, {a, b, {c, d, e}2}3 − {{a, b, c}3, d, e}2 + {c, {b, a, d}3, e}2 − {c, d, {a, b, e}1}2 ≡ 0, {a, b, {c, d, e}3}3 − {{a, b, c}3, d, e}3 + {c, {b, a, d}3, e}3 − {c, d, {a, b, e}3}3 ≡ 0. We replace {a, b, c}3 by the opposite of {a, b, c}1; to save space we omit "≡ 0": (20) (21) (22) (23) (24) {a, b, {c, d, e}1}1 − {{a, b, c}1, d, e}1 + {c, {b, a, d}2, e}2 − {{a, b, e}1, d, c}1, {a, b, {c, d, e}1}2 − {{a, b, c}2, d, e}1 + {c, {b, a, d}1, e}2 − {{a, b, e}2, d, c}1, {{c, d, e}1, b, a}1 − {{c, b, a}1, d, e}1 + {c, {b, a, d}1, e}1 − {c, d, {a, b, e}1}1, {{c, d, e}2, b, a}1 − {{c, b, a}1, d, e}2 + {c, {d, a, b}1, e}2 − {c, d, {a, b, e}1}2, {{e, d, c}1, b, a}1 − {e, d, {c, b, a}1}1 + {e, {d, a, b}1, c}1 − {{e, b, a}1, d, c}1. Part 2: We obtain the following 12 identities: {a, {b, c, d}1, e}1 ≡ {a, {b, c, d}2, e}1 ≡ {a, {b, c, d}3, e}1, {a, b, {c, d, e}1}1 ≡ {a, b, {c, d, e}2}1 ≡ {a, b, {c, d, e}3}1, {{a, b, c}1, d, e}2 ≡ {{a, b, c}2, d, e}2 ≡ {{a, b, c}3, d, e}2, {a, b, {c, d, e}1}2 ≡ {a, b, {c, d, e}2}2 ≡ {a, b, {c, d, e}3}2, {{a, b, c}1, d, e}3 ≡ {{a, b, c}2, d, e}3 ≡ {{a, b, c}3, d, e}3, {a, {b, c, d}1, e}3 ≡ {a, {b, c, d}2, e}3 ≡ {a, {b, c, d}3, e}3. We replace {a, b, c}3 by the opposite of {a, b, c}1, obtaining (25) (26) (27) (28) {a, {b, c, d}1, e}1 ≡ {a, {b, c, d}2, e}1 ≡ {a, {d, c, b}1, e}1, {a, b, {c, d, e}1}1 ≡ {a, b, {c, d, e}2}1 ≡ {a, b, {e, d, c}1}1, {{a, b, c}1, d, e}2 ≡ {{a, b, c}2, d, e}2 ≡ {{c, b, a}1, d, e}2, {a, b, {c, d, e}1}2 ≡ {a, b, {c, d, e}2}2 ≡ {a, b, {e, d, c}1}2, and other equivalent identities; note that (28) follows from (27) by using (19). We now see that (22) becomes (24) by the transposition ce and using (25) and (26). We retain (24), which we write as a derivation property: (29) {{e, d, c}1, b, a}1 ≡ {{e, b, a}1, d, c}1 − {e, {d, a, b}1, c}1 + {e, d, {c, b, a}1}1. We also see that (23) becomes another derivation property by using (28): (30) {{c, d, e}2, b, a}1 ≡ {{c, b, a}1, d, e}2 − {c, {d, a, b}1, e}2 + {c, d, {e, b, a}1}2. This completes the proof. (cid:3) 8 MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA Definition 4.3. A Jordan triple disystem (JTD) is a vector space D over a field of characteristic not 2 with two trilinear operations {−, −, −}i : D × D × D → D (i = 1, 2) satisfying the following identities: (J1) (J2) (J3) (J4) (J5) (J6) (J7) (J8) {a, b, c}2 ≡ {c, b, a}2, {a, {b, c, d}1, e}1 ≡ {a, {b, c, d}2, e}1, {a, b, {c, d, e}1}1 ≡ {a, b, {c, d, e}2}1, {{a, b, c}1, d, e}2 ≡ {{a, b, c}2, d, e}2, {{e, d, c}1, b, a}1 ≡ {{e, b, a}1, d, c}1 − {e, {d, a, b}1, c}1 + {e, d, {c, b, a}1}1, {{e, d, c}2, b, a}1 ≡ {{e, b, a}1, d, c}2 − {e, {d, a, b}1, c}2 + {e, d, {c, b, a}1}2, {a, b, {c, d, e}1}1 ≡ {{a, b, c}1, d, e}1 − {c, {b, a, d}2, e}2 + {{a, b, e}1, d, c}1, {a, b, {c, d, e}1}2 ≡ {{a, b, c}2, d, e}1 − {c, {b, a, d}1, e}2 + {{a, b, e}2, d, c}1. (We have omitted the redundant second identities in lines 2, 3 and 4.) We conclude this section with some examples of Jordan triple disystems. Example 4.4. Let T be a Jordan triple system with product {−, −, −} over a field of characteristic not 2. It is straightforward to check that T becomes a Jordan triple disystem by setting {−, −, −}1 = {−, −, −}2 = {−, −, −}. In particular, every associative algebra gives rise to a Jordan triple disystem by defining {a, b, c}1 = {a, b, c}2 = abc + cba. (For details see Section 5.) Example 4.5. Let A be a differential associative algebra in the sense of Loday [16]: that is, A is an associative algebra with product a · b together with a linear map d : A → A such that d2 = 0 and d(a · b) = d(a) · b + a · d(b) for all a, b ∈ A. One endows A with a dialgebra structure by defining a ⊣ b = a · d(b) and a ⊢ b = d(a) · b. It follows from Section 5 that A becomes a Jordan triple disystem by defining {a, b, c}1 = a · d(b) · d(c) + d(c) · d(b) · a, {a, b, c}2 = d(a) · b · d(c) + d(c) · b · d(a). Example 4.6. Let L be a Leibniz algebra over a field of characteristic not 2. If an element x ∈ L satisfies [x, [x, [x, L]]] = {0} then we define L(x) = { y ∈ L [x, [x, y]] = 0 }. By Vel´asquez and Felipe [24] (see also Gubarev and Kolesnikov [9]) we know that the quotient space Lx = L/L(x) becomes a Jordan dialgebra with the product ab = [a, [b, x]]. It follows from Section 7 that Lx has the structure of a Jordan triple disystem with the trilinear operations defined as follows for all a, b, c ∈ L: {a, b, c}1 = [[a, [b, x]], [c, x]] − [[a, [c, x]], [b, x]] + [a, [b, [c, x]]], {a, b, c}2 = [[b, [a, x]], [c, x]] + [[b, [c, x]], [a, x]] − [b, [a, [c, x]]]. 5. Jordan triple diproducts in an associative dialgebra In this section, we study two trilinear operations in an associative dialgebra. We use computer algebra to determine the identities satisfied by these operations of degree ≤ 5 and prove that these identities are equivalent to those of Theorem 4.2. We start by recalling the algorithm applied by Bremner and S´anchez-Ortega [5] to the alternating ternary sum. In the general case it converts a multilinear operation of degree n in an associative algebra into a family of n multilinear operations of degree n in an associative dialgebra. JORDAN TRIPLE DISYSTEMS 9 Definition 5.1. BSO algorithm. We start with a multilinear operation ω of degree n in an associative algebra over a field F, which we can identify with an element of the group algebra FSn: ω(a1, a2, . . . , an) = X σ∈Sn xσ aσ(1)aσ(2) · · · aσ(n) (xσ ∈ F). For all i, j = 1, 2, . . . , n we collect the terms in which ai is in position j; we write Sj,i n for the set of permutations σ with σ(j) = i: ωi(a1, a2, . . . , an) = nX j=1 X Sj,i n xσ aσ(1) · · · aσ(j−1)aiaσ(j+1) · · · aσ(n). We define n new multilinear operations in an associative dialgebra; bωi is obtained from ω by making ai the center of each dialgebra monomial: bωi(a1, a2, . . . , an) = nX j=1 X (i) n S xσ aσ(1) · · · aσ(j−1)baiaσ(j+1) · · · aσ(n). Definition 5.2. The Jordan triple product in an associative algebra A over a field of characteristic not 2 is the trilinear operation (a, b, c) = abc + cba. Definition 5.3. The Jordan triple diproducts are obtained by applying the BSO algorithm to the Jordan triple product: (a, b, c)1 = babc + cbba, (a, b, c)2 = abbc + cbba, (a, b, c)3 = abbc +bcba. It is clear that (a, b, c)3 = (c, b, a)1, so we will only consider (a, b, c)1 and (a, b, c)2. In the rest of this section, we use computer algebra to determine the multilinear polynomial identities of degrees 3 and 5 satisfied by the Jordan triple diproducts (· · · )1 and (· · · )2 over a field of characteristic 0. 5.1. Degree 3: operation 1. In this case a polynomial identity is a linear com- bination of the six permutations of (a, b, c)1: x1(a, b, c)1 + x2(a, c, b)1 + x3(b, a, c)1 + x4(b, c, a)1 + x5(c, a, b)1 + x6(c, b, a)1. We expand each diproduct to obtain a linear combination of the 18 multilinear dialgebra monomials of degree 3 ordered as follows: babc,bacb,bbac,bbca,bcab,bcba, abbc, abcb, bbac, bbca, cbab, cbba, abbc, acbb, babc, bcba, cabb, cbba. We construct the 18 × 6 matrix E in which the (i, j) entry is the coefficient of the i-th dialgebra monomial in the expansion of the j-th diproduct monomial:   . . . . . . . . . . . . . . . . 1 1 . . . . . . . . . . . . . 1 Et = . . 1 . . . . . . . . . . . . . . . . 1 . . . . . . . . . 1 . . . . 1 . . . . . . . . . . . 1 . . . . . 1 . . . . . . . . 1 . . . . . . 1 . . . . . 1 .   The coefficient vectors of the polynomial identities satisfied by (−, −, −)1 are the vectors in the nullspace of E, which is the zero subspace since rank(E) = 6. Lemma 5.4. The diproduct (· · · )1 satisfies no polynomial identity of degree 3. 10 MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA 5.2. Degree 3: operation 2. Replacing (· · · )1 by (· · · )2 gives the matrix Et =   . . . . . . 1 . . . . 1 . . . . . . . . . . . . . 1 . . . . . . . . . . . . . . . 1 . . . . . . . . . 1 . 1 1 . 1 . . . . . . . . . 1 . . . . . . . . . . . . . . . 1 . . . . . . . . . . . . . 1 . . . . 1 . . . . . .   We compute the row canonical form and obtain the canonical basis of the nullspace: (cid:2) 0 −1 0 1 0 0 (cid:3) , (cid:2) 0 0 −1 0 1 0 (cid:3) , (cid:2) −1 0 0 0 0 1 (cid:3) . Lemma 5.5. Every polynomial identity of degree 3 satisfied by the diproduct (· · · )2 follows from the symmetry in the first and third arguments: (a, b, c)2 ≡ (c, b, a)2. 5.3. Degree 5: operation 1. Since (· · · )1 satisfies no polynomial identity of de- gree 3, we consider three association types of degree 5: ((a, b, c)1, d, e)1, (a, (b, c, d)1, e)1, (a, b, (c, d, e)1)1. There are 3 · 5! multilinear diproduct monomials of these three types, and 5 · 5! dialgebra monomials of the forms babcde, abbcde, abbcde, abcbde, abcdbe. We expand the diproduct monomials in an associative dialgebra and obtain ((a, b, c)1, d, e)1 = babcde + cbbade + edbabc + edcbba, (a, (b, c, d)1, e)1 = babcde +badcbe + ebcdba + edcbba, (a, b, (c, d, e)1)1 = babcde +babedc + cdebba + edcbba. We construct the 600×360 matrix E in which the (i, j) entry is the coefficient of the i-th dialgebra monomial in the expansion of the j-th diproduct monomial. Using a computer algebra system, we find that rank(E) = 150, and hence the nullspace has dimension 210. We compute the canonical basis of the nullspace, and find that all the components are ±1. We sort these vectors by increasing number of nonzero components: there are 30 with two, 120 with four, and 60 with six. We construct the 480 × 360 matrix M with a 360 × 360 upper block and a 120 × 360 lower block. For each nullspace basis vector, we perform the following computations: (1) Apply all permutations of a, b, c, d, e to the corresponding linear combina- tion of diproduct monomials, and store the results in the lower block. (2) Compute the row canonical form; the lower block is now zero, and the upper block contains a basis for the subspace of the nullspace generated by the nullspace basis vectors up to the current vector. (3) If the rank of the matrix has increased from the previous vector to the current vector, then we record the current vector as a generator. We obtain three generators of the nullspace, corresponding to these identities: (a, (b, c, d)1, e)1 − (a, (d, c, b)1, e)1 ≡ 0, ((e, b, a)1, d, c)1 − ((e, d, c)1, b, a)1 − (e, (b, a, d)1, c)1 + (e, d, (c, b, a)1)1 ≡ 0, ((e, b, a)1, d, c)1 − ((e, d, c)1, b, a)1 − (e, (b, a, d)1, c)1 + (e, d, (a, b, c)1)1 ≡ 0. A similar computation shows that no two of these identities generate the entire nullspace. The difference of the second and third identities is (e, d, (c, b, a)1)1 − (e, d, (a, b, c)1)1 ≡ 0, and this gives a simpler set of three generating identities. JORDAN TRIPLE DISYSTEMS 11 Proposition 5.6. Every polynomial identity of degree 5 satisfied by the diproduct (· · · )1 is a consequence of these three independent identities: (a, (b, c, d)1, e)1 ≡ (a, (d, c, b)1, e)1, ((e, d, c)1, b, a)1 ≡ ((e, b, a)1, d, c)1 − (e, (d, a, b)1, c)1 + (e, d, (c, b, a)1)1. (a, b, (c, d, e)1)1 ≡ (a, b, (e, d, c)1)1, 5.4. Degree 5: operation 2. Lemma 5.5 implies that we need to consider only two association types for the diproduct (· · · )2 of degree 5: ((a, b, c)2, d, e)2 = abcbde + cbabde + ebdabc + ebdcba, (a, (b, c, d)2, e)2 = abbcde + adbcbe + ebbcda + edbcba. The first type is symmetric in a and c, giving 5!/2 monomials; the second is sym- metric in a and e and in b and d, giving 5!/4 monomials. We construct the 600 × 90 matrix E in which the (i, j) entry is the coefficient of the i-th dialgebra monomial in the expansion of the j-th diproduct monomial. We find that rank(E) = 90. Proposition 5.7. Every polynomial identity of degree 5 satisfied by the diproduct (· · · )2 is a consequence of the symmetry of Lemma 5.5. 5.5. Degree 5: operations 1 and 2. We now consider multilinear polynomial identities which involve both diproducts (· · · )1 and (· · · )2. In addition to the three association types for (· · · )1 and the two association types for (· · · )2, we must also consider the five association types involving both operations: 1: 2: 3: 4: 5: ((a, b, c)1, d, e)1 (a, (b, c, d)1, e)1 (a, b, (c, d, e)1)1 ((a, b, c)2, d, e)2 (a, (b, c, d)2, e)2 6: 7: 8: 9: 10: ((a, b, c)2, d, e)1 (a, (b, c, d)2, e)1 (a, b, (c, d, e)2)1 ((a, b, c)1, d, e)2 (a, (b, c, d)1, e)2 Lemma 5.5 and Proposition 5.6 imply the following results: symmetries − b ↔ d c ↔ e a ↔ c a ↔ e; b ↔ d monomials 5! = 120 5!/2 = 60 5!/2 = 60 5!/2 = 60 5!/4 = 30 1: 2: 3: 4: 5: symmetries a ↔ c b ↔ d c ↔ e − a ↔ e monomials 5!/2 = 60 5!/2 = 60 5!/2 = 60 5! = 120 5!/2 = 60 6: 7: 8: 9: 10: The total number of multilinear diproduct monomials is 690. In this case, the ma- trix E has size 600 × 690, and as before the (i, j) entry is the coefficient of the i-th dialgebra monomial in the expansion of the j-th diproduct monomial. For such a large matrix we use modular arithmetic to do Gaussian elimination in order to con- trol the memory requirements. Since we consider multilinear monomials, all vector spaces are representations of S5; if use a modulus p > 5 we will obtain dimensions equivalent to those which we would have obtained using rational arithmetic. We find that rank(E) = 250, and hence the nullspace has dimension 440. We compute the canonical basis and sort it by increasing number of nonzero components. We now construct an 810 × 690 matrix M with a 690 × 690 upper block and a 120 × 690 lower block. Before processing the 440 nullspace identities, we first process the third identity of Proposition 5.6, which we rewrite as follows: ((a, b, c)1, d, e)1 − ((a, d, e)1, b, c)1 + (a, (b, e, d)1, c)1 − (a, b, (c, d, e)1)1 ≡ 0. 12 MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA This is the only known identity that has not already been taken into account; the other identities from Lemma 5.5 and Proposition 5.6 express symmetries which have been assumed in our enumeration of the diproduct monomials. This identity generates a 90-dimensional subspace of the nullspace. Continuing with the 440 nullspace vectors, we obtain six additional generators (we omit "≡ 0"): (31) (32) (33) (34) (35) (36) ((c, b, a)2, d, e)2 − ((a, b, c)1, d, e)2, (a, (b, c, d)1, e)1 − (a, (b, c, d)2, e)1, ((e, d, c)2, b, a)2 − ((a, b, e)2, d, c)1 − ((a, b, c)2, d, e)1 + (e, (b, a, d)1, c)2, ((e, b, a)2, d, c)2 + ((c, b, a)2, d, e)2 − ((e, d, c)2, b, a)1 − (e, (d, a, b)1, c)2, ((e, b, a)1, d, c)1 − ((e, d, c)1, b, a)1 − (e, (b, a, d)1, c)1 + (e, d, (a, b, c)2)1, ((a, b, c)1, d, e)1 + ((a, d, c)1, b, e)1 − (a, (b, c, d)1, e)1 − (c, (b, a, d)2, e)2. Further calculations show that if we take any proper subset of these six identities, and combine it with the third identity of Proposition 5.6, then the resulting set of identities does not generate the entire nullspace. It follows that identities (31) -- (36) are independent, modulo the third identity of Proposition 5.6. Using Lemma 5.5 and Proposition 5.6, we see that (31), (32), (35) are equivalent to ((a, b, c)2, d, e)2 ≡ ((a, b, c)1, d, e)2, (a, (b, c, d)1, e)1 ≡ (a, (b, c, d)2, e)1, ((e, b, a)1, d, c)1 − ((e, d, c)1, b, a)1 − (e, (d, a, b)1, c)1 + (e, d, (c, b, a)2)1 ≡ 0. From the last identity we subtract the third identity of Proposition 5.6 and obtain (a, b, (c, d, e)1)1 ≡ (a, b, (c, d, e)2)1. We now see that identities (31) -- (36) are equivalent to the following six identities (we omit "≡ 0" in the last three): ((a, b, c)2, d, e)2 ≡ ((a, b, c)1, d, e)2, (a, (b, c, d)1, e)1 ≡ (a, (b, c, d)2, e)1, (a, b, (c, d, e)1)1 ≡ (a, b, (c, d, e)2)1, ((e, d, c)2, b, a)2 − ((a, b, e)2, d, c)1 − ((a, b, c)2, d, e)1 + (e, (b, a, d)1, c)2, ((e, b, a)2, d, c)2 + ((c, b, a)2, d, e)2 − ((e, d, c)2, b, a)1 − (e, (d, a, b)1, c)2, ((a, b, c)1, d, e)1 + ((a, d, c)1, b, e)1 − (a, (b, c, d)1, e)1 − (c, (b, a, d)2, e)2. The last three identities are equivalent (assuming known symmetries) to ((e, d, c)1, b, a)2 ≡ ((e, b, a)2, d, c)1 − (e, (b, a, d)1, c)2 + ((c, b, a)2, d, e)1. ((e, d, c)2, b, a)1 ≡ ((e, b, a)1, d, c)2 − (e, (d, a, b)1, c)2 + (e, d, (c, b, a)1)2. (a, (b, c, d)2, e)2 ≡ ((c, b, a)1, d, e)1 + ((c, d, a)1, b, e)1 − (c, (b, a, d)1, e)1. We have proved the following result. Theorem 5.8. Every polynomial identity of degree ≤ 5 satisfied by the Jordan triple diproducts (· · · )1 and (· · · )2 in an associative dialgebra is a consequence of these eight independent identities: (a, b, c)2 ≡ (c, b, a)2, (a, (b, c, d)1, e)1 ≡ (a, (b, c, d)2, e)1, JORDAN TRIPLE DISYSTEMS 13 (a, b, (c, d, e)1)1 ≡ (a, b, (c, d, e)2)1, ((a, b, c)1, d, e)2 ≡ ((a, b, c)2, d, e)2, ((e, d, c)1, b, a)1 ≡ ((e, b, a)1, d, c)1 − (e, (d, a, b)1, c)1 + (e, d, (c, b, a)1)1, ((e, d, c)2, b, a)1 ≡ ((e, b, a)1, d, c)2 − (e, (d, a, b)1, c)2 + (e, d, (c, b, a)1)2, ((e, d, c)1, b, a)2 ≡ ((e, b, a)2, d, c)1 − (e, (b, a, d)1, c)2 + ((c, b, a)2, d, e)1, (a, (b, c, d)2, e)2 ≡ ((c, b, a)1, d, e)1 + ((c, d, a)1, b, e)1 − (c, (b, a, d)1, e)1. Remark 5.9. These identities are equivalent to those of Theorem 4.2. We have (a, (b, c, d)1, e)1 ≡ (a, (b, c, d)2, e)1 ≡ (a, (d, c, b)2, e)1 ≡ (a, (d, c, b)1, e)1, (a, b, (c, d, e)1)1 ≡ (a, b, (c, d, e)2)1 ≡ (a, b, (e, d, c)2)1 ≡ (a, b, (e, d, c)1)1, ((a, b, c)1, d, e)2 ≡ ((a, b, c)2, d, e)2 ≡ ((c, b, a)2, d, e)2 ≡ ((c, b, a)1, d, e)2, corresponding to the identities in the first four lines displayed in Theorem 4.2. The fifth and sixth identities of Theorem 5.8 coincide with the fifth and sixth lines displayed in Theorem 4.2. Using the known identities, ((e, d, c)1, b, a)2 ≡ ((c, d, e)1, b, a)2 ≡ (a, b, (c, d, e)1)2, ((e, b, a)2, d, c)1 ≡ ((a, b, e)2, d, c)1, (e, (b, a, d)1, c)2 ≡ (c, (b, a, d)1, e)2, ((c, b, a)2, d, e)1 ≡ ((a, b, c)2, d, e)1, we see that the seventh identity of Theorem 5.8 becomes the eighth line displayed in Theorem 4.2. To conclude, we apply the transposition ac to the eighth identity of Theorem 5.8: (c, (b, a, d)2, e)2 ≡ ((a, b, c)1, d, e)1 + ((a, d, c)1, b, e)1 − (a, (b, c, d)1, e)1 Using this equivalent form of the fifth identity of Theorem 5.8, ((a, d, c)1, b, e)1 − (a, (b, c, d)1, e)1 ≡ −(a, b, (c, d, e)1)1 + ((a, b, e)1, d, c)1, we obtain (c, (b, a, d)2, e)2 ≡ ((a, b, c)1, d, e)1 − (a, b, (c, d, e)1)1 + ((a, b, e)1, d, c)1, which is equivalent to the seventh line displayed in Theorem 4.2. Theorem 5.10. If D is a subspace of an associative dialgebra over a field of char- acteristic not 2, 3, 5 which is closed under the Jordan diproducts (· · · )1 and (· · · )2, then D is a Jordan triple disystem with respect to these operations. We exclude characteristics 2, 3, 5 since our computer algebra methods require modular arithmetic with a prime greater than the degree of the identities. 6. Conjecture: dialgebra operations and identities The results of Sections 4 and 5 suggest a general conjecture relating multilinear operations in associative dialgebras and their polynomial identities. Fix a coefficient field F and an integer n ≥ 2. Let ω be a multilinear n-ary operation in an associative algebra over F; we can identify ω with an element of the group algebra FSn. Fix a degree d and consider the multilinear polynomial identities of degree e ≤ d satisfied by ω; we may assume that d ≡ 1 (mod n−1). To be precise, let Ae be the multilinear subspace of degree e in the free nonassociative n-ary algebra over F on e generators. Let Ie be the subspace of Ae consisting of those nonassociative n-ary polynomials which vanish identically when the n-ary 14 MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA ω y Id(ω) BSO−−−−→ ω1, . . . , ωn y KP−−−−→ Jd(ω1, . . . , ωn) ?= KPd(ω) Figure 1. Diagrammatic formulation of Conjecture 6.1 operation is replaced by ω; thus Ie is the kernel of the expansion map from Ae to the group algebra FSe regarded as the multilinear subspace of degree e in the free associative algebra on e generators. By definition, the multilinear polynomial identities of degree ≤ d satisfied by ω are the direct sum Id(ω) = M 1≤e≤d Ie. We apply the KP algorithm to the identities in Id(ω) and obtain a set of multilinear polynomial identities involving n new n-ary operations. To be precise, let Be be the multilinear subspace of degree e in the free nonassociative multioperator algebra with n operations of arity n. Let KP(Ie) be the subspace of Be obtained by applying the KP algorithm to the identities in Ie, and define KPd(ω) = M 1≤e≤d KP(Ie). We now consider a different path to the same destination. We apply the BSO algorithm to ω and obtain n multilinear operations ω1, . . . , ωn of arity n in an associative dialgebra. Consider the multilinear polynomial identities of degree e ≤ d satisfied by ω1, . . . , ωn. To be precise, let Je be the subspace of Be consisting of those nonassociative polynomials in the n operations which vanish identically when the n operations are replaced by ω1, . . . , ωn; thus Je is the kernel of the expansion map from Be to the direct sum of e copies of the group algebra FSe regarded as the multilinear subspace of degree e in the free associative dialgebra on e generators. We define Jd(ω1, . . . , ωn) = M Je. 1≤e≤d Conjecture 6.1. If the field F has characteristic 0 or p > d then KPd(ω) = Jd(ω1, . . . , ωn). Expressed less formally, the conjecture says that the following two processes produce the same results when the group algebra FSd is semisimple: • Find the identities satisfied by ω, and apply the KP algorithm. • Apply the BSO algorithm, and find the identities satisfied by ω1, . . . , ωn. The conjecture is equivalent to the commutativity of the diagram in Figure 1. 7. Jordan triple diproducts in a Jordan dialgebra Every Jordan algebra J with operation a ◦ b gives rise to a Jordan triple system by considering the same underlying vector space with respect to the triple product (37) ha, b, ci = (a ◦ b) ◦ c − (a ◦ c) ◦ b + a ◦ (b ◦ c). JORDAN TRIPLE DISYSTEMS 15 In this section we consider a Jordan dialgebra D with operation ab and the two trilinear operations which give rise to the structure of a Jordan triple disystem on the same underlying vector space. The first trilinear operation h· · · i1 can be obtained by replacing the Jordan algebra product a ◦ b in (37) by the the Jordan dialgebra product ab in D: (38) ha, b, ci1 = (ab)c − (ac)b + a(bc). The second trilinear operation h· · · i2 can be obtained from the first by transposing a and b and changing the signs of the second and third terms: (39) ha, b, ci2 = (ba)c + (bc)a − b(ac). It is straightforward to verify that in a special Jordan dialgebra, where the product is ab = a ⊣ b + b ⊢ a, these operations reduce to twice the first and second Jordan diproducts in an associative dialgebra: (a ⊣ b + b ⊢ a) ⊣ c + c ⊢ (a ⊣ b + b ⊢ a) − (a ⊣ c + c ⊢ a) ⊣ b − b ⊢ (a ⊣ c + c ⊢ a) + a ⊣ (b ⊣ c + c ⊢ b) + (b ⊣ c + c ⊢ b) ⊢ a = babc + bbac + cbab + cbba −bacb − cbab − bbac − bcba +babc +bacb + bcba + cbba = 2(cid:0)babc + cbba(cid:1), (b ⊣ a + a ⊢ b) ⊣ c + c ⊢ (b ⊣ a + a ⊢ b) + (b ⊣ c + c ⊢ b) ⊣ a + a ⊢ (b ⊣ c + c ⊢ b) − b ⊣ (a ⊣ c + c ⊢ a) − (a ⊣ c + c ⊢ a) ⊢ b =bbac + abbc + cbba + cabb +bbca + cbba + abbc + acbb −bbac −bbca − acbb − cabb = 2(cid:0)abbc + cbba(cid:1). Lemma 7.1. In a Jordan dialgebra D with operation ab, every polynomial identity of degree 3 satisfied by ha, b, ci1 and ha, b, ci2 is a consequence of the symmetry of ha, b, ci2 in its first and third arguments: ha, b, ci2 ≡ hc, b, ai2. Proof. We use computer algebra. We construct an 18 × 24 matrix E in which • the upper left 6 × 12 block contains the right commutative identities • the lower left 12 × 12 block contains the expansions of h· · · i1 and h· · · i2 • the lower right 12 × 12 block contains the identity matrix • the upper right 6 × 12 block contains the zero matrix More precisely, columns 1 -- 12 of the matrix correspond to the 12 multilinear mono- mials of degree 3 in the free nonassociative algebra, (ab)c, (ac)b, (ba)c, (bc)a, (ca)b, (cb)a, a(bc), a(cb), b(ac), b(ca), c(ab), c(ba), and columns 13 -- 24 correspond to the 12 multilinear monomials of degree 3 in the trilinear operations h· · · i1 and h· · · i2, ha, b, ci1, ha, c, bi1, hb, a, ci1, hb, c, ai1, hc, a, bi1, hc, b, ai1, ha, b, ci1, ha, c, bi1, hb, a, ci1, hb, c, ai1, hc, a, bi1, hc, b, ai1. There are six permutations of the right-commutative identity: a(bc)−a(cb), a(cb)−a(bc), b(ac)−b(ca), b(ca)−b(ac), c(ab)−c(ba), c(ba)−c(ab). 16 MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA   . . . . . . . . . . . + − . . − + . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . + − . . − + . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . + − . − + . . . . . . . . . . . . . . . . . . . . . . . . + − . − + . . . . + . . . . . . . . . + . . . . + . . + . . . . . . . . . . . . . . . . . . . . . . . + − . . − + . . . . + . . . . . + . . . . . . . . . . . . . + . . . . . + . . . . . . . . . . . . . . . + − . . − + . . . . . . + . . + . . . . . + . . . . . . . . . . . . + . . . . . . . . + + . . . . − . . . . . . . . . + . . . . . . . . + + . . . . + + . . + + . . . . − . . . . . . . . + . . . . . . . − . . . . . . . . . . . . . + . . . . + + . . . . . − . . . . . . . . . + . . . . . . − . . . . . . . . . . . . . . . + + . . . . . − . . . . . . . . . . . . + . . + .     Figure 2. The 18 × 24 matrix for the proof of Lemma 7.1 + . . + . . . . . . . . . . . ∗ . . . . . . . . . . . . . . . . . . . ∗ . . . . . . . . ∗ ∗ . . + . . . . . . . . . . . ∗ . . . . . . . . . . . + . . . . . . . . . . . ∗ . . . . . . . . . ∗ ∗ . . . . + . . . . . . . . . . . ∗ . . . . . . . . . + . . . . . . . . . . . ∗ . . . ∗ ∗ . . . . . . . . . . + . . . . . . . . . . . . + . . . . ∗ ∗ ∗ ∗ . . . . . . . . + . . . . . . . . . . . . . . + . . . . ∗ ∗ ∗ ∗ . . . . . . . . . . + . . . . . . . . . . . . . . . . + . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ∗ ∗ ∗ ∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . + . . . . − . . . . . . . . . . . . . . . . . . . + . − . . . . . . . . . . . . . . . . . . . . . . + . − . Figure 3. The row canonical form of the matrix of Figure 2   Entry (i, j) of the upper left block is the coefficient of the j-th nonassociative mono- mial in the i-th permutation of the right-commutative identity. Entry (12+i, j) of the lower left block contains is the coefficient of the j-th nonassociative monomial in the expansion of the i-th trilinear monomial using equations (38) and (39). This matrix is displayed in Figure 2, using the symbols ·, +, − for 0, 1, −1. The rank is 15, and the row canonical form is displayed in Figure 3, using the symbol ∗ for 1 2 . The dividing line between the upper and lower parts of the matrix lies immediately above row 13, since that is the uppermost row whose leading 1 is in the right part of the matrix. These rows represent the dependence relations among the expansions of the trilinear monomials which hold as a result of the right-commutative identities. The rows of the lower right 3 × 12 block of the row canonical form represent the permutations of the identity ha, b, ci2 − hc, b, ai2 ≡ 0. (cid:3) Theorem 7.2. In a Jordan dialgebra D with operation ab, every polynomial iden- tity of degree 5 satisfied by ha, b, ci1 and ha, b, ci2 is a consequence of the identity of Lemma 7.1 together with these seven independent identities: ha, hb, c, di1, ei1 − ha, hd, c, bi2, ei1 ≡ 0, ha, b, hc, d, ei1i1 − ha, b, he, d, ci2i1 ≡ 0, hha, b, ci1, d, ei2 − hhc, b, ai2, d, ei2 ≡ 0, hha, b, ci2, d, ei1 − hhe, d, ai2, b, ci2 + ha, hb, e, di1, ci2 − hhe, d, ci2, b, ai2 ≡ 0, hha, b, ci2, d, ei1 − ha, hb, c, di1, ei2 − he, hb, a, di1, ci2 + hha, d, ci2, b, ei2 ≡ 0, JORDAN TRIPLE DISYSTEMS 17 hha, b, ci1, d, ei1 + hha, b, ei1, d, ci1 − ha, b, hc, d, ei2i1 − hc, hd, a, bi2, ei2 ≡ 0, hha, b, ci1, d, ei1 + hha, d, ci1, b, ei1 − ha, hb, c, di2, ei1 − he, hb, a, di2, ci2 ≡ 0. Proof. The basic strategy is the same as in the proof of Lemma 7.1, but the matrix is much larger and some further computations are required. There are 14 association types for a nonassociative binary operation of degree 5, (((ab)c)d)e, ((a(bc))d)e, ((ab)(cd))e, (a((bc)d))e, (a(b(cd)))e, ((ab)c)(de), (a(bc))(de), (ab)((cd)e), (ab)(c(de)), a(((bc)d)e), a((b(cd))e), a((bc)(de)), a(b((cd)e)), a(b(c(de))). and 5! permutations of the variables a, b, c, d, e, giving 1680 multilinear monomials, which correspond to the columns in the left part of the matrix; we order them first by association type and then by lexicographical order of the permutation. We need to generate all the consequences of degree 5 of the defining identities for Jordan dialgebras. A multilinear identity I(a1, . . . , an) of degree n produces n+2 identities of degree n+1 (we have n substitutions and two multiplications): I(a1an+1, . . . , an), . . . , I(a1, . . . , anan+1), I(a1, . . . , an)an+1, an+1I(a1, . . . , an). The right-commutative identity of degree 3 produces 5 identities of degree 4, and each of these produces 6 identities of degree 5, for a total of 30. The linearized versions of the two identities of degree 4 from Definition 3.5 each produce 6 identities of degree 5, for a total of 12. Altogether we have 42 identities of degree 5, and each allows 5! permutations of the variables, for a total of 5040 identities. The last two paragraphs show that the upper left block has size 5040 × 1680; its (i, j) entry is the coefficient of the j-th nonassociative monomial in the i-th identity. There are 10 association types for two trilinear operations of degree 5, assuming that the second operation is symmetric in its first and third arguments: 1: 2: 3: 4: 5: hha, b, ci1, d, ei1 ha, hb, c, di1, ei1 ha, b, hc, d, ei1i1 hha, b, ci2, d, ei2 ha, hb, c, di2, ei2 6: 7: 8: 9: 10: hha, b, ci2, d, ei1 ha, hb, c, di2, ei1 ha, b, hc, d, ei2i1 hha, b, ci1, d, ei2 ha, hb, c, di1, ei2 The symmetry of h· · · i2 gives the number of multilinear monomials in each type: 120 + 120 + 120 + 60 + 60 + 60 + 120 + 60 + 60 + 30 = 810; these monomials correspond to the columns in the right part of the matrix. The lower left block has size 810 × 1680; its (i, j) entry is the coefficient of the j-th nonassociative monomial in the expansion, using equations (38) and (39), of the i-th monomial in the operations h· · · i1 and h· · · i2. The lower right block is the 810 × 810 identity matrix, and the upper right block is the 5040 × 810 zero matrix. (See Figure 4.) We compute the row canonical form of this matrix and find that its rank is 2215. We ignore the first 1655 rows since their leading ones are in the left part, and retain only the next 560 rows which have have their leading ones in the right part. We sort these rows by increasing number of nonzero components. We construct another matrix with an upper block of size 810 × 810 and a lower block of size 120 × 810. For each of the 560 identities satisfied by the operations h· · · i1 and h· · · i2, we apply all 5! permutations of the variables, store the permuted identities in the lower block, and compute the row canonical form of the matrix; 18 MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA   consequences in degree 5 of the Jordan dialgebra identities from Definition 3.5 expansions using (38) and (39) of the monomials of degree 5 for h· · · i1 and h· · · i2 zero matrix identity matrix   Figure 4. Matrix for the proof of Theorem 7.2 after each iteration, the lower block is the zero matrix. We record the index numbers of the identities which increase the rank, and obtain these results: identity rank 1 121 241 301 331 342 451 454 120 240 360 390 450 470 530 560 Further computations show that identity 301 is superfluous: the other seven identi- ties generate the entire 560-dimensional space. These seven identities are displayed in standard notation in the statement of this Theorem. (cid:3) Theorem 7.3. If D is a subspace of a Jordan dialgebra over a field of character- istic not 2, 3, 5 which is closed under the trilinear operations h· · · i1 and h· · · i2 of equations (38) and (39), then D is a Jordan triple disystem with respect to these operations. Proof. Theorem 7.1 tells us that identity (J1) is satisfied: ha, b, ci2 ≡ hc, b, ai2. Identities (J2), (J3) and (J4) can be easily obtained from (J1) and the first three identities of Theorem 7.2: ha, hb, c, di1, ei1 ≡ ha, hd, c, bi2, ei1 ≡ ha, hb, c, di2, ei1, ha, b, hc, d, ei1i1 ≡ ha, b, he, d, ci2i1 ≡ ha, b, hc, d, ei2i1, hha, b, ci1, d, ei2 ≡ hhc, b, ai2, d, ei2 ≡ hhc, b, ai2, d, ei2. Identity (J7) is a consequence of the sixth identity of Theorem 7.2 by applying (J1) and (J3). To obtain (J6), we apply the transpositions ae and bd to the fourth identity of Theorem 7.2: hhe, d, ci2, b, ai1 − hha, b, ei2, d, ci2 + he, hd, b, ai1, ci2 − hha, b, ci2, d, ei2 ≡ 0. Using identities (J1) and (J4) we obtain these two identities, hha, b, ei2, d, ci2 ≡ hhe, b, ai2, d, ci2 ≡ hhe, b, ai1, d, ci2, hha, b, ci2, d, ei2 ≡ hhc, b, ai2, d, ei2 ≡ hhc, b, ai1, d, ei2, and substituting these in the previous identity gives (J6). To obtain identity (J8), we apply (J1) and (J4) to the fifth identity of Theorem 7.2: (40) hha, b, ci2, d, ei1 − ha, hb, c, di1, ei2 − hc, hb, a, di1, ei2 + hha, d, ci1, b, ei2 ≡ 0. JORDAN TRIPLE DISYSTEMS 19 Identity (J6) gives −hc, hb, a, di1, ei2 + hha, d, ci1, b, ei2 ≡ −ha, b, he, d, ci1i2 + hha, b, ei2, d, ci1. On the other hand, using identities (J1) and (J4) we obtain ha, b, he, d, ci1i2 ≡ hhe, d, ci1, b, ai2 ≡ hhe, d, ci2, b, ai2 ≡ hhc, d, ei2, b, ai2 ≡ hhc, d, ei1, b, ai2 ≡ ha, b, hc, d, ei1i2. Applying these considerations to identity (40) we obtain hha, b, ci2, d, ei1 − hc, hb, a, di1, ei2 + hha, b, ei2, d, ci1 − ha, b, hc, d, ei1i2 ≡ 0, which is equivalent to (J8). It remains to show (J5). We apply (J1) and (J2) to the last identity of Theorem 7.2 and obtain (41) hha, b, ci1, d, ei1 + hha, d, ci1, b, ei1 − ha, hb, c, di1, ei1 − hc, hb, a, di2, ei2 ≡ 0. From (J7) we have: hc, hb, a, di2, ei2 ≡ −ha, b, hc, d, ei1i1 + hha, b, ci1, d, ei1 + hha, b, ei1, d, ci1, and substituting this in (41) gives (42) hha, d, ci1, b, ei1 − ha, hb, c, di1, ei1 + ha, b, hc, d, ei1i1 − hha, b, ei1, d, ci1 ≡ 0. To finish, we observe that ha, b, hc, d, ei1i1 ≡ ha, b, hc, d, ei2i1 ≡ ha, b, he, d, ci2i1 ≡ ha, b, he, d, ci1i1, and applying these to (42) yields hha, d, ci1, b, ei1 − ha, hb, c, di1, ei1 + ha, b, he, d, ci1i1 − hha, b, ei1, d, ci1 ≡ 0, which is equivalent to (J5). (cid:3) Second proof of Theorem 7.3. We conclude with an alternative proof, with- out using computer algebra, of the defining identities (J1) -- (J8) for Jordan triple disystems with respect to the operations h· · · i1 and h· · · i2 in a Jordan dialgebra: ha, b, ci1 = (ab)c − (ac)b + a(bc), ha, b, ci2 = (ba)c + (bc)a − b(ac). Definition 7.4. Let D be a Jordan dialgebra with operation ab. The annihilator Dann is the ideal spanned by { [a, b] = ab − ba a, b ∈ D }. The right center Z r(D) is the ideal consisting of the elements b ∈ D such that ab = 0 for all a ∈ D. A derivation is a linear map δ : D → D satisfying δ(ab) = δ(a)b + aδ(b) for all a, b ∈ D. A left derivation is a linear map µ : D → D satisfying µ(ab) = µ(a)b+µ(b)a for all a, b ∈ D. For a ∈ D the right and left multiplication operators Ra : D → D, La : D → D are defined by Rab = ba, Lab = ab for all b ∈ D. Lemma 7.5. We have Dann ⊆ Z r(D) for any Jordan dialgebra D. Proof. Right commutativity gives c[a, b] = c(ab) − c(ba) = 0. (cid:3) Lemma 7.6. If D is a Jordan dialgebra then for all a, b, c ∈ D we have ha, b, ci1 − hc, b, ai1 ∈ Dann, ha, b, ci1 − ha, b, ci2 ∈ Dann. 20 MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA Proof. Right commutativity gives ha, b, ci1 − hc, b, ai1 = (ab)c − (ac)b + a(bc) − (cb)a + (ca)b − c(ba) = (ab)c − c(ab) + (ca − ac)b + a(cb) − (cb)a = [ab, c] + [c, a]b + [a, cb], ha, b, ci1 − ha, b, ci2 = (ab)c − (ac)b + a(bc) − (ba)c − (bc)a + b(ac) = (ab − ba)c + a(bc) − (bc)a + b(ac) − (ac)b = [a, b]c + [a, bc] + [b, ac]. Both expressions belong to Dann since the annihilator is an ideal. (cid:3) Proposition 7.7. If D is a Jordan dialgebra then identities (J1) -- (J4) are satisfied by the operations h· · · i1 and h· · · i2: ha, b, ci2 ≡ hc, b, ai2, ha, hb, c, di1, ei1 ≡ ha, hb, c, di2, ei1 ≡ ha, hd, c, bi1, ei1, ha, b, hc, d, ei1i1 ≡ ha, b, hc, d, ei2i1 ≡ ha, b, he, d, ci1i1, hha, b, ci1, d, ei2 ≡ hha, b, ci2, d, ei2 ≡ hhc, b, ai1, d, ei2. Proof. The first line follows directly from right commutativity. For the second line, using Lemma 7.6 it suffices to show ha, x, ei1 ≡ 0 for x ∈ Dann, and again this follows from right commutativity. The third and fourth lines are similar. (cid:3) Right commutativity is equivalent to both Rab = Rba and LaLb = LaRb. The operations h· · · i1 and h· · · i2 can be expressed using multiplication operators: ha, b, ci1 = RcRb a − RbRc a + Rcb a = (cid:0) Rcb + [Rc, Rb](cid:1)a = Lab c − RbLa c + LaLb c = (cid:0) Lab + [La, Rb](cid:1)c, ha, b, ci2 = Lba c + RaLb c − LbLa c = (cid:0)Lba − [Lb, Ra](cid:1)c. It has been shown by Felipe and Vel´asquez [8, 26] that [Ra, Rb] is a derivation, that [La, Rb] is a left derivation, and that for all a, b, c ∈ D we have (43) (44) R[Ra,Rb]c = [[Ra, Rb], Rc], L[La,Rb]c = [[La, Rb], Rc]. Lemma 7.8. If D is a Jordan dialgebra then for i = 1, 2 and a, b, c, d ∈ D we have (1) Rd ha, b, cii = h Rd a, b, c ii − h a, Rd b, c ii + h a, b, Rd c ii, (2) Ld ha, b, cii = h Ld a, b, c i1 − h a, Rd b, c i2 + h Ld c, b, a i1. Proof. (1) For i = 1 we use the linearization of the right Jordan identity, (45) [Ra, Rbc] + [Rb, Rca] + [Rc, Rab] = 0. By definition of h· · · i1 we have Rdha, b, ci1 = ((ab)c)d − ((ac)b)d + (a(bc))d, hRda, b, ci1 = ((ad)b)c − ((ad)c)b + (ad)(bc), ha, Rdb, ci1 = (a(bd))c − (ac)(bd) + a((bd)c), ha, b, Rdci1 = (ab)(cd) − (a(cd))b + a(b(cd)). Using right commutativity and equations (43) and (45) we obtain Rdha, b, ci1 − hRda, b, ci1 + ha, Rdb, ci1 − ha, b, Rdci1 = (a(bc))d − (ad)(bc) + (a(cd))b − (ab)(cd) + (a(bd))c − (ac)(bd) JORDAN TRIPLE DISYSTEMS 21 + ((ab)c)d − ((ac)b)d − ((ad)b)c + ((ad)c)b + a((bd)c) − a(b(cd)) = (cid:0)[Rd, Rcb] + [Rb, Rcd] + [Rc, Rbd] + [Rd, [Rc, Rb]] + R[Rc,Rb]d(cid:1)a = 0. For i = 2 the proof is similar, using (44) and this equation proved in [26]: (46) [La, Rbc] + [Rb, Lac] + [Rc, Lab] = 0, (2) For i = 1, right commutativity and the definitions of h· · · i1 and h· · · i2 give Ldha, b, ci1 − hLda, b, ci1 + ha, Ldb, ci2 − hLdc, b, ai1 = (cid:0)LdLab − LdRbLa + LdLaLb − LRbRad + RbLda − LdaRb + LRaRbd + RaLdb − LdbLa − RaRbLd + RbRaLd − RbaLd(cid:1)c = (cid:0)LdRab − LdRbRa + LdRaRb + L[Ra,Rb]d + RbLda − LdaRb + RaLdb − LdbRa − [Ra, Rb]Ld − RabLd(cid:1)c = (cid:0)[Ld, Rab] + [Rb, Lda] + [Ra, Ldb] + L[Ra,Rb]d(cid:1)c = 0. We have used (46) and (44) and the fact that L[Ra,Rb]d = [[Ra, Rb], Ld], since [Ra, Rb] is a derivation of D. For i = 2, it suffices to observe that right commutativity implies Ldha, b, ci1 = Ldha, b, ci2. (cid:3) Lemma 7.9. If D is a Jordan dialgebra with derivation δ and left derivation µ then for i = 1, 2 and a, b, c, d ∈ D we have δha, b, cii = hδa, b, cii + ha, δb, cii + ha, b, δcii, µha, b, cii = hµa, b, ci1 + ha, µb, ci2 + hµc, b, ai1. Proof. Straightforward. (cid:3) Proposition 7.10. If D is a Jordan dialgebra then identities (J5) -- (J8) are satisfied by the operations h· · · i1 and h· · · i2. Proof. For a, b, c, d, e ∈ D we apply Lemmas 7.8 (1) and 7.9 to get hhe, d, ci1, b, ai1 = (cid:0)Rab + [Ra, Rb](cid:1)he, d, ci1 = h(Rab+[Ra, Rb])e, d, ci1 − he, (Rba+[Rb, Ra])d, ci1 + he, d, (Rab+[Ra, Rb])ci1 = hhe, b, ai1, d, ci1 − he, hd, a, bi1, ci1 + he, d, hc, b, ai1i1, which is identity (J5). The proof of identity (J6) is similar, replacing the outermost operation h· · · i1 by h· · · i2. Finally, Lemmas 7.8 (2) and 7.9 give ha, b, hc, d, ei1i1 = (cid:0)Lab + [La, Rb](cid:1)hc, d, ei1 = h(Lab+[La, Rb])c, d, ei1 − hc, (Lab−[La, Rb])d, ei2 + h(Lab+[La, Rb])e, d, ci1 = hha, b, ci1, d, ei1 − hc, hb, a, di2, ei2 + hha, b, ei1, d, ci1, ha, b, hc, d, ei1i2 = (cid:0)Lba − [Lb, Ra](cid:1)hc, d, ei1 = h(Lba−[Lb, Ra])c, d, ei1 − hc, (Lba+[Lb, Ra])d, ei2 + h(Lba−[Lb, Ra])e, d, ci1 = hha, b, ci2, d, ei1 − hc, hb, a, di1, ei2 + hha, b, ei2, d, ci1, which are identities (J7) and (J8). The proof is complete. (cid:3) 22 MURRAY R. BREMNER, RA ´UL FELIPE, AND JUANA S ´ANCHEZ-ORTEGA Acknowledgements Murray Bremner was supported by a Discovery Grant from NSERC; he thanks the Centro de Investigaci´on en Matem´aticas in Guanajuato (Mexico) for its hos- pitality during his visit in February 2011. Ra´ul Felipe was supported by CONA- CyT grant 106923. Juana S´anchez-Ortega was supported by the Spanish MEC and Fondos FEDER jointly through project MTM2010-15223, and by the Junta de Andaluc´ıa (projects FQM-336 and FQM2467). She thanks the Department of Mathematics and Statistics at the University of Saskatchewan (Canada) for its hospitality during her visit from March to June 2011. References [1] H. Braun, M. Koecher: Jordan-Algebren. Springer-Verlag, 1966. [2] M. R. Bremner: On the definition of quasi-Jordan algebra. Communications in Algebra 38 (2010) 4695 -- 4704. [3] M. R. Bremner, L. A. Peresi: Classification of trilinear operations. Communications in Algebra 35 (2007) 2932 -- 2959. [4] M. R. Bremner, L. A. Peresi: Special identities for quasi-Jordan algebras. Communications in Algebra (to appear). arXiv:1008.2723v1 [math.RA] [5] M. R. Bremner, J. S´anchez-Ortega: The partially alternating ternary sum in an associa- tive dialgebra. Journal of Physics A: Mathematical and Theoretical 43 (2010) 455215. [6] C. Cuvier: Alg`ebres de Leibnitz: d´efinitions, propri´et´es. Annales Scientifiques de l' ´Ecole Normale Sup´erieure (4) 27 (1994) 1 -- 45. [7] J. Faraut, S. Kaneyuki, A. Kor´anyi, Q. Lu, G. Roos: Analysis and geometry on complex homogeneous domains. Birkhauser Boston, 2000. [8] R. Felipe: Restrictive, split and unital quasi-Jordan algebras. Journal of Algebra 335 (2011) 1 -- 17. [9] V. Y. Gubarev, P. S. Kolesnikov: The Tits-Kantor-Koecher construction for Jordan dial- gebras. Communications in Algebra 39 (2011) 497 -- 520. [10] N. Jacobson: Lie and Jordan triple systems. American Journal of Mathematics 71 (1949) 149 -- 170. [11] N. Jacobson: Structure and Representations of Jordan Algebras. American Mathematical Society, Providence, 1968. [12] P. Jordan, J. von Neumann, E. Wigner: On an algebraic generalization of the quantum mechanical formalism. Annals of Mathematics (2) 35 (1934) 29 -- 64. [13] P. S. Kolesnikov: Varieties of dialgebras and conformal algebras. Sibirskiı Matematicheskiı Zhurnal 49 (2008) 322 -- 339; translation in Siberian Mathematical Journal 49 (2008) 257 -- 272. les alg`ebres de Leibniz. [14] J. L. Loday: Une version non commutative des alg`ebres de Lie: L'Enseignement Math´ematique 39 (1993) 269 -- 293. [15] J. L. Loday: Alg`ebres ayant deux op´erations associatives: les dig`ebres. Comptes Rendus de l'Acad´emie des Sciences. S´erie I. Math´ematique 321 (1995) 141 -- 146. [16] J. L. Loday: Dialgebras. Dialgebras and Related Operads, pages 7 -- 66. Lectures Notes in Mathematics, 1763. Springer-Verlag, 2001. [17] O. Loos: Jordan Pairs. Lecture Notes in Mathematics, 460. Springer-Verlag, 1975. [18] K. McCrimmon: A Taste of Jordan Algebras. Springer-Verlag, 2004. [19] E. Neher: Jordan Triple Systems by the Grid Approach. Lecture Notes in Mathematics, 1280. Springer-Verlag, 1987. [20] J. M. Osborn: Lie triple algebras with one generator. Mathematische Zeitschrift 110 (1969) 52 -- 74. [21] H. Petersson: Zur Theorie der Lie-Tripel-Algebren. Mathematische Zeitschrift 97 (1967) 1 -- 15. [22] A. P. Pozhidaev: Dialgebras and related triple systems. Sibirskiı Matematicheskiı Zhurnal 49 (2008) 870 -- 885; translation in Siberian Mathematical Journal 49 (2008) 696708. [23] A. P. Pozhidaev: 0-dialgebras with bar-unity, Rota-Baxter and 3-Leibniz algebras. Contem- porary Mathematics 499, 245 -- 256. American Mathematical Society, 2009. JORDAN TRIPLE DISYSTEMS 23 [24] R. Vel´asquez, R. Felipe: Quasi-Jordan algebras. Communications in Algebra 36 (2008) 1580 -- 1602. [25] R. Vel´asquez, R. Felipe: Split dialgebras, split quasi-Jordan algebras and regular elements. Journal of Algebra and its Applications 8 (2009) 191 -- 218. [26] R. Vel´asquez, R. Felipe: On K-B quasi-Jordan algebras and their relation with Leibniz algebras. www.cimat.mx/reportes/enlinea/I-10-10.pdf [27] V. Voronin: Special and exceptional Jordan dialgebras. arXiv:1011.3683v1 [math.RA] [28] K. A. Zhevlakov, A. M. Slinko, I. P. Shestakov, A. I. Shirshov: Rings that are Nearly Associative. Academic Press, 1982. Department of Mathematics and Statistics, University of Saskatchewan, Canada E-mail address: [email protected] CIMAT, Centro de Investigaci´on en Matem´aticas, Guanajuato, M´exico E-mail address: [email protected] Departmento de ´Algebra, Geometr´ıa y Topolog´ıa, Universidad de M´alaga, Espana E-mail address: [email protected]
1805.04652	3	1805	2019-11-22T05:28:20	Geometrically Partial actions	[ "math.RA" ]	We introduce "geometric" partial comodules over coalgebras in monoidal categories, as an alternative notion to the notion of partial action and coaction of a Hopf algebra introduced by Caenepeel and Janssen. The name is motivated by the fact that our new notion suits better if one wants to describe phenomena of partial actions in algebraic geometry. Under mild conditions, the category of geometric partial comodules is shown to be complete and cocomplete and the category of partial comodules over a Hopf algebra is lax monoidal. We develop a Hopf-Galois theory for geometric partial coactions to illustrate that our new notion might be a useful additional tool in Hopf algebra theory.	math.RA	math	GEOMETRICALLY PARTIAL ACTIONS JIAWEI HU AND JOOST VERCRUYSSE Abstract. We introduce "geometric" partial comodules over coalgebras in monoidal categories, as an alternative notion to the notion of partial action and coaction of a Hopf algebra introduced by Caenepeel and Janssen. The name is motivated by the fact that our new notion suits better if one wants to describe phenomena of partial actions in algebraic geometry. Under mild conditions, the category of geometric par- tial comodules is shown to be complete and cocomplete and the category of partial comodules over a Hopf algebra is lax monoidal. We develop a Hopf-Galois theory for geometric partial coactions to illustrate that our new notion might be a useful additional tool in Hopf algebra theory. Contents Introduction 1 4 1. A categorical reformulation of partial group actions 4 1.1. The classical definition of a partial group action 5 1.2. Lax and quasi partial actions 5 1.3. Partial actions and spans 12 2. Partial comodules over a coalgebra 12 2.1. Geometrically partial comodules 18 2.2. Partial comodule morphisms 20 2.3. Coassociativity 24 2.4. Completeness and cocompleteness of the category of partial comodules 31 3. Partial comodules over a bialgebra 3.1. Lax monoidal categories 31 3.2. The lax monoidal category of geometric partial comodules over a bialgebra 34 4. Partial comodule algebras and Hopf-Galois theory 40 40 4.1. Algebras in the category of partial comodules 42 4.2. Partial comodules in the category of algebras 44 4.3. Partial Hopf modules 4.4. Hopf-Galois theory 46 55 Acknowledgements References 56 Introduction The coordinate algebras of algebraic groups provide classical examples of Hopf algebras and the interaction between Hopf algebra theory and algebraic geometry that arises from this construction has showed to be very fruitful for both worlds. With the rise of quantum groups in 80s of the the 20th century, deformations of Hopf algebras associated to algebraic groups have inspired the field of non-commutative (algebraic) geometry, 1 2 JIAWEI HU AND JOOST VERCRUYSSE where non-commutative algebras play the role of non-commutative spaces and (non- commutative, non-cocommutative) Hopf algebras coacting on these algebras play the role of symmetry groups of these spaces. The aim of the present paper is to introduce a new type of symmetries in (non- commutative) algebraic geometry, that correspond to partial group actions. It is well-known that (usual) actions of a (discrete) group G on a k-algebra A are in correspondence with semi-direct product structures, or smash product structures, on A⊗kG. In order to describe certain algebras (such as Toeplitz algebras) as a generalized smash product, the notion of a partial group action was introduced in the setting of C ∗- algebras by Exel [11]. Roughly, a partial action of a group G on an object X associates to each element of X an isomorphism between two appropriate subobjects of X. In case these subobjects always coincide with the whole object X, the action is a usual (or as we will call them from now on: global) group action. Immediate examples of these partial actions can be obtained by restricting a (global) action to an arbitrary subobject of X. Since its introduction, this notion of partial group action and the related notion of partial representation, has been investigated from a purely algebraic point of view and many interesting results have been obtained, see e.g. [9], [10]. A first attempt to bring partial actions from the setting of groups to the setting of Hopf algebras was made by Caenepeel and Janssen in [8]. This approach has shown to be very successful in the sense that many classical Hopf-algebraic results appear to have a partial counterpart. However, in this initial approach, several aspects of the theory remained unclear. For example, the definition of Caenepeel and Janssen only allowed to describe partial (co)actions of Hopf algebras on other (co)algebras. It was not possible to define partial actions on vector spaces nor to define partial actions of algebras other than Hopf alge- bras. A next step was made in [1], where it was shown that, in analogy with classical actions of Hopf algebras, partial actions can be viewed as internal algebras in an ap- propriate monoidal category. However, in contrast to the classical case, the monoidal category in play is no longer the usual monoidal category of representations (or mod- ules) of the Hopf algebra H, but rather the category of partial representations which coincides with the category of representations over a newly constructed Hopf algebroid Hpar. Lately, it was shown in [2] how partial representations can be globalized and the partial representations of Sweedler's 4-dimensional Hopf algebra were completely classified. A recent development in the theory of partial actions, is the approach of [15], where the initial theory of partial actions over C ∗-algebras is merged with the Hopf-algebra setting, in the study of partial actions of C ∗-quantum groups. Furthermore, it turns out that if one studies partial actions of Hopf algebras that arise from algebraic groups, the partial actions are not what one would expect. Indeed, it was observed in [5] that a partial coaction of a Hopf algebra O(G), which is the coordinate algebra of an algebraic group G, on an algebra O(X), which is the coordinate algebra of an algebraic space X, is always global unless X is a disjoint union of non-empty sub- spaces. The spirit of partial actions would however also ask for more involved examples, where the elements of the algebraic group G act as an isomorphism between arbitrary algebraic subspaces of X. Indeed, as we mentioned before, examples of partial actions can be constructed by restricting global actions. If the algebraic group G acts (globally) on an algebraic variety X, we expect that the same group acts partially on arbitrary (not necessarily irreducible) subvarieties of X. For a more concrete example, one could consider two circles in the real plane intersecting in two points. From the global point GEOMETRICALLY PARTIAL ACTIONS 3 of view, such a configuration has only few symmetries (or more precisely there are very few isometries of the plane that restrict to this union of circles). Nevertheless, each of the individual circles has a lot of symmetries. Partial actions allow to describe at once the (few) global symmetries of the pair of circles, and the (many) symmetries of the individual circles, as well as combinations of these. To describe this kind of phenomena from a Hopf-algebraic point of view, we propose an alternative definition of partial (co)actions of Hopf algebras, that we call geometric partial (co)actions and that also allows us to bring partial action into the realm of non-commutative geometry as the algebraic structure to describe partial symmetries. To arrive at this goal, we will give in Section 1 a detailed study of partial actions of groups on sets, and provide a new approach to these. This approach is motivated by category theory, where partial morphisms have an interpretation as spans where one of the legs is a monomorphism. Given any category C, one can build this way a bicategory of partial morphisms, which is a subbicategory of the category of spans over C. A partial action of a group G on an object X is then noting else than a lax functor from G into the endo-hom category of partial morphisms from X to X. Based on this viewpoint, we generalize the notion of partial action of a group to partial (co)actions of (co)algebras in arbitrary categories with pullbacks (respectively pushouts). More precisely, given a coalgebra H in the monoidal category C, a partial comodule datum for H is a quadruple (X, X • H, π, ρ), where π : X ⊗ H → X • H is an epimorphism and ρ : X → X •H is a morhpism in C. By considering 3 levels of strictness for the coassociativity condition on a given partial comodule datum, we consider then 3 versions of partial comodules: quasi, lax and geometric partial comodules. The name for the latter version is motivated by the fact that the above mentioned examples of partial actions of algebraic groups arise exactly as those 'geometric partial comodules'. The initial partial actions of groups coincide with geometric partial actions of groups, viewed as coalgebras in the opposite of the category of sets. In case of arbitrary (Hopf) algebras, this new notion covers the one of Caenepeel and Janssen, but allows to go far beyond the notions of partial actions and partial representations as discussed above. Fi- nally, our definition allows to consider partial (co)modules over arbitrary (co)algebras, where before it was only possible to consider such structures over Hopf algebras (with bijective antipode). Although partial comodules are only a laxified version of classical comodules, they share surprisingly many properties with classical comodules. In particular, we show that a version of the fundamental theorem for comodules is still valid for geometric partial comodules and the category of partial comodules is complete and cocomplete. All this is shown in Section 2. One of the key features of Hopf algebras, is that their categories of (co)modules have a natural monoidal structure, inherited by the monoidal structure of the base category wherein the considered Hopf algebra is defined. At this point, the theory of (geometric) partial modules becomes different from the global theory. Indeed, although the category of quasi partial comodules over a bialgebra is shown to posses an associative monoidal structure (with an oplax unit), the more interesting category of geometric partial co- modules has only an oplax monoidal structure [14] (see Section 3). By definition an oplax monoidal structure requires the existence of n-fold tensor products, along with suitable coherence conditions. Where the tensor product of global comodules over a Hopf k-algebra is given by the tensor product of the underlying vector spaces, the vec- tor space tensor product of two geometric partial comodules is in general no longer a 4 JIAWEI HU AND JOOST VERCRUYSSE geometric partial comodule. Therefore, their tensor product is defined as the biggest geometric quotient of the underlying vector space product. Using this oplax monoidal structure, one can give meaning to an algebra in the category of geometric partial comodules. We discuss these 'geometric partial comodule algebras' and initiate a Hopf-Galois theory for them in Section 4. Some remarks on notation: given an object X in a category C, we denote the identity morphism on X in C by idX or shortly by X. If C is a monoidal category with tensor product ⊗ and monoidal unit k, then we denote the right unit constraint as rX : X → X ⊗ k (classically, the reversed direction is used, but it turns out that this direction is more convenient for our work). 1. A categorical reformulation of partial group actions 1.1. The classical definition of a partial group action. Let G be a group and X a set. A partial action datum of G on X is a couple (Xg, αg)g∈G, where • {Xg}g∈G, a family of subsets of X indexed by the group G; • {αg : Xg−1 → X}g∈G a family of maps indexed by the group G; Recall from [11] that a partial action α of G on X is a partial action datum (Xg, αg)g∈G that satisfies the following axioms (PA1) Xe = X and αe = idX, where e denotes the unit of G; (PA2) αg(Xg−1 ∩ Xh) ⊂ Xg ∩ Xgh; (PA3) αh ◦ αg = αhg on Xg−1 ∩ X(hg)−1. Remark that thanks to the second axiom (PA2), the third axiom (PA3) makes sense, since αh ◦ αg(Xg−1 ∩ X(hg)−1) ⊂ αh(Xg ∩ Xh−1) ⊂ Xhg ∩ Xh and Furthermore, combining (PA2) and (PA3), we find that αhg(Xg−1 ∩ X(hg)−1) ⊂ Xh ∩ Xhg. Xg ∩ Xgh = αg ◦ αg−1(Xg ∩ Xgh) ⊂ αg(Xg−1 ∩ Xh) and therefore, we can deduce the stronger axiom (PA2') αg(Xg−1 ∩ Xh) = Xg ∩ Xgh. If we take in particular h = e, then we find that αg(Xg−1) = Xg. Moreover, since αg ◦ αg−1(x) = x for all x ∈ Xg, we find that each map αg induces a bijection αg : Xg−1 → Xg. This last fact is often supposed as part of the definition of a partial action. Many examples of partial actions have been observed in recent literature. It makes no sense to repeat them here, however, we will gave a few exemplary ones, which will make the transition to some of the new results in this paper more easy. Examples 1.1. (1) Consider a (global) action of the group G on a set Y , and let X ⊂ Y be any (non-empty) subset of Y . Then G acts partially on X, by defining Xg := {x ∈ X g−1 · x ∈ X} and defining αg : Xg−1 → Xg, αg(x) = g · x. (2) As a particular case of the previous one, we consider the following geometric example. Let G = (A2, +) be the group of 2-dimensional affine translations. This group acts strictly transitive on the affine plane A2. Consequently, this group acts partially on any subset of A2. In one of the next sections, we will discuss in more detail the case of the partial action of this group on two intersecting lines. GEOMETRICALLY PARTIAL ACTIONS 5 (3) Consider the additive group Z. For any z ∈ Z with z ≥ 0 we define its domain X−z = N and its action αz : Z → Z, x 7→ x + z. On the other hand for each z < 0 we define its domain X−z = {x ∈ Z x ≥ −z} and its action αz : Z → Z, x 7→ x + z. Then one easily verifies this defines a partial action which is obtained by restricting the action of Z on itself to N. 1.2. Lax and quasi partial actions. As we explained, the axiom (PA2) in the defi- nition of partial actions is designed to make sense of axiom (PA3) which expresses the associativity. However, this axiom can be weakened further. Definition 1.2. Let G be a group, X a set and α = (Xg, αg) be a partial action datum. We say that α is a lax partial action of the following axioms hold (LPA1) Xe = X and αe = idX, where e denotes the unit of G; (LPA2) Xg−1 ∩ α−1 g (Xh−1) ⊂ X(hg)−1. (LPA3) αh ◦ αg = αhg on Xg−1 ∩ α−1 g (Xh−1). Axiom (LPA2) tells that if x ∈ Xg−1 and α(g)(x) ∈ Xh−1, then x ∈ X(hg)−1 and therefore axiom (LPA3) makes sense. As one can easily verify, any partial action is a lax partial action and the converse holds if and only if αg(Xg−1) ⊂ Xg (see Proposition 1.10). The following example shows that lax partial actions are a proper generalization of partial actions. Example 1.3. This example is a variation of Example 1.1 (3). Consider the additive group Z. For any z ∈ Z with z ≥ 0 we define its domain X−z = Z and its action αz : Z → Z, x 7→ x + z. On the other hand for each z < 0 we define its domain X−z = {x ∈ Z x ≥ −z} and its action αz : X−z → Z, x 7→ x + z. Then one can verify that this is indeed a lax partial action and moreover it is not a partial action, since ρz : X−z → Xz = Z is not a bijection for any z < 0. For sake of completeness, we also state another weakening of the definition of partial action, which is, by our opinion, naturally the most general version of a partial action. Definition 1.4. Let G be a group, X a set and α = (Xg, αg) be a partial action datum. We say that α is a quasi partial action of the following axioms hold (QPA1) Xe = X and αe = idX, where e denotes the unit of G; (QPA2) αh ◦ αg = αhg on Xg−1 ∩ α−1 g (Xh−1) ∩ X(gh)−1. Remark that in this definition, we ask associativity to hold exactly there where both αh ◦ αg and αhg make sense. The following construction shows that quasi partial actions properly generalize lax and usual partial actions. Example 1.5. Let G be a group acting (globaly) on a set X. For any g ∈ G consider an arbitrary subset Xg ⊂ X and let αg : Xg−1 → X be the restriction of the action of g on X. Then this defines a quasi partial action of G on X. 1.3. Partial actions and spans. We will now reformulate the definition of a partial action, making no explicit reference to the elements of the set or the group, but stating everything internally in the category Set of sets. This way, the definition can be easily lifted to any (monoidal) category (with pullbacks). As we will show, quasi and partial actions arise naturally in this context. 6 JIAWEI HU AND JOOST VERCRUYSSE Recall that in any category C, a span from X to Y is a triple (A, f, g), where A is an object of C and f : A → X and g : A → Y are two morphisms of C. If C has pullbacks and (A, f, g), (B, h, k) are spans from X to Y and from Y to Z respectively, then one constructs a new span, called the composition span, from X to Z by the following pullback construction: p ⑦⑦⑦⑦⑦⑦⑦⑦ ❅❅❅❅❅❅❅❅ g A f ~⑦⑦⑦⑦⑦⑦⑦⑦ X P ❄❄⑧⑧ q ❅❅❅❅❅❅❅❅ ⑦⑦⑦⑦⑦⑦⑦⑦ h B k ❅❅❅❅❅❅❅❅ Z which we will denote as (B, h, k)•(A, f, g). Given two spans (A, f, g), (B, h, k) : X → Y , a morphism of spans α : (A, f, g) → (B, h, k) is a map α : A → B such that the following diagram commutes Y A B X f ~⑦⑦⑦⑦⑦⑦⑦⑦ `❆❆❆❆❆❆❆❆ h α g ❅❅❅❅❅❅❅❅ ?⑦⑦⑦⑦⑦⑦⑦⑦ k Y In this way, we obtain a bicategory Span(C), whose 0-cells are the objects of C, 1-cells are spans and 2-cells are morphisms of spans. We can also consider the (usual) category span(C), whose objects are the objects of C and whose morphisms are isomorphism classes of spans. In what follows, we will use the following variation on the usual category of spans. Definition 1.6. By a partial morphism from X to Y in a category C, we mean a morphism (A, f, g) in the category Span(C), with the additional property that f : A → X is a monomorphism. By Par(C) we denote the subbicategory of Span(C), with the 0-cells as Span(C) (and C), whose 1-cells are given by partial morphisms in C. By par(C) we denote the corresponding subcategory of span(C). Remark that the above definition of Par(C) makes sense since the pullback of a monomor- phism is a monomorphism. Moreover, if α, β : (A, f, g) → (B, h, k) are 2-cells in Par(C) then α = β since h ◦ α = f = h ◦ β and h is a monomorphism. Hence Par(C) is locally a poset. In the particular case of Par(Set), a there is a morphism of spans α : (A, f, g) → (B, h, k) if and only if A is a subset of B and g is the restriction of k to A. / Y . When We will denote a partial morphism from X to Y by a dotted arrow X we consider a partial map as a triple (A, f, g), we will often omit to write explicitly the first map f , as it is an inclusion and supposed to be known if we know the object A, ie. we will write (A, f, g) = (A, g) = g. Lemma 1.7. Let G be a group and X a set. Then there is a bijective correspondence between (i) partial action data of G on X; ~ ~ ` ? / GEOMETRICALLY PARTIAL ACTIONS 7 (ii) partial morphisms G × X → X; (iii) maps G → Par(X, X). Proof. (i) ⇔ (ii). Let (Xg, αg)g∈G be a partial action datum of G on a set X. Then we can construct the set (1) G • X = {(g, x) x ∈ Xg−1} ⊂ G × X, which is the set of all "compatible pairs" in G × X. Clearly, the partial action then induces a well-defined map α : G • X → X, α(g, x) = αg(x). Hence we obtain a partial morphism G × X / X , G • X G × X / X ιX yssssssssss α #●●●●●●●●● Conversely, consider any partial morphism α = (G • X, ι, α) : G × X → X, where G • X is a subset of G × X, ι : G • X → G × X is the canonical inclusion and α : G • X → X is a map. Then for any g ∈ G, we can define Xg−1 = {x ∈ X (g, x) ∈ G • X} and we recover formula (1). (i) ⇔ (iii). Let (Xg, αg)g∈G be a partial action datum, then for any g ∈ G we have Xg−1 ιg ③③③③③③③③ X αg "❉❉❉❉❉❉❉❉ / X where ιg : Xg → X is the canonical inclusion, is a partial endomorphism of X which defines a map G → Par(X, X). Conversely, any map G → Par(X, X) gives in the same way a family (Xg, αg)g∈G, i.e. a partial action datum. (cid:3) The natural question that now arises is what are the conditions on a partial morphism α : G × X → X for the associated partial action datum to become an actual partial action. A first naive guess would be to impose the usual associativity and unitality conditions of an action expressed in the category par(C), or equivalently to impose that the map G → par(X, X) is a morphism of monoids (where the the later is the endomorphism monoid of X in the (1-)category par(Set)). However, as we will point out now, this leads to a global action. Lemma 1.8. Let G be a group with multiplication m : G × G → G, m(g, h) = gh and the unit e : {∗} → G, e(∗) = e. Consider a partial action datum (Xg, αg)g∈G. (1) the following assertions are equivalent (i) The partial action datum satisfies axiom (PA1); (ii) The associated partial morphism α : G × X → X satisfies α ◦ (e × X) ≃ X in Par(Set). (iii) The associated map α′ : G → Par(X, X) preserves the unit. (2) The following assertions are equivalent (i) The partial action datum defines a global action of G on X; / K k y # / N n " / 8 JIAWEI HU AND JOOST VERCRUYSSE (ii) The associated partial morphism α : G × X → X satisfies the following identities in par(Set) (i.e. isomorphism of spans) α • (e × X) ≃ X (G × α) • α ≃ (m × G) • α (iii) The associated map α′ : G → Par(X, X) is a morphism of monoids. Proof. (1). Let us compute the composition of spans α • (e × X). This leads to the following pullback X ∼= {∗} × X ∼= v❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧ X ιX v♠♠♠♠♠♠♠♠♠♠♠♠♠ (❘❘❘❘❘❘❘❘❘❘❘❘❘ e×X {∗} • X ❄❄⑧⑧ / G × X e•X (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ v❧❧❧❧❧❧❧❧❧❧❧❧❧❧ ιX G • X α (❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘ / X where {∗} • X = {x ∈ X x ∈ Xe} ∼= Xe. Hence α ◦ (e × X) is the identity morphism on X in the category Par(Set), if and only if Xe = X and αe = idX , which is exactly axiom (PA1). Furthermore, it is clear that this is equivalent to saying that α′(e) = (Xe, ιe, αe) is the span (X, idX, idX). (2). By part (1), we only have to prove the equivalence of the associativity constraints. Let us compute the composition of spans (G × α) • α in Par(Set), which is given by the following pullback. G × (G • X) G • X G • (G • X) ❄❄⑧⑧ v♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗ G×α / G × X G•α (◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ ι G×ι v♠♠♠♠♠♠♠♠♠♠♠♠ G × G × X α (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ / X Explicitly we find G • (G • X) = {(h, g, x) ∈ G × G × X x ∈ Xg−1, gx ∈ Xh−1}. Similarly, we can compute the composition (m × G) • α in Par(Set), which is again given by a pullback (G × G) • X ❄❄⑧⑧ G × G × X ♠♠♠♠♠♠♠♠♠♠♠♠♠ ♠♠♠♠♠♠♠♠♠♠♠♠♠ ι′ v♠♠♠♠♠♠♠♠♠♠♠♠ (❘❘❘❘❘❘❘❘❘❘❘❘❘ m×X m•X (◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠♠ ι G • X G × G × X / G × X α (❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘ / X where now (G × G) • X = {(h, g, x) ∈ G × G × X x ∈ X(hg)−1}. I i v ( v ( H h v ( / / ( I i v ( I i v ( I i v / / I i v ( ( I i v ( / / GEOMETRICALLY PARTIAL ACTIONS 9 We then find that (g, g−1, x) ∈ G • (G • X) if and only if x ∈ Xg (and g−1x ∈ Xg−1). On the other hand, (g, g−1, x) ∈ (G × G) • X if and only if x ∈ Xe = X. Hence, we obtain that the action is global if and only if G • (G • X) and (G × G) • X are isomorphic spans. In the same way, if the action is global, then clearly α′ is a morphism of monoids. Conversely, if α′ is a morphism of monoids then we obtain in particular that α′(g−1) • α′(g) = α′(e) = (X, idX , idX). Since the underlying set of the span of α′(g−1) • α′(g) is given by {x ∈ Xg−1 gx∈Xg, we find that α′(g−1) ◦ α′(g) = α′(e) implies that Xe = Xg for all g ∈ G and hence we have a global action. (cid:3) As we have just observed, partial actions are not just actions in the category of partial morphisms. The monoid morphism G → par(X, X) can also be viewed as a func- tor between 2 one-object categories. However, since the Par(Set) is a bicategory, the Par(X, X) becomes a (monoidal) category, or a one-object bicategory. Consequently, there is a natural laxified version of a partial action considering only a lax functor be- tween G and Par(X, X). In the next proposition, we show that this coincides exactly with the lax partial actions we introduced above. Recall that a lax functor F : B → B′ between 2 bicategories consists of • a map from the 0-cells of B to the 0-cells of B′, • for any pair of 0-cells X, Y of B, a functor FX,Y : B(X, Y ) → B′(X ′, Y ′) • for any 0-cell X in B a 2-cell uX : idF X → F (idX)' • for any two 1-cells a ∈ B(X, Y ) and b ∈ B(Y, Z) a 2-cell αa,b : F (a) • F (b) → F (a • b) (where • denotes the horizontal composition), which in natural in a and b; satisfying the usual coherence axioms. If the category B′ is locally a poset, then these coherence conditions follow automatically from the above information. Proposition 1.9. Let G be a group with multiplication m : G × G → G, m(g, h) = gh and the unit e : {∗} → G, e(∗) = e. Consider a partial action datum (Xg, αg)g∈G. The following assertions are equivalent (i) The partial action datum defines a lax partial action of G on X; (ii) For the associated partial morphism α : G × X → X, there exist morphisms of spans u : X → α • (e × X) and θ : (G × α) • α → (m × G) • α; (iii) The associated map α′ : G → Par(X, X) is a lax functor where G is considered as a locally discrete 2-category with one 0-cell. Proof. (i) ⇔ (ii). As we have shown in Lemma 1.8, α • (e × X) is given by the span (Xe, ιe, αe) : X → X. There existence of a morphism of spans u : X → α • (e × X), means that X ⊂ Xe ⊂ X, hence X = Xe, and αe = idX . Furthermore, we also know from Lemma 1.8 the explicit form of (G×α)•α and (m×G)• α. The existence of the morphism θ then means that G•(G•X) ⊂ (G×G)•X, which is exactly axiom (LPA2) and the restriction of the partial action αgh to Xg−1 ∩ α−1 g (Xh−1) coincides with αh ◦ αg, which is exactly axiom (LPA3). (i) ⇒ (iii). Recall that the map α′ : G → Par(X, X) is given by α′(g) = (Xg−1, ιg, αg). Both G and Par(X, X) are considered as one-object bicategories and moreover G has only trivial 2-cells, Par(X, X) is a poset. Hence, α′ : G → Par(X, X) induces a lax functor if and only if there exists morphism of spans u′ : (X, idX , idX) → (Xe, ιe, αe) and θ′ : α′(h) • α′(g) → α′(hg). As in the first part of the proof, the existence u′ is 10 JIAWEI HU AND JOOST VERCRUYSSE equivalent axiom (LPA1). Furthermore, remark that α′(h) • α′(g) is given by the span Xg−1 ∩ α−1 g (Xh−1) ιg t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ X Xg−1 u❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ αg X αg )❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ιh Xh−1 αh ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ X Hence the existence of θ′ means that axioms (PLA2) and (PLA3) hold. (cid:3) As we have pointed out before, partial actions are a special instance of lax partial actions. In the next result we provide equivalent conditions for a lax partial action to be partial. Let us first make the following observation. Given a partial action datum, we can consider the pullback G × (G • X) (G × G) • X (G • G) • X v♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗ (◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠ G × G × X which is nothing else than the intersection G × (G • X) ∩ (G × G) • X. If the partial action datum defines a lax partial action, then existence of the morphism of spans θ : (G × α) • α → (m × G) • α implies that the following diagram commutes v♠♠♠♠♠♠♠♠♠♠♠♠ G × (G • X) v♠♠♠♠♠♠♠♠♠♠♠♠ G × G × X h◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ θ G • (G • X) G•α (◗◗◗◗◗◗◗◗◗◗◗◗◗ G • X G • X 6♠♠♠♠♠♠♠♠♠♠♠♠♠ m•X (G × G) • X α (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ 6❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧ α X and therefore the image of θ lies in (G • G) • X, i.e. we can corestrict θ to a morphism ¯θ : G • (G • X) → (G • G) • X. Proposition 1.10. Let α be a lax partial action of the group G on the set X. Then the following statements are equivalent (i) α is a partial action; (ii) ¯θ : G • (G • X) → (G • G) • X is an isomorphism; (iii) for each g ∈ G, we have that αg : Xg−1 → Xg. H h u ) G g t * G g t * I i v u ( u ( I i v I i v ( I i v ( 6 5 U h 6 GEOMETRICALLY PARTIAL ACTIONS 11 Proof. (ii) ⇔ (i) ⇒ (iii). By definition, partial actions and lax partial actions only differ in their second axiom. From the above discussion, we know that (G • G) • X = G × (G • X) ∩ (G × G) • X = {(h, g, x) ∈ G × G × X x ∈ Xg−1 ∩ X(gh)−1} Therefore, ¯θ is an isomorphism we obtain that if (h−1g−1, g, x) ∈ (G • G) • X, i.e. x ∈ Xg−1 ∩ Xh then also ((gh)−1, g, x) ∈ G • (G • X), i.e. x ∈ Xg−1 and gx ∈ Xhg Hence we find that αg(Xg−1 ∩ Xh) ⊂ Xgh. In particular, taking h = e, then we find that αg(Xg−1) ⊂ Xg. Combining both, we recover exactly axiom (PA2). (iii) ⇒ (i). For any g ∈ G and x ∈ Xg we find that g−1x ∈ Xg−1, we can apply g on g−1x and find that x = g · g−1x. So x ∈ Xg if and only if x = gy for some y ∈ Xg−1. Now take any x ∈ Xg−1 ∩ Xh. Then by the above, we can write x = hy for some y ∈ Xh−1. Since we have that y ∈ Xh−1 and x = hy ∈ Xg−1, it follows by axiom (LPA2) that y ∈ X(gh)−1 and gh · y = g · (hy) = gx. In particular, we find that gx = gh · y ∈ Xgh. Hence we obtain exactly axiom (PA2). (cid:3) Finally, we also restate the definition of quasi partial action in terms of spans, the proof of which is clear. Proposition 1.11. A partial action datum (Xg, αg) defines a quasi partial action of G on X if and only if the equivalent statements of Lemma 1.8 (1) hold and the associativity constraint αh ◦ αg = αhg holds on all elements of the following pullback G • (G • X) (G • G) • X θ1 v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗ Θ θ2 (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠ G × G × X Consequently, a quasi partial action is lax if and only if the span (Θ, θ1, θ2) : G • (G • X) → (G • G) • X is induced by a morphism, and the quasi partial action is a partial action if and only if Θ is an isomorphism. Proof. Let us just remark that the associativity on Θ means that the following diagram commutes Θ θ1 v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ G • (G • X) (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ θ2 (G • G) • X G•α G • X α X m•X G • X α X (cid:3) v ( u ( I i v v ( 12 JIAWEI HU AND JOOST VERCRUYSSE 2. Partial comodules over a coalgebra Let C be a braided monoidal category with pullbacks that are preserved by all endofunc- tors on C of the form −⊗X and X ⊗−. Then the observations from the previous section allow us to define partial actions of a Hopf algebra in C on any object in C such that taking C = Set we recover the classical definition of partial actions of groups on sets. Since we will rather be interested in examples inspired by algebraic geometry, hence in coactions rather than actions, we will take a dual point of view and consider from now on a braided monoidal category C with pushouts that are preserved by all endofunctors of the form − ⊗ X and X ⊗ −, and Hopf algebras mentioned below are Hopf algebras in C. Remark that in such a category, the tensor product of two epimorphisms is an epimorphism. Since pushouts are colimits, any braided closed monoidal category will serve as an example, in particular any category of modules over a commutative ring k. In what follows the latter will be our standard example, were we in fact mostly will restrict to the case where k is a field. Inspired by this example we will denote the unit of the monoidal category C by k. 2.1. Geometrically partial comodules. In [1], the notion of a "partial module" over a Hopf algebra H was introduced, by means of partial representations of Hopf algebras and similarly, "partial comodules" can be introduced by means to partial corepresen- tations, see [3]. In this section, we introduce alternative notions of partial (co)module over any (co)algebra. To prevent a clash of terminologies in case C = H, we will call our notions (in rising order of generality) a quasi, lax and geometric partial (co)modules. We show that in the case of Hopf algebras, the partial modules of [1], and in particular, the partial actions of [8], appear as special cases of our quasi partial comodules. Examples arising from (usual) partial actions of (algebraic) groups on (algebraic) sets give rise to geometric partial comodules. Definition 2.1. Let (H, ∆, ǫ) be a coalgebra in a monoidal category C. A partial comodule datum is a quadruple X = (X, X • H, πX , ρX), where X and X • H are objects in C, πX : X ⊗ H ։ X • H is an epimorphism and ρX : X → X • H is a morphism in C. Remark that a partial comodule datum can be viewed as the following cospan in C X ρX &▲▲▲▲▲▲▲▲▲▲▲ πX xrrrrrrrrrr X • H X ⊗ H Suppose now that the category C has pushouts. Then to any partial comodule datum induces canonically four pushouts, that we denote by X • k, (X • H) • H, X • (H ⊗ H) and (X • H) • H, and that are defined respectively by the following diagrams: X ⊗ H X ⊗ H X • H X ⊗ k X • H (X • H) ⊗ H πX yrrrrrrrrrr %▲▲▲▲▲▲▲▲▲▲ X•ǫ ⑧⑧ ❄❄ X • k X⊗ǫ %▲▲▲▲▲▲▲▲▲▲ yrrrrrrrrrr πX,ǫ πX v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ ρX •H ρX ⊗H (◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠ πX•H ⑧⑧ ❄❄ (X • H) • H / / & x x x y y y % % y y y v v v ( ( v v v GEOMETRICALLY PARTIAL ACTIONS 13 X • H X ⊗ H ⊗ H X ⊗ H πX v❧❧❧❧❧❧❧❧❧❧❧❧❧ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ X•∆ ⑧⑧❄❄ X • (H ⊗ H) X⊗∆ (❘❘❘❘❘❘❘❘❘❘❘❘❘ v♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗ πX,∆ π′ X πX ⊗H (◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠ X,∆ π′ ⑧⑧❄❄ X • (H • H) (X • H) ⊗ H Finally, we consider a last pushout that we denote as Θ and that is given by the following diagram (X • H) ⊗ H (X • H) • H πX•H v♠♠♠♠♠♠♠♠♠♠♠♠ (❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘ θ1 ⑧⑧❄❄ Θ π′ X,∆ (◗◗◗◗◗◗◗◗◗◗◗◗ v❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧ θ2 X • (H • H) We will call Θ the coassociativity pushout. We are now ready to state the exact definitions of a partial comodule. Definition 2.2. Let (H, ∆, ǫ) be a coalgebra in a monoidal category with pushouts C. A quasi partial comodule is a partial comodule datum (X, X • H, πX, ρX ) that satisfies the following conditions [QPC1] (X • ǫ) ◦ ρX = πX,ǫ ◦ rX : X → X • k are identical isomorphisms. I.e. the following diagram commutes X ρX %▲▲▲▲▲▲▲▲▲▲▲ πX yrrrrrrrrrr %▲▲▲▲▲▲▲▲▲▲ X•ǫ X • H idX X⊗ǫ %▲▲▲▲▲▲▲▲▲▲ yrrrrrrrrrr πX,ǫ X ⊗ k idX X ⊗ H ⑧⑧❄❄ X • k / X ∼= yrrrrrrrrrrr rX [QPC2] θ1 ◦ (ρX • H) ◦ ρX = θ2 ◦ π′ X ◦ (X • ∆) ◦ ρX, i.e. the following diagram commutes ∼= X X • H / (X • H) • H X ρX ;①①①①①①①①①① #❋❋❋❋❋❋❋❋❋ ρX ρX •H / X • (H ⊗ H) X•∆ π′ X X • H / X • (H • H) θ1 %▲▲▲▲▲▲▲▲▲▲▲ 9sssssssssss θ2 Θ v v v ( ( v v v ( ( ( ( v v v v v v ( ( ( ( ( ( v v v / / $ $ % y y y % / y z z % y / % ; # / / 9 14 JIAWEI HU AND JOOST VERCRUYSSE A quasi partial comodule will be called a lax partial comodule when the cospan ΘX : X • (H • H) 99K (X • H) • H is induced by a morphism θ (in C). Furthermore a lax partial comodule is called a geometric partial comodule if θ is an isomorphism. Remarks 2.3. (1) As we have remarked before, partial morphisms and hence the structure of partial comodules, have a typical bicategorical behaviour. For ex- ample, both the unitality and coassociativity axiom of a (geometric) partial comodule consist of 2 parts. Firstly, it concerns an (iso)morphism of certain pushouts (to be precise, X • k ∼= X for unitality and (X • H) • H ∼= X • (H • H) for coassociativity of geometric partial comodules), and then an identity between certain maps, involving and uniquely characterizing the previous isomorphisms. An useful consequence of the unitality axiom is therefore that (X•ǫ)◦πX ≃ X⊗ǫ, where the symbol "≃" means that both maps are identical up to the (canonical) isomorphism X • k ∼= X. (2) Remark that by uniqueness of colimits, the pushout ΘX = (Θ, θ1, θ2) is unique up to isomorphism and hence is not part of the structure of a quasi partial comodule. Similarly, if ΘX is induced by a morphism θX , then this morphism is uniquely determined by its property θX ◦ πX•H = π′ X,∆ since πX•H is an epimorphism. Also, whenever there exists a morphism θX with this property, then Θ ∼= (X • H) • H. (3) We will often denote a (quasi) partial comodule by (X, πX, ρX ) or just by X. (4) Of course, one can make state dual definitions of a quasi, lax and geometric partial module over an algebra. We leave the details to the reader, it suffices to apply the above definition to the opposite category Cop. (5) When working in the base category C = Mk We will sometimes use the following Sweedler notation for quasi partial comodules. For any x ⊗ h ∈ X ⊗ H, we write πX (x ⊗ h) = x • h ∈ X • H. Remark that X • H is no longer a tensor product (see below for an interpretation of X • H as a monoidal product when H is a bialgebra). Hence x • h represents a certain class of tensors in X ⊗ H and by the surjectivity of πX , any element of X • H can be represented in such a way, although non-uniquely. We then write ρX (x) = x[0] •x[1], which means that there exists an element x[0] ⊗ x[1] ∈ X ⊗ H such that ρX (x) = πX (x[0] ⊗ x[1]). Again, the element x[0] ⊗ x[1] ∈ X ⊗ H is not unique, so some care is needed in this notation. However, the class x[0] •x[1] ∈ X •H is well-defined since ρX is a proper map. Axiom [QPC1] tells us then that, as for usual coactions, x[0]ǫ(x[1]) = x for all x ∈ X, and in particular this expression makes sense. We will treat axiom [QPC2] in a similar way by the expression x[0][0] • x[0][1] • x[1] = x[0] • x[1](1) • x[1](2) However, this expression now holds in the pushout Θ, and by definition, the left hand side in the above expression is the notation for θ1 ◦ (ρX • H) ◦ ρX(x) and the right hand side is θ2 ◦ π′ X ◦ (X • ∆) ◦ ρX (x), for the same x ∈ X. A first class of examples is obtained from the results of the previous section by taking C = Setop. Indeed, quasi, lax and (usual) partial actions of a group coincide in this way with quasi, lax and geometric partial (co)modules. Remark that in the above formation, these notions also allow to consider partial actions of arbitrary monoids rather than groups. GEOMETRICALLY PARTIAL ACTIONS 15 Before we give some more examples, let us first state the following (well-known) lemma that will be useful for our purposes. Lemma 2.4. Consider vector spaces U, V , W and linear maps f : U → V , g : U → W , where g is surjective. Then the pushout of the pair (f, g) is given by P = V /f (ker g), where g : V → P is the canonical surjection and f : W → P is given by f (w) = f (u) + f (ker g), where u is any element of U such that g(u) = w ∈ W . Let us now provide some examples. Example 2.5 (Quotient of a global comodule). Consider a global H-comodule X with coaction ρ : X → X ⊗ H and any epimorphism π : X → Y in C. Then we can define a partial comodule datum over Y by taking the pushout of the pair (π, (π ⊗ H) ◦ ρ) (2) Y ρ &▲▲▲▲▲▲▲▲▲▲▲ π yrrrrrrrrrrrrrrrrrrrrrrrrrrrr ❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ ρY X ⑧⑧❄❄ Y • H X ⊗ H π⊗id %▲▲▲▲▲▲▲▲▲▲ t❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤ πY Y ⊗ H Consider the following diagram. X π Y (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ ρ ⑧⑧❄❄ ρ ⑧⑧ ❄❄ X π Y v♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠ / X ⊗ H ρ⊗H / X ⊗ H ⊗ H X⊗∆ π⊗H π⊗H⊗H X ⊗ H π⊗H Y ⊗ H Y ⊗ H ⊗ H Y ⊗ H Y • H v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ ⑧⑧ ❄❄ (Y • H) • H ρY ⊗H (◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠ ◗◗◗◗◗◗◗◗◗◗◗◗ ◗◗◗◗◗◗◗◗◗◗◗◗ πY ⊗H v♠♠♠♠♠♠♠♠♠♠♠♠♠ ◗◗◗◗◗◗◗◗◗◗◗◗ ◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠ ⑧⑧❄❄ (Y • H) ⊗ H ⑧⑧❄❄ (Y • H) • H ⑧⑧❄❄ (Y • H) ⊗ H ◗◗◗◗◗◗◗◗◗◗◗◗◗ ◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗ ⑧⑧❄❄ Y ⊗ H ⊗ H ⑧⑧❄❄ Y • (H • H) (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ Y • H ⑧⑧❄❄ Y • (H ⊗ H) v♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠ By composing pushouts in the diagram, we see that (Y • H) • H is the pushout of the pair (π, (ρY ⊗ H) ◦ (π ⊗ H) ◦ ρ). Moreover, diagram chasing and the coassociativity of (X, ρ) tells us that (ρY ⊗ H) ◦ (π ⊗ H) ◦ ρ = (πY ⊗ H) ◦ (π ⊗ H ⊗ H) ◦ (ρ ⊗ H) ◦ ρ = (πY ⊗ H) ◦ (π ⊗ H ⊗ H) ◦ (X ⊗ ∆) ◦ ρ = (πY ⊗ H) ◦ (Y ⊗ ∆) ◦ (π ⊗ H) ◦ ρ And hence (Y • H) • H is has to be isomorphic to Y • (H • H), which is exactly the pushout of (π, (πY ⊗ H) ◦ (Y ⊗ ∆) ◦ (π ⊗ H) ◦ ρ). We can conclude that (Y, ρY , πY ) is a geometric partial comodule. Performing this construction in C = Setop, we recover Example 1.1 (1) Example 2.6 (Quotient of a partial comodule). The previous example can be general- ized in the following way. Let (X, X • H, πX , ρX) be a partial H-comodule datum, and y & % t / / o o o o / / ( / / v v v ( v v v o o v ( ( ( o o v ( v v v v v v ( ( ( v v v v ( ( ( v v v 16 JIAWEI HU AND JOOST VERCRUYSSE p : X → Y an epimorphism. Then consider the pushout P of the pair (πX, p ⊗ H): X ⊗ H X • H πX yssssssssss %▲▲▲▲▲▲▲▲▲▲▲ p1 P p⊗H %▲▲▲▲▲▲▲▲▲▲ yrrrrrrrrrrr p2 Y ⊗ H Moreover, we can then define a partial comodule datum (Y, Y •H, πY , ρY ) by considering the following pushout p xrrrrrrrrrrrrrrrrrrrrrrrrrrrr ❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ ρY Y X ⑧⑧❄❄ Y • H ρX &▲▲▲▲▲▲▲▲▲▲▲ X ⊗ H p1 &▲▲▲▲▲▲▲▲▲▲▲ t❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤ π′ Y P and taking πY = π′ quasi or geometric partial comodule if X is so. Y ◦ p2. Similar to the previous example, one can show that Y is a Example 2.7 (Partial action in the affine plane). Since the affine group (A2, +) acts strictly transitive on the affine plane, the algebra A = k[x, y] is a Galois object over the bialgebra H = k[x, y]. In particular, A is an H-comodule with coaction ρ : k[x, y] → k[x, y] ⊗ k[x, y] ∼= k[x, y, x′, y′], ρ(f )(x, y, x′, y′) = f (x + x′, y + y′) where f ∈ k[x, y]. Considering the quotient B = k[x, y]/(xy) we find by the previous example that B is a partial H-comodule with B • H = k[x, y, x′, y′]/ρ((xy)). Remark that ρ((xy)) is not an ideal in k[x, y, x′, y′], hence B • H is not an algebra quotient of B ⊗ H. Furthermore, ρB : B → B • H given by ρB(f )(x, y, x′, y′) = f (x + x′, y + y′) for all f ∈ B. As we have remarked in the introduction, it follows from the results of [5] that this example cannot be described by means of partial actions in the sense of Caenepeel- Janssen (see Example 2.9). Example 2.8 (A partial action on the quantum plane). By a similar construction as in the previous example, we obtain a partial action on the quantum plane. Consider the tensor algebra T (V ) where V is a 2-dimensional vector space. Then this tensor algebra is known to be a Hopf algebra and it coacts on itself by the comultiplication. We can view T (V ) as the free algebra k hx, yi with two generators x, y and the coaction is then given by the comultiplication ∆ : k hx, yi → k hx, yi ⊗ k hx, yi , ∆(x) = x ⊗ 1 + 1 ⊗ x, ∆(y) = y ⊗ 1 + 1 ⊗ y. Now consider the quantum plane kq[x, y] = k hx, yi /(xy − qyx). By Example 2.5, the quantum plane is a partial comodule over the tensor algebra. Example 2.9. Consider a partial coaction in the sense of Caenepeel-Janssen [8]. This means that H is a Hopf algebra, A is an algebra and is a linear map satisfying the following axioms: ρ : A → A ⊗ H, ρ(a) = a[0] ⊗ a[1] y % % y x & & t GEOMETRICALLY PARTIAL ACTIONS 17 (CJ1) (ab)[0] ⊗ (ab)[1] = a[0]b[0] ⊗ a[1]b[1] (CJ2) a[0][0] ⊗ a[0][1] ⊗ a[1] = a[0]1[0] ⊗ a[1](1)1[1] ⊗ a[1](2) (CJ3) a[0]ǫ(a[1]) = a Then we define e = 1[0] ⊗ 1[1] ∈ A ⊗ H, which is an idempotent, because of the first axiom. Then we get that A ⊗ H = (A ⊗ H)e ⊕ (A ⊗ H)e′ where e′ = 1 − e. If we put A • H = (A ⊗ H)e, then we have that the map π : A ⊗ H → A • H, a ⊗ h 7→ a1[0] ⊗ h1[1] is surjective with right inverse the inclusion map and kernel (A ⊗ H)e′ = {a ⊗ h − a1[0] ⊗ h1[1] a ⊗ h ∈ A ⊗ H} = N and A • H = (A ⊗ H)/N. This allows us to define the partial action datum over A: A #❋❋❋❋❋❋❋❋❋ ρ=π◦ρ ρ / A ⊗ H ysssssssss π A • H To check the coassociativity, we consider the diagram A ⊗ H ρ⊗H A ⊗ H ⊗ H A⊗∆ A ⊗ H ρ A • H A v♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠ ρ A (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ ρ ρ A • H π v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ ρ•H ⑧⑧❄❄ (A • H) • H ρ⊗H (◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠ ◗◗◗◗◗◗◗◗◗◗◗◗ ◗◗◗◗◗◗◗◗◗◗◗◗ πA•H (A • H) ⊗ H A ⊗ H ⊗ H ⑧⑧❄❄ (A • H) ⊗ H π⊗H v♠♠♠♠♠♠♠♠♠♠♠♠♠ ◗◗◗◗◗◗◗◗◗◗◗◗ ◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ πA•H θ1 π⊗H ◗◗◗◗◗◗◗◗◗◗◗◗◗ ◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠ A,∆ π′ θ2 ⑧⑧❄❄ Θ ⑧⑧❄❄ (A • H) • H ⑧⑧ ❄❄ A • (H • H) π (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ A•∆ ⑧⑧❄❄ A • (H ⊗ H) A⊗∆ v♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠ πA,∆ π′ X Using Lemma 2.4, we find that (A • H) • H ∼= ((A • H) ⊗ H)/K, A • (H • H) = ((A • H) ⊗ H)/L and Θ = ((A • H) ⊗ H)/(K + L) where K = {a[0] ⊗ a[1] ⊗ h − a[0]1[0][0] ⊗ a[1]1[0][1] ⊗ h1[1]a ⊗ h ∈ A ⊗ H} and L = {a1[0] ⊗ h(1)1[1] ⊗ h(2) − a1[0]1[0′] ⊗ h(1)1[1](1)1[1′] ⊗ h(2)1[1](2)a ⊗ h ∈ A ⊗ H} Although in general K and L are not necessarily isomorphic subspaces of (A • H) ⊗ H, we see because of axiom (CJ2) that (π ⊗ H) ◦ (ρ ⊗ H) ◦ ρ(a) = (π ⊗ H) ◦ (A ⊗ ∆) ◦ ρ(a) in (A • H) ⊗ H. Hence the coassociativity holds in particular in the quotient Θ. We can conlude that a Caenepeel-Janssen partial action induces a quasi (and not geometric) partial comodule. Example 2.10. Let (X, X • H, πX , ρX) be a quasi partial H-comodule. We know that X ⊗ H is a (global) right H-comodule with coaction X ⊗ ∆. By applying the result of Example 2.5, we find that the epimorphism πX : X ⊗ H → X • H induces X • H with the structure of a partial H-comodule under the partial coaction X • H (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ π′ X ◦X•∆ v♠♠♠♠♠♠♠♠♠♠♠♠ X,∆ π′ X • (H • H) (X • H) ⊗ H # / y / / ( / / v v v ( v v v o o v ( ( ( v o o ( v v v v v v ( ( ( v v v v ( ( ( v v v ( v ( v v v 18 JIAWEI HU AND JOOST VERCRUYSSE which is geometric by Example 2.5. Therefore, we obtain that the following pushouts are isomorphic, where we denote X • ∆ = π′ X ◦ X • ∆. X • (H • H) (X • (H • H)) ⊗ H (3) (4) (X • H) ⊗ H X•∆⊗H ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ X,∆,H π′ X • ((H • H)) • H) π′ X,∆ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ X•(∆•H) (X • H) ⊗ H X • (H • H) (X • H) ⊗ H ⊗ H π′ X,∆ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ X•(H•∆) (X•H)⊗∆ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ πX,H,∆ π′′ X,1 π′ X,∆⊗H ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ X,H,∆ π′ X • (H • (H • H)) X • (H • (H ⊗ H)) (X • (H • H)) ⊗ H Denote (X • (H • ∆)) = π′′ are identical up-to-isomorphism of their codomains. X,1 ◦(X •(H •∆)). Then we find that the following morphisms (X • (H • ∆)) ◦ (X • ∆) ≃ (X • (∆ • H)) ◦ (X • ∆) When X itself is a geometric partial comodule, one can use the isomorphism θ : X • (H • H) ∼= (X • H) • H to rewrite the above pushouts as (X • (H • H)) • H ∼= X • ((H • H)) • H) ∼= (X • H) • (H • H) ∼= X • (H • (H • H)) we will explain this in more detail in Section 2.3. 2.2. Partial comodule morphisms. Definition 2.11. If (X, πX , ρX) and (Y, πY , ρY ) are two partial H-comodule data, then a morphism of partial H-comodule data is a couple (f, f • H) of morphisms in C, where f : X → Y and f • H : X • H → Y • H such that the following two squares commute X ρX X • H πX X ⊗ H f f •H f ⊗H Y ρY / Y • H πY Y ⊗ H A morphism of a quasi, lax or geometric partial comodule is a morphism of the un- derlying partial comodule data. We denote the categories of quasi, lax and geometric partial H-comodules respectively by qPModH , lPModH and gPModH. When we denote t t t * * t t t t t t * * t t t * * * * * * t t t / / / / / O O O O O O O O GEOMETRICALLY PARTIAL ACTIONS 19 PModH , we mean any of the three partial comodule categories, without specifying which one. If H is an algebra in C, then H is a coalgebra in Cop and one defines the categories of partial modules as the opposite of the corresponding categories of partial comodules over the coalgebra H in Cop Remark 2.12. If (f, f • H) is a morphism of partial comodule data, then f • H is determined by f . Indeed, suppose that both (f, f •H), (g, g•H) : X → Y are morphisms of comodule data with f = g, then using the fact that πX is an epimorphism, it follows that f • H = g • H. This justifies that from now on we will denote a partial morphism (f, f • H) just as f . If moreover πX is a regular epimorphism (that is, it is a coequalizer) in C, then one can express the property of the existence of f • H more explicitly. We spell this out in the abelian case (were all epimorphisms are regular) in the next lemma. Lemma 2.13. Suppose that the category C is abelian. Let (X, πX , ρX) and (Y, πY , ρY ) be partial H-comodule data in C. Then a morphism f : X → Y satisfies (f ⊗ H)(ker πX ) ⊂ ker πY if and only if there exists a unique morphism f • H : X • H → Y • H such that πY ◦ (f ⊗ H) = (f • H) ◦ πX . Proof. The existence and uniqueness of f • H follows directly by universal property of (X • H, πX ) = coker (ker (πX )) in the abelian category C. Conversely, if f • H with the stated property exists, then πY ◦ (f ⊗ H)(ker πX ) = (f • H) ◦ πX (ker πX) = 0 and by the universal property of ker πY we find then that (f ⊗ H)(ker πX ) ⊂ ker πY . (cid:3) Remark 2.14. In case C = Vect, one then finds that a map f : X → Y between two geometric partial modules is a morphism of partial comodules if and only if the following conditions hold: (1) f (x) • h = 0 if x • h = 0; (2) f (x[0]) • x[1] = f (x)[0] • f (x)[1]; where we used the notation introduced in Remark 2.3, and where the second condition make sense thanks to the first one. Lemma 2.15. If f : (X, πX , ρX , θX) → (Y, πY , ρY , θY ) is a morphism of quasi partial H-comodules, then there exist unique morphisms (f • H) • H, f • (H ⊗ H), f • (H • H) and θf such that the following diagrams commute ρX •H X • H (X • H) • H πX•H (X • H) ⊗ H X • H X•∆ X • (H ⊗ H) πX,∆ X ⊗ H ⊗ H f •H (f •H)•H (f •H)⊗H f •H f •(H⊗H) f ⊗H⊗H ρY •H Y • H / (Y • H) • H πY •H (Y • H) ⊗ H Y • H Y •∆ / Y • (H ⊗ H) πY,∆ Y ⊗ H ⊗ H π′ X ◦X•∆ X • H X • (H • H) π′ X,∆ (X • H) ⊗ H (X • H) • H f •H Y • H f •(H•H) (f •H)⊗H (f •H)•H π′ Y ◦Y •∆ / Y • (H • H) π′ Y,∆ (Y • H) ⊗ H (Y • H) • H θX 1 θY 1 θX 2 θY 2 ΘX θf / ΘY X • (H • H) f •(H•H) Y • (H • H) / / o o / o o / / o o / o o / / o o / o o / / o o / o o 20 JIAWEI HU AND JOOST VERCRUYSSE If moreover X and Y are lax, then the following diagram commutes as well. (X • H) • H (f •H)•H (Y • H) • H θX θY X • (H • H) f •(H•H) / Y • (H • H) Proof. This follows by the universal property of the considered pushouts. For example, (f • H) • H : (X • H) • H → (Y • H) • H is defined as the unique morphism that makes the following diagrams commute, where the inner and outer diamond are pushouts X • H (X • H) ⊗ H πY πX f •H v❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧❧ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ / Y • H ρX •H ρY •H X ⊗ H f ⊗H Y ⊗ H ρY ⊗H ρX ⊗H (◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠ (❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘ v♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠ (Y • H) ⊗ H (f •H)⊗H πX•H πY •H (Y • H) • H (f •H)•H (X • H) • H (cid:3) 2.3. Coassociativity. For a usual H-comodule (M, ρ), it is well-known that the coas- sociativity condition implies a generalized coassociativity condition saying that all mor- phisms from M to M ⊗H ⊗n that is constructed out of a combination of ρ, ∆ and identity maps are identical. Our next aim is to prove a similar theorem for geometric partial comodules. To this end, let consider the following compositions of partial mappings from X to X ⊗ H ⊗ H ⊗ H. Let us first construct ρ1 = ((ρ • H) • H) ◦ (ρ • H) ◦ ρ : X → ((X • H) • H) • H which is done in the following diagram, where all quadrangles are pushouts. X ρX X • H ρ πX t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ ρX •H X ⊗ H (a) (X • H) • H ((X • H) • H) ⊗ H ρ⊗H / X ⊗ H ⊗ H X ⊗ H ⊗ H ⊗ H ρ⊗H⊗H ρX ⊗H⊗H ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ πX•H⊗H πX ⊗H⊗H (X • H) ⊗ H ⊗ H (X • H) ⊗ H ρX ⊗H ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ (ρX •H)•H πX•H πX ⊗H t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ (ρX •H)⊗H π(X•H)•H ((X • H) • H) • H In the same way, we can construct ρ2 : ((ρ • H) • H) ◦ (X • ∆) ◦ ρ : X → (X • H) • (H • H), / / / v ( v ( ( / ( v v o o O O / / t t t / / / t t t * * * t t t t t t * t t t GEOMETRICALLY PARTIAL ACTIONS 21 where we denote as before X • ∆ = π′ diagram. X ◦ (X • ∆) and which is defined by the following X ρX X • H ρ πX t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ X•∆ X ⊗ H (b) X⊗∆ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ πX,∆ π′ X X⊗∆ / X ⊗ H ⊗ H X ⊗ H ⊗ H ⊗ H ρ⊗H⊗H ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ ρX ⊗H⊗H πX ⊗H⊗H X ⊗ H ⊗ H πX ⊗H (X • H) ⊗ H ⊗ H πX ⊗H ✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ ρX •(H•H) X,∆ π′ (ρX •H)⊗H ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ X,∆ π′′ (X • H) • (H • H) X • (H • H) ((X • H) • H) ⊗ H X • (H ⊗ H) (c) (X • H) ⊗ H πX•H⊗H Since we know by the coassociativity on X that the pushouts (X • H) • H given by the diagram (a) is isomorphic to the pushout X • (H • H) which is the combinination of diagrams (b) and (c). Therefore it follows that the pushouts ((X • H) • H) • H and (X • H) • (H • H) constructed above are isomorphic as well, in such a way that the constructed maps ρ1 and ρ2 from X into these pushouts are identical up to this isomorphism. Next, we construct a morphism ρ3 : ((X • H) • ∆) ◦ (ρ • H) ◦ ρ : X → (X • H) • (H • H) denoting ((X • H) • ∆) = π′ X•H ◦ ((X • H) • ∆), as in the following diagram. X ρX X • H ρ πX t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ ρX •H X ⊗ H (a) ρX ⊗H ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ (X•H)•∆ πX•H ρ⊗H / X ⊗ H ⊗ H X ⊗ H ⊗ H ⊗ H X⊗H⊗∆ (X • H) ⊗ H X ⊗ H ⊗ H ⊗ H πX ⊗H (X•H)⊗∆ t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ πX•H,∆ X•H π′ X⊗H⊗∆ πX ⊗H⊗H ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ πX•H⊗H X•H,∆ π′ (X • H) • (H • H) (X • H) • (H ⊗ H) ((X • H) • H) ⊗ H (X • H) • H (X • H) ⊗ H ⊗ H Let us first remark that the constructed pushout is the same as the one from the previous diagram. Indeed, we had constructed (X • H) • (H • H) as the pushout of πX with ((ρX • H) ⊗ H) ◦ (πX ⊗ H) ◦ (X ⊗ ∆) = (πX•H ⊗ H) ◦ (ρX ⊗ H ⊗ H) ◦ (X ⊗ ∆) = (πX•H ⊗ H) ◦ ((X • H) ⊗ ∆) ◦ (ρX ⊗ H) It follows that the morphism ρ3 is identical to ρ2 (and to ρ1). Furthermore, one sees that the pushout (a) appears again in the last diagram, by a same argument as before, this can be replaced by the combination of the pushouts (b) and (c), since θX : X • (H • H) → (X • H) • H is an isomorhpism. This leads us to the map ρ4 : ((X • H) • ∆) ◦ X • ∆ ◦ ρ : X → X • (H • (H • H)) / / t t t * / / / * * * * * t t t * * * t t t * * t t t / / t t t * / / / t t t * * * t t t t t t * t t t * * * * * * t t t JIAWEI HU AND JOOST VERCRUYSSE ρ X ⊗ H ρ⊗H / X ⊗ H ⊗ H X ⊗ H ⊗ H ⊗ H X⊗H⊗∆ 22 X ρX X • H πX t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ X•∆ (b) + (c) (X • H) ⊗ H X ⊗ H ⊗ H ⊗ H X • (H • H) (X • H) ⊗ H ⊗ H (πX ⊗H)◦(X⊗∆) ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ X•(H•∆) X,∆ π′ πX ⊗H (X•H)⊗∆ t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ πX,H,∆ π′′ X,1 X⊗H⊗∆ πX ⊗H⊗H ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ π′ X,∆⊗H X,H,∆ π′ X • (H • (H • H)) X • (H • (H ⊗ H)) (X • (H • H)) ⊗ H Remark that (X •H)•(H •H) is the pushout of the pair (πX•H , (πX•H ⊗H)◦((X •H)⊗∆) and X • (H • (H • H)) is the pushout of the pair (π′ X,∆ ⊗ H) ◦ ((X • H) ⊗ ∆). Since πX•H = θX ◦ π′ X,∆ and θX is an isomorphism, it follows that both pushouts are isomorphic and φ3 and φ4 are identical up to this isomorphism. Let us now consider the morphism X,∆, (π′ ρ5 : ((X • ∆) • H) ◦ (ρ • H) ◦ ρ : X → (X • (H • H)) • H where ((X • ∆) • H) = (π′ diagram X • H) ◦ ((X • ∆) • H) and that is given by the following X ρX X • H ρ πX t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ ρX •H X ⊗ H (a) ρX ⊗H ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ (X•∆)•H πX•H ρ⊗H / X ⊗ H ⊗ H X ⊗ H ⊗ H ⊗ H X⊗∆⊗H (X • H) ⊗ H X ⊗ H ⊗ H ⊗ H πX ⊗H (X•∆)⊗H t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ πX•(H⊗H) X •H π′ X⊗∆⊗H πX,∆⊗H ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ πX•(H •H) π′ X ⊗H (X • (H • H)) • H (X • (H ⊗ H)) • H (X • (H • H)) ⊗ H (X • H) • H (X • (H ⊗ H)) ⊗ H And again, by replacing the pullback (a) by the pullback (b)+(c), we obtain a map that is the same up-to-isomorphism the same as ρ5: ρ6 : (X • (∆ • H)) ◦ X • ∆ ◦ ρ : X → X • ((H • H) • H) / / t t t * / / / t t t * * * t t t t t t * t t t * * * * * * t t t / / t t t * / / / t t t * * * t t t t t t * t t t * * * * * * t t t GEOMETRICALLY PARTIAL ACTIONS 23 where (X • (∆ • H)) = π′′ diagram. X,2 ◦ (X • (∆ • H)). This map ρ6 is defined by the following ρ X ⊗ H ρ⊗H / X ⊗ H ⊗ H X ⊗ H ⊗ H ⊗ H πX ⊗H (X•∆)⊗H t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ πX,∆,H π′′ X,2 X⊗∆⊗H X⊗∆⊗H πX,∆⊗H ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ X ⊗H X,∆,H π′ π′ X • ((H • H)) • H) X • ((H ⊗ H)) • H) (X • (H • H)) ⊗ H X ρX X • H πX t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ X•∆ (πX ⊗H)◦(X⊗∆) ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ X•(∆•H) X,∆ π′ (b) + (c) (X • H) ⊗ H X ⊗ H ⊗ H ⊗ H X • (H • H) (X • (H ⊗ H)) ⊗ H By Example 2.10, we know that X • ((H • H)) • H) ∼= X • (H • (H • H)) and the maps ρ4 and ρ6 are identical up to this isomorphism. Hence we have hereby proven that the all above constructed pushouts are isomorphic and the maps ρi (i = 1, . . . , 6) are identical up to these isomorphisms. All this is summarized in the following result. Theorem 2.16 (generalized coassociativity). Let (X, πX , ρX, θX ) be a geometric partial comodule. Then the pushouts introduced above are all isomorphic X • (H • (H • H)) ∼= (X • H) • (H • H) ∼= ((X • H) • H) • H ∼= (X • (H • H)) • H ∼= X • ((H • H) • H) Moreover up to these isomorphisms, the following morphisms X → X • H • H • H are identical (X • H • ∆) ◦ (X • ∆) ◦ ρ ≃ (ρ • H • H) ◦ (X • ∆) ◦ ρ ≃ (X • H • ∆) ◦ (ρ • H) ◦ ρ ≃ (ρ • H • H) ◦ (ρ • H) ◦ ρ ≃ (X • ∆ • H) ◦ (ρ • H) ◦ ρ ≃ (X • ∆ • H) ◦ (X • ∆) ◦ ρ Corollary 2.17. If (X, πX , ρX) is a geometrically partial H-comodule, then (X •H, (X • H) • H, πX•H , ρX • H) is a geometrically partial H-comodule. Corollary 2.18. All higher coassociativity conditions follow now by an induction argu- ment from the previous two results. Remarks 2.19. (1) A lax version of the above results on generalized coassociativity Indeed, analysing the reasoning at the start can be proven in the same way. of this section, each of the isomorphisms between the pullbacks obtained in Theorem 2.16 follows from the isomorphism θ : X • (H • H) → (X • H) • H at appropriate places. When θX is only assumed to be a morphism (not an isomorphism), then we also obtain only morphisms (and not isomorphisms) between the constructed pullbacks. The coassociativity will then hold up to composition with the induced morphisms onto ((X • H) • H) • H. (2) As the isomorphisms between the respective pullbacks are constructed by apply- ing the universal property of the pullback, one can moreover easily see, that these / / t t t / / / t t t * * * t t t t t t * t t t * * * * * * t t t 24 JIAWEI HU AND JOOST VERCRUYSSE isomorphisms are compatible in a way that the following diagram commutes X • (H • (H • H)) (X • H) • (H • H) / ((X • H) • H) • H X • ((H • H) • H) / (X • (H • H)) • H where all arrows are isomorphisms in the geometric case, and just morphisms in the lax case. The commutativity of this diagrams seems to suggest that there is an underlying (skew) monoidal structure with tensor product − • −. In the next section, we will show that in case H is a bialgebra, there is at least a lax monoidal structure on the category of geometric partial modules, which coincides with the •-product that we encountered so far. 2.4. Completeness and cocompleteness of the category of partial comodules. For global comodules, the forgetful functor U : ModH → C allows a right adjoint given by the free functor − ⊗ H : C → ModH. Since every global comodule is also a partial module, the free functor − ⊗ H : C → PModH still makes sense, however it no longer serves as a right adjoint for the forgetful functor U : PModH → C, which is defined as U(X, ρX , πX, θX ) = X on objects and U(f, f • H) = f on morphisms. We now show that the forgetful functor still has a right adjoint. Troughout this section, we suppose that the counit ǫ of the coalgebra H is an epimorphism in C (this is for example the case if C = Vect or if H is a bialgebra). Proposition 2.20. Let V be any object in C, then V can be endowed with a partial H-comodule structure putting V • H = V , π = V ⊗ ǫH and ρ = idV . We call this the "trivial partial comodule structure" on V . Moreover a trivial partial comodule is always geometric and the functor T : C → PMod that assigns to each C-object the trivial partial comodule structure, is fully faithful and a right adjoint for the forgetful functor U : PModH → C. Proof. It can be easily verified that (V, V, V ⊗ǫH , idV ) is a geometric partial H-comodule with (V • H) • H = V • (H • H) = V . Given a partial comodule (X, X •H, πX , ρX), we find that T U(X) = (X, X, X ⊗ǫH , idX) and we define the unit of the adjunction as ηX = (idX, X • ǫH ) : X → T U(X). For any object V in C, we see that UT (V ) = V . Then the unit-counit conditions become trivial. Since the counit is the identity, we obtain that T is fully faithful. (cid:3) Since the forgetful functor has a right, it preserves all colimits that exist in PModH . The main aim of this section is to show that colimits and limits indeed exist in PModH. Let us first show that thanks to the observation of the previous proposition, the category PModH is well-copowered. Recall that a category is called well-copowered if and only if for any object X, there exist up-to-isomorphism only a set of epimorphisms f : X → Y . Corollary 2.21. A morphism f ∈ PModH is an epimorphism if and only if U(f ) = f is an epimorphism in C. Furthermore, the category PModH is well-copowered if C is so. Proof. Since the forgetful functor U : PModH → C has a right adjoint, U preserves epimorphisms. Conversely, if f : X → Y in PModH is such that U(f ) is an epimorphism, then f is an epimorphism as well. Indeed, suppose that we have g, h : Y → Z in PModH such that g◦f = h◦f . Then also U(g)◦U(f ) = U(g)◦U(f ) in C and hence U(g) = U(f ). / / / / O O GEOMETRICALLY PARTIAL ACTIONS 25 But in Remark 2.12, we remarked that for a morphism f ∈ PModH, f • H is completely determined by f (or by U(f ) to be precise). Hence we find that g = h in PModH. Let (X, X • H, πX, ρX) be a partial comodule datum. Since C is well-copowered, there exists up-to-isomorphism only a set of epimorphisms f : X → Y in C. Moreover, for each Y , there exist again since C is well-copowered, only a set of epimorphisms Y ⊗ H → Y • H, hence also only a set of partial comodule data over Y . We conclude that there will be only a set of epimorphisms f : X → Y in PModH and hence PModH is also well-copowered. (cid:3) Theorem 2.22. Suppose that the endofunctor − ⊗ H : C → C preserves colimits. Then the following statements hold. (i) If the category C is k-linear then PModH is also k-linear and the forgetful functor is k-linear. (ii) If the category C has all colimits of a shape Z, then PModH also has colimits of shape Z. Hence, if C is cocomplete then PModH is cocomplete and the forgetful functor U : PModH → C preserves colimits. Consequently, in case the above conditions hold, PModH is addive. Proof. (i) Let X = (X, X • H, πX, ρX ) and (Y, Y • H, πY , ρY ) be a two partial comodule data and (f, f • H), (g, g • H) : X → Y two morphisms. Let us verify that (f + g, f • H + g • H) is again a morphism. Then we have ρY ◦ (f + g) = ρY ◦ f + ρY ◦ g = (f • H) ◦ ρX + (g • H) ◦ ρX = ((f • H) + (g • H)) ◦ ρX And similarly, ((f •H)+(g•H))◦πX = πY ◦((f ⊗H)+(g⊗H)). Hence (f +g, f •H +g•H) is indeed a morphism in PModH . Similarly, for any a ∈ k, we define a(f, f •H) = (af, af •H). One easily verifies that this is again a morphism, and using this addition and scalar multiplication, the Hom-sets in PModH are k-modules and composition is k-bilinear. (ii) Let Z be any small category and F : Z → PModH a functor, where we denote for each Z ∈ Z, F Z = (F Z, F Z • H, ρF Z, πF Z), i.e. we denote UF Z = F Z for short. Consider the functor UF : Z → C and denote (C, γZ) = colim UF , where γZ : F Z → C are such that γZ = γZ ′ ◦ F f for any f : Z → Z ′ in Z. Consider now the functor (UF )H : Z → C given by (UF )H Z = F Z ⊗ H. Then by assumption we have that colim (UF )H = (C ⊗ H, γZ ⊗ H). Finally consider the functor (UF )• : Z → C given by (UF )•Z = F Z • H for all Z ∈ Z, and denote colim (UF )• = (C • H, δZ) where δZ : F Z •H → C •H are such that δZ = δZ ′ ◦F f •H. Let us verify that (C •H, δZ ◦ρF Z) is a cocone for UF . Indeed, for any morphism f : Z → Z ′ in Z, F f is a morphism in PModH and hence the following diagram commutes F Z ρF Z F Z • H &▲▲▲▲▲▲▲▲▲▲ δZ C • H F f F f •H F Z ′ ρF Z′ F Z ′ • H xrrrrrrrrrr δZ′ By the universal property of colim UF , we then obtain a unique morphism ρC : C → C • H such that δZ ◦ ρF Z = ρC ◦ γZ for all Z ∈ Z. In the same way, one shows that (C • H, δZ ◦ πF Z) is a cocone for UF H , and hence there exists a morphism πC : C ⊗ H → / / / / & x 26 JIAWEI HU AND JOOST VERCRUYSSE C • H such that δZ ◦ πF Z = πC ◦ γZ ⊗ H for all Z ∈ Z. The situation is summarized in the next diagram. F Z ρF Z F Z • H πF Z F Z ⊗ H γZ δZ γZ ⊗H C ρC / C • H πC / C ⊗ H Let us show that (C, C • H, ρC, πC) is a partial comodule datum, i.e. that πC : C ⊗ H → C • H is an epimorphism in C. To this end, consider f, g : C • H → X in C such that f ◦ πC = g ◦ πC. Then for all Z ∈ Z we have that f ◦ πC ◦ (γZ ⊗ H) = f ◦ δZ ◦ πF Z = g ◦ πC ◦ (γZ ⊗ H) = g ◦ δZ ◦ πF Z Since each πF Z is epi, we find f ◦ δZ = g ◦ δZ for all Z and since the δZ are jointly epi, we obtain that f = g and therefore πC is indeed an epimorphism. Furthermore, by the interchange law for colimits, it follows that the pushouts C •H(•H), (C • H) • H and ΘC can be computed as the colimits of the respective functors Z → C that construct the pushouts Z • H(•H), (Z • H) • H and ΘZ. Hence, it follows that if all F Z are quasi, lax or geometric comodules, then Z will be such as well. The last statement follows from the above, since it is well-know that a preadditive category with binary coproducts is additive. (cid:3) As we will show further in this section, there exist monomorphisms f in PModH such that U(f ) is not a monomorphism in C. In particular, U does not have a left adjoint. Nevertheless, we have the following result. Lemma 2.23. Consider a morphism f : X → Y in PModH. If Uf : UX → UY is a monomorphism in C, then f is also a monomorphism in PModH. Proof. Consider two morphisms g, h : Z → X in PModH such that f ◦ g = f ◦ h. Since Uf is a monomorphism, we obtain Ug = Uh. Then by Remark 2.12, we find that also g • H = h • H, i.e. g = h in PMod. (cid:3) Definition 2.24. A subcomodule of a partial comodule (X, X • H, ρX, πX ) is a partial comodule datum (Y, Y • H, ρY , πY ), together with a morphism f : Y → X for which both f and f • H are monomorphisms in C. From now on, we restrict to our case of interest C = Vectk where k is a field. Proposition 2.25. Let (X, X • H, ρX , πX) be a partial comodule datum and j : Y → X a subobject of X in Vectk. Consider the epi-mono factorization of πX ◦(j⊗H) : Y ⊗H → X • H, which we denote as follows: Y ⊗ H πY / Y • H j•H / X • H Then (i) ker πY ∼= Im (j ⊗ H) ∩ ker πX ; / / / O O / O O / / / / GEOMETRICALLY PARTIAL ACTIONS 27 (ii) Denote as usual by Y • (H • H) the pushout of the pair (πY , (πY ⊗ H) ◦ Y ⊗ ∆) Y,∆ : (Y • H) ⊗ H → Y • (H • H) associated pushout of the morphism πY . X,∆ and therefore Y • (H • H) is isomorphic ∼= Im ((j • H) ⊗ H) ∩ ker π′ and by π′ Then ker π′ to the image of the map π′ Y,∆ X,∆ ◦ (j • H) ⊗ H; If moreover Y allows a partial comodule datum of the form (Y, Y • H, ρY , πY ) such that j is a morphism of partial comodule data, then (iii) Y is a partial subcomodule of X. (iv) (Y • H) • H is isomorphic to the image of the map πX•H ◦ (j • H) ⊗ H; (v) if X is a lax (resp. geometric) partial comodule, then Y is as well a lax (resp. geometric) partial comodule. Proof. (i). By construction we have the following commutative diagram Y ⊗ H πY Y • H j⊗H j•H / X ⊗ H πX / X • H Since j • H is injective and by the commutativity of the diagram, we then find that ker πY = ker ((j • H) ◦ πY ) = ker (πX ◦ (j ⊗ H)) ∼= Im (j ⊗ H) ∩ ker πX . (ii). By definition of a partial comodule, we know that (X•ǫ)◦πX (P xi⊗hi) = P xiǫ(hi) for all P xi ⊗ hi ∈ X ⊗ H, so in particular also for all P xi ⊗ hi ∈ Y ⊗ H. Hence, it follows that the map (X • ǫ) ⊗ H is a left inverse for the map (πX ⊗ H) ◦ (X ⊗ ∆). Also, remember that the kernel of the map π′ X,∆ is given by (πX ⊗ H) ◦ (X ⊗ ∆)(ker πX ). Combining these observations, we find that (πX ⊗ H) ◦ (X ⊗ ∆) is an isomorphism between ker πX and ker π′ X,∆. As all maps restrict to Y , we find in the same way that (πY ⊗ H) ◦ (Y ⊗ ∆) : Y ⊗ H → (Y • H) ⊗ H has a left inverse and this map defines an isomorphism between ker πY and ker π′ Y,∆. This is summarized in the next diagram. ker π′ Y,∆ ∼= '◆◆◆◆◆◆◆◆◆◆ ker πY ker πX ker π′ X,∆ ∼= w♣♣♣♣♣♣♣♣♣♣ 7♦♦♦♦♦♦♦♦♦♦♦ w♦♦♦♦♦♦♦♦♦♦♦ (Y • H) ⊗ H (j•H)⊗H Y ⊗ H X ⊗ H g❖❖❖❖❖❖❖❖❖❖❖ '❖❖❖❖❖❖❖❖❖❖❖ (X • H) ⊗ H As part (i) states exactly that the inner square of this diagram is a pullback, a simple ∼= diagram chasing argument shows that the outer square is also a pullback, i.e. ker π′ Im ((j • H) ⊗ H) ∩ ker π′ Y,∆ X,∆. / / / / _ ' _ w / / _ _ / / j J w t ' / / 7 7 7 g g g 28 JIAWEI HU AND JOOST VERCRUYSSE From this, it follows immediately that the map (j • H) • H is injective, which proofs the statement. ker π′ Y,∆ (j•H)⊗H / (ker π′ X,∆) (Y • H) ⊗ H (j•H)⊗H (X • H) ⊗ H π′ Y,∆ π′ X,∆ Y • (H • H) j•(H•H) / X • (H • H) (iii). It is clear by construction that (j, j • H) is a morphism of partial comodule data and j • H is injective. (iv). This is proven in the same way as in part (ii). We have to show that (j • H) • H : (Y • H) • H → (X • H) • H is injective. So suppose that (y • h) • h′ ∈ (Y • H) • H is such that (j(y) • h) • h′ = 0 in (X • H) • H. Since πY •H is surjective, we find that (y • h) • h′ = πY •H ((y • h) ⊗ h′) and (j(y) • h) ⊗ h′ ∈ ker πX•H = (ρX ⊗ H)(ker πX ). Hence, (j(y) • h) ⊗ h′ = (xi[0] • xi[1]) ⊗ hi for some xi ⊗ hi ∈ ker πX . Applying (X • ǫ) ⊗ H to the last identity, we obtain by part (i) that xi ⊗ hi = j(y) ⊗ ǫ(h)h′ ∈ j(Y ) ⊗ H ∩ ker πX ∼= ker πY . Hence (j(y) • h) ⊗ h′ = (xi[0] • xi[1]) ⊗ hi ∈ (j • H) ⊗ H ◦ (ρY ⊗ H)(ker πY ) ∼= ker πY •H , so (y • h) • h′ = 0. (v). Suppose that X is a lax partial module. Then by part (iii) and (iv) above, we can restrict and corestrict θX to obtain a morphism θY : Y • (H • H) → (Y • H) • H. If moreover X is geometric, than we can also restrict and corestrict θ−1 X to obtain an inverse θ−1 (cid:3) Y of θY and Y is again geometric. The following corollary describes a phenomenon that was also observed in [2] for the case of partial representations. Corollary 2.26. Any partial subcomodule of a global comodule is again global. Proof. By Proposition 2.25, we know that for partial subcomodule Y of partial comodule X that ker πY ⊂ ker πX. Moreover, if X is global then ker πX = 0 and therefore also ker πY = 0 so Y is global. (cid:3) We are now ready to prove the 'fundamental theorem for partial comodules'. Theorem 2.27 (Fundamental theorem for partial comodules). Let X = (X, X•H, ρX , πX) be a quasi partial comodule over the k-coalgebra H, and consider any x ∈ X. Then there exists a finite dimensional (quasi) partial subcomodule Y ⊂ X such that x ∈ Y . Consequently, if X is moreover lax (resp. geometric), then Y is as well lax (resp. geometric). Proof. Take x ∈ X and write ρX (x) = Pi yi • hi = πX (P yi ⊗ hi), where hi is a base of H. We know that in the tensor Pi yi • hi only a finite number of the elements yi are non-zero. By coassociativity in the partial comodule X, the identity (5) θ1 ◦ (ρX • H)(ρX(x)) = θ2 ◦ π′ X ◦ (X • ∆)(ρX (x)) _ / _ / / / GEOMETRICALLY PARTIAL ACTIONS 29 holds in the coassociativity pushout Θ. Write as before ρX (x) = πX(P yi ⊗ hi), and let us denote ∆(hi) = P ai jk ∈ k. Then we can rewrite the right hand side of (5) as jkhj ⊗ hk for certain ai θ2 ◦ π′ X ◦ (X • ∆)(ρX(x)) = θ2 ◦ π′ X ◦ (X • ∆)(πX (X yi ⊗ hi)) X,∆(X πX (yi ⊗ hi(1)) ⊗ hi(2)) X,∆(X πX (yk ⊗ ak jihj) ⊗ hi). On the other hand, the left hand side of (5) can be rewritten as = θ2 ◦ π′ = θ2 ◦ π′ θ1 ◦ (ρX • H)(ρX(x)) = θ1 ◦ (ρX • H)(πX(X yi ⊗ hi)) = θ1 ◦ πX•H (X ρ(yi) ⊗ hi). Furthermore, by definition of the coassociativity pushout, we have the identity θ1 ◦ πX•H = θ2 ◦ π′ X,∆ : (X • H) ⊗ H → Θ. Moreover, these maps are surjective, and their kernel is given by ker πX•H + ker π′ X,∆. From Lemma 2.4 it follows that ker πX•H = (ρ ⊗ H)(ker πX) and ker π′ X,∆ = (πX ⊗ H) ◦ (X ⊗ ∆)(ker πX ). Hence we know that there exist elements P zi ⊗ hi,P z′ i ⊗ hi ∈ ker πX such that X ρ(yi) ⊗ hi + X ρ(zi) ⊗ hi = X πX (yk ⊗ ak jihj) ⊗ hi + X πX (z′ k ⊗ ak jihj) ⊗ hi in (X • H) ⊗ H. When we apply now (X • ǫ) ⊗ H to this identity, we obtain that X yi ⊗ hi + X zi ⊗ hi = X yi ⊗ hi + X z′ i ⊗ hi from which it follows that P zi ⊗ hi = P z′ that ρ(x) = πX ((yi + zi) ⊗ hi). Combining the above, we find that i ⊗ hi. Since P zi ⊗ hi ∈ ker πX , we still have X ρ(yi + zi) ⊗ hi = X πX ((yk + zk) ⊗ ak and using the linear independence of the hi's, we obtain that jihj) ⊗ hi ρ(yi + zi) = πX ((yk + zk) ⊗ ak jihj). Hence, when we define Y as the subspace of X generated by the (finite number of non- zero) elements yi +zi, we see that x ∈ Y and ρ(Y ) ⊂ πX(Y ⊗H). As in Proposition 2.25, define Y •H and πY : Y ⊗H → Y •H by means of the epi-mono factorisation of the map πX restricted to Y , then by construction we have the following commutative diagram Y ⊗ H πY Y • H X ⊗ H πX / X • H and moreover, the above shows that (Y, Y • H, πY , ρY ), where ρY is the restriction of ρX to Y , is a partial comodule datum and the inclusion map j : Y → X is a morphism of partial comodules. Then, by Proposition 2.25, we know that Y is a partial subcomodule of X. The last statement follows from the fact that any quasi partial subcomodule of a lax (resp. geometric) partial comodule is itself lax (resp. geometric) by Proposition 2.25. (cid:3) Corollary 2.28. The category of geometric partial comodules has a generator. / / / 30 JIAWEI HU AND JOOST VERCRUYSSE Proof. Let I be the set of isomorphism classes of finite dimensional geometric partial comodules over H. Since there exists clearly only a set of partial comodule structures over a given finite dimensional vector space, it follows that I is indeed a set. For any i ∈ I choose one comodule Mi and denote by G the coproduct `i∈I Mi. Then the fundamental theorem implies there is a surjective morphism G → X for any geometric partial comodule. Hence, G is a generator for gPModH. (cid:3) Corollary 2.29. The category of geometric partial comodules is complete and cocom- plete. Proof. This follows from the known fact that a cocomplete well-copowered category with a generator is complete. (cid:3) Remark 2.30. Although the category PModH is complete, its limits are not preserved by the forgetful functor U to Vect. More precisely, if L is a limit of a diagram D in PModH , then it is clear that U(L) is a cone for the diagram U(D) in Vect. Hence there is a morphism u : U(L) → L′ in Vect where L′ is the limit in Vect of U(D). In general however, this morphism u is not a bijection. Rather, L can be understood as the biggest partial comodule inside L′ that allows a cone on D. Remark however, that in order to be able to speak about the 'biggest' partial comodule inside L′, we already use implicitly the existence of limits in PModH. This can be seen more explicitly by considering the kernel of a morphism f : X → Y in PModH which can be understood as the biggest partial subcomodule K of X such that U(K) is contained in the vector space kernel of f . Thanks to the completeness and cocompleteness of PModH , we can construct from two partial subcomodules v : V → X and w : W → X the pushout of the pullback of v and w, which is then a partial subcomodule of X containing both V and W . The following result will be important in the next section. Corollary 2.31. The forgetful functor gPModH → PCD (PCD denotes the category of partial comodule data), is fully faithful and has a left adjoint B. Proof. Let X be a partial comodule datum. Then BX is the biggest partial subcomodule of X which is geometric. As we explained in the previous remark this construction makes sense. It is easily verified that this provides a left adjoint to the forgetful functor. (cid:3) In the remaining part of this section, we will show that the category of Partial modules is not abelian. To this end, we will construct an example of a morphism f such that ker coker f and coker ker f are not isomorphic. Consider a global comodule X and Y a linear subspace of X which is not a (global) sub- comodule (recall that by Corollary 2.26 any subcomodule of global module is global). We can then construct the induced partial comodule X/Y as in Example 2.5 and con- sider the canonical projection p : X → X/Y which is a morphism of partial comodules. Then the vector space kernel of p is just Y . However as we assumed that Y was not a subcomodule of X, Y is also not a partial subcomodule of X and hence it can not be the kernel of p in PModH . Rather, this kernel is the biggest (global) subcomodule of X contained in Y . Suppose that Y was a one-dimensional subspace of X, then it follows that the kernel of p has to be 0. Then p is both a monomorphism (as morphisms with a zero kernel in additive categories are monomorphisms) and an epimorphism (as p is surjective and Corollary 2.21) but not an isomorphism. Hence PModH is not abelian. We also see as mentioned earlier that there exist monomorphisms in P ModH whose underlying map in Vect is not injective. GEOMETRICALLY PARTIAL ACTIONS 31 3. Partial comodules over a bialgebra 3.1. Lax monoidal categories. A category C is called lax monoidal [14] if • for each n ∈ N there exists an n-fold tensor functor ⊗n : C × · · · × C } {z n → C; • for each for each (k1, . . . , kn) ∈ Nn, there exists a natural transformation γk1,...,kn : ⊗n ◦ (⊗k1 × . . . × ⊗kn) → ⊗k1+...+kn • there exists natural transformation ι : idC → ⊗1, that satisfy the following associativity and unitality conditions. ⊗n ◦ (⊗k1 × . . . × ⊗kn) ◦ ((⊗ℓ11 × . . . × ⊗ℓ1k1 ) × . . . × (⊗ℓn1 × . . . × ⊗ℓnkn ) ⊗k1+...+kn ◦ ((⊗ℓ11 × . . . × ⊗ℓ1k1 ) × . . . × (⊗ℓn1 × . . . × ⊗ℓnkn ) γk1,...,kn ∗id u❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦ %❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑ ,...,ℓn1,...,ℓnkn id∗(γ %❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑ u❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦ ⊗n ◦ (⊗ℓ11+...+ℓ1k1 ℓ11+...+ℓ1k1 γ ⊗ℓ11+...+ℓ1k1 +...+ℓn1+...+ℓnkn ℓ11 ,...,ℓ1k1 ×...×γℓn1,...,ℓnkn ) × . . . × ⊗ℓn1+...+ℓnkn) ,...,ℓn1+...+ℓnkn ℓ11,...,ℓ1k1 γ id⊗n ∗(ι,...,ι) ⊗n ❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ ❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ γ1,...,1 ⊗n ⊗n ◦ (⊗1, . . . , ⊗1) ⊗n ⊗1 ◦ ⊗n ι∗idιn ◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ ◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ γn ⊗n Remark that the last two conditions imply in particular that the functor ⊗1 is idempo- tent. There is an obvious notion of oplax monoidal categories, where the direction of the natural transformations γ and ι is reversed. If the natural transformations γ and ι are invertible, then a lax monoidal category is just a monoidal category. A (lax) monoidal functor between lax monoidal categories is a functor F : C → D that comes equipped with natural transformations for each n ∈ N, satisfying the following compatibility conditions with γ and ι ζn : ⊗D n F n → F ⊗C n : Cn → D ⊗D n (⊗D k1 × . . . × ⊗D kn)F k1+...+kn γ k1,...,kn D F k1+...+kn ⊗D k1+...+knF k1+...+kn F ⊗n(ζk1 ×...×ζkn ) ⊗D n F n(⊗C k1 × . . . × ⊗C kn) ζn(⊗k1 ×...×⊗kn ) F ⊗C n (⊗C k1 × . . . × ⊗C kn) ζk1+...+kn F γ k1,...,kn C / F ⊗C k1+...+kn F ιC !❇❇❇❇❇❇❇❇❇ ιDF ⊗D 1 F F ⊗C 1 <①①①①①①①① ζ1 u % % u / / / / / / / / / ! < 32 JIAWEI HU AND JOOST VERCRUYSSE Similarly, a functor G : C → D is called opmonoidal if there are natural transformations δn : F ⊗C n → ⊗D n F n Cn → D. satisfying appropriate compatibility conditions with γ and ι. The following result might be well-known, but as we didn't found a reference we state it and give a sketch of the proof, which is quite elementary, but because of notational problems becomes quite technical. This result allows to construct many lax monoidal categories. Theorem 3.1. (i) Let D be a lax monoidal category and consider a pair (L, R) of adjoint functors C L R / D then C is also a lax monoidal category such that R is a monoidal functor and L is an opmonoidal functor. (ii) Let D be an oplax monoidal category and consider a pair (L, R) of adjoint functors D L R / E then E is also an oplax monoidal category such that R is a monoidal functor and L is an opmonoidal functor. Proof. We only give a sketch of the proof of part (i), the second follows by duality. Denote the n-fold tensor products in D by ⊗D by γD and ιD. For any n-tuple (c1, . . . , cn) of objects in C, define the n-fold tensor product in C as n and its associativity and unity constraints ⊗C n(c1, . . . , cn) = R(⊗D n (Lc1, . . . , Lcn)). More precisely, the n-fold tensor product in C is defined as the following composition of functors ⊗C n = R ◦ ⊗D n ◦ Ln : Cn → C. Let us denote by η : idC → RL and ǫ : LR → idD the unit and counit of the adjunction (L, R). For any n-tuple (c1, . . . , cn) in C, we can consider the morphism (6) δc1,...,cn n = ǫ⊗n(Lc1,...,Lcn) : L ⊗C n (c1, . . . , cn) = LR ⊗D n (Lc1, . . . , Lcn) → ⊗D n (Lc1, . . . , Lcn) which is natural in each of the entries ci, defining in this way for each n ∈ N a natural transformation δn = ǫ ⊗D n Ln : L⊗C n = LR ⊗D n L → ⊗D n Ln : Cn → D. Similarly, for each n-tuple (d1, . . . , dn) in D we put (7) ζ d1,...,dn n = R ⊗D R ⊗D n (ǫd1, . . . , ǫdn) : n (LRd1, . . . LRdn) = ⊗C n(Rd1, . . . Rdn) → R ⊗D (d1, . . . , dn) which defines a natural transformation ζn = R ⊗D n ǫn : ⊗C nRn = R ⊗D n (LR)n → R⊗D n : Dn → C. / o o / o o GEOMETRICALLY PARTIAL ACTIONS 33 To define the associativity constraint of C, first remark that ⊗n ◦ (⊗k1 × . . . × ⊗kn) = (R ⊗D = R ⊗D n Ln) ◦ ((R ⊗D n (LR)n(⊗D k1 Lk1) × . . . × (R ⊗D kn)Lk1+...+kn k1 × . . . × ⊗D kn Lkn)) We now define the associativity constraint γC of C as the following composition k1+...+kn))Lk1+...+kn = ⊗C R ⊗D k1 × . . . × ⊗D kn)Lk1+...+kn n (LR)n(⊗D R(⊗D k1,...,kn C γ k1+...+kn (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ )Lk1+...+kn ) (ζn(⊗D k1 ×...×⊗D kn 6♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠ k1 ,...,kn D (Rγ Lk1+...+kn ) R ⊗D n (⊗D k1 × . . . × ⊗D kn)Lk1+...+kn The unitality constraint of C is defined as the composition ιC = (RιDL) ◦ η : idC → R ⊗D 1 L = ⊗C 1 . The associativity conditions for the lax monoidal structure on C then follow directly from the naturality of ǫ and the associativity in D. The unitality conditions in C follow from the unit-counit condition of the adjunction (L, R) and the unitality conditions in D. The monoidal structure on the functor R is given by (7), the op-monoidal structure on L is given by (6). (cid:3) The previous proposition can be applied in particular to a monoidal category category D and allows to produce in this way many natural examples of (op)lax monoidal categories. As an intermediate notion between lax monoidal categories and monoidal categories, one can consider a monoidal category with lax unit. This is a category C endowed with a monoidal tensor product ⊗ : C × C → C, endowed with an associativity constraint αC,C ′,C ′′ : (C ⊗ C ′) ⊗ C ′′ → C ⊗ (C ′ ⊗ C ′′) which is a natural isomorphism that satisfies the usual pentagon condition. A lax unit I for such an associative tensor product is an object I in C such that for any C ∈ C there are natural transformations satisfying the usual compatibility constraints with α: ℓC : I ⊗ C → C, r : C ⊗ I → C (C ⊗ I) ⊗ C ′ αC,I,C′ / C ⊗ (I ⊗ C ′) '❖❖❖❖❖❖❖❖❖❖❖ w♦♦♦♦♦♦♦♦♦♦♦ C ⊗ C ′ The following is now an easy observation. Lemma 3.2. If (C, ⊗, I) is a monoidal category with lax unit, then C is a lax monoidal category by defining • ⊗0 = I, ⊗1 = idC, ⊗2 = ⊗; • for all n > 2, ⊗n = ⊗ ◦ (id × ⊗n−1); • ι = id : idC → ⊗1; • for all (k1, . . . , kn) ∈ Nn 0 , γk1,...,kn is canonically obtained from combinations of α and identities and are therefore invertible; ( 6 / / ' / w 34 JIAWEI HU AND JOOST VERCRUYSSE • for any (k1, . . . , kn) ∈ Nn, where ki1 = . . . = kim = 0 (m < n), γk1,...,kn is canonically obtained from combinations of ℓ, r, α and identities and are not invertible; Similarly, one introduces the notion of a monoidal category with an oplax unit, which gives rise to an oplax monoidal category. 3.2. The lax monoidal category of geometric partial comodules over a bi- algebra. The following result is essentially due to Johnstone [13], who formulated the proof in case of cartesian closed categories, but the argument easily generalizes to closed monoidal categories. Let us recall first that a monoidal category is called left closed monoidal if for each object X in C, the endofunctor X ⊗ − : C → C has a right adjoint, that we denote by [X, −] and that is called the internal hom. In other words, if C is right closed, then for any triple of objects X, Y, Z in C we have isomorphisms HomC(X ⊗ Y, Z) ∼= HomC(Y, [X, Z]) for any f ∈ HomC(X ⊗ Y, Z) we denote the corresponding element in HomC(Y, [X, Z]) by f , and conversely for any g ∈ HomC(Y, [X, Z]), we have g ∈ HomC(X ⊗ Y, Z) with f = f and g = g. If one considers the evaluation and coevaluation maps evX Y : X ⊗ [X, Y ] → Y ; coevX Y : Y → [X, X ⊗ Y ], then we can write f = [X, f ] ◦ coevX Y ; g = evX Z ◦ (X ⊗ f ). Given any morphism f : X → X ′, we can consider the composition X ⊗ [X ′, Y ] f ⊗id / X ′ ⊗ [X ′, Y ] evX′ Y / Y . We denote the corresponding morphism [X ′, Y ] → [X, Y ] by [f, Y ]. It is well-known that in this way, [−, Y ] : C → C becomes a contravariant functor for any object Y in C. Suppose that C is moreover right closed, and where the right internal hom denoted by {−, −}. Then we find for any three objects X, Y, Z in C that HomC(Y, [X, Z]) ∼= HomC(X ⊗ Y, Z) ∼= HomC(X, {Y, Z}) Hence HomCop([X, Z], Y ) ∼= HomC(X, {Y, Z}) and the (contravariant) functor [−, Z] : C → Cop has a right adjoint {−, Z}, and there- fore [−, Z] : C → C sends epimorphisms to monomorphisms. Lemma 3.3. Let C be a bi-closed monoidal category, f : A → B and epimorphism and g : C → D a regular epimorphism. Then the pushout of the pair (f ⊗ C, A ⊗ g) is given / / GEOMETRICALLY PARTIAL ACTIONS 35 by (B ⊗ D, B ⊗ g, f ⊗ D). B ⊗ C A ⊗ D A ⊗ C f ⊗C yssssssssss %❑❑❑❑❑❑❑❑❑❑ B⊗g A⊗g %❑❑❑❑❑❑❑❑❑❑ yssssssssss f ⊗D ⑧⑧ ❄❄ B ⊗ D h k u T Proof. Suppose that g is the coequalizer of the pair r, s : R → C. Consider any object T and maps h : B ⊗ C → T , k : A ⊗ D → T such that ℓ = h ◦ (f ⊗ C) = k ◦ (A ⊗ g) : A ⊗ C → T . Using the left closure on C, we find that ℓ corresponds uniquely to a morphism ℓ ∈ Hom(C, [A, T ]) and one checks that ℓ = [A, k] ◦ coevA D ◦ g = [f, T ] ◦ [B, h] ◦ coevB C . R r s / C g :✉✉✉✉✉✉✉✉✉✉✉✉ $■■■■■■■■■■ coevB C [A, A ⊗ D] coevA D D u [A,k] &▲▲▲▲▲▲▲▲▲▲ 8rrrrrrrrrr [f,T ] [A, T ] [B, B ⊗ C] / [B, T ] [B,h] Since g coequalizes the pair (r, s), it follows from the first equality that ℓ also coequalizes the pair (r, s). Furthermore, since f is an epimorphism, [f, T ] is a monomorphism and we find that [B, h] ◦ coevB C ◦ r = [B, h] ◦ coevB C ◦ s. Therefore, the universal property of the coequalizer g leads to a unique morphism u : D → [B, T ] such that u ◦ g = [B, h] ◦ coevB C : C → [B, T ]. Moreover, since g is an epimorphism, u also satisfies [f, T ] ◦ u = [A, k] ◦ coevA D Consequently the induced morphism u = u : B ⊗ D → T satisfies u ◦ (B ⊗ g) = h, u ◦ (f ⊗ D) = k and is unique in this sense, which proves the universal property of the pushout (B ⊗ D, B ⊗ g, f ⊗ D). (cid:3) Let C be a braided monoidal category with pushouts and consider be a bialgebra H in C. Let (X, X • H, πX, ρX ) and (Y, Y • H, πY , ρY ) be two partial comodule data over H. Then we can construct a new partial comodule datum (X ⊗Y, (X ⊗Y )•H, πX⊗Y , ρX⊗Y ) in the following way. Consider the map µX,Y = (X ⊗ Y ⊗ µH) ◦ (X ⊗ σH,Y ⊗ H), where y % % ! ! y } } / / & / / / : : : $ / 8 8 8 36 JIAWEI HU AND JOOST VERCRUYSSE σ denotes the braiding of the category. Then define (X ⊗ Y ) • H and πX⊗Y by the following pushout. (X • H) ⊗ (Y • H) X ⊗ Y ⊗ H X ⊗ H ⊗ Y ⊗ H πX ⊗πY t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ )❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ µ µX,Y )❘❘❘❘❘❘❘❘❘❘❘❘❘❘ v❧❧❧❧❧❧❧❧❧❧❧❧❧❧ πX⊗Y ⑧⑧❄❄ (X ⊗ Y ) • H By taking ρX⊗Y = µ ◦ (ρX ⊗ ρY ), we obtain the desired partial comodule datum. This construction lead to the following result. Proposition 3.4. Let C be a braided closed monoidal category where all epimorphisms are regular and let H be a bialgebra in C. Then by the above defined tensor product, the category of partial comodule data over H is a monoidal category with an op-lax unit, such that the following is a diagram of strict monoidal functors. ModH "❊❊❊❊❊❊❊❊❊ PCDH ②②②②②②②②②② C Proof. Let us first verify the associativity of the defined tensor product for PCD. Con- sider 3 partial comodule data X, Y , Z and consider the following diagram. (X • H) ⊗ (Y • H) ⊗ Z ⊗ H X ⊗ Y ⊗ H ⊗ Z ⊗ H (X • H) ⊗ (Y • H) ⊗ (Z • H) (X ⊗ Y ) • H ⊗ Z ⊗ H X ⊗ Y ⊗ Z ⊗ H X•H⊗Y •H⊗πZ v♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗ µ⊗(Z•H) ⑧⑧ ❄❄ ((X⊗Y )•H)⊗πZ ((X ⊗ Y ) • H) ⊗ (Z • H) X ⊗ H ⊗ Y ⊗ H ⊗ Z ⊗ H πX ⊗πY ⊗Z⊗H µ⊗Z⊗H v♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗ ⑧⑧❄❄ ⑧⑧❄❄ µX,Y ⊗Z⊗H (◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠♠ πX⊗Y ⊗Z⊗H µ(X⊗Y ),Z (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ v♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠ ((X ⊗ Y ) ⊗ Z) • H The upper square is a pushout by definition of the tensor product and the fact that the functor − ⊗ Z ⊗ H preserves pushouts since it has a right adjoint. The down square is a pushout by definition of the tensor product. The left square is a pushout by the Lemma 3.3. hence, by combining these pushouts we find that ((X ⊗ Y ) ⊗ Z) • H is the pushout of (X • H ⊗ Y • H ⊗ πZ) ◦ (πX ⊗ πY ⊗ Z) ≃ πX ⊗ πY ⊗ πZ along µ(X⊗Y ),Z ◦ (µX,Y ⊗ Z ⊗ H). In the same way, (X ⊗ (Y ⊗ Z)) • H is shown to be the pushout of πX ⊗ πY ⊗ πZ and µX,Y ⊗Z ◦ (X ⊗ H ⊗ µY,Z). One can easily verify that by the properties of the braiding in C and associativity of the multiplication of H, the ) t t t ) v v v / / " ( v v v ( v v v v v v ( ( v v ( GEOMETRICALLY PARTIAL ACTIONS 37 maps µ(X⊗Y ),Z ◦ (µX,Y ⊗ Z ⊗ H) and µX,Y ⊗Z ◦ (X ⊗ H ⊗ µY,Z) are the same up-to- isomorphism. Hence we find that ((X ⊗ Y ) ⊗ Z) • H ∼= (X ⊗ (Y ⊗ Z)) • H, which induces the associativity constraint for the monoidal product in PCD. Next, let us verify that the partial comodule datum k = (k, H, idH, η) is an oplax unit for this monoidal product. Consider any partial comodule datum X = (X, X • H, πX, ρX ) and construct the tensor product X ⊗ k. We know that the underlying C object is just X ⊗ k ∼= X via the isomorphism rX : X → X ⊗ k. Futhermore, the object (X ⊗ k) • H is constructed by the following pushout. (X • H) ⊗ H X ⊗ H X ⊗ H ⊗ H πX ⊗H v♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗ µ X⊗µ '◆◆◆◆◆◆◆◆◆◆◆◆ x♣♣♣♣♣♣♣♣♣♣♣♣ πX⊗k ⑧⑧❄❄ (X ⊗ k) • H Then consider the map rX • H := µ ◦ (X • H) ⊗ η : X • H → (X ⊗ k) • H. One easily verifies that rX • H ◦ πX = πX⊗k and therefore (rX, rX • H) : X → X ⊗ k is a morphism of partial comodule data. (cid:3) Let us remark that the category of partial comodule data can not have a (strong) monoidal unit. Indeed, since the forgetful functor is strict monoidal, the underlying C- object of the monoidal unit needs to be the monoidal unit k of C. Hence the monoidal unit should be of the form (k, K, π, ρ), where K is a quotient of H. However, the pushout (X • H) ⊗ K X ⊗ H X ⊗ H ⊗ H πX ⊗π v♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗ µ X⊗µ '◆◆◆◆◆◆◆◆◆◆◆◆ x♣♣♣♣♣♣♣♣♣♣♣♣ πX⊗k ⑧⑧❄❄ (X ⊗ k) • H can be computed explicitly in the case C = Vect via Lemma 2.4 and in this case we see that (X ⊗ k) • H is the quotient of X ⊗ H with respect to (X ⊗ µ)(ker πX ⊗ H + X ⊗ H ⊗ ker π). However, this last set is strictly larger then ker πX , so we can never get that (X ⊗ k) • H ∼= X • H. Nevertheless, the oplax monoidal unit from Proposition 3.4 becomes a strong unit for a suitable subcategory that we will define now. Definition 3.5. We call a partial comodule datum X over a bialgebra H left equivariant if the morphism πX : X ⊗ H → X • H is left H-linear by the free H-action on X ⊗ H. More explicitly, in case C = Vectk, this means that if x ⊗ h ∈ ker πX , then also x ⊗ h′h ∈ ker πX for all h′ ∈ H. In the same way, one defines right equivariant partial comodule data. When X is both left and right equivariant, we simply say it is equivariant. Example 3.6. Consider the Example 2.9 of a Caenepeel-Janssen partial action. Since A • H = (A ⊗ H)e it is a left A ⊗ H-module, hence in particular left equivariant. Corollary 3.7. The oplax monoidal unit of PCD is a left monoidal unit for all left equivariant partial comodule data, and a right monoidal unit for all right equivariant partial comodule data. ' v v v ( x x x ' v v v ( x x x 38 JIAWEI HU AND JOOST VERCRUYSSE Consequently, the category of equivariant partial comodule data over a k-bialgebra H is monoidal. Proof. Consider the pushout (X • H) ⊗ H X ⊗ H X ⊗ H ⊗ H πX ⊗H v♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ p1 X⊗µ '◆◆◆◆◆◆◆◆◆◆◆◆ w♣♣♣♣♣♣♣♣♣♣♣♣♣ p2 ⑧⑧❄❄ P By Lemma 2.4, we know that P can be computed as the quotient of X ⊗ H by the subspace (X ⊗ µ)(ker πX). However, under the stated assumptions, ker πX is a right H-submodule of X ⊗ H and hence (X ⊗ µ)(ker πX ) = ker πX . Therefore, P is exactly X • H and p2 is nothing else than πX . From this observation and the definition of the tensor product, it now follows that X ⊗ k ∼= X in PCD. Similarly, using that ker πX is a left H-submodule of X ⊗ H, we find that k is a left unit for the monoidal structure on PCD. (cid:3) The result from Proposition 3.4 leads to the natural question whether the full subcate- gories of the category of partial comodule data, consisting of quasi, lax and geometric partial comodules inherit a monoidal structure. To answer this question, let us compute the pushout (X ⊗ Y ) • (H • H), which is given by the following composition of pushouts X ⊗ H ⊗ Y ⊗ H µX,Y / X ⊗ Y ⊗ H X⊗Y ⊗∆ X ⊗ Y ⊗ H ⊗ H πX⊗Y ⊗H ((X ⊗ Y ) • H) ⊗ H πX ⊗πY πX⊗Y πX⊗Y,∆ π′ X⊗Y,∆ (X • H) ⊗ (Y • H) µX,Y / (X ⊗ Y ) • H / (X ⊗ Y ) • (H ⊗ H) (X⊗Y )•∆ / (X ⊗ Y ) • (H • H) π′ X⊗Y Furthermore, checks that the composition of the upper morphisms can be rewritten as (πX⊗Y ⊗ H) ◦ (X ⊗ Y ⊗ ∆) ◦ µX,Y = (µX,Y ⊗ H) ◦ µX•H,Y •H ◦ (πX ⊗ H ⊗ πY ⊗ H) ◦ (X ⊗ ∆ ⊗ Y ⊗ ∆) In the same way, the pushout ((X ⊗ Y ) • H) • H is given by the following composition of pushouts X ⊗ H ⊗ Y ⊗ H µX,Y / X ⊗ Y ⊗ H ρX ⊗ρY ⊗H (X • H) ⊗ (Y • H) ⊗ H / ((X ⊗ Y ) • H) ⊗ H µX,Y ⊗H πX ⊗πY πX⊗Y (X • H) ⊗ (Y • H) / (X ⊗ Y ) • H µX,Y ρX⊗Y •H π(X⊗Y )•H / ((X ⊗ Y ) • H) • H and again we can rewrite the composition of the upper morphisms (µX,Y ⊗ H) ◦ (ρX ⊗ ρY ⊗ H) ◦ µX,Y = (µX,Y ⊗ H) ◦ µX•H,Y •H ◦ (ρX ⊗ H ⊗ ρY ⊗ H) ' v v v ( w w w / / / / / / / / / / / / / / GEOMETRICALLY PARTIAL ACTIONS 39 Hence, to study the relation between (X ⊗ Y ) • (H • H) and ((X ⊗ Y ) • H) • H, we have to compare the following pushouts X ⊗ H ⊗ Y ⊗ H X⊗∆⊗Y ⊗∆ / X ⊗ H ⊗ H ⊗ Y ⊗ H ⊗ H πX ⊗H⊗πY ⊗H / (X • H) ⊗ H ⊗ (Y • H) ⊗ H πX ⊗πY (X • H) ⊗ (Y • H) and p1 p2 / P X ⊗ H ⊗ Y ⊗ H ρX ⊗H⊗ρY ⊗H / (X • H) ⊗ H ⊗ (Y • H) ⊗ H πX ⊗πY (X • H) ⊗ (Y • H) q1 q2 / Q Using Lemma 2.4, we find that P and Q are isomorphic to quotients of (X • H) ⊗ H ⊗ (Y • H) ⊗ H by the respective subspaces (πX ⊗ H ⊗ πY ⊗ H) ◦ (X ⊗ ∆ ⊗ Y ⊗ ∆)(ker (πX ⊗ πY )) = (πX ⊗ H ⊗ πY ⊗ H) ◦ (X ⊗ ∆ ⊗ Y ⊗ ∆)(ker πX ⊗ Y ⊗ H + X ⊗ H ⊗ ker πY ) and (ρX ⊗ ρY )(ker (πX ⊗ πY )) = (ρX ⊗ ρY )(ker πX ⊗ Y ⊗ H + X ⊗ H ⊗ ker πY ) Since in general (πX ⊗ H) ◦ (X ⊗ ∆)(X ⊗ H) 6∼= (ρX ⊗ H)(X ⊗ H), we find that P and Q are non-isomorphic, and therefore also (X ⊗ Y ) • (H • H) and ((X ⊗ Y ) • H) • H are non-isomorphic (even if X and Y are geometric partial comodules). Hence, we can conclude that when X and Y are geometric (respectively lax) partial comodules, then X ⊗ Y is in general no longer a geometric (respectively lax) partial comodule. However, when X and Y are both quasi comodules, then θ1 ◦ (ρX • H) and θ2 ◦ (X • ∆) do have identical images when restricted to the image of ρX , we find that X ⊗ Y is still a quasi partial comodule. We can then conclude on the following. Theorem 3.8. Let C be a category satisfying the conditions of Proposition 3.4. The category of quasi partial comodules over a bialgebra in C is monoidal with an oplax monoidal unit, and the forgetful functor to C is strict monoidal. The category of equivariant quasi partial comodules over a bialgebra is a monoidal cat- egory. Although the above introduced tensor product is not well-defined on the category geo- metric partial comodules, in case work with a bialgebra over a field, we can combine Proposition 3.4 with Theorem 3.1 and obtain immediately the following result. Theorem 3.9. The category of geometric partial comodules over a bialgebra H over a field k is an op-lax monoidal category and the forgetful functor U : gPModH → Vectk is monoidal. Remark 3.10. Let us describe the oplax tensor product of the category gPModH a bit more explicitly. Consider two geometric partial comodules M and N, and let M ⊗ N be the tensor product partial comodule datum (which we know is a quasi partial comodule). Then M • N is the geometric partial comodule that is uniquely defined by the following universal property. There exists a morphism p : M ⊗ N → M • N and for every other / / / / / 40 JIAWEI HU AND JOOST VERCRUYSSE geometric partial comodule T with a morphism M ⊗ N → T , there exists a unique morphism u : M • N → T such that t = u ◦ p. Since the zero module is a geometric partial comodule that is a minimal solution for the above problem, M • N will always exist. Given two p : M ⊗ N → P and q : M ⊗ N → Q, Let R be the pullback of the pushout of p and q (which exist since we proved that geometric partial comodules are bi-complete). Then there is a unique morphism M ⊗ N → R compatible with both p and q. In this way, we can construct the "biggest" quotient M • N of M ⊗ N that is still a geometric partial comodule. Remark that if one of the geometric partial comodules X and Y is global, then X • Y = X ⊗ Y . 4. Partial comodule algebras and Hopf-Galois theory In the partial case, there turn out to be two kinds of 'comodule algebras': those which arise as partial comodules in the category of algebras, and those that arise as algebras in the (oplax monoidal) category of partial comodules. While these notions coincide in the global case, for partial coactions they differ, as a consequence of the fact that pushouts in the category of algebras are different from pushouts in the category of vector spaces. Throughout this section, we assume that k is a field. 4.1. Algebras in the category of partial comodules. Let C be an oplax monoidal category and denote ⊗0(∅) = I and ⊗n(X1, . . . , Xn) = (X1 ⊗ . . . ⊗ Xn). An algebra C, is an object A endowed with morphisms m : (A ⊗ A) → A and u : I → A satisfying the following conditions ((A) ⊗ (A ⊗ A)) (A ⊗ A ⊗ A) γ 5❦❦❦❦❦❦❦❦❦❦❦❦❦❦ )❙❙❙❙❙❙❙❙❙❙❙❙❙❙ γ ((A ⊗ A) ⊗ (A)) ι⊗m / (A ⊗ A) m⊗ι / (A ⊗ A) m )❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ 5❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦ m A (A) γ ;✈✈✈✈✈✈✈✈✈ #❍❍❍❍❍❍❍❍❍ γ m #❋❋❋❋❋❋❋❋❋❋ ;①①①①①①①①①① m / A (A⊗u) (A ⊗ I) / (A ⊗ A) ιA (I ⊗ A) / (A ⊗ A) (u⊗A) One can verify that just as in the case of algebras in usual strong monoidal categories, all higher associativity conditions follow form the one given here. Then we obtain the following natural definitions. Definition 4.1. Let H be a k-bialgebra. A quasi (resp. geometric) partial H-comodule algebra is a an algebra in the oplax monoidal category of quasi (resp. geometric) partial H-comodules. / ) 5 ) / 5 / # # ; / / ; GEOMETRICALLY PARTIAL ACTIONS 41 Since the forgetful functors gPModH → qPModH → Vectk are monoidal, each geometric partial comodule algebra is also a quasi partial comodule algebra and a quasi or geomet- ric partial comodule algebra is also a k-algebra. More precisely, we have the following result. Proposition 4.2. Let C be a category satisfying the conditions of Proposition 3.4 and H a Hopf algebra in C. If (A, A • H, πA, ρA) is an algebra in the monoidal category with oplax unit qPModH, then A and A • H are algebras in C and the morphisms πA and ρA are algebra morphisms. Proof. We already remarked that A is an algebra in C, since the forgetful functor qPModH → C is monoidal. To see that A • H is an algebra consider the following, which expresses that the multiplication µ : A ⊗ A → A is a morphism of partial comod- ules, and the construction of (A ⊗ A) • H as pushout. ρA⊗ρA t❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤ µA,A A ⊗ A ρA⊗A µA µA•H (A • H) ⊗ (A • H) / (A ⊗ A) • H πA⊗πA πA⊗A A ⊗ H ⊗ A ⊗ H µA,A A ⊗ A ⊗ H µA⊗H A ρA / A • H πA A ⊗ H One can then easily verify that the morphism (µA • H) ◦ µA,A defines an associative multiplication on A • H, and by construction πA and ρA are then multiplicative. In a similar way, the unit morphism u : k → A is a morphism of partial comodules if the following diagram commutes k ηH H H u / A ρA u•H / A • H πA u⊗H / A ⊗ H which means in Sweedler notation that (8) ρA(1A) = 1A • 1H Then ρA ◦ u : k → A • H is a unit for the algebra A • H and the morphisms πA and ρA are unital. (cid:3) Again, since the functor gPModH → qPModH is monoidal, it follows that for a geometric partial comodule algebra A, the vector space A • H is naturally a k-algebra and piA is a k-algebra morhpism. This implies that ker πA is a two-sided ideal in A ⊗ H. Remark 4.3. In contrast to what might think naively, the C-objects (A • H) • H, A • (H • H) or ΘA do not posses a natural algebra structure in general. The main reason for this, is that these objects are defined as pushouts in C without any interaction with the multiplication µA. This is the main motivation to introduce a second type of comodule algebras in the next section. / / t / / / / O O / / O O O O / / / O O 42 JIAWEI HU AND JOOST VERCRUYSSE Examples 4.4. Let A be a (global) H-comodule algebra, and p : A → B a surjective algebra morphism. Then just as in Example 2.5, we can endow B with the structure of a (geometric) partial comodule. Taking however the pushout (2) in the category of algebras, one endowes B with the structure of a partial comodule algebra. Let us make this construction a bit more explicit in the following example. Consider the example of geometric partial k[x, y]-comodule from Example 2.7, whose un- derlying object is B = k[x, y]/(xy). Then B has a natural algebra structure, however B• H = k[x, y, x′, y′]/ρ((xy)) is not an algebra since ρ((xy)) is not an ideal. Hence, (B, B • H, πB, ρB) is not a partial comodule algebra. However, consider k[x, y, x′, y′]/(ρ(xy)), where (ρ(xy)) is the ideal generated by ρ((xy)), then there is a surjective algebra morphism π : B • H → k[x, y, x′, y′]/(ρ(xy)). The partial comodule datum (B′, B′ • H, πB′, ρB′), where B′ = B, B′ •H = k[x, y, x′, y′]/(ρ(xy)), πB′ = π ◦πB and ρB′ = π ◦ρB is a partial k[x, y]-comodule algebra. Since the partial comodule B′ is geometric (being a quotient of a global one), B′ is also a geometric comodule algebra. In the same way, one can turn the partial comodule from Example 2.7 and Example 2.8 into a geometric partial comodule algebra and the partial comodule from Example 2.9 into quasi partial comodule algebra. For this last example, and using the notation of Example 2.9, it is easy to see that if e ∈ A ⊗ H is central, then already has that A • H = (A ⊗ H)e is an algebra, so the partial comodule datum of Example 2.9 is already a partial comodule algebra. 4.2. Partial comodules in the category of algebras. Recall that the category of algebras Alg(C) in a braided monoidal category C, is again a monoidal category. Furthermore, a coalgebra H in Alg(C) is exactly a bialgebra in C and a comodule over H in Alg(C) is exactly an H-comodule algebra in C. Following this point of view, we introduce the following definitions. Definition 4.5. Let H be a bialgebra in the braided monoidal category C, and consider H as a coalgebra in the category Alg(C). A quasi (resp. lax, geometric) partial algebra- comodule over H is a quasi (resp. lax, geometric) partial H-comodule (A, A • H, πA, ρA) in Alg(C). Remark 4.6. In the global case, algebra-comodules and comodule-algebras are identical structures. In the partial setting this however is no longer the case. Firstly, given a partial comodule datum (A, A • H, πA, ρA) in Alg(C), then applying the forgetful functor U : Alg(C) → C yields a partial H-comodule datum UA = (A, A • H, πA, ρA) in C, provided that U(πA) is an epimorphism in C. This last condition is not necessarily the case if we take C = Modk where k is an arbitrary commutative ring, but it holds if k is a field. However, even in case of C = Vectk, the forgetful functor U : Algk → Vectk does not preserve pushouts. In other words, the canonical morphism ΘU A → U(ΘA) is not an isomorphism in general. If A is a quasi partial H-comodule algebra, then coassociativity holds in UΘA, but not necessarily in ΘU A, and UA is not necessarily a quasi partial H-comodule. The next result tells however that conversely, partial comodule-algebras are still algebra- comodules. Proposition 4.7. If (A, A • H, πA, ρA) is a quasi partial H-comodule algebra, then (A, A • H, πA, ρA) is also a quasi partial algebra-comodule over H. Proof. It follows from Proposition 4.2 that (A, A•H, πA, ρA) is a partial comodule datum in the category Alg(C). If A is a quasi partial comodule, then the coassociativity holds GEOMETRICALLY PARTIAL ACTIONS 43 in the sense that (9) θ1 ◦ (ρA • H) • ρA = θ2 ◦ X • ∆ ◦ ρA where (ΘA, θ1, θ2) is the coassociativity pushouts in C. On the other hand, we can also consider the coassociativity pushout (Θ′ 2) in Alg(C). Since the forgetful functor Alg(C) → C does not preserve pushouts, ΘA and Θ′ A can be non-isomorphic objects in C, but by the universal property of the pushouts (A • H) • H, A • (H • H) and ΘA, we will obtain a morphism π : ΘA → Θ′ A. By composing both sides of (9) with π, we find that the coassociativity also holds in Alg(C), and hence A is a quasi partial algebra-comodule. (cid:3) A, θ′ 1, θ′ The difference between the pushouts in Alg(C) and C can be understood very well in the situation where C = Vectk. Indeed, consider k-algebra morphisms a ⑦⑦⑦⑦⑦⑦⑦⑦ A R b ❅❅❅❅❅❅❅❅ B where b is surjective. Then we know from Lemma 2.4 that the pushout of a and b in Vect is given by the quotient A/a(ker b). In general (or more precisely, when a is not surjective) a(ker b) is not an ideal in A, and hence A/a(ker b) is not an algebra. However, if we denote by I the ideal generated by a(ker b), then one can easily see that A/I is the pushout of (a, b) in Algk. As a consequence, we find the following. Corollary 4.8. If (A, A • H, πA, ρA) is a geometric partial H-comodule k-algebra, then (A, A • H, πA, ρA) is also a geometric partial algebra-comodule over H. Proof. By Proposition 4.7 we know already that A is a quasi partial algebra-comodule. On the other hand, since A is geometric as comodule algebra, we find that the pushouts (A • H) • H and A • (H • H) are isomorphic in Vectk. Because of the explicit description of these pushouts recalled above, this means that the following subspaces of (A• H)⊗H are isomorphic (even identical): (ρA ⊗ H)(ker πA) = (πA ⊗ H) ◦ (A ⊗ ∆)(ker πA) Hence the ideals generated by these subspaces will also be the same, and therefore the corresponding coassociativity pushouts in Algk will be isomorphic, which means exactly that A is geometric as a partial algebra-comodule. (cid:3) As the following examples illustrate, the converse of the previous corollary does not hold. Examples 4.9. All examples from Example 4.4 will give rise to examples of algebra- comodules. Since the examples obtained from Example 2.7 and Example 2.8 are geo- metric as comodule-algebra, they are by the previous proposition also geometric as algebra-comodules. We remarked before that the example from Example 2.9 is not geo- metric as comodule-algebra, however we will show now that is does become geometric as algebra-comodule. Let A be a partial coaction of a Hopf algebra H in the sense of [8]. Consider as in Example 2.9 the partial comodule datum (A, A•H, π, ρ), where A•H = {a1[0] ⊗h1[1]} = (A ⊗ H)e, which is a direct summand of A ⊗ H and the left A ⊗ H-module generated 44 JIAWEI HU AND JOOST VERCRUYSSE by the idempotent e = ρ(1) = 1[0] ⊗ 1[1] and which can be seen as the quotient of A ⊗ H by the left ideal (A ⊗ H)e′ where e′ = 1 ⊗ 1H − e (we denote 1 = 1A the unit of A). As explained in Example 4.4, in order to obtain a partial comodule algebra one has to consider an alternative partial comodule datum, where A •′ H is the quotient of A ⊗ H by the two-sided ideal (A ⊗ H)e′(A ⊗ H). The ideal in A ⊗ H ⊗ H generated by (ρ ⊗ H)((A ⊗ H)e′(A ⊗ H)) is then nothing else than the ideal generated by (ρ ⊗ H)(e′). Similarly, the ideal in A ⊗ H ⊗ H generated by the image of (A ⊗ H)e′(A ⊗ H) under (π ⊗ H) ◦ A ⊗ ∆ is the ideal generated by (π ⊗ H) ◦ (A ⊗ ∆)(e′). Using axiom (CJ2), we find (ρ ⊗ H)(e′) = (ρ ⊗ H)(1 ⊗ 1H − e) = 1[0] ⊗ 1[1] ⊗ 1H − 1[0][0] ⊗ 1[0][1] ⊗ 1[1] = 1[0] ⊗ 1[1] ⊗ 1H − 1[0]1[0′] ⊗ 1[1](1)1[1′] ⊗ 1[1](2) = (π ⊗ H) ◦ A ⊗ ∆(e′) Hence it follows that both elements generate the same ideals, which implies that A is geometric as a partial algebra-comodule. 4.3. Partial Hopf modules. Let C be an oplax monoidal category and denote as in Section 4.1 ⊗0(∅) = I and ⊗n(X1, . . . , Xn) = (X1 ⊗ . . . ⊗ Xn). Let (A, m, u) be an algebra in C. Then a (right) A-module in C is an object M endowed with a morphism µM : (M ⊗ A) → M satisfying the following conditions (M ⊗ A ⊗ A) γ 5❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦ )❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ γ ((M) ⊗ (A ⊗ A)) ((M ⊗ A) ⊗ (A)) ι⊗m / (M ⊗ A) µM ⊗ι / (A ⊗ A) µM )❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ 5❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦ µM M (M ⊗ I) γ :✉✉✉✉✉✉✉✉✉ (M) ιA (A⊗u) / (M ⊗ A) µM $❍❍❍❍❍❍❍❍❍❍ / M Then we obtain the following natural definitions. Definition 4.10. Let H be a Hopf k-algebra and (A, A • H, πA, ρA) a quasi (resp. geometric) partial H-comodule algebra, i.e. an algebra in the lax monoidal category of quasi (resp. geometric) partial H-comodules. A quasi (resp. geometric) partial (A, H)- relative Hopf module is a right A-module in the lax monoidal category of quasi (resp. geometric) partial H-comodules. When A and H are clear from the context, we will just call this a partial Hopf module. We will denote by PModH A the category whose objects are quasi partial (A, H)-relative Hopf modules and whose morphisms are morphism of partial H-modules that are at the same time A-linear. As it is the case for partial comodule-algebras, since the forgetful functors gPModH → qPModH → Vectk are monoidal, each geometric partial relative Hopf module is also a quasi partial relative Hopf module and a quasi or geometric partial relative Hopf module is also a module for the k-algebra A. / ) 5 ) / 5 / $ : / GEOMETRICALLY PARTIAL ACTIONS 45 Let us make the previous definition a bit more explicit. Lemma 4.11. Let (A, A • H, πA, ρA) be a quasi partial H-comodule algebra. A quasi partial (A, H)-relative Hopf module is a quasi partial H-comodule (M, M • H, πM , ρM ) endowed with an A-module structure µM : M ⊗ A → M such that the following compat- ibility conditions hold: [PRHM1] ker (πM ⊗ πA) ⊂ ker (πM ◦ µM ⊗H); [PRHM2] (ma)[0] • (ma)[1] = m[0]a[0] • m[1]a[1] for all m ∈ M and a ∈ A. where µM ⊗H : M ⊗ H ⊗ A ⊗ H → M ⊗ H, µM ⊗H((m ⊗ h)(a ⊗ k)) = ma ⊗ hk. is the induced A ⊗ H-module on M ⊗ H. Under these conditions, M • H is a right A • H-module. Proof. Similarly as in the proof of Proposition 4.2, consider the following diagram which expresses that the A-action µM : M ⊗ A → M is a morphism of partial comodules, and the construction of (M ⊗ A) • H as pushout. ρM ⊗ρA t❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤ µM,A (M • H) ⊗ (A • H) / (M ⊗ A) • H M ⊗ A ρM ⊗A µM µM •H πM ⊗πA πM ⊗A M ⊗ H ⊗ A ⊗ H µM,A M ⊗ A ⊗ H µM ⊗H By construction, we know that ker πM ⊗A = µM,A(ker (πM ⊗ πA)) M ρM / M • H πM M ⊗ H Then by Lemma 2.13, in order for the linear map µM to be a morphism of partial H- comodules it is needed that (µM ⊗ H)(ker πM ⊗A) ⊂ ker πM . Since µM ⊗H = (µM ⊗ H) ◦ µM,A, this means furthermore that πM ◦ (µM ⊗ H)(ker πM ⊗A) = πM ◦ µM ⊗H(ker (πM ⊗ πA)) = 0 or equivalently, ker (πM ⊗ πA) ⊂ ker (πM ◦ µM ⊗H), i.e. [PRHM1] holds. This condition implies that the map µM •H = (µM • H) ◦ µM,A : (M • H) ⊗ (A • H) → M • H, µM •H((m • h)(a • k)) = ma • hk is well-defined and defines an action of A • H on M • H. Moreover, µM will be a morphism of partial H-modules if and only if ρM ◦ µM = µM •H ◦ (ρM ⊗ ρA) which gives exactly condition [PRHM2]. (cid:3) Example 4.12. Clearly any algebra in a lax monoidal category is a module over itself, hence any quasi partial H-comodule algebra (A, A • H, πA, ρA) is also a quasi partial (A, H)-relative Hopf module. The following observation will be useful later. / / t / / / / O O / / O O O O 46 JIAWEI HU AND JOOST VERCRUYSSE Lemma 4.13. If M is a quasi partial (A, H)-relative Hopf module, then πM (ma ⊗ h) = 0 for all (i) m ⊗ h ∈ ker πM and a ∈ A; (ii) m ∈ M and a ⊗ h ∈ ker πA. Consequently, the isomorphism M ⊗A (A ⊗ H) ∼= M ⊗ H induces an surjective map M ⊗A (A • H) → M • H. Proof. Since ker (πM ⊗ πA) = ker πM ⊗ A ⊗ H + M ⊗ H ⊗ ker πA, we know that for all m ⊗ h ∈ ker πM and a ∈ A, m ⊗ h ⊗ a ⊗ 1H ∈ ker (πM ⊗ πA). Hence by axiom (PRHM1), we find that πM (ma ⊗ h) = 0. Similarly, for m ∈ M and a ⊗ h ∈ ker πA, we have m ⊗ 1H ⊗ a ⊗ h ∈ ker (πM ⊗ πA) and we can follow the same reasoning. For the last statement, it follows from part (ii) that M ⊗A (ker πA) is included in ker πM via the canonical isomorphism M ⊗A (A ⊗ H) ∼= M ⊗ H. Consequently, we obtain the stated surjection M ⊗A (A • H) → M • H. (cid:3) 4.4. Hopf-Galois theory. Definition 4.14. Let H be a Hopf k-algebra, (A, A • H, πA, ρA) a quasi partial H- comodule algebra and (M, M • H, πM , ρM , µM ) a quasi partial (A, H)-relative Hopf module. The H-coinvariants of M are defined as the following equalizer in Vectk (10) M coH ρM / M πM ◦(M ⊗ηH ) / M • H i.e. M coH = {m ρM (m) = m • 1H}. Proposition 4.15. Let H be a Hopf k-algebra, (A, A • H, πA, ρA) a quasi partial H- comodule algebra. (i) The coinvariants AcoH of A form a subalgebra of A; (ii) The coinvariants M coH of a quasi partial (A, H)-relative Hopf module M form a module over AcoH; (iii) This induces a functor (−)coH : PModH A → ModAcoH . Proof. (i). Since ρA and πA ◦ (A ⊗ ηH) are both algebra morphisms and the forgetful functor U : Algk → Vectk creates limits, AcoH is a subalgebra of A. Alternatively, this follows by a direct computation similar to the next part. (ii). Consider the restriction of the multiplication map µM : M coH ⊗AcoH → M, µM (m⊗ a) = ma. Then ρ(ma) = m[0]a[0] • m[1]a[1] = (m[0] • m[1])(a[0] • a[1]) = (m • 1H)(a • 1H) = (ma • 1H) hence ρM (ma) ∈ M coH. (iii). This part is easily verified. (cid:3) Let us remark that the coinvariant functor is representable. Lemma 4.16. For any quasi partial (A, H)-relative Hopf module, we have that M coH ∼= HomH A (A, M), the set of partial H-comodule morphisms from A into M that are right A-linear. / / / / GEOMETRICALLY PARTIAL ACTIONS 47 Proof. Let f : A → M be a right A-linear morphism of partial H-comodules. Then the following diagram commutes A ρA A • H πA A ⊗ H f f •H f ⊗H / M ρM / M • H πM / M ⊗ H Since we know that ρA(1A) = πA(1A ⊗ 1H), it follows from the commutativity of this diagram that f (1A) ∈ M coH. Conversely, given any m ∈ M coH we clearly have that the map fm : A → M, fm(a) = ma is right A-linear. Moreover, thanks Lemma 4.13(ii), we also know that (fm ⊗ H)(ker πA) ⊂ ker πM . Hence the morphism fm • H : A • H → M • H, fm(a • h) = ma • h is well-defined. Finally, we find for any a ∈ A that ρM (fm(a)) = ρM (ma) = ρM (m)ρA(a) = (m • 1H)(a[0] • a[1]) = ma[0] • a[1] = fm • H(a[0] • a[1]) I.e. fm ∈ HomH between M coH and HomH A (A, M). The above constructions provide well-defined mutual inverses (cid:3) A (A, M). Proposition 4.17. For any AcoH-module N, the right A-module N ⊗AcoH A can be endowed with the structure of a quasi partial (A, H)-relative Hopf module by means of the following partial H-comodule structure πN ⊗AcoH A = N ⊗AcoH πA : ρN ⊗AcoH A = N ⊗AcoH ρA : Moreover, this construction yields a functor − ⊗AcoH A : ModAcoH → PModH left adjoint to the coinvariant functor (−)coH : PModH → ModAcoH . Proof. The construction of the functor − ⊗AcoH A : ModAcoH → PModH A is clear from the statement. To verify the adjunction property, we will define a counit ζ and a unit ν. For any quasi partial (A, H)-relative Hopf module M we define N ⊗AcoH A ⊗ H → N ⊗AcoH (A • H) =: (N ⊗AcoH A) • H N ⊗AcoH A → N ⊗AcoH (A • H) A that is a ζM : M coH ⊗AcoH A → M ζM (m ⊗AcoH a) = ma Clearly, ζM is a right A-linear map. Let us check that it is also a morphism of partial H-comodules. Firstly, we need to verify that πM ◦ (ζM ⊗ H)(ker πM coH ⊗AcoH A) = 0 (see Lemma 2.13). Since by construction ker πM coH ⊗AcoH A = M coH ⊗AcoH ker πA, this follows directly by Lemma 4.13(ii). Secondly, we should check that ρM ◦ ζM = (ζM • H) ◦ (M coH ⊗AcoH ρA), where we know that ζM • H is well-defined by the first part. Indeed, take any m ⊗ a ∈ M coH⊗AcoH A then ρM (ζM (m ⊗ a)) = (ma)[0] • (ma)[1] = (m[0] • m[1])(a[0] • a[1]) = (m • 1H)(a[0] • a[1]) = ma[0] • a[1] = ζM (m ⊗AcoH a[0]) • a[1] = (ζM • H)(m ⊗AcoH ρA(a)). On the other hand, for any AcoH-module N, we define νN : N → (N ⊗AcoH A)coH, νN (n) = n ⊗ 1A. / / O O / O O 48 JIAWEI HU AND JOOST VERCRUYSSE Since A is an algebra in the category of partial H-modules we have that ρA(1) ∈ AcoH (see (8)). It is now easily verified that ζ and ν are indeed the counit and unit for this adjunction. (cid:3) Remarks 4.18. If A is lax (respectively geometric) as partial H-comodule, then we find for any N ∈ ModAcoH that the partial Hopf module N ⊗AcoH A constructed in the previous proposition is also lax (resp. geometric). Let ι : B → AcoH be any ring morphism, then of course the adjunction from Proposi- tion 4.17 can be combined with the extension-restriction of scalars functors, to obtain a pair of adjoint functors − ⊗B A : ModB ⇄ PModH A : (−)coH . Definition 4.19. Let A be a quasi partial H-comodule algebra and B → AcoH a ring morphism. We call the morphism ι : B → A a partial Hopf-Galois extension if and only if the following canonical map is bijective can : A ⊗B A → A • H, can(a ⊗B a′) = aa′ [0] • a′ [1] denotes the product (a • 1H )(a′ [0] • a′ [1] Remark that here aa′ since m : A ⊗ A → A is a morphism of partial H-comodules. [0] • a′ [1]) which is well-defined The following examples show how partial Hopf-Galois extensions can be interpreted as "partial principle bundles". Example 4.20. Let A be a global H-comodule algebra, and suppose that A/AcoH is Galois, i.e. the canonical map A ⊗AcoH A → A ⊗ H is bijective. Consider a surjective algebra morphism p : A → B and endow B with the induced structure of a partial comodule algebra. Then we obtain a canonical algebra morphism AcoH → BcoH and in fact BcoH ∼= AcoH/(AcoH ∩ ker p). We find that the following diagram commutes A ⊗AcoH A p⊗p B ⊗BcoH B can A can B / A ⊗ H πB◦(p⊗H) / B • H If canA is an isomorphism, it is clear that canB is surjective. Moreover, consider any b ⊗ b′ ∈ ker canB. Since p ⊗ p is surjective, we can write b ⊗ b′ = p(a) ⊗ p(a′), such that can(a ⊗ a′) ∈ ker πB ◦ (p ⊗ H). From the construction of B • H as a pushout, we know that ker πB = (p ⊗ H) ◦ ρA(ker p). Hence, we find that can(a ⊗ a′) = u[0] ⊗ u[1] + v ⊗ h for some u ∈ ker p and v ⊗ h ∈ ker p ⊗ H. Applying can−1 to this equation, we obtain that a ⊗ a′ = 1 ⊗ u + can−1(v ⊗ h). Moreover, since can is left A-linear, can−1 is left A-linear as well and therefore can−1(v ⊗ h) = vcan−1(1 ⊗ h) ∈ ker p ⊗AcoH A, since v ∈ ker p which is an ideal in A as p is an algebra map. We can conlude that a ⊗ a′ ∈ A ⊗AcoH ker p + ker p ⊗AcoH A and therefore b ⊗ b′ = p(a) ⊗ p(a′) = 0. So canB is bijective as well, i.e. B is partially Hopf Galois. Example 4.21. It is known that if an algebraic group G acts strictly transitive on an algebraic space X (i.e. X is a principal homogeneous G-space), then the coordinate algebra A = O(G) is O(G)-Galois with trivial coinvariants. If we take any subvariety Y ⊂ X then we know that O(Y ) will be a partial O(G)-comodule algebra. Applying the previous example, we find that that O(Y ) will be partially Hopf-Galois. For example, the partial comodule algebras from Example 2.7, Example 2.8 (see Example 4.4) provide / / GEOMETRICALLY PARTIAL ACTIONS 49 examples of partial principle homogeneous G-spaces (where G is respectively k[x, y] and k hx, yi). More generally, X is a principal O(G)-bundle if and only if O(X) is an O(G)-Galois extension (with possible non-trivial coinvariants), see e.g. [16]. Again, any subvariety Y of the principle bundle X will give rise to a partial principle bundle. In the global case, we know that if A/AcoH is Hopf-Galois and A is faithfully flat as left AcoH-module, then the category of relative (A, H)-Hopf modules is equivalent to the category of AcoH-modules. This theorem has a nice interpretation in terms of corings, as this equivalence factors through the category of comodules over the canonical Sweedler A-coring A ⊗B A and the canonical map is an A-coring morphism. For more details we refer to e.g. [6], where it is also pointed out that the faithful flatness is in fact too strong. Since in the partial setting it follows from earlier observations in this paper that the category of partial comodules is not abelian, the category of partial relative (A, H)-Hopf modules cannot be expected to be equivalent with a module category. Nevertheless, let us show that under the same mild conditions as in the global case, we can characterize when the functor − ⊗AcoH A : ModAcoH → PModH A is fully faithful. The following is an adaptation of the approach from [6] (see also [7]). Recall that a morphism of left B-modules f : N → M is called pure if and only if for any right B-module P , the map P ⊗B f : P ⊗B N → P ⊗B M is injective. In particular, if ι : B → A be a ring morphism, then ι is said to be pure (as left B-module morphism) if for any right B-module P the map ιP : P → P ⊗B A, ιP (p) = p ⊗B 1A is injective. Lemma 4.22. Let ι : B → A be a ring morphism. Then the following statements are equivalent: (i) ι is pure as left B-module morphism (ii) For any right B-module N, the fork (11) ιN N / N ⊗B A ιN ⊗B A N ⊗B ιA / N ⊗B A ⊗B A is an equalizer in ModB. In particular, if A is faithfully flat as left B-module, then A is left pure. Proof. (i) ⇒ (ii) Denote by E the equalizer of (11), and define P = E/ιN (N). Take any e ∈ E, the we can write e = ni ⊗B ai and ni ⊗B 1A ⊗B ai = ni ⊗B ai ⊗B 1A. Apply π ⊗B A to this identity, then we have that ιP (π(e)) = π(e) ⊗B 1A = π(ni ⊗B ai) ⊗B 1A = π(ni ⊗B 1A) ⊗B ai = 0, since ni ⊗B 1A ∈ ιN (N). Since ιP is injective, it follows that π(e) = π(ni ⊗B ai) = 0 in P = E/ι(N), hence ni ⊗B ai ∈ ι(N). (ii) ⇒ (i). Since (11) is an equalizer, we have in particular that ιN is injective. (cid:3) Proposition 4.23. Let ι : B → A be a partial H-Galois extension, then the functor − ⊗B A : ModB → PModH Under these conditions, B ∼= AcoH. A is fully faithful if and only if ι is pure. / / / / 50 JIAWEI HU AND JOOST VERCRUYSSE Proof. Consider the following commutative diagram N ⊗Bι N ⊗B A N νN (N ⊗B A)coH / N ⊗B A N ⊗Bι⊗B A N ⊗BA⊗B ι ρN ⊗B A=N ⊗B ρA N ⊗B A•ηH / N ⊗B A ⊗B A N ⊗B can / N ⊗B A • H The lower row is an equalizer by the definition of the coinvariants (N ⊗B A)coH. Since can is an isomorphism, it then follows that the upper row is an equalizer if and only if νN is an isomorphism. The upper row in the above diagram is exactly (11). By the previous lemma, this means that ι : B → A is pure if and only if the unit ν of the adjunction from Proposition 4.17 is a natural isomorphism, i.e. − ⊗B A is fully faithful. For the last statement, just remark that νB : B → (B ⊗B A)coH = AcoH is an isomor- phism. (cid:3) As we have remarked before, since partial comodules do not provide an abelian category, one cannot expect that the functor − ⊗B A : ModB → PModH A is an equivalence in general. The following observation shows that as soon as H is non-trivial, this functor will never be an equivalence. Indeed, by construction any induced partial Hopf module M = N ⊗B A satisfies M • H = M ⊗A (A • H). It follows however from Lemma 4.11 that in general we only have an surjection M ⊗A (A • H) ։ M • H. This motivates the following definition. Definition 4.24. A partial relative Hopf module M is called minimal iff M • H = M ⊗A (A • H). Example 4.25. Let A be a partial H-coaction as in Example 2.9, considered as a partial comodule algebra, see Example 4.4. Then we have that A • H = {a1[0] ⊗ h1[1] a ⊗ h ∈ A ⊗ H} where e = 1[0] ⊗ 1[1] is supposed to be central in A ⊗ H. Let M be a relative Hopf module in the sense of [8], that is M is a right A-module, endowed with a coaction ρ : M → M ⊗ H, ρ(m) = m[0] ⊗ m[1] satisfying • m = m[0]ǫ(m[1]); • ρ(m[0]) ⊗ m[1] = m[0]1[0] ⊗ m[1](1)1[1] ⊗ m[1](2); • ρ(ma) = m[0]a[0] ⊗ m[1]a[1]. Then by defining M • H = {m1[0] ⊗ h1[1] m ⊗ h ∈ M ⊗ H} we find that M can be endowed with the structure of a partial Hopf module in the sense defined here. Moreover, one then easily checks that this partial Hopf module is minimal: M • H ∼= M ⊗A (A • H), m1[0] ⊗ h1[1] 7→ m ⊗A (1[0] ⊗ h1[1]). A key tool in [8] is the observation that for a partial action as in the previous example, the A-bimodule A • H is an A-coring, whose comodules correspond exactly with the partial relative Hopf modules. We now extend this result to the present setting. Lemma 4.26. Let A be a quasi partial H-comodule algebra. (i) There is a canonical epimorphism p : (A • H) • H → (A • H) ⊗A (A • H); (ii) For any minimal partial Hopf module M, we have a canonical epimorphism pM : (M • H) • H → (M • H) ⊗A (A • H). / / / / / / / / / GEOMETRICALLY PARTIAL ACTIONS 51 Consequently, if A is geometric as H-comodule, then C = A • H is an A-coring and can : A • H → A ⊗B A is a morphism of A-corings. Moreover, there is a functor from the category of geometric minimal partial Hopf modules to the category of C-comodules. Proof. (i). Consider the following diagram. (A • H) ⊗ H φ∼= πA•H (A • H) • H (A • H) ⊗A (A ⊗ H) (A•H)⊗AπA / (A • H) ⊗A (A • H) / 0 / 0 where φ is the isomorphism given by φ((a • h) ⊗ h) = (a • h) ⊗A (1A ⊗ h). We know that ker ((A • H) ⊗A πA) = (A • H) ⊗A ker πA and ker πA•H = (ρA ⊗ H)(ker πA). Consider any a ⊗ h ∈ ker πA. Then we find that (a[0] • a[1]) ⊗A (1A ⊗ h) = (1A • 1H) · a ⊗A (1A ⊗ h) = (1A • 1H) ⊗A (a ⊗ h) Hence we find that φ(ker πA•H ) ⊂ ker ((A • H) ⊗A πA). Consequently, φ induces an epimorphism p : (A • H) • H → (A • H) ⊗A (A • H). (ii). By part (i), we know that the following diagram commutes A ⊗ H A • H πA v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ p◦(ρA•H) ρA⊗H )❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ u❦❦❦❦❦❦❦❦❦❦❦❦❦❦ (id⊗AπA)◦φ (A • H) ⊗A (A • H) (A • H) ⊗ H Applying the functor M ⊗A − to this diagram, and using M ⊗A (A • H) = M • H, we find that the following diagram commutes as well (we avoid naming the arrows to not overload the diagram) M ⊗ H M • H v❧❧❧❧❧❧❧❧❧❧❧❧❧❧ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ )❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ u❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦ (M • H) ⊗A (A • H) (M • H) ⊗ H Hence, by the universal property of the pushout, we obtain an epimorphism (M • H) • H → (M • H) ⊗A (A • H). Using the above, we can define a coring structure on A • H by means of the counit A • ǫ and comultiplication p ◦ (A • ∆), and one can easily observe that can is a coring morphism for this structure. Similarly, for a minimal partial Hopf module M, the partial H-coaction is also an A • H-coaction since ρM : M → M • H = M ⊗A (A • H). Furthermore, it is enough to remark that for a geometric partial Hopf module, the coassociativity holds in M •H •H. If M is minimal then by the above (M •H)⊗A (A•H) is a quotient of M • H • H, so coassociativity also holds there. (cid:3) / / / / / v v v ) ( u u u v v v ) ( u u u 52 JIAWEI HU AND JOOST VERCRUYSSE Remark 4.27. Given a comodule M over the coring A • H, one can construct a partial comodule datum (M, M ⊗A (A • H), M ⊗A πA, ρM ). However, one cannot expect that all such comodule data provide partial H-comodules. Indeed, if A/B is Galois, then the canonical map is a coring isomorphism A ⊗B A ∼= A • H. and the categories A ⊗B A-comodules and A • H-comodules are isomorphic as well. When A is flat as left B-module, then the category of A ⊗B A-comodules is abelian. If ModA•H coincides with the category of partial Hopf modules, than this would imply that the latter category is Abelian, which we know is not the case. If A is flat as left B-module, then the functor − ⊗B A preserves all equalizers. Recall from [6] (see [7] for a corrected version of the statement) that the functor − ⊗B A preserves the equalizers of the form (10) provided A is pure as left A-module and B lies in the center of A. Proposition 4.28. Let M be a partial Hopf module M. (i) If ζM (counit of the adjunction Proposition 4.17) is an isomorphism, then M is minimal. (ii) If A/B is partially Hopf-Galois and ζM is a monomorphism, then M is minimal. (iii) If M is minimal and geometric, A is partially Hopf Galois and the functor − ⊗B A preserves the equalizers of the form (10) (e.g. A is flat as left B-module, or A is pure as left A-module and B ⊂ Z(A)), then ζM is an isomorphism. Proof. (i) If ζM is an isomorphism of partial Hopf modules, then we find a composition of isomorphisms M ⊗A (A • H) ζ −1 M ⊗A(A•H) / (M coH ⊗B A) ⊗A (A • H) ∼= M coH ⊗B (A • H) = (M coH ⊗B A) • H ζM •H / M • H and hence M is minimal. (ii). Consider the following commutative diagram M coH ⊗B A M ⊗B A ζM M ρM / M • H Since M ⊗B A ∼= M ⊗A (A ⊗B A) ∼= M ⊗A (A • H), we know that the right vertical arrow is an epimorphism (see Lemma 4.11). If A is flat as left B-module then the upper horizontal arrow is injective, and the lower horizontal arrow is injective as it has a left inverse M •ǫ. Hence if ζM is injective, we find that M ⊗B A → M •H is an isomorphism, i.e. M is minimal. (iii). Since A/B is Hopf-Galois, we find obtain an isomorphism M ⊗B A ∼= / M ⊗A (A ⊗B A) M ⊗Acan / M ⊗A (A • H) And therefore, if M is minimal we find that M ⊗B A ∼= M ⊗A (A•H) = M •H. Consider now the following diagram M coH ⊗B A M ⊗B A / M • H ⊗B A ρM ⊗BA πM ◦(M ⊗ηH )⊗B A ζ ′ M ζM M ρM ∼= M • H pM / M • H • H ρM •H M •∆ / / / / / / / / / / / / / / O O / / / O O O O GEOMETRICALLY PARTIAL ACTIONS 53 By assumption, the upper row in this diagram is an equalizer. Since M is geometric, the fork on the lower row splits and hence is also an equalizer. The surjective morphism pM is obtained from Lemma 4.26 and induces the morphism ζ ′ M . Then a diagram chasing argument shows that ζM and ζ ′ (cid:3) M are mutual inverses. The following result subsumes the Hopf-Galois theory for partial coactions in the sense of Caenepeel-Janssen [8]. Corollary 4.29. Suppose that A a partial H-comodule algebra that is geometric as partial comodule. If A/B is a partial Hopf-Galois extension and either • A is pure as left B-module and B ⊂ Z(A); • A is faithfully flat as left B-module; then ModAcoH is equivalent to the category full subcategory of PModH imal geometric partial Hopf modules. A consisting of min- Proof. Since A is geometric as partial comodule, the functor − ⊗B A : ModAcoH → PModH A lands in the category of minimal geometric partial Hopf modules. By Proposi- tion 4.23 and Proposition 4.28 we then obtain the stated equivalence of categories. (cid:3) As we have remarked before, the functor − ⊗B A : ModAcoH → PModH A cannot be expected to become an equivalence of categories. More precisely, it follows from Propo- sition 4.28 that we cannot expect that the functor (−)coH is full whenever it is applied to non-minimal partial Hopf modules. We will finish our work by characterizing under which conditions this functor remains however faithful. Proposition 4.30. Under the same conditions as in Corollary 4.29, A is a generator in PModH A if and only if the functor (−)coH : PModH A → ModAcoH is faithful. Proof. Recall that A is a generator in PModH the canonical morphism A if and only if for any object M in PModH A φM : a A → M, φ(af ) = X f (af ) HomH A (A,M ) f A (A,M ) A denotes a coproduct of copies of A indexed by the set is surjective. Here `HomH HomH Recall from Lemma 4.16 that M coH = HomH we obtain a well-defined morphism A (A, M). A (A, M) for any object M in PModH A . Hence αM : a A → M coH ⊗AcoH A, αM (af ) = f (1A) ⊗AcoH af , HomH A (A,M ) which is clearly surjective. One now easily sees that φM = ζM ◦ αM . Hence φM is surjective if and only if ζM is surjective. Finally, it is well-know that a right adjoint functor is faithful if and only if the counit of the adjunction is a natural epimorphism. (cid:3) Remark 4.31. Under the conditions of Proposition 4.30, we know that for an object M in PModH A , the partial Hopf module morphism ζM : M coH ⊗B A → M is surjective. Hence, we find that M ∼= M coH ⊗B A/ker ζM as right A-module. However, as we remarked earlier, ker ζM is not necessarily a partial H-comodule. We then know from Example 2.6 that M ⊗B A/ker ζM is a geometric partial comodule. Then ζM induces a morphism of partial Hopf modules ζ ′ M : M ⊗B A/ker ζM → M, such that the underlying 54 JIAWEI HU AND JOOST VERCRUYSSE A-module morphism is an isomorphism. However, ζ ′ M is not necessarily an isomorphism of partial Hopf modules, since in general (M ⊗B A/ker ζM ) • H and M • H can be different. Therefore consider the following definition. Let (X, X • H, πX , ρX) and (Y, Y • H, πY , ρY ) be two partial H-comodule data. Let f : X → Y be a morphism in C. Then consider the pushout X ⊗ H f ⊗H / Y ⊗ H πY / / Y • H πX X • H pY / PX,Y pX With this notation, f is said to be a weak morphism of partial comodules, if the following diagram commutes f X ρX X • H pX Y ρY Y • H pY / PX,Y Then the A-linear inverse of ζ ′ M will be a weak morphism of partial comodules that is moreover a 2-sided inverse of the (strong) morphism ζ ′ M . This motivates that weak morphisms of (geometric) partial comodules might be better behaved than the strong morphisms we studied in this work. We will investigate this further in future work. Let us finish by proving result which completely characterizes the image of the functor − ⊗AcoH A : ModAcoH → PModH A . Theorem 4.32. (i) If A and A • H are flat as left AcoH-module (e.g. A is flat as left AcoH-module and A/AcoH is H-Galois), then the functor − ⊗AcoH A : ModAcoH → PModH A preserves equalizers. (ii) If A is faithfully flat as left AcoH-module then the functor − ⊗AcoH A : ModAcoH → PModH A reflects isomorphisms. (iii) If A is faithfully flat as left AcoH-module and A/AcoH is partially Hopf-Galois, then the category of AcoH-modules is equivalent to the Eilenberg-Moore category (PModH A )C, where C is the comonad associated to the adjoint pair of Proposi- tion 4.17. Proof. (i). Consider the following equalizer diagram in ModAcoH : E e / N f g / M By the flatness of A as a left AcoH-module, we then know that (E ⊗AcoH A, e ⊗AcoH A) is the equalizer of the pair (f ⊗AcoH A, g ⊗AcoH A) in ModA. However, we have to show that this is also an equalizer in PModH A . To this end, consider any partial relative Hopf module T with a morphism t : T → N ⊗AcoH A such that (f ⊗AcoH A)◦t = (g ⊗AcoH A)◦t. Then we can apply the forgetful functor PModH A → ModA and we find that there exists a unique right A-linear map u : T → E ⊗AcoH A such that t = e ⊗AcoH A ◦ u. We will / / / / / / / / / GEOMETRICALLY PARTIAL ACTIONS 55 be done if we can show that u is a morphism of partial H-comodules. Firstly we will verify that πE⊗AcoH A ◦ (u ⊗ H)(ker πT ) = 0 (cf. Lemma 2.13). Since A • H is flat as a left AcoH-module and e is an injective map (being an equalizer in a module category), it is equivalent to check that (e ⊗AcoH (A • H)) ◦ πE⊗AcoH A ◦ (u ⊗ H)(ker πT ) = 0. By the functoriality of − ⊗AcoH A : ModAcoH → PModH A , we obtain that e ⊗AcoH A is a morphism of partial comodules and (e ⊗AcoH (A • H)) = (e ⊗AcoH A) • H). This allows us to rewrite the left hand side of the last equality as πN ⊗AcoH A ◦ (e ⊗AcoH A ⊗ H) ◦ (u ⊗ H)(ker πT ) = πN ⊗AcoH A ◦ (t ⊗ H)(ker πT ) = 0 where the second equality is the defining property of u and the last equality follows from the fact that t is a morphism of partial comodules. Hence the map u • H : T • H → E ⊗AcoH (A • H) is well defined and the unique map satisfying (u • H) ◦ πT = πE⊗AcoH A ◦ (u ⊗ H). Then we find ((e ⊗AcoH A) • H) ◦ (u • H) ◦ πT = ((e ⊗AcoH A) • H) ◦ πE⊗AcoH A ◦ (u ⊗ H) = πN ⊗AcoH A ◦ (e ⊗AcoH A) ⊗ H ◦ (u ⊗ H) = πN ⊗AcoH A ◦ (t ⊗ H) = (t • H) ◦ πT where we used the fact that e ⊗AcoH A is a morphism of partial comodules in the second equality, t = e ⊗AcoH A ◦ u in the third equality and the fact that t is a morphism of partial comodules in the last equality. By the surjectivity of πT , it follows now that (t • H) = (e ⊗AcoH A • H) ◦ (u • H). For u to be a partial comodule morphism, it remains to check that (u • H) ◦ ρT = ρE⊗AcoH A ◦ u. Again, using the flatness of A • H a left A-module and the injectivity of e it is sufficient to check that the compositions of the se maps with e ⊗AcoH (A • H) are equal. Using the fact that t and e are partial module morphisms, we can indeed prove that (e ⊗AcoH (A • H)) ◦ (u • H) ◦ ρT = (t • H) ◦ ρT = ρN ⊗AcoH A ◦ t = ρN ⊗AcoH ◦ e ⊗AcoH A ◦ u = (e ⊗AcoH (A • H)) ◦ ρE⊗AcoH A ◦ u A and therefore (E ⊗AcoH A, e ⊗AcoH A) satisfies the uni- Hence u lives already in PModH versal property of the equalizer in PModH A . (ii). Let f : M → N be a morphism in ModAcoH such that f ⊗AcoH A is an isomorphism in PModH A . Then f ⊗AcoH A is also an isomorphism in ModA, and since A is faithfully flat as left AcoH-module, we find that f is an isomorphism in ModAcoH . (iii). This follows immediately from the previous two parts by the dual of Beck's monadicity theorem, see e.g. [4] or [12, Theorem 2.7]. (cid:3) Remark 4.33. The proof of part (i) in the previous theorem can be adapted to show that the functor − ⊗AcoH A : ModAcoH → PModH A preserves arbitrary limits. Acknowledgements. The first author wants to thank CSC (China Scholarship Coun- cil) for a PhD-student fellowship. He also thanks the "Fonds David et Alice Van Buuren" for the prize that allowed him to finish his PhD in Brussels. The second author wants to thank the "F´ed´eration Wallonie-Bruxelles" for the ARC grant "Hopf algebras and the Symmetries of Non-commutative Spaces" that supported this work. 56 JIAWEI HU AND JOOST VERCRUYSSE The second author wants to thank Ana Agore for a discussion concerning the complete- ness of comodule categories and Mitchell Buckley for useful discussions about categorical aspects of spans, bicategories and pushouts. Both authors thank Eliezer Batista and Paolo Saracco for interesting discussions on partial actions and coactions. We thank the anonymous referee for this careful reading of the paper and the many useful suggestions that improved the presentation of this paper, as well for pointing out some inaccuracies. References [1] M. M. S. Alves, E. Batista, J. Vercruysse, Partial representations of Hopf algebras, J. Algebra 426 (2015) 137 -- 187. 2, 12 [2] M. M. S. Alves, E. Batista, J. Vercruysse, Dilations of partial representations of Hopf algebras, J. Lond. Math. Soc. 100 (2019) 273 -- 300. 2, 28 [3] M. M. S. Alves, E. Batista, F. L. Castro, G. Quadros, J. Vercruysse, Partial co-representations of Hopf algebras, arXiv:1911.09141, 2019. 12 [4] M. Barr, C. Wells, Toposes, triples and theories, Grundlehren der Mathematischen Wissenschaften 278, Springer-Verlag, New York, 1985. 55 [5] E. Batista, J. Vercruysse, Dual constructions for partial actions of Hopf algebras, J. Pure Appl. Algebra 220 (2016) 518 -- 559. 2, 16 [6] S. Caenepeel, Galois corings from the descent point of view, Fields Inst. Comm. 43 (2004), 163 -- 186. 49, 52 [7] S. Caenepeel, E. De Groot and J. Vercruysse, Galois theory for Comatrix Corings: Descent theory, Morita theory, Frobenius and separability properties, Trans. Amer. Math. Soc. 359 (2007), 185 -- 226. 49, 52 [8] S. Caenepeel, K. Janssen, Partial (co)actions of Hopf algebras and partial Hopf-Galois theory, Comm. Algebra 36 (2008) 2923 -- 2946. 2, 12, 16, 43, 50, 53 [9] M. Dokuchaev, R. Exel, Associativity of Crossed Products by Partial Actions, Enveloping Actions and Partial Representations, Trans. Amer. Math. Soc. 357 (5) (2005) 1931 -- 1952. 2 [10] M. Dokuchaev, R. Exel, P. Piccione, Partial Representations and Partial Group Algebras, J. Algebra 226 (2000) 505 -- 532. 2 [11] R. Exel, Circle Actions on C ∗-Algebras, Partial Automorphisms and Generalized Pimsner- Voiculescu Exect Sequences, J. Funct. Anal. 122 (1994) 361 -- 401. 2, 4 [12] J. G´omez-Torrecillas, Comonads and Galois corings, Appl. Categ. Structures 14 (2006), 579 -- 598. 55 [13] P. Johnstone, An answer to "Products of epimorphisms", discussion on the category theory mailing list, January 2018. 34 [14] T. Leinster, Higher operads, higher categories, London Mathematical Society Lecture Note Series, 298, Cambridge University Press, Cambridge, 2004. 3, 31 [15] P. Quast F. Kraken, T. Timmermann, Partial actions of C ∗-quantum groups I: Restriction and Globalization, Banach J. Math. Anal. (2018). 2 [16] H.-J. Schneider, Principal homogeneous spaces for arbitrary Hopf algebras. Israel J. Math. 72 (1990), 167 -- 195. 49 Departement of Mathematical Sciences, East China Normal University, NO.500,Dongchuan Road, Shanghai, China D´epartement de Math´ematiques, Facult´e des sciences, Universit´e Libre de Bruxelles, Boulevard du Triomphe, B-1050 Bruxelles, Belgium E-mail address: [email protected] D´epartement de Math´ematiques, Facult´e des sciences, Universit´e Libre de Bruxelles, Boulevard du Triomphe, B-1050 Bruxelles, Belgium E-mail address: [email protected]
1904.07708	1	1904	2019-04-14T00:30:48	Injective Semimodules - Revisited	[ "math.RA", "math.CT" ]	Injective modules play an important role in characterizing different classes of rings (e.g. Noetherian rings, semisimple rings). Some semirings have no non-zero injective semimodules (e.g. the semiring of non-negative integers). In this paper, we study some of the basic properties of the so called e-injective semimodules introduced by the first author using a new notion of exact sequences of semimodules. We clarify the relationships between the injective semimodules, the e-injective semimodule, and the i-injective semimodules through several implications, examples and counter examples. Moreover, we provide partial results for the so called Embedding Problem (of semimodules in injective semimodules).	math.RA	math	Injective Semimodules - Revisited* Jawad Abuhlail† Rangga Ganzar Noegraha‡ [email protected] [email protected] Department of Mathematics and Statistics King Fahd University of Petroleum & Minerals 31261 Dhahran, KSA Universitas Pertamina Jl. Teuku Nyak Arief Jakarta 12220, Indonesia April 17, 2019 Abstract Injective modules play an important role in characterizing different classes of rings (e.g. Noetherian rings, semisimple rings). Some semirings have no non-zero injective semimod- ules (e.g. the semiring of non-negative integers). In this paper, we study some of the basic properties of the so called e-injective semimodules introduced by the first author using a new notion of exact sequences of semimodules. We clarify the relationships between the injective semimodules, the e-injective semimodule, and the i-injective semimodules through several implications, examples and counter examples. Moreover, we provide partial results for the so called Embedding Problem (of semimodules in injective semimodules). Introduction Semirings (defined, roughly, as rings not necessarily with subtraction) generalize both rings and distributive bounded lattices. Semirings and their semimodules (defined, roughly, as modules not necessarily with subtraction) have many applications in Mathematics, Computer Science and Theoretical Science (e.g., [HW1998], [Gla2002], [LM2005]). Our main reference on semirings and their applications is Golan's book [Gol1999], and Our main reference in rings and modules is [Wis1991]. *MSC2010: Primary 16Y60; Secondary 16D50 Key Words: Semirings; Semimodules; Injective Semimodules; Exact Sequences The authors would like to acknowledge the support provided by the Deanship of Scientific Research (DSR) at King Fahd University of Petroleum & Minerals (KFUPM) for funding this work through projects No. RG1304-1 & RG1304-2 †Corresponding Author ‡The paper is extracted from his Ph.D. dissertation under the supervision of Prof. Jawad Abuhlail. 1 The notion of injective objects can be defined in any category relative to a suitable factor- ization system. Injective semimodules have been studied intensively (see [Gla2002] for details). Recently, several papers established homological characterizations of special classes of semirings using (cf., [KNT2009], [Ili2010], [KN2011], [Abu2014], [KNZ2014], [AIKN2015], [IKN2017], [AIKN2018]). For example, left (right) V -semirings, all of whose congruence-simple left (right) semimodules are injective have been completely characterized in [AIKN2015]. In addition to the classical notions of injective semimodules over a semiring, several other notions were considered in the literature, e.g. the so called i-injective semimodules [Alt2003] and the k-injective semimodules [KNT2009]. One reason for the interest of such notions is the phenomenon that assuming that all semimodules of a given semiring S to be injective forces the semiring to be a (semisimple) ring (cf. [Ili2010, Theorem 3.4]). Using a new notion of exact sequences of semimodules over a semiring, Abuhlail [Abu2014-CA] introduced a homological notion of exactly injective semimodules (e-injective semimodules for short) assuming that an ap- propriate Hom functors preserve short exact sequences. Such semimodules were called initially uniformly injective semimodules and used in [Abu2014-SF] under the name normally injective semimodules; the terminology e-injective semimodules was used first in [AIKN2018]. The paper is divided into three sections. In Section 1, we collect some basic definitions, examples and preliminaries used in this paper. In particular, we recall the definition and basic properties of exact sequences in the sense of Abuhlail [Abu2014]. In Section 2, we investigate mainly the e-injective semimodules over a semiring and clarify their relationships with the injective semimodules and the i-injective semimodules. In Lemma 2.12 and Proposition 2.14 we provide homological detailed proofs of the fact that the class of injective left semimodules is closed under retracts and direct products. It was shown in [AIKN2018, Proposition-Example 4.6.] that, for an additively idempotent division semiring D, the class of e-injective D-semimodules is strictly larger than the class of injective D-semimodules. Subsection 2.1 is devoted to showing that for the semiring S := M2(R+), the class of S-i-injective left semimodules is strictly larger than the class of S-e-injective left S-semimodules: Lemma 2.18 shows that all left S-semimodules are S-I-injective, while Example 2.19 provides a left S-semimodule which is not S-e-injective. In Section 3, we investigate the so called Embedding Problem. While every module over a ring R can be embedded in an injective semimodules, and a module M is injective if M is R- injective (using the Baer's Criterion), any semiring whose category of semimodules has both of these nice properties is a ring [Ili2008, Theorem 3]. Call a left S-semimodule c-i-injective if it is M-i-injective for every cancellative left S-semimodule M. We prove in Theorem 3.18 that every left S-semimodule can be embedded subtractively in a c-i-injective left S-semimodule. 2 1 Preliminaries In this section, we provide the basic definitions and preliminaries used in this work. Any notions that are not defined can be found in our main reference [Gol1999]. We refer to [Wis1991] for the foundations of Module and Ring Theory. Definition 1.1. ([Gol1999]) A semiring is a datum (S, +, 0, ·, 1) consisting of a commutative monoid (S, +, 0) and a monoid (S, ·, 1) such that 0 6= 1 and a · 0 = 0 = 0 · a for all a ∈ S; a(b + c) = ab + ac and (a + b)c = ac + bc for all a, b, c ∈ S. A semiring S with (S, ·, 1) a commutative monoid is called a commutative semiring. A semiring S with a + a = a for all a ∈ S is said to be an additively idempotent semiring. A semiring with no non-zero zero-divisors is called entire. We set V (S) := {s ∈ S s + t = 0 for some t ∈ S}. (1) If V (S) = {0}, then we say that S is zerosumfree. Examples 1.2. ([Gol1999]) • Every ring is a cancellative semiring. • Any distributive bounded lattice L = (L, ∨, 1, ∧, 0) is a commutative idempotent semiring and 1 is an infinite element of L . • The sets (Z+, +, 0, ·, 1) (resp. (Q+, +, 0, ·, 1), (Q+, +, 0, ·, 1)) of non-negative integers (resp. non-negative rational numbers, non-negative real numbers) is a commutative can- cellative semiring which is not a ring. • Mn(S), the set of all n × n matrices over a semiring S, is a semiring. • B := {0, 1} with 1 + 1 = 1 is an additively idempotent semiring called the Boolean semir- ing. • Rmax := (R ∪ {−∞}, max, −∞, +, 0) is an additively idempotent semiring. 1.3. [Gol1999] Let S and T be semirings. The categories SSM of left S-semimodules with mor- phisms the S-linear maps, SMT of right S-semimodules with morphisms the T -linear maps, and SSMT of (S, T )-bisemimodules with morphisms the S-linear T -linear maps are defined as for left (right) modules and bimodules over rings. The set of cancellative elements of a (bi)semimodule M is K+(M) := {m ∈ M m + m1 = m + m2 =⇒ m1 = m2 for any m1, m2 ∈ M}; and we say that M is cancellative, if K+(M) = M. We write L ≤S M to indicate that L is a subsemimodule of the S-semimodule M. 3 Example 1.4. The category of Z+-semimodules is nothing but the category of commutative monoids. Example 1.5. ([Gol1999, page 150, 154]) Let S be a semiring, M be a left S-semimodule and L (cid:0)S M. The subtractive closure of L is defined as L := {m ∈ M m + l = l′ for some l, l′ ∈ L}. (2) −→ M/L), where π is the canonical projection. We say that One can easily check that L = Ker(M L is subtractive, if L = L. We say that M is a subtractive semimodule, if every S-subsemimodule L ≤S M is subtractive. π Following [BHJK2001], we use the following definitions. Definition 1.6. Let S be a semiring. A left S-semimodule M is ideal-simple, if 0 and M are the only S-subsemimodules of M. 1.7. (cf., [AHS2004]) The category SSM of left semimodules over a semiring S is a closed under homomorphic images, subobjects and arbitrary products (i.e. a variety in the sense of Universal Algebra). In particular, SSM is complete, i.e. has all limits (e.g., direct products, equalizers, kernels, pullbacks, inverse limits) and cocomplete, i.e. has all colimits (e.g., direct coproducts, coequalizers, cokernels, pushouts, direct colimits). For the construction of the pullbacks and the pushouts, see [AN]. 1.8. Let M be a left S-semimodule. We say that N ≤S M is a retract of M, if there exists a (surjective) S-linear map θ : M −→ N and an (injective) S-linear map ψ : N −→ M such that θ ◦ ψ = idN; direct summand of M, if there exists L ≤S M such that M = L ⊕ N. Exact Sequences Throughout, (S, +, 0, ·, 1) is a semiring and, unless otherwise explicitly mentioned, an S-module is a left S-semimodule. Definition 1.9. A morphism of left S-semimodules f : L → M is k-normal, if whenever f (m) = f (m′) for some m, m′ ∈ M, we have m + k = m′ + k′ for some k, k′ ∈ Ker( f ); i-normal, if Im( f ) = f (L) (:= {m ∈ M m + l ∈ L for some l ∈ L}). normal, if f is both k-normal and i-normal. Remark 1.10. Among others, Takahashi ([Tak1981]) and Golan [Gol1999] called k-normal (resp., i-normal, normal) S-linear maps k-regular (resp., i-regular, regular) morphisms. We changed the terminology to avoid confusion with the regular monomorphisms and regular epi- morphisms in Category Theory which have different meanings when applied to categories of semimodules. 4 The following technical lemma is helpful in several proofs in this and forthcoming related papers. Lemma 1.11. ([AN]) Let L f → M g → N be a sequence of semimodules. (1) Let g be injective. (a) f is k-normal if and only if g ◦ f is k-normal. (b) If g ◦ f is i-normal (normal), then f is i-normal (normal). (c) Assume that g is i-normal. Then f is i-normal (normal) if and only if g ◦ f is i-normal (normal). (2) Let f be surjective. (a) g is i-normal if and only if g ◦ f is i-normal. (b) If g ◦ f is k-normal (normal), then g is k-normal (normal). (c) Assume that f is k-normal. Then g is k-normal (normal) if and only if g ◦ f is k- normal (normal). There are several notions of exactness for sequences of semimodules. In this paper, we use the relatively new notion introduced by Abuhlail: Definition 1.12. ([Abu2014, 2.4]) A sequence f −→ M g −→ N L (3) of left S-semimodules is exact, if g is k-normal and f (L) = Ker(g). 1.13. We call a sequence of S-semimodules L f → M g → N proper-exact if f (L) = Ker(g) (exact in the sense of Patchkoria [Pat2003]); semi-exact if f (L) = Ker(g) (exact in the sense of Takahashi [Tak1981]); quasi-exact if f (L) = Ker(g) and g is k-normal (exact in the sense of Patil and Doere [PD2006]). 1.14. We call a (possibly infinite) sequence of S-semimodules · · · → Mi−1 fi−1→ Mi fi→ Mi+1 fi+1→ Mi+2 → · · · (4) chain complex if f j+1 ◦ f j = 0 for every j; exact (resp., proper-exact, semi-exact, quasi-exact) if each partial sequence with three terms f j→ M j+1 A short exact sequence (or a Takahashi extension [Tak1982b]) of S-semimodules is an f j+1→ M j+2 is exact (resp., proper-exact, semi-exact, quasi-exact). M j exact sequence of the form 0 −→ L f −→ M g −→ N −→ 0 5 Remark 1.15. In the sequence (3), the inclusion f (L) ⊆ Ker(g) forces f (L) ⊆ f (L) ⊆ Ker(g), whence the assumption f (L) = Ker(g) guarantees that f (L) = f (L), i.e. f is i-normal. So, the definition puts conditions on f and g that are dual to each other (in some sense). The follows examples show some of the advantages of the new definition of exact sequences over the old ones: Lemma 1.16. Let L, M and N be S-semimodules. (1) 0 −→ L f −→ M is exact if and only if f is injective. (2) M g −→ N −→ 0 is exact if and only if g is surjective. (3) 0 −→ L f −→ M (4) 0 −→ L f −→ M g −→ N is semi-exact and f is normal if and only if L ≃ Ker(g). g −→ N is exact if and only if L ≃ Ker(g) and g is k-normal. (5) L f −→ M (6) L f −→ M g −→ N −→ 0 is semi-exact and g is normal if and only if N ≃ M/ f (L). g −→ N −→ 0 is exact if and only if N ≃ M/ f (L) and f is i-normal. (7) 0 −→ L f −→ M g −→ N −→ 0 is exact if and only if L ≃ Ker(g) and N ≃ M/L. Corollary 1.17. The following assertions are equivalent: (1) 0 → L f → M g → N → 0 is an exact sequence of S-semimodules; (2) L ≃ Ker(g) and N ≃ M/ f (L); (3) f is injective, f (L) = Ker(g), g is surjective and (k-)normal. In this case, f and g are normal morphisms. Remark 1.18. An S-linear map is a monomorphism if and only if it is injective. Every surjective S-linear map is an epimorphism. The converse is not true in general; for example the embedding ι : Z+ ֒→ Z is an epimorphism of Z+-semimodules. Proposition 1.19. (cf., [Bor1994, Proposition 3.2.2]) Let C, D be arbitrary categories and C F −→ D G −→ C be functors such that (F , G ) is an adjoint pair. (1) F preserves all colimits which turn out to exist in C. (2) G preserves all limits which turn out to exist in D. Corollary 1.20. Let S, T be semirings and T FS a (T, S)-bisemimodule. 6 (1) HomT (−, G) : T SM −→ SMS converts all colimits in to limits. (2) For every family of left T -semimodules {Yλ}Λ, we have a canonical isomorphism of right S-semimodules HomT (Mλ∈Λ Yλ, G) ≃ ∏ λ∈Λ HomT (Yλ, G). (3) For any directed system of left T -semimodules (X j, { f j j′})J, we have an isomorphism of right S-semimodules HomT (lim −→ X j, G) ≃ lim ←− HomT (X j, G). (4) HomT (−, G) converts coequalizers into equalizers; (5) HomT (−, G) converts cokernels into kernels. Proof. By [KN2011], (G ⊗S −, HomT (G, −)) is an adjoint pair of covariant functors, where G ⊗S − : SSM −→ T SM and HomT (G, −) : T SM −→ SMS. It follows directly from Proposition 1.19 that G := HomT (G, −) preserves limits, whence the contravariant functor HomT (−, G) : T SM −→ SMS converts colimits to limits. In particular, HomT (−, G) converts direct coproducts (resp. coequalizers, cokernels, pushouts, direct colimits) to direct products (resp. equalizers, kernels, pullbacks, inverse limits).(cid:4) Corollary 1.20 allows us to improve [Tak1982a, Theorem 2.6]. Proposition 1.21. Let T GS be a (T, S)-bisemimodule and consider the functor HomT (−, G) : T SM −→ SMS. Let f → M g → N −→ 0 L be a sequence of left T -semimodules and consider the sequence of right S-semimodules 0 −→ HomT (N, G) (g,G) → HomT (M, G) ( f ,G) −→ HomT (L, G). (5) (6) (1) If M g → N −→ 0 is exact and g is normal, then 0 −→ HomT (N, G) (g,G) → HomT (M, G) is exact and (g, G) is normal. (2) If (5) is semi-exact and g is normal, then (6) is proper-exact (semi-exact) and (g, G) is normal. (3) If (5) is exact and HomT (−, G) converts i-normal morphisms into k-normal ones, then (6) is exact. 7 Proof. (1) The following implications are clear: M g → N −→ 0 is exact =⇒ g is surjective =⇒ (g, G) is injective =⇒ 0 −→ HomT (N, G) is normal and consider the exact sequence of S-semimodules (g,G) → HomT (M, G) is exact. Assume that g 0 −→ Ker(g) ι −→ M g −→ N −→ 0. Notice that N ≃ Coker(ι). By Corollary 1.20, HomT (−, G) converts cokernels into kernels, we conclude that (g, G) = ker(( f , G)) whence normal. (2) Apply Lemma 1.16 (5): L g → N −→ 0 is semi-exact and g is normal ⇐⇒ M ≃ Coker( f ). Since the contravariant functor HomT (−, G) converts cokernels into kernels, it follows that HomT (N, G) = Ker(( f , G)) which is in turn equivalent to (6) being semi-exact and (g, G) being normal. Notice that f → M (g, G)(HomS(N, G)) = (g, G)(HomS(N, G)) = Ker(( f , G)), i.e. (6) is proper-exact (whence semi-exact). (3) This follows immediately from "2" and the assumption on HomT (−, G).(cid:4) 2 Injective Semimodules There are several notions of injectivity for a semimodules M over a semiring S which coincide if it were a module over a ring. In this section, we consider some of these and clarify the relationships between them. In particular, we investigate the so called e-injective semimodules which turn to coincide with the so called normally injective semimodules (both notions introduced by Abuhlail and called uniformly injective semimodules in [Abu2014-CA, 1.25, 1.24], the terminology "e- injective" was first used in [AIKN2018]. We also clarify their relations ships with injective semimodules [Gol1999] and i-injective semimodules [Alt2003]. As before, (S, +, 0, ·, 1) is a semiring and, unless otherwise explicitly mentioned, and S- module is a left S-semimodule. Exact sequences here are in the sense of Abuhlail [Abu2014] (see Definition 1.12). Definition 2.1. ([Abu2014-CA, 1.24]) Let M be a left S-semimodule. A left S-semimodules J is M-e-injective, if the contravariant functor transfers every short exact sequence of left S-semimodules HomS(−, J) : SSM −→ Z+SM 0 −→ L f −→ M g −→ N −→ 0 into a short exact sequence of commutative monoids 0 −→ HomS(N, J) −→ HomS(M, J) −→ HomS(L, J) −→ 0. We say that J is e-injective, if J is M-e-injective for every left S-semimodule M. 8 2.2. Let I be a left S-semimodule. For a left S-semimodule M, we say that I is M-injective [Gol1999, page 197] if for every injective S-linear map f : L → M and any S- linear map g : L → I, there exists an S-linear map h : M → I such that h ◦ f = g; 0 f M h / L g I M-i-injective [Alt2003] if for every normal monomorphism f : L → M and any S-linear map g : L → I, there exists an S-linear map h : M → I such that h ◦ f = g; 0 / L g uI f (normal) M h normally M-injective [Abu2014-CA, 1.24] if for every normal monomorphism f : L −→ M and any S-linear map g : L −→ I, there exists an S-linear map h : M −→ I such that h ◦ f = g 0 / L f (normal) / M h2 h1 I h′ h g I and whenever an S-linear map h′ : M → I satisfies h′ ◦ f = g, there exist S-linear maps h1, h2 : M → I such that h1 ◦ f = 0 = h2 ◦ f and h + h1 = h′ + h2. We say that I is injective (resp., i-injective, normally injective) if I is M-injective (resp., M-i-injective, normally M-projective) for every left S-semimodule M. Proposition 2.3. Let I be a left S-semimodule. (1) Let M be a left S-semimodule. Then I is M-e-injective if and only if I is normally M- injective. (2) SI is e-injective if and only if SI is normally injective. Proof. We only need to prove (1). Let M be a left S-semimodule. (=⇒) Assume that I is M-e-injective. Let L ≤S M be a subtractive S-subsemimodule. By Lemma 1.16, we have a short exact sequence of left S-semimodules 0 −→ L ι −→ M π −→ M/L −→ 0 (7) 9 / / / / / / u / / / / / / z z where ι is the canonical embedding and π is the canonical projection. By our assumption, the contravariant functor HomS(−, I) : SSM −→ Z+SM preserves exact sequences, whence the following sequence of commutative monoids 0 −→ HomS(M/L, I) (π,I) −→ HomS(M, I) (ι,I) −→ HomS(L, I) −→ 0 is exact. In particular, (ι, I) : HomS(M, I) −→ HomS(L, I) is a normal epimorphism, i.e. I is normally M-injective. (⇐=) Let 0 −→ L f −→ M g −→ N −→ 0 (8) be an exact sequence of left S-semimodules. Applying the contravariant functor HomS(−, I) to (8) it follows by Lemma 1.21 (2) and our assumption that the following sequence of commutative monoids 0 −→ HomS(N, I) is exact, i.e. SI is injective.(cid:4) (g,I) −→ HomS(M, I) ( f ,I) −→ HomS(L, I) −→ 0 (9) (10) (11) The proof of the following result is similar to that of [Alt2003, Theorem 3.7] Proposition 2.4. Let f −→ M g −→ N L be a sequence of left S-semimodules, I a left S-semimodule and consider the sequence HomS(N, I) (g,I) −→ HomS(M, I) ( f ,I) −→ HomS(L, I) of commutative monoids. (1) If (10) is exact with g normal and I is i-injective, then (11) is proper-exact. (2) If (10) is exact with g normal and I is e-injective, then (11) is exact and (g, I) is normal. (3) If (10) is exact and I is injective, then (11) is proper exact. Proof. By Corollary 1.17, we have a short exact sequence of left S-semimodules 0 −→ Ker(g) ι −→ M π −→ M/Ker(g) −→ 0 where ι and π are the canonical S-linear maps. Since (10) is proper exact, f (M) = Ker(g) and M/Ker(g) = M/ f (M) ≃ Coker( f ). By the Universal Property of Kernels, there exists a unique S-linear map ef : L −→ Ker(g) such that ι◦ef = f (and ef is surjective). On the other hand, by the 10 Universal Property of Cokernels, there exists a unique S-linear map eg : M/Ker(g) −→ N such thateg ◦ π = g. So, we have a commutative diagram of left S-semimodules z✉✉✉✉✉✉✉✉✉✉ w♦♦♦♦♦♦♦♦♦♦♦♦♦ (12) ef L 0 0 / Ker(g) ι M/Ker(g) / 0 ②②②②②②②②② 0 f / M g v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ N π eg Applying the contravariant functor HomS(−, I), we get the sequence 0 −→ HomS(M/Ker(g), I) (π,I) −→ HomS(M, I) (ι,I) −→ HomS(Ker(g), I) −→ 0 (13) and we obtain the commutative diagram HomS(N, I) (g,I) w♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣ (eg,I) (π,I) 0 ⑦⑦⑦⑦⑦⑦⑦⑦⑦⑦⑦⑦⑦⑦⑦⑦⑦ / HomS(M/Ker(g), I) HomS(M, I) HomS(Ker(g), I) / 0 0 0 }④④④④④④④④④④④④④④④④④④ (ι,I) ( f ,I) xqqqqqqqqqqqqqqqqqqqqqqq (ef ,I) HomS(L, I) of commutative monoids. (14) (1) Since g is k-normal, we conclude thateg is injective. Moreover, π is surjective, and g =eg◦π is normal, whence eg is normal by Lemma 1.11 (2). Since SI is i-injective, the sequence (13) is proper-exact and (eg, I) is surjective (see Proposition 1.21 (2)). It follows that = = Ker(( f , I)) = Ker((ef , I) ◦ (ι, I)) = im((π, I) ◦ (eg, I)) Ker((ι, I)) im((π, I)) im((g, I)). = ((13) is proper exact) ((ef , I) is injective) ((eg, I) is surjective) exact. (2) By (1), the sequence (11) is proper-exact. Since (ef , I) is injective and (ι, I) is k-normal, it follows by Lemma 1.11 (1-a) that ( f , I) = (ef , I) ◦ (ι, I) is k-normal. Consequently, (11) is Notice that, moreover, (π, I) is a normal monomorphism and (eg, I) is i-normal, whence (g, I) = (π, I) ◦ (eg, I) is normal by Lemma 1.11 (1-c). 11 w z / / / / v / w / / / } / / x / (3) The proof is similar to that of (1). Notice that by our assumption (10) is exact; in particular, g is k-normal, which is need to show thateg is injective, whence (eg, I) is surjective since SI is injective.(cid:4) Theorem 2.5. Let M be a left S-Semimodule. The following are equivalent for a left S-semimodule I : (1) SI is normally M-injective; (2) SI is M-e-injective; (3) For every exact sequence of left S-semimodules (10) with g normal, the induced sequence of commutative monoids (11) is exact and (g, I) is normal. Proof. (1) ⇒ (2) This follows by Proposition 2.3. (2) ⇒ (3) This follows by Proposition 2.4. (3) ⇒ (1) This follows directly by applying the assumption to the exact sequences of left S-semimodules the form 0 −→ M −→ N with g normal.(cid:4) g Using Propositions 1.21 and 2.4, one can easily recover the following characterizations of injective and i-injective semimodules proved in [Alt2003] which inspired our characterizations of e-injective semimodules in Theorem 2.5. Theorem 2.6. Let M be a left S-Semimodule. The following are equivalent for a left S-semimodule I : (1) SI is M-i-injective; (2) for every proper-exact sequence of left S-semimodules (8) in which f is normal and g is k-normal, the induced sequence of commutative monoids (9) is proper-exact; (3) for every proper-exact sequence of left S-semimodules (10) with g normal, the induced sequence of commutative monoids (11) is proper-exact. Theorem 2.7. Let M be a left S-Semimodule. The following are equivalent for a left S-semimodule I : (1) SI is M-injective; (2) for every proper-exact sequence of left S-semimodules (8) in which f is k-normal and g is normal, the induced sequence of commutative monoids (9) is proper-exact. (3) for every exact sequence of left S-semimodules (10), the induced sequence of commutative monoids (11) is proper-exact. 12 It follows directly from the definitions that, for any semiring S and any left S-semimodule M, S(M) of M-i-injective left S-semimodules contains both the class IS(M) of injective the class I i left S-semimodules and the class I e S (M) of S-e-injective left S-semimodules, i.e. IS(M) ∪ I e S (M) ⊆ I i S(M). (15) While every projective semimodule is e-projective (see [AIKN2018]), it is not evident that every injective semimodule is e-injective if the base semiring is arbitrary. However, we have a partial result: Proposition 2.8. ([AIKN2018, Theorem 4.5]) Let S be an additively idempotent semiring. Then every injective left S-semimodule is e-injective. The following examples shows that the converse of Proposition 2.8 is not true in general. Example 2.9. ([AIKN2018, 4.6]) Let D be an additively idempotent division semiring (e.g., D = B, the Boolean semiring). Then D has an e-injective left S-semimodule which is not injective. We illustrate first that relative injectivity and relative e-injectivity are not related even when the base semiring is commutative and additively idempotent. The following example shows that the relative version of Proposition 2.8 is not valid, i.e. relative injectivity does not guarantee relative e-injectivity. Example 2.10. Consider the semiring S := {0, 1, a} [Alt2003] with addition and multiplication given by + 0 1 a 0 0 1 a 1 1 1 1 a a 1 a addition · 0 1 a 0 0 0 0 1 0 1 a a 0 a a multiplication We show that S is S-injective but not S-e-injective. The ideals of S are {0}, {0, a, } and S. Clearly, {0, a} is subtractive, whence S is a subtractive semiring and our example shows that the inclusion I e S(S) = IS(S) is strict. S (S) ⊆ I i Claim I: S is S-injective. We need to consider only the canonical embedding {0, a} ι ֒→ S and the S-linear map ϕ : {0, a} −→ S, 0 7→ 0 and a 7→ a. (16) Notice that ϕ(a) 6= 1 : if so, then 1 = ϕ(a) = ϕ(a · 1) = aϕ(1), i.e. a has a multiplicative inverse, a contradiction. Notice that ϕ can be extended to an S-linear map through idS. It follows that S is S-injective. Claim II: S is not S-e-injective. The S-linear map (16) can be extended through another S-linear map, namely eh : S −→ S, 0 7→ 0, a 7→ a, 1 7→ a. 13 However, the only S-linear map h : S −→ S such that (h ◦ ι)({0, a}) = 0 is the h = 0 : indeed, h(1) = a implies 0 = h(a) = h(a · 1) = ah(1) = a · a = a, a contradiction; and h(1) = 1 implies 0 = h(a) = h(a · 1) = ah(1) = a · 1 = a, a contradiction. So, we cannot find h1, h2 ∈ HomS(S, S) such that idS + h1 =eh + h2. Consequently, S is not S-e-injective.(cid:4) The following example shows that relative e-injectivity does not guarantee relative injectivity. In fact, we given an example for which IS(S) $ I e S (S) = SSM = I i S(S). Example 2.11. Consider the commutative additively idempotent semiring S := (Z+, max, 0, ·, 1). Then S has no non-trivial proper subtractive ideals, whence every S-semimodule is S-e-injective (S-i-injective). By [Alt2003, Example 2.7], S is not S-injective. In particular, our example shows that the inclusion IS(S) ⊆ I i S(S) is strict. Next, we provide detailed homological proofs rather than compact categorical ones of the facts that, for a given left S-semimodule M, the class of M-e-injective semimodules is closed under retracts and direct products (cf., [AIKN2018, Corollary 3.3]). Proposition 2.12. (1) Let M be a left S-semimodule. Every retract of a left M-e-injective S- semimodule is M-e-injective. (2) A retract of an e-injective S-semimodule is e-injective. 2.13. We need to prove (1) only. Let J be an M-e-injective left S-semimodule and I a retract of J along with S-linear maps ι : I −→ J and π : J −→ I such that π◦ι = idI. Let f : L → M be a normal S-monomorphism and g : L → I be an S-linear map. 0 M ✎ f ✎ ✎ h∗ ✎ / L g I ι ✎ J ✎ ✎ Since J is M-e-injective, there is an S-linear map h∗ : M → J such that h∗ ◦ f = ι ◦ g. Consider h := π◦ h∗. Then we have h ◦ f = (π◦ h∗) ◦ f = π◦ (h∗ ◦ f ) = π◦ (ι◦ g) = (π◦ ι) ◦ g = idI ◦ g = g. Suppose that h′ : M → I is an S-linear map such that h′ ◦ f = g. Notice that ι ◦ h′ ◦ f = ι ◦ g. Since J is M-e-injective, there exist S-linear maps h∗ 2 ◦ f 2 : M → J such that h∗ 1 ◦ f = 0 = h∗ 1, h∗ 14 / / / and h∗ + h∗ 1 = ι◦ h′ + h∗ 2. 0 / L g I ι J h∗ 2 h∗ 1 J f / M h′ ⑧⑧⑧⑧⑧⑧⑧⑧ h h∗ Consider h1 := π◦ h∗ Moreover, we have 1 and h2 := π◦ h∗ 2. Then we have, for i = 1, 2, hi ◦ f = π◦ h∗ i ◦ f = π◦ 0 = 0. h + h1 = π◦ ι◦ h + π◦ h∗ = π◦ (ι◦ h′ + h∗ = h′ + h2.(cid:4) 1 = π◦ (ι◦ h + h∗ 1) 2) = π◦ ι◦ h′ + π◦ h∗ 2 Proposition 2.14. Let M be a left S-semimodule and {Jλ}λ∈Λ be a collection of left S-semimodules. Then ∏ λ∈Λ Jλ is (M)-e-injective if and only if Jλ is M-e-injective for every λ ∈ Λ. Proof. Let J := ∏ λ∈Λ S-linear maps. Jλ and, for each λ ∈ Λ, let ιλ : Jλ −→ J and πλ : J −→ Jλ be the canonical (=⇒) For each λ ∈ Λ, we have πλ ◦ ιλ = idJλ, i.e. Jλ is a retract of J. The result follows from Lemma 2.12. (⇐=) Assume that Jλ is M-e-injective for every λ ∈ Λ. Let f : L → M be a normal monomor- phism and g : L → J an S-linear map. 0 M ✍ f ✍ ✍ ✍ h∗ λ ✍ ✍ ✍ / L g J πλ ✍ Jλ Since Jλ is M-e-injective for each λ ∈ Λ, there is an S-linear map h∗ πλ ◦ g. By the Universal Property of Direct Products, there exists an S-linear map λ : M → Jλ such that h∗ λ ◦ f = h : M −→ J, m 7→ ∏ λ∈Λ (ιλ ◦ h∗ λ)(m). Notice that for every l ∈ L, we have (h ◦ f )(l) = ∏ λ∈Λ (ιλ ◦ πλ)(g(l)) = g(l). (ιλ ◦ h∗ λ)( f (l)) = ∏ λ∈Λ 15 / / / / / / / / / Suppose that there exists an S-linear map h′ : M → J such that h′ ◦ f = g. It follows that : πλ ◦ h′ ◦ f = πλ ◦ g for every λ ∈ Λ. Since Jλ is M-e-injective, there exist S-linear maps h∗ 1λ M → J such that h∗ 1λ = πλ ◦ h′ + h∗ 2λ ◦ f = 0 = h∗ 2λ ◦ f and h∗ λ + h∗ 1λ , h∗ 2λ . h∗ 1λ / h∗ 2λ Jλ f / M h′ ~⑦⑦⑦⑦⑦⑦⑦⑦⑦ h h∗ λ 0 / L g J πλ Jλ For i = 1, 2, there exists by the Universal Property of Direct Products an S-linear map hi : M −→ J, m 7→ ∏ λ∈Λ (ιλ ◦ h∗ iλ )(m). For i = 1, 2 and every l ∈ L we have (hi ◦ f )(l) = ∏ λ∈Λ (ιλ ◦ h∗ iλ )( f (l)) = ∏ λ∈Λ ιλ(0) = 0. Moreover, we have for every m ∈ M : )(m) (ιλ ◦ h∗ 1λ (h + h1)(m) = ∏ λ∈Λ = ∏ λ∈Λ = ∏ λ∈Λ = ∏ λ∈Λ = ∏ λ∈Λ (ιλ ◦ πλ ◦ h)(m) + ∏ λ∈Λ (ιλ ◦ (πλ ◦ h + h∗ ))(m) 1λ (ιλ ◦ (h∗ (ιλ ◦ (πλ ◦ h′ + h∗ 2λ (h′ + ιλ ◦ h∗ 1λ = (h′ + h2)(m).(cid:4) λ + h∗ 1λ ))(m) )(m) ))(m) Lemma 2.15. Let 0 −→ L p −→ M q −→ N → 0 be a short exact sequence of left S-semimodules. If a left S-semimodule J is M-e-injective, then J is L-e-injective and N-e-injective. Proof. Let J be a left S-semimodule. Step I: J is L-e-injective. Let f : K → L be a normal monomorphism and g : K → J an S-linear map. Clearly, p ◦ f is a normal monomorphism. f L p / M 0 / K g J w♦ ♦ ♦ ♦ ♦ ♦ ♦ h∗ 16 / / / / / ~ / / / / w Since J is M-e-injective, there exists an S-linear map h∗ : M → J such that h∗ ◦ p ◦ f = g. Consider h := h∗ ◦ p : L −→ J. Then h ◦ f = h∗ ◦ p ◦ f = g. Suppose now that h′ : L → J is an S-linear map such that h′ ◦ f = g. Since p : L −→ M is a normal monomorphism and J is M-e-injective, there exists an S-linear map h : M → J such that h ◦ p = h′. Since h ◦ p = h∗ ◦ p ◦ f = g, there exist S-linear maps h∗ 2 : M → J such that 1 ◦ p ◦ f = 0 = h∗ h∗ 2 ◦ p, we have h1 ◦ f = h∗ 2. Considering h1 := h∗ 2 ◦ p ◦ f and h∗ + h∗ 1 ◦ p and h2 := h∗ 1 = h + h∗ 2 ◦ p ◦ f = h2 ◦ f and 1 ◦ p ◦ f = 0 = h∗ 1, h∗ h + h1 = h∗ ◦ p + h∗ = (h + h∗ = h′ + h2. 1 ◦ p = (h∗ + h∗ 2) ◦ p = h ◦ p + h∗ 1) ◦ p 2 ◦ p Step II: J is N-e-injective. Let f : K → N be a normal monomorphism and g : K → J an S-linear map. U q′ / K g 0 J ❡❝ f ′ f M ✵ q N ✲ ✫ ✜ h∗ ✍ ③ ❧ Let (U ; q′, f ′) be a pullback of (q, f ) (see [Tak1982b, 1.7]). Clearly, f ′ is a normal S-monomorphism. Since J is M-e-injective, there exists an S-linear map h∗ : M → J such that h∗ ◦ f ′ = g ◦ q′. Let n ∈ N. Since q is surjective, there exists mn ∈ M such that n = q(mn). Define h : N → J, n 7→ h∗(mn). Claim: h is well-defined. Suppose that there exists another m ∈ M such that q(m) = n = q(mn). Since q is k-normal, there exist m1, m2 ∈ Ker(q) such that m+m1 = mn +m2. Since m1, m2 ∈ Ker(q), (m1, 0), (m2, 0) ∈ U and so for i = 1, 2 we have h∗(mi) = (h∗ ◦ f ′)(mi, 0) = (g ◦ q′)(mi, 0) = g(0) = 0, whence h∗(m) = h∗(mn). Thus h well defined as a map. Clearly, h is S-linear. Moreover, for every k ∈ K we have f (k) = q(m f (k)) for some m f (k) ∈ M, thus (m f (k), k) ∈ U and it follows that (h ◦ f )(k) = (h ◦ f ◦ q′)(m f (k), k) = (h ◦ q ◦ f ′)(m f (k), k) = (h∗ ◦ f ′)(m f (k), k) = (g ◦ q′)(m f (k), k) = g(k), i.e. h ◦ f = g. Suppose that there exists an S-linear map h′ : N → J such that h′ ◦ f = g. Notice that h′ ◦ q ◦ 1 ◦ f ′ = 0 = f ′ = h′ ◦ f ◦ q′ = g ◦ q′. Since J is M-e-injective, there exist h∗ 2 ◦ f ′ and h∗ + h∗ h∗ 2 : M → J such that h∗ 1 = h′ ◦ q + h∗ 2. 1, h∗ 17 / / p p / / / Let n ∈ N. Since q is surjective, there exists mn ∈ M such that q(mn) = n. Define h1 : N → J, n 7→ h∗ 1(mn) and h2 : N → J, n 7→ h∗ 2(mn). One can prove as above that h1 and h2 are well-defined. It is clear that both h1 and h2 are S-linear. Notice that for every k ∈ K, we have (m f (k), k) ∈ U whence, for i = 1, 2, we have (hi ◦ f )(k) = (hi ◦ f ◦ q)(m f (k), k) = (hi ◦ q ◦ f ′)(m f (k), k) = (h∗ i ◦ f ′)(m f (k), k) = 0. Moreover, for every n ∈ N, we have h(n) + h1(n) (h + h1)(n) = (h∗ + h∗ = 1)(mn) = (h′ ◦ q)(mn) + h∗ (h′ + h2)(n).(cid:4) = = h∗(mn) + h∗ = (h′ ◦ q + h∗ 1(mn) 2)(mn) 2(mn) = h′(n) + h2(n) Remark 2.16. The converse of Lemma 2.15 is not true in general as will be shown in Example 2.20. 2.1 A Counter Example This subsection is devoted to studying the left self-injectivity of S := M2(R+). We show in par- ticular that I i S(S) = SSM and that the inclusion I e S(S) is strict. S (S) & I i Lemma 2.17. The only non-trivial proper subtractive left ideals of S are E1 = Span(cid:18)(cid:26)(cid:20) 1 0 E2 = Span(cid:26)(cid:20) 0 0 Nr = (cid:26)(cid:20) ra a 0 0 (cid:21)(cid:27)(cid:19) =(cid:26)(cid:20) a 0 0 1 (cid:21)(cid:27) =(cid:26)(cid:20) 0 c b 0 (cid:21) a, b ∈ R+(cid:27) 0 d (cid:21) c, d ∈ R+(cid:27) rb b (cid:21) a, b ∈ R+(cid:27) , r ∈ R+\{0}. Proof. We give the proof is three steps. Step I: E1, E2 and Nr are subtractive left ideals of S. Notice that E1 is a left ideal of S, since for every a, b, c, d, p, q, r, s ∈ R+ we have (cid:20) p q s (cid:21)(cid:20) a 0 b 0 (cid:21) +(cid:20) c 0 d 0 (cid:21) =(cid:20) pa + qb + c 0 ra + sb + d 0 (cid:21) ∈ E1. r Moreover, E1 is subtractive since (cid:20) p q s (cid:21) +(cid:20) a 0 d 0 (cid:21) b 0 (cid:21) =(cid:20) c 0 r 18 implies q = 0 = s, i.e.(cid:20) p q (cid:20) k m n (cid:21)(cid:20) ra a s (cid:21) ∈ E1. Similarly, we have E2 is a subtractive ideal. r(ma + nb + d) ma + nb + d (cid:21) ∈ Nr rd d (cid:21) =(cid:20) r(ka + lb + c) ka + lb + c For any non-zero r ∈ R+, Nr is a left ideal since rb b (cid:21) +(cid:20) rc c r l for all a, b, c, d, k, l, m, n ∈ R+. Moreover, Nr is subtractive since m n (cid:21) +(cid:20) ra a (cid:20) k rb b (cid:21) =(cid:20) rc rd d (cid:21) c l implies c = a + k/r = a + l and d = b + m/r = b + n, whence k = rl and m = rn, i.e.(cid:20) k m n (cid:21) ∈ l Nr. Step II: The only subtractive left ideal containing E1, E2 or Nr for some r ∈ R+\{0})} strictly is I = S. Let I be a subtractive left ideal of M2(R+). Case 1: E1 $ I. In this case, there exists(cid:20) p q as(cid:20) p 0 r 0 (cid:21) ∈ I and r s (cid:21) ∈ I such that q 6= 0 or s 6= 0, which implies(cid:20) 0 q 0 s (cid:21) ∈ I Either way(cid:20) 0 0 0 1 (cid:21) ∈ I, whence E2 ⊆ I and I = S. Case 2: E2 $ I. One can show, in a was similar to that of Case 1, that I = S. Case 3: Nr $ I for some r ∈ R+\{0}. l In this case, there exists some(cid:20) k generality, that k < rl. Then k + p = rl for some p ∈ R+\{0}. Thus(cid:20) p 0 m n (cid:21) ∈ I with k 6= rl or m 6= rn. Assume, without loss of 0 q (cid:21) ∈ I q 0 (cid:21) ∈ I or(cid:20) p 0 for some q ∈ R+ as (cid:20) p 0 q 0 (cid:21) +(cid:20) k m n (cid:21) =(cid:20) rl rn n (cid:21) ∈ I l l 19 If q 6= 0, then If s 6= 0, then r (cid:20) p 0 r 0 (cid:21) +(cid:20) 0 q 0 1 (cid:21) =(cid:20) 0 (cid:20) 0 0 (cid:20) 0 0 0 1 (cid:21) =(cid:20) 0 0 s (cid:21) =(cid:20) p q 1/q 0 (cid:21)(cid:20) 0 q 0 1/s (cid:21)(cid:20) 0 q s (cid:21) ∈ I 0 s (cid:21) ∈ I. 0 s (cid:21) ∈ I. 0 0 or Thus or l l m n (cid:21) =(cid:20) rl (cid:20) p 0 0 q (cid:21) +(cid:20) k 0 0 (cid:21) =(cid:20) 1/p 0 (cid:20) 1 0 (cid:20) 1 0 0 0 (cid:21) =(cid:20) 1/p 0 0 (cid:21)(cid:20) p 0 0 (cid:21)(cid:20) p 0 0 0 (cid:21) ∈ I, whence E1 $ I and so I = S. m m/r (cid:21) ∈ I. q 0 (cid:21) ∈ I 0 q (cid:21) ∈ I. 0 0 Either way we have(cid:20) 1 0 Step III. Let I be any non-zero subtractive left ideal of S. Then(cid:20) k m n (cid:21) ∈ I\{0} for some l k, l, m, n ∈ R+. Case 1: k 6= 0. In this case, we have 0 (cid:21)(cid:20) k (cid:20) 1/k 0 0 (cid:21) ∈ I. 0 0 (cid:21) ∈ I, whence E1 ⊆ I. Otherwise, (cid:20) k/l 1 m n (cid:21) =(cid:20) 1 l/k 0 0 0 l If l = 0, then (cid:20) 1 0 0 (cid:21) ∈ I, whence Nk/l ⊆ I. In either case, it follows by Step II that I ∈ {E1, Nk/l, S}. Case 2: l 6= 0. In this case, we have 1/l 0 (cid:21)(cid:20) k (cid:20) 0 m n (cid:21) =(cid:20) 0 k/l 1 (cid:21) ∈ I. 0 0 l If k = 0, then(cid:20) 0 0 0 1 (cid:21) ∈ I, whence E2 ⊆ I. Otherwise Nk/l ⊆ I. In either case, it follows by Step II that I ∈ {E2, Nk/l, S}. Case 3: m 6= 0. In this case, we have 0 (cid:21)(cid:20) k (cid:20) 0 1/m 0 (cid:21) ∈ I. 0 0 (cid:21) ∈ I, whence E1 ⊆ I. Otherwise,(cid:20) m/n 1 m n (cid:21) =(cid:20) 1 n/m 0 0 0 l If n = 0, then(cid:20) 1 0 0 (cid:21) ∈ I, whence Nm/n ⊆ I. In either case, it follows by Step II that I ∈ {E1, Nm/n, S}. Case 4: n 6= 0. In this case, we have 0 1/n (cid:21)(cid:20) k (cid:20) 0 m n (cid:21) =(cid:20) 0 m/n 1 (cid:21) . 0 0 l If m = 0, then(cid:20) 0 0 0 1 (cid:21) ∈ I, whence E2 ⊆ I. Otherwise, Nm/n ⊆ I. In either case, it follows by Step II that I ∈ {E2, Nm/n, S}.(cid:4) 20 Lemma 2.18. Every left S-semimodule is S-i-injective. Proof. Let M be a left S-semimodule, f : N → S a normal S-monomorphism, and g : N → M an S-linear map. Then f (N) is a subtractive left ideal of S, whence f (N) ∈ {0, E1, E2, S} or f (N) = Nr for some r ∈ R+\{0}. Case I: f (N) = 0. In this case, choose h = 0 : S → M. Clearly, g = h ◦ f . Case II: f (N) = S. In this case, f is an S-isomorphism. Choose h = g◦ f −1, whence g = h◦ f . Case III: f (N) = E1. In this case, there exists a unique n0 ∈ N such that 0 0 (cid:21) . f (n0) =(cid:20) 1 0 Consider the S-linear map Let n ∈ N. It follows that h : S → M, (cid:20) p q s (cid:21) 7−→(cid:20) p q s (cid:21) g(n0). r r f (n) =(cid:20) a 0 b 0 (cid:21) =(cid:20) a 0 for some a, b ∈ R+. Since f is injective, n =(cid:20) a 0 b 0 (cid:21) f (n0) = f(cid:18)(cid:20) a 0 b 0 (cid:21) n0(cid:19) b 0 (cid:21) n0. It follows that b 0 (cid:21)) h((cid:20) a 0 b 0 (cid:21) n0(cid:19) b 0 (cid:21) g(n0) = g(cid:18)(cid:20) a 0 h( f (n)) = = (cid:20) a 0 (h ◦ f )(n) = = g(n). Case IV: f (N) = E2. The proof is similar to Case III. Case V: f (N) = Nr for some r ∈ R+\{0}. In this case, there exists a unique n0 ∈ N such that 0 f (n0) =(cid:20) 1 1/r 0 (cid:21) . l m (cid:21) g(n0). l m (cid:21) 7−→(cid:20) j k h : S → M, (cid:20) j k Define For every n ∈ N, we have f (n) =(cid:20) ra a rb b (cid:21) =(cid:20) ra a rb b (cid:21) f (n0) = f(cid:18)(cid:20) ra a rb b (cid:21) n0(cid:19) 21 for some a, b ∈ R+. Since f is injective, n =(cid:20) ra a rb b (cid:21) n0 and so (h ◦ f )(n) = h( f (n)) = h(cid:18)(cid:20) ra a rb b (cid:21) g(n0) = g(cid:18)(cid:20) ra a rb b (cid:21)(cid:19) rb b (cid:21) n0(cid:19) = (cid:20) ra a = g(n).(cid:4) We are now ready to provide an example of an S-i-injective semimodule which is not S-e- injective. Example 2.19. The left S-semimodule N1 =(cid:26)(cid:20) a a b b (cid:21) a, b ∈ R+(cid:27) (17) is S-i-injective but not S-e-injective. Proof. Let ι : N1 → S be an embedding and id : N1 → N1 be the identity map. Since N1 is subtractive, ι is a normal S-monomorphism. Let h1, h2 : S → N1 with Then and h1(cid:18)(cid:20) p q r r s (cid:21)(cid:19) =(cid:20) p p (h1 ◦ ι)(cid:18)(cid:20) a a r s = r (cid:21) and h2(cid:18)(cid:20) p q s (cid:21)(cid:19) =(cid:20) q q s (cid:21) . b b (cid:21)(cid:19) = h1(cid:18)(cid:20) a a b b (cid:21)(cid:19) b b (cid:21) (cid:20) a a b b (cid:21)(cid:19) = id(cid:18)(cid:20) a a b b (cid:21)(cid:19) = h2(cid:18)(cid:20) a a b b (cid:21)(cid:19) b b (cid:21) (cid:20) a a b b (cid:21)(cid:19) . = id(cid:18)(cid:20) a a = (h2 ◦ ι)(cid:18)(cid:20) a a Suppose that there exist k1, k2 : S → N1 such that k1 ◦ ι = 0 = k2 ◦ ι and h1 + k1 = h2 + k2. Write k1(cid:18)(cid:20) 1 0 0 1 (cid:21)(cid:19) =(cid:20) l m n o (cid:21) and k2(cid:18)(cid:20) 1 0 0 1 (cid:21)(cid:19) =(cid:20) p q s (cid:21) r 22 for some k, l, m, n, o, p, q, r, s ∈ R+. Then k1(cid:18)(cid:20) a b k2(cid:18)(cid:20) a b c d (cid:21)(cid:19) = (cid:20) a b c d (cid:21)(cid:19) = (cid:20) a b c d (cid:21)(cid:20) l m n o (cid:21) c d (cid:21)(cid:20) p q s (cid:21) r for every a, b, c, d ∈ R+. It follows that 0 = (k1 ◦ ι)(cid:18)(cid:20) 1 1 0 0 (cid:21)(cid:19) =(cid:20) 1 1 0 0 (cid:21)(cid:20) l m n o (cid:21) =(cid:20) l + n m + o 0 0 (cid:21) , which implies that l = m = n = o = 0 as 0 is the only element of R+ which has additive inverse. So, 0 = (k2 ◦ ι)(cid:18)(cid:20) 1 1 0 0 (cid:21)(cid:20) p q = (cid:20) 1 1 0 0 (cid:21)(cid:19) = k2(cid:18)(cid:20) 1 1 0 0 (cid:21)(cid:19) (cid:21) , s (cid:21) = (cid:20) p + r q + s 0 0 r which implies that p = q = r = s = 0 as 0 is the only element of R+ which has additive inverse. Thus k1 = 0 = k2, a contradiction with h1 + k1 = h2 + k2 as h1 6= h2. Hence, N1 is not S-e- injective.(cid:4) The following example shows that the converse of Lemma 2.15 is not true in general. Example 2.20. Consider the short exact sequence ιE1−→ S 0 → E1 πE2−→ E2 → 0 of left S-semimodules. Then N1 is E1-e-injective and E2-e-injective but not S-e-injective. Proof. Let f : M → E1, g : M → N1 be S-linear maps where f is a normal monomorphism. If f = 0, then we are done. If f 6= 0. then f is an isomorphism as E1 is ideal-simple. Define h = g ◦ f −1. Then h ◦ f = (g ◦ f −1) ◦ f = g. If h′ : E1 → N1 is an S-linear map satisfies h′ ◦ f = g, then h′ = h′ ◦ ( f ◦ f −1) = g ◦ f −1 = h. Hence N1 is E1-e-injective. Similarly, N1 is E2-e-injective. However, N1 is not S-e-injective as shown in Example 2.19. 3 The Embedding Problem It is well-known that the category of left (right) modules over a ring R has enough injectives, i.e. every left (right) R-module M can be embedded in an injective left (right) R-module (e.g., E(M), the injective hull of M). This is true only for the left (right) semimodules over some special semirings which are not rings (e.g., the additively idempotent semirings [Wan1994], [Gol1999, Corollary 17.34]). In [Ili2016], Il'in conjectured that a semiring S has the property that every left (right) S-semimodule has an injective hull if and only if S is additively regular (i.e. for every a ∈ S, there exists some x ∈ S such that a + x + a = a). In fact, the situation over some semirings can be extremely bad: 23 Lemma 3.1. If S is an entire, cancellative, zerosumfree semiring, then the only injective left S-semimodule is {0} (cf., [Gol1999, Proposition 17.21]). Example 3.2. The category of commutative monoids (i.e., Z+-semimodules) has no non-zero injective objects. Another significant difference is that Baer's Criterion (a left module M over a ring R is in- jective if M is R-injective) is not valid for semimodules over arbitrary semirings (which are not rings). Definition 3.3. Let S be a semiring. We say that the category SSM has enough injectives (resp. enough e-injectives, enough i-injectives), if every left S-semimodule can be embedded in an injective (resp. e-injective, i-injective) left S-semimodule. Lemma 3.4. ([Ili2008, Theorem 3]) If SS satisfies the Baer's criterion and SSM has enough injectives, then S is a ring. Proposition 3.5. Let S be a semiring. If SSM has enough e-injectives, then every injective left S-semimodule is e-injective. ι ֒→ E, Proof. Let I be an injective left S-semimodule. By assumption, there is an embedding I where SE is e-injective. Since SI is injective, there exists and S-linear map π : E −→ I such that π◦ ι = idI. It follows that SI is a retract of an e-injective left S-semimodule, whence e-injective by Proposition 2.12.(cid:4) Proposition 3.6. (compare with [AIKN2018, Theorem 4.5]) Let S be an additively idempotent semiring. Then SSM has enough e-injectives, and every injective left S-semimodule is e-injective. 3.7. We define a left S-semimodule N to be divisible, if for every s ∈ S, which is not a zero divisor, there exists for every n ∈ N some mn ∈ N such that smn = n. As in the case of modules over a ring, every injective semimodule over a semiring is divisible. The proof of the following observation is similar to that in the case of modules over rings [Wis1991, 16.6]. Lemma 3.8. Every S-injective left S-semimodule is divisible. Proof. Let N be an injective left S-semimodule and n ∈ N. Let s ∈ S be a non zero-divisor. Claim: −→ S and the there exists mn ∈ N such that smn = n. Consider the canonical embedding 0 −→ Ss S-linear map ι h : Ss −→ N, ts 7→ tn. By our assumption, N is S-injective, whence there exists an S-linear map g : S −→ N such that g ◦ ι = h. Let mn := g(1S). Then we have n = h(s) = (g ◦ ι)(s) = g(s) = g(s · 1S) = sg(1S) = smn.(cid:4) The converse of Lemma 3.8 is not true in general as the following example shows. 24 Example 3.9. Q is a divisible commutative monoid which is not injective. 3.10. Let R be a ring. Every left R-module can be embedded in an injective module HomZ(R, D), (cf., [Gri2007, page 407, 421]). For a semiring S, we prove that every left S-semimodule can be embedded into HomZ+(S, D) for some divisible commutative monoid D. However, it is unknown whether HomZ+(S, D) is necessarily e-injective. Lemma 3.11. Every commutative monoid can be embedded subtractively in a divisible commu- tative monoid. Proof. Let B be a commutative monoid. Then there exists a surjective morphism of monoids f : Z+(Λ) → B for some index set Λ. Let g be the embedding of Z+(Λ) into Q+(Λ). Let (g′, f ′; P) be a pushout of ( f , g) (see [AN, Theorem 2.3, Corollary 2.4]). Q+(Λ) g Z+(Λ) f ′ f / P g′ / B Notice that g′ is subtractive since g is subtractive. Moreover, the commutative monoid P is λ)Λ ∈ Q+ such that p = f ′((qλ)Λ) divisible since for every n ∈ Z+ and p ∈ P we have (qλ)Λ, (q′ and nq′ λ = qλ. Thus n f ′((q′ λ)Λ) = f ′((nq′ λ)Λ) = f ′((qλ)Λ). Let C := {q ∈ Q+0 ≤ q < 1}. Then B ⊕C(Λ) is a commutative monoid with (b, (cλ)) + (b′, (c′ λ⌋)). λ)) = (b + b′ + f ((⌊cλ + c′ λ⌋)Λ), (cλ + c′ λ − ⌊cλ + c′ B ⊕C(Λ) ① <① ϕ ① ① g∗ f ∗ ① P g′ / B Q+(Λ) g Z+(Λ) f ′ f g∗ : B −→ B ⊕C(Λ), b 7→ (b, 0) The map is a Z+-monomorphism. The map f ∗ : Q+(Λ) −→ B ⊕C(Λ), (qλ) 7→ ( f ((⌊qλ⌋)Λ), (qλ − ⌊qλ⌋)Λ) is a Z+-homomorphism. Since f ∗ ◦g = g∗ ◦ f , there exists, by the Universal Property of Pushouts, a unique map ϕ : P → B ⊕C(Λ) such that ϕ◦ f ′ = f ∗ and ϕ◦ g′ = g∗. Since g∗ is injective, g′ is in- jective. Hence g′ : B −→ P is a normal Z+-monomorphism from B into the divisible commutative monoid P.(cid:4) 25 / O O / O O / / 1 1 < O O / O O L L Lemma 3.12. Every left S-semimodule can be embedded into HomZ+(S, D) for some divisible commutative monoid D. Proof. Let M be a left S-semimodule. By Lemma 3.11 there exists a normal monomorphism of commutative monoids µ : M → D for some divisible commutative monoid D. Consider the canonical S-linear map ε : M −→ HomZ+(S, D), m 7→ [s 7→ µ(sm)]. Suppose that ε(m) = ε(m′) for some m, m′ ∈ M. Then, in particular, ε(m)(1S) = ε(m′)(1S), i.e. µ(m) = µ(m′). Since µ is injective, we conclude that m = m′.(cid:4) The embedding into an injective R-module (where R is a ring) implies a nice result in the cate- gory of R-modules: an R-module P is projective if and only if P is J-projective for every injective R-module J [Gri2007, page 411]. For semimodules, we have so far the following implication. Proposition 3.13. Let γ : T −→ S be a morphism of semirings and M a left S-semimodule. If T A is T M-i-injective, then HomT (S, A) is SM-i-injective. Proof. Let ι : K → M be a normal S-monomorphism and f : K → HomT (SS, A) an S-linear map. 0 ι M / K f HomT (SS, A) Recall the canonical isomorphism of commutative monoids HomS(K, HomT (SS, A)) θK,A ≃ HomT (K, A), f 7→ [k 7→ f (k)(1S)]. Consider the T -linear map θK,A( f ) : K −→ A. 0 / K θK,A( f ) ~⑦ A M ⑦ ι ⑦ h ⑦ Since ι : K → M is also a normal T -monomorphism and T A is M-i-injective, there exists a T - linear map h : M −→ A such that h ◦ ι = θK,A( f ). Notice that θ−1 M,A(h) : M → HomT (SS, A) is S-linear and that for all k ∈ K and every s ∈ S we have ((θ−1 M,A(h) ◦ ι)(k))(s) = θ−1 h(sι(k)) = θK,A( f )(sk) = (s f (k))(1S) = f (k)(s). M,A(h)(sι(k)) = (h ◦ ι)(sk) f (sk)(1S) f (k)(1S · s) = = = Hence, HomT (SS, A) is M-i-injective as a left S-semimodule.(cid:4) 26 / / / / / / ~ The following result is a combination of Proposition 3.13 and [AIKN2018, Corollary 3.5]. Corollary 3.14. Let γ : T −→ S be a morphism of semirings. The functor preserves injective, e-injective and i-injective objects. HomT (SS, −) : T SM −→ SSM Lemma 3.15. Every divisible commutative monoid is Z+-i-injective. Proof. Let D be a divisible commutative monoid, f : I → Z+ a normal monomorphism of com- mutative monoids and g : I → D a morphism of commutative monoids. Since f (I) is subtractive, f (I) = kZ+ for some k ∈ Z+. Let i0 ∈ I be such that f (i0) = k and notice that i0 is unique as f is injective. By our choice, D is divisible and so there exists d ∈ D such that kd = g(i0). The map h : Z+ → D, n 7→ nd is a well-defined morphism of monoids. Moreover, for every i ∈ I, we have f (i) = nk for some n ∈ Z+ whence i = ni0 as f is injective. It follows that f (i) = f (ni0) = n f (i0) = nk, and so (h ◦ f )(i) = h(nk) = h(n)k = ndk = ng(i0) = g(ni0) = g(i). It follows that h f = g.(cid:4) Definition 3.16. We say that a left S-semimodule I is c-injective (resp. c-e-injective, c-i- injective), if I is M-injective (resp., M-e-injective, M-i-injective) for every cancellative left S- semimodule M. Proposition 3.17. Every divisible commutative monoid is c-i-injective. Proof. Let D be a divisible commutative monoid, N a cancellative left S-semimodule, f : M → N a normal Z+-monomorphism and g : M → J a morphism of commutative monoids. 0 / M f N g J Define S = {(A,α) : A ≤Z+ N, M ⊆ A, α : A → J with α(m) = g(m) ∀ m ∈ M}. Notice that S is not empty, since (M, g) ∈ S . Define an order on S as follows: (A,α) ≤ (B,β) ⇔ A ⊆ B and β(a) = α(a) ∀a ∈ A. Let ((Aλ,αλ))Λ be a chain in S . Set A := Sλ∈Λ Aλ and define α : A → J such that, if x ∈ Aλ, then α(x) = αλ(x). Notice that α is well-defined, thus the chain has an upper bound (A,α). By Zorn's Lemma, S has a maximal element (C,γ). 27 / / / Claim: If A 6= N, then (A,α) ∈ S is not maximal. Let (A,α) ∈ S with A $ N. Choose b ∈ N\A and set B := A + Z+b. Notice that L := {r ∈ Z+ rb ∈ A} is an ideal of Z+ and κ : L −→ J, r 7→ α(rb) is a morphism of monoids. By Lemma 3.15 there exists a morphism of monoids χ : Z+ → J such that χ(r) = α(rb) ∀ r ∈ L. Define β : B → J, a + rb 7→ α(a) + χ(r). We claim that β is well-defined. Suppose that a + rb = a′ + r′b for some r ∈ L and a ∈ A. Assume, without loss of generality, that r′ > r, whence r′ = r + r for some r ∈ Z+. It follows that a + rb = a′ + r′b = a′ + rb + rb, whence a = a′ + rb as N is cancellative. It follows that β(a′ + r′b) = β((a′ + rb) + rb) = α( rb + a′) + χ(r) = α(a) + χ(r) = β(a + rb). Thus β is well-defined as morphism of monoids with β(a) = α(a) ∀ a ∈ A. Thus (A,α) is not maximal in . It follows that there exists a morphism of monoids h : N −→ J such that (N, h) is maximal in S . Clearly, h : N → J such that h ◦ f = g.(cid:4) The following result is, in some sense, a generalization of the fact (mentioned without proof in [Gol1999, 17.35]) that any cancellative semimodule over semiring can be embedded in a c- injective module. While c-i-injectivity is formally weaker than c-injectivity, our result works for arbitrary, not necessarily cancellative, semimodules over semirings. Theorem 3.18. Every left S-semimodule can be embedded as a subtractive subsemimodule of a c-i-injective left S-semimodule. Proof. Let M be a left S-semimodule. By Lemma 3.12, M can be embedded as a subtractive subsemimodule of the left S-semimodule HomZ+(S, D) for some divisible commutative monoid D. Let N be a cancellative left S-semimodule; then N is, in particular, a cancellative commutative monoid. By Proposition 3.17, D is an N-i-injective Z+-semimodule, whence HomZ+(S, D) is N-i-injective by Proposition 3.13.(cid:4) The following examples shows one of the advantages of Theorem 3.18. Example 3.19. Let S be an entire, cancellative, zerosumfree semiring. By Theorem 3.18, every left S-semimodule L can be embedded subtractively in a c-i-injective left S-semimodule. On the other hand, if L 6= 0, then L cannot be embedded in an injective S-semimodule since the only injective left S-semimodule is {0} (by Lemma 3.1). This is the case, in particular, for S := Z+. References [AN] J. Abuhlail and R. G. Noegraha, Pushouts and e-Projective Semimodules, sub- mitted. (available at: https://arxiv.org/abs/1904.01549) 4, 5, 25 28 [Abu2014] J. Abuhlail, Exact sequence of commutative monoids and semimodules, Homol- ogy Homotopy Appl. 16 (1) (2014), 199 -- 214. 2, 5, 8 [Abu2014-CA] J. Abuhlail, Semicorings and semicomodules, Commun. Alg. 42(11) (2014), 4801 -- 4838. 2, 8, 9 [Abu2014-SF] J. Abuhlail, Some remarks on tensor products and flatness of semimodules, Semigroup Forum 88(3) (2014) 732 -- 738. 2 [AIKN2015] J. Abuhlail , S. Il'in , Y. Katsov, and T. Nam, On V-semirings and semirings all of whose cyclic semimodules are injective, Commun. Alg. 43 (11) (2015), 4632 -- 4654. 2 [AIKN2018] J. Abuhlail, S. Il'in , Y. Katsov, and T. Nam, Toward homological characteriza- tion of semirings by e-injective semimodules, J. Algeb. Appl. 17(4) (2018). 2, 8, 13, 14, 24, 27 [AHS2004] J. Ad´amek, H. Herrlich and G. E. Strecker, Abstract and Concrete Cate- gories; The Joy of Cats 2004. Dover Publications Edition (2009) (available at: http://katmat.math.uni-bremen.de/acc). 4 [Alt2003] [Bor1994] H. Al-Thani, Injective semimodules, J. Inst. Math. Comput. Sci. 16 (3), 143-152 (2003). 2, 8, 9, 10, 12, 13, 14 F. Borceux, Handbook of Categorical Algebra. I, Basic Category Theory, Cam- bridge Univ. Press (1994). 6 [BHJK2001] R. El Bashir, J. Hurt, A. Jancarik, and T. Kepka, Simple commutative semirings, J. Algebra, 236 (2001), 277 - 306. 4 [Gla2002] [Gol1999] [Gri2007] [HW1998] K. Głazek, A Guide to the Literature on Semirings and their Applications in Mathematics and Information Sciences. With Complete Bibliography, Kluwer Academic Publishers, Dordrecht (2002). 1, 2 J. Golan, Semirings and their Applications, Kluwer Academic Publishers, Dor- drecht (1999). 1, 3, 4, 8, 9, 23, 24, 28 P. A. Grillet, Abstract Algebra, Second Edition, Springer, New York (2007). 25, 26 U. Hebisch and H. J. Weinert, Semirings: Algebraic Theory and Applications in Computer Science, World Scientific Publishing Co., Inc., River Edge, NJ (1998). 1 [Ili2008] S. N. Il'in, On the applicability of two theorems from the theory of rings and modules to semirings, Math. Notes 83 (3-4) (2008), 492-499. 2, 24 29 [Ili2010] [Ili2016] [IKN2017] S. N. Il'in, Direct sums of injective semimodules and direct products of projec- tive semimodules over semirings; Russian Math. 54 (10) (2010), 27 -- 37. 2 S. N. Il'in, On injective envelopes of semimodules over semirings; J. Alg. Appl. 15 (7) (2016), 27-37. 23 S. Il'in , Y. Katsov, and T. Nam, Toward homological structure theory of semi- modules: On semirings all of whose cyclic semimodules are projective, J. Alge- bra 476 (2017), 238-266. 2 [KN2011] Y. Katsov, T. G. Nam, Morita equivalence and homological characterization of rings, J. Alg. Appl. 10 (3), 445-473 (2011). 2, 7 [KNT2009] Y. Katsov, T. G. Nam, N. X. Tuyen, On subtractive semisimple semirings, Alge- bra Colloq.16 (3) (2009), 415-426. 2 [KNZ2014] Y. Katsov; T. Nam and J. Zumbragel, On simpleness of semirings and complete semirings, J. Algebra Appl. 13 (6) (2014), 29 pages. 2 [LM2005] [Pat2003] [PD2006] G. L. Litvinov and V. P. Maslov (editors), Idempotent Mathematics and Mathe- matical Physics, Papers from the International Workshop held in Vienna, Febru- ary 3 -- 10, 2003. Contemporary Mathematics, 377. American Mathematical So- ciety, Providence, RI (2005). 1 A. Patchkoria, Extensions of semimodules and the Takahashi functor ExtΛ(C, A), Homology Homotopy Appl. 5 (1) (2003), 387-406. 5 K. B. Patil and R. P. Deore, Some results on semirings and semimodules, Bull. Calcutta Math. Soc. 98(1) (2006), 49-56. 5 [Tak1982b] M. Takahashi, Extensions of Semimodules I, Math. Sem. Notes Kobe Univ. 10 (1982), 563 -- 592. 5, 17 [Tak1981] M. Takahashi, On the bordism categories. II. Elementary properties of semi- modules. Math. Sem. Notes Kobe Univ. 9 (2) (1981), 495-530. 4, 5 [Tak1982a] M. Takahashi, On the bordism categories. III. Functors Hom and for semimod- ules. Math. Sem. Notes Kobe Univ. 10 (2) (1982), pp. 551-562. 7 [Wan1994] Huaxiong Wang, Injective hulls of semimodules over additively-idempotent semirings, Semig. Forum 48 (1994), 377-379. 23 [Wis1991] R. Wisbauer, Foundations of Module and Ring Theory, Gordon and Breach, Reading (1991). 1, 3, 24 30
1601.00777	3	1601	2017-11-23T09:11:38	$*$-isomorphism of Leavitt path algebras over $\mathbb{Z}$	[ "math.RA", "math.OA" ]	We characterise when the Leavitt path algebras over $\mathbb{Z}$ of two arbitrary countable directed graphs are $$-isomorphic by showing that two Leavitt path algebras over $\mathbb{Z}$ are $$-isomorphic if and only if the corresponding graph groupoids are isomorphic (if and only if there is a diagonal preserving isomorphism between the corresponding graph $C^$-algebras). We also prove that any $$-homomorphism between two Leavitt path algebras over $\mathbb{Z}$ maps the diagonal to the diagonal. Both results hold for slight more general subrings of $\mathbb{C}$ than just $\mathbb{Z}$.	math.RA	math	∗-ISOMORPHISM OF LEAVITT PATH ALGEBRAS OVER Z TOKE MEIER CARLSEN ABSTRACT. We characterise when the Leavitt path algebras over Z of two arbitrary countable directed graphs are ∗-isomorphic by showing that two Leavitt path algebras over Z are ∗-isomorphic if and only if the corresponding graph groupoids are isomor- phic (if and only if there is a diagonal preserving isomorphism between the correspond- ing graph C∗-algebras). We also prove that any ∗-homomorphism between two Leavitt path algebras over Z maps the diagonal to the diagonal. Both results hold for more general subrings of C than just Z. Keywords: Leavitt path algebras, graph groupoids, graph C∗-algebras. 1. INTRODUCTION Leavitt path algebras were introduced independently in [2] and [5] as algebraic ana- logues of graph C∗-algebras and have since then attracted a lot of attention, both in connection with graph C∗-algebras and as interesting algebraic objects on their own (they are called Leavitt path algebras because they generalise certain algebras studied by Leavitt in [16, 17, 18]). The Leavitt path algebra of a directed graph E over a unital commutative ring R is a universal R-algebra LR(E) whose generators and relations are determined by E (see Section 2.3 for the precise definition of LR(E)). Each involution (for example the identity map) of R gives rise to an involution of LR(E) which is therefore a ∗-algebra. It is natural to ask when two Leavitt path algebras are (∗-)isomorphic. Abrams and Tomforde showed in [3] that if two Leavitt path algebras LC(E) and LC(F) over C are ∗-isomorphic, then so are the corresponding graph C∗-algebras C∗(E) and C∗(F) (their proof is easily generalised to subrings of C that are closed under complex conjugation and contains 1). Johansen and Sørensen gave in [14] the first example of two Leavitt path algebras that are not ∗-isomorphic in spite of the corresponding graph C∗-algebras being isomor- phic, when they showed that the Leavitt path algebras over Z of E2 and E2− are not ∗-isomorphic (that C∗(E2) and C∗(E2−) are isomorphic was proved by Rørdam in [20] as an important step towards classifying simple Cuntz-Krieger algebras). Each Leavitt path algebra LR(E) contains a certain abelian subalgebra DR(E) called the diagonal. Johansen and Sørensen obtained their result by showing that when E is a finite graph and R is a subring of C satisfying certain conditions, then every projection in LR(E) belongs to DR(E). We generalise this result to arbitrary graphs E and more Date: September 24, 2018. 1 ∗-ISOMORPHISM OF LEAVITT PATH ALGEBRAS OVER Z 2 general subrings R of C (Proposition 4). It follows than any ∗-homomorphism between two Leavitt path algebras LZ(E) and LZ(F) over Z is diagonal preserving in the sense that it maps DZ(E) to DZ(F) (Corollary 5). From a countable directed graph E, one can construct a topological groupoid GE such that the C∗-algebra of GE is isomorphic to the graph C∗-algebra C∗(E), and the Steinberg algebra AR(GE) is isomorphic to LR(E). It is proved in [8] that two graph C∗-algebras C∗(E) and C∗(F) are isomorphic in a diagonal preserving way if and only the corresponding graph groupoids GE and GF are isomorphic, and it is proved in [21] (see also [4], [7], and [10]) that if R is indecomposable and reduced (in particular if R is a integral domain), then there is a diagonal preserving isomorphism between LR(E) and LR(F) if and only if GE and GF are isomorphic. By using Corollary 5, we show in Theorem 1 that two Leavitt path algebras LZ(E) and LZ(F) of countable graphs are ∗-isomorphic if and only if the groupoids GE and GF are isomorphic (if and only if there is a diagonal preserving isomorphism between the graph C∗-algebras C∗(E) and C∗(F)). As is the case with Proposition 4 and Corollary 5, the result in Theorem 1 holds for more general subrings of C than just Z. The rest of the paper is organised in the following way. In Section 2 we recall the definitions of directed graphs, the Leavitt path algebras, graph C∗-algebras, and graph groupoids; and introduce notation. In Section 3 we present our main result (Theorem 1) and discuss how it is related to orbit equivalence of graphs and results in [4], [6], [7], [8], [10], [12], and [21] (Remarks 2 and 3). We also ask the question if there exist Leav- itt path algebras that are isomorphic without the corresponding graph groupoid being isomorphic. If the answer to this question is "No", then both The Isomorphism Conjec- ture for Graph Algebras and The Morita Conjecture for Graph Algebras introduced in [3] are true. Finally we present and prove in Section 4, Proposition 4 and Corollary 5 before we give the proof of Theorem 1. 2. DEFINITIONS AND NOTATION We recall in this section the definition of a directed graph, as well as the definitions of the Leavitt path algebra, the graph C∗-algebra, and the graph groupoid of a graph; and introduce some notation. Most of this section is copied from [8]. 2.1. Directed graphs. A directed graph is a quadruple E = (E0, E1, s, r) where E0 and E1 are sets, and s and r are maps from E1 to E0. A graph E is said to be countable if E0 and E1 are countable. A path µ of length n in E is a sequence of edges µ = µ1 . . .µn such that r(µi) = s(µi+1) for 1 ≤ i ≤ n− 1. The set of paths of length n is denoted E n. We denote by µ the length of µ. The range and source maps extend naturally to paths: s(µ) := s(µ1) and r(µ) := r(µn). We regard the elements of E0 as path of length 0, and for v ∈ E0 we set s(v) := r(v) := v. For v ∈ E0 and n ∈ N0 we denote by vE n the set of paths of length n with source v. We define E∗ :=Sn∈N0 E n to be the collection of all paths with finite length. We define E0 sing := E0 \ E0 reg. If µ = µ1µ2···µm,ν = ν1ν2···νn ∈ E∗ and r(µ) = s(ν), then we let µν denote the reg := {v ∈ E0 : vE1 is finite and nonempty} and E0 ∗-ISOMORPHISM OF LEAVITT PATH ALGEBRAS OVER Z 3 path µ1µ2···µmν1ν2···νn. A loop (also called a cycle) in E is a path µ ∈ E∗ such that µ ≥ 1 and s(µ) = r(µ). An edge e is an exit to the loop µ if there exists i such that s(e) = s(µi) and e 6= µi. A graph is said to satisfy condition (L) if every loop has an exit. An infinite path in E is an infinite sequence x1x2 . . . of edges in E such that r(ei) = s(ei+1) for all i. We let E∞ be the set of all infinite paths in E. The source map extends to E∞ in the obvious way. We let x = ∞ for x ∈ E∞. The boundary path space of E is the space ∂E := E∞ ∪{µ ∈ E∗ : r(µ) ∈ E0 sing}. If µ = µ1µ2···µm ∈ E∗, x = x1x2··· ∈ E∞ and r(µ) = s(x), then we let µx denote the infinite path µ1µ2···µmx1x2··· ∈ E∞. For µ ∈ E∗, the cylinder set of µ is the set Z(µ) := {µx ∈ ∂E : x ∈ r(µ)∂E}, where r(µ)∂E := {x ∈ ∂E : r(µ) = s(x)}. Given µ∈ E∗ and a finite subset F ⊆ r(µ)E1 we define Z(µ\ F) := Z(µ)\ [e∈F Z(µe)! . The boundary path space ∂E is a locally compact Hausdorff space with the topology given by the basis {Z(µ\ F) : µ ∈ E∗, F is a finite subset of r(µ)E1}, and each such Z(µ\ F) is compact and open (see [23, Theorem 2.1 and Theorem 2.2]). 2.2. Graph C∗-algebras. The graph C∗-algebra of a directed graph E is the univer- sal C∗-algebra C∗(E) generated by mutually orthogonal projections {pv : v ∈ E0} and partial isometries {se : e ∈ E1} satisfying (CK1) s∗ese = pr(e) for all e ∈ E1; (CK2) ses∗e ≤ ps(e) for all e ∈ E1; ses∗e for all v ∈ E0 (CK3) pv = ∑ reg. e∈vE1 If µ = µ1···µn ∈ E n and n ≥ 2, then we let sµ := sµ1 ··· sµn. Likewise, we let sv := pv if v ∈ E0. Then spanC{sµs∗ν : µ,ν ∈ E∗, r(µ) = r(ν)} is dense in C∗(E). We define D(E) to be the closure in C∗(E) of spanC{sµs∗µ : µ∈ E∗}. Then D(E) is an abelian C∗- subalgebra of C∗(E), and it is isomorphic to the C∗-algebra C0(∂E). We furthermore have that D(E) is a maximal abelian sub-algebra of C∗(E) if and only if E satisfies condition (L) (see [19, Example 3.3]). 2.3. Leavitt path algebras. Let E be a directed graph and R a commutative ring with a unit. The Leavitt path algebra of E over R is the universal R-algebra LR(E) generated by pairwise orthogonal idempotents {v : v ∈ E0} and elements {e, e∗ : e ∈ E1} satisfying (LP1) e∗ f = 0 if e 6= f ; (LP2) e∗e = r(e); (LP3) s(e)e = e = er(e); (LP4) e∗s(e) = e∗ = r(e)e∗; ∗-ISOMORPHISM OF LEAVITT PATH ALGEBRAS OVER Z 4 (LP5) v = ∑e∈vE1 ee∗ if v ∈ E0 reg. If µ = µ1···µn ∈ E n and n ≥ 2, then we let µ be the element µ1···µn ∈ LR(E) and µ∗ the element µ∗n ···µ∗1 ∈ LR(E). For v∈ E0, we let v∗ := v. Then LR(E) = spanR{µν∗ : µ,ν∈ E∗, r(µ) = r(ν)}. There is a Z-grading Ln∈Z LR(E)n of LR(E) given by LR(E)n = spanR{µν∗ : µ,ν ∈ E∗, r(µ) = r(ν), µ−ν = n} (see [22, Section 3.3]). We define DR(E) := spanR{µµ∗ : µ ∈ E∗}. Then DR(E) is an abelian subalgebra of LR(E), and it is maximal abelian if and only if E satisfies condition (L) (see [9, Proposition 3.14 and Theorem 3.22]). If R is a subring of C that is closed under complex conjugation and contains 1, then µν∗ 7→ νµ∗ extends to a conjugate linear involution on LR(E), i.e. LR(E) is a ∗-algebra. There is an injective ∗-homomorphism ιLR(E) → C∗(E) mapping v to pv and e to se for v ∈ E0 and e ∈ E1 (see [22, Theorem 7.3]). 2.4. Graph groupoids. Let E be a directed graph. For n ∈ N0, let ∂E≥n := {x ∈ ∂E : x ≥ n}. Then ∂E≥n = ∪µ∈E n Z(µ) is an open subset of ∂E. We define the shift map on E to be the map σE : ∂E≥1 → ∂E given by σE(x1x2x3··· ) = x2x3··· for x1x2x3··· ∈ ∂E≥2 and σE (e) = r(e) for e ∈ ∂E ∩ E1. For n ≥ 1, we let σn E be the n-fold composition of σE with itself. We let σ0 E is a local homeomorphism for all n ∈ N. When we write σn E (x), we implicitly assume that x ∈ ∂E≥n. étale topological groupoid The graph groupoid of a countable directed graph is the locally compact, Hausdorff, E denote the identity map on ∂E. Then σn GE = {(x, m− n, y) : x, y ∈ ∂E, m, n ∈ N0, and σm(x) = σn(y)}, with product (x, k, y)(w, l, z) := (x, k + l, z) if y = w and undefined otherwise, and inverse given by (x, k, y)−1 := (y,−k, x). The topology of GE is generated by subsets of the form E (x) = σn Z(U, m, n,V ) := {(x, k, y) ∈ GE : x ∈ U, k = m− n, y ∈ V, σm E(y)} where m, n ∈ N0, U is an open subset of ∂E≥m such that the restriction of σm E to U is injective, and V is an open subset of ∂E≥n such that the restriction of σn E to V is injective, and σm E(V ). The map x 7→ (x, 0, x) is a homeomorphism from ∂E to the unit space G 0 E of GE. There is a ∗-isomorphism from the C∗-algebra of GE to C∗(E) that maps C0(G 0 E ) onto D(E) (see [8, Proposition 2.2] and [15, Proposition 4.1]), and a ∗-isomorphism from the Steinberg algebra AR(GE) of GE to LR(E) that maps spanR{1Z(Z(µ),0,0,Z(µ)) : µ ∈ E∗} onto DR(E) (see [7, Theorem 2.2] and [13, Example 3.2]). E (U ) = σn 3. THE MAIN RESULT We say that a subring R of C that is closed under complex conjugation and contains i=0λi2, then λ1 = ··· = λn = 0. 1 is kind if whenever λ0,λ1, . . . ,λn ∈ R satisfy λ0 = ∑n Notice that if a subring R of C is closed under complex conjugation and contains 1 and has an essentially unique partition of the unit as defined in [14], then it is kind i=0λi2 = 1). In particular, Z (because if λ0 = ∑n i=0λi2, then λ0 − 12 + ∑n i=1λi2 + ∑n ∗-ISOMORPHISM OF LEAVITT PATH ALGEBRAS OVER Z 5 is kind. The subring of C generated by 1,π−1 and √1− π−2 is an example of a kind subring that does not have an essentially unique partition of the unit. Theorem 1. Let E and F be countable directed graphs, and let R and S be subrings of C that are closed under complex conjugation and contain 1, and assume that R is kind. Then the following are equivalent. DS(F). algebras. (1) The Leavitt path algebras LR(E) and LR(F) of E and F are isomorphic as ∗- (2) There is a ∗-algebra isomorphism π : LS(E) → LS(F) such that π(DS(E)) = (3) There is a ∗-isomorphism φ : C∗(E) → C∗(F) such that φ(D(E)) = D(F). (4) The graph groupoids GE and GF are isomorphic as topological groupoids. Notice that if R = Z, then any ∗-ring isomorphism between LR(E) and LR(F) is auto- matically a ∗-algebra isomorphism. The proof of Theorem 1 is given in the next section. Remark 2. Let PE and PF be the pseudogroups introduced in [8, Section 3]. It is shown in [6, Theorem 5.3] that Conditions (3) and (4) each are equivalent to the follow- ing two conditions. odic points such that {h◦ α◦ h−1 : α ∈ PE} = PF. isolated eventually periodic points. (5) There is a homeomorphism h : ∂E → ∂F that preservs isolated eventually peri- (6) There is an orbit equivalence h : ∂E → ∂F as in [8, Definition 3.1] that preservs It follows from [12, Corollary 4.6] and the discussion right before [12, Proposition 3.1] that if either E and F each satisfy condition (L), or E and F each have only finitely many vertices and no sinks, then any homeomorphism h : ∂E → ∂F automatically preserves isolated eventually periodic points. Remark 3. Suppose T is a commutative ring with a unit and an involution t 7→ t that fixes the unit and the zero element (this could for instance be the identity map). Then there is involution on LT (E) given by tµν∗ 7→ tνµ∗ for t ∈ T and µ,ν ∈ E∗. Thus LT (E) is a ∗-algebra. Condition (4) implies the following condition (see [10, Theorem 4.1]). (7) There is a ∗-algebra isomorphism η : LT (E) → LT (F) such that η(DT (E)) = DT (F). Obviously, condition (7) implies the following condition. (8) There is a ring isomorphism ζ : LT (E) → LT (F) such that ζ(DT (E)) = DT (F). It is shown in [21, Theorem 6.1] (see also [4, Corollary 4.4], [7, Theorem 6.2], and [10, Corollary 4.1]) that if T is indecomposable and either E satisfies condition (L), or T is reduced, then (8) implies (4) and (1) -- (8) are all equivalent (notice that if T is an integral domain, then it is indecomposable and reduced). In the light of Theorem 1 and the above remarks, the following question seems natu- ral. ∗-ISOMORPHISM OF LEAVITT PATH ALGEBRAS OVER Z 6 Question. Do there exist directed graphs E and F and a commutative ring R with a unit such that LR(E) and LR(F) are isomorphic as rings, but GE and GF are not isomorphic? It follows from Theorem 1, [11, Theorem 4.2], and Theorem 5 and part 2 of the remarks following Corollary 7 in [1] that if the answer to the above question is "No", then both The Isomorphism Conjecture for Graph Algebras and The Morita Conjecture for Graph Algebras introduced in [3] are true (we cannot rule out the possibility that the conjectures are true even if the answer to the above question is "Yes"). 4. PROOF OF THE MAIN RESULT Let E be a directed graph and R a subring of C that is closed under complex conjuga- tion and contains 1. As in [14], we say that p ∈ LR(E) is a projection if p = p∗ = p2. For the proof of Theorem 1 we need the following generalisation of [14, Theorem 5.6]. Proposition 4. Let E be a directed graph and let R be a subring of C that is closed under complex conjugation, contains 1 and is kind. If p ∈ LR(E) is a projection, then p ∈ DR(E). Proof. This proof is inspired by the proof of [14, Proposition 4.4]. For µ,ν ∈ E∗, we shall write µ ≤ ν to indicate that there is an η ∈ E∗ such that Since LR(E) = spanZ{αβ∗ : α,β ∈ E∗}, it follows that there are finite subsets A, B µη = ν, and µ < ν to indicate that µ ≤ ν and µ 6= ν. of E∗ and a family (λ(α,β))(α,β)∈A×B of elements of R such that p = ∑ λ(α,β)αβ∗. (α,β)∈A×B By repeatedly replacing αβ∗ by ∑e∈r(α)E1 αee∗β∗ if necessary, we can assume that there is a k such that B ⊆ E k ∪{µ ∈ E∗ : µ < k and r(µ) ∈ E0 sing}. We can also, by letting some of the λ(α,β) be 0 if necessary, assume that B ⊆ A. Notice that αβ∗ = 0 unless r(α) = r(β). For β ∈ B, let Aβ := {α ∈ A : r(α) = r(β)}. We shall also assume that if β ∈ B, then there is a least one α ∈ Aβ such that λ(α,β) 6= 0 (otherwise we just remove β from B). We claim that λ(α,β) = 0 for all (α,β) ∈ A× B with α ∈ Aβ \{β}, and that λ(β,β) = (−1)mβ for all β ∈ B where mβ is the number of β′s in B such that β′ < β. Let B′ = {β ∈ B : λ(α,β) = 0 for all α∈ Aβ\{β} and λ(β,β) = (−1)mβ}, and suppose B′ 6= B. Choose β ∈ B\ B′ such that β′ < β for no β′ ∈ B\ B′. Let Fβ = {e ∈ r(β)E1 : βe ≤ β′ for some β′ ∈ B\{β}} and γβ = β− β ∑ e∈Fβ ee∗ (Fβ = /0 and γβ = β unless β < k and r(β)E1 is infinite). Then β′∗γβ = 0 for β′ ∈ B unless β′ ≤ β. ∗-ISOMORPHISM OF LEAVITT PATH ALGEBRAS OVER Z 7 Since p = p∗ p, it follows that (a) Recall that LR(E) is Z-graded. The degree 0 part of the left-hand side of (a) is γ∗β pγβ = γ∗β p∗ pγβ. (b) λ(β′,β′) r(β)− ∑ e∈Fβ ee∗! ∑ β′∈B≤β where B≤β := {β′ ∈ B : β′ ≤ β}, and the degree 0 part of the right-hand side of (a) is λ(β′,β′)λ(β,β) ∑ β′∈B<β (c) λ(β′,β′)! ∑ λ(β′,β′)! + ∑ β′∈B<β β′∈B<β β′∈B<β λ(α,β)2!(cid:18)r(β)− ∑ e∈Fβ + ∑ λ(β′,β′)λ(β,β) + ∑ α∈Aβ ee∗(cid:19) where B<β := {β′ ∈ B : β′ < β} (we are using here that λ(α,β′) = 0 for β′ ∈ B<β and α ∈ A\{β′}). Suppose mβ is even. Then ∑β′∈B<β λ(β′,β′) = 0 (because λ(β′,β′) = (−1)mβ′ for each β′ ∈ B<β). Since (b) = (c), it follows that λ(β,β) = ∑α∈Aβ λ(α,β)2. Since R is kind, it follows that λ(α,β) = 0 for α ∈ Aβ \{β} and λ(β,β) = 1 (recall that λ(α,β) 6= 0 for at least one α ∈ Aβ), but this contradicts the assumption that β /∈ B′. If mβ is uneven, then ∑β′∈B<β λ(β′,β′) = 1, so it follows from the equality of (b) and (c) that 1 + λ(β,β) + λ(β,β) + ∑α∈Aβ λ(α,β)2 = 1 + λ(β,β) from which we deduce that λ(α,β) = 0 for α ∈ Aβ \{β} and λ(β,β) = −1, and thus that β ∈ B′. So we also reach a contradiction in this case. We conclude that we must have that B′ = B, and thus that λ(α,β) = 0 for all (α,β) ∈ A×B with α∈ Aβ\{β}. Since αβ∗ = 0 for α /∈ Aβ, it follows that p = ∑β∈B λ(β,β)ββ∗ ∈ DR(E). (cid:3) Corollary 5. Let E and F be directed graphs, and let R be a subring of C that is closed under complex conjugation, contains 1 and is kind. If π : LR(E) → LR(F) is a ∗-algebra homomorphism, then π(DR(E)) ⊆ DR(F). Proof. Follows from Proposition 4 and [14, Proposition 6.1]. (cid:3) Proof of Theorem 1. The equivalence of (3) and (4) is proved in [8], and that (4) implies (1) and (2) follows from [13, Example 3.2]. We shall prove (1) =⇒ (3) and (2) =⇒ (3). (2) =⇒ (3): We shall closely follow the proof of [14, Lemma 3.5]. Suppose π : LS(E) → LS(F) is a ∗-algebra isomorphism such that π(DS(E)) = DS(F). As in the proof of [3, Theorem 4.4], π extends to a ∗-isomorphism φ : C∗(E) → C∗(F) satisfying φ◦ ιLS(E) = ιLS(F) ◦ π. If µ ∈ E∗, then φ(sµs∗µ) = φ(ιLS(E)(µµ∗)) = ιLS(F)(π(µµ∗)) ∈ ιLS(F)(DS(F)) ⊆ D(F). ∗-ISOMORPHISM OF LEAVITT PATH ALGEBRAS OVER Z 8 Since D(E) is generated by {sµs∗µ : µ ∈ E∗}, it follows that φ(D(E)) ⊆ D(F). That φ−1(D(F)) ⊆ D(E) follows in a similarly way. Thus φ(D(E)) = D(F). (1) =⇒ (3): Suppose π : LR(E) → LR(F) is a ∗-algebra isomorphism. It follows from Corollary 5 that π(DR(E)) = DR(F), so an application of the implication (2) =⇒ (3) with S = R shows that (3) holds. (cid:3) REFERENCES [1] G. Abrams, P.N. Áhn and L. Márki, A topological approach to Morita equivalence for rings with local units, Rocky Mountain J. Math. 22 (1992), 405 -- 416. [2] G. Abrams and G. Aranda-Pino, The Leavitt path algebra of a graph, J. Algebra 293 (2005), 319 -- 334. [3] G. Abrams and M. Tomforde, Isomorphism and Morita equivalence of graph algebras, Trans. Amer. Math. Soc. 363 (2011), 3733 -- 3767. [4] P. Ara, J. Bosa, R. Hazrat and A. Sims, Reconstruction of graded groupoids from graded Steinberg algebras, Forum Math. 29 (2017), 1023 -- 1038. [5] P. Ara, M.A. Moreno and E. Pardo, Nonstable K-theory for graph algebras, Algebr. Represent. Theory 10 (2007), 157 -- 178. [6] S.E. Arklint, S. Eilers, and E. Ruiz, A dynamical characterization of diagonal preserving ∗- isomorphisms of graph C∗-algebras, to appear in Ergodic Theory Dynam. Systems, published online doi:10.1017/etds.2016.141, (2017), 21 pages. [7] J.H. Brown, L. Clark and A. an Huef, Diagonal-preserving ring ∗-isomorphisms of Leavitt path [8] N. Brownlowe, T.M. Carlsen and M.F. Whittaker, Graph algebras and orbit equivalence, Ergodic algebras, J. Pure Appl. Algebra 221(2017), 2458 -- 2481. Theory Dynam. Systems 37 (2017), 389 -- 417. [9] C. Gil Canto and A. Nasr-Isfahani, The maximal commutative subalgebra of a Leavitt path algebra, arXiv:1510.03992v2, 21 pages. [10] T.M. Carlsen and J. Rout, Diagonal-preserving graded isomorphisms of Steinberg algebras, to ap- pear in Commun. Contemp. Math., published online doi:10.1142/S021919971750064X (2017), 24 pages. [11] T.M. Carlsen, E. Ruiz and A. Sims, Graph algebras and orbit equivalence, Proc. Amer. Math. Soc. 145 (2017), 1581 -- 1592. [12] T.M. Carlsen and M.L. Winger, Orbit equivalence of graphs and isomorphism of graph groupoids, to appear in Math. Scand., arXiv:1610.09942v3, 9 pages. [13] L.O. Clark and A. Sims, Equivalent groupoids have Morita equivalent Steinberg algebras, J. Pure Appl. Algebra 219 (2015), 2062 -- 2075. [14] R. Johansen, and A.P.W Sørensen, The Cuntz splice does not preserve ∗-isomorphism of Leavitt [15] A. Kumjian, D. Pask, I. Raeburn and J. Renault, Graphs, groupoids, and Cuntz-Krieger algebras, J. path algebras over Z, J. Pure Appl. Algebra 220 (2016), 3966 -- 3983. Funct. Anal. 144 (1997) 505- -- 541. [16] W.G. Leavitt, Modules over rings of words, Proc. Amer. Math. Soc. 7 (1956) 188 -- 193. [17] W.G. Leavitt, Modules without invariant basis number, Proc. Amer. Math. Soc. 8 (1957) 322- -- 328. [18] W.G. Leavitt, The module type of a ring, Trans. Amer. Math. Soc. 42 (1962) 113- -- 130. [19] G. Nagy and S. Reznikoff, Pseudo-diagonals and uniqueness theorems, Proc. Amer. Math. Soc. 142 (2014), 263- -- 275. [20] M. Rørdam, Classification of Cuntz-Krieger algebras, K-Theory 9 (1995) 31- -- 58. [21] B. Steinberg, Diagonal-preserving isomorphisms of étale groupoid algebras, arXiv:1711.01903v1, 24 pages. ∗-ISOMORPHISM OF LEAVITT PATH ALGEBRAS OVER Z 9 [22] M. Tomforde, Uniqueness theorems and ideal structure for Leavitt path algebras. J. Algebra 318 (2007) 270- -- 299. [23] S. Webster, The path space of a directed graph, Proc. Amer. Math. Soc. 142 (2014), 213 -- 225. UNIVERSITY OF THE FAROE ISLANDS, DEPARTMENT OF SCIENCE AND TECHNOLOGY, NÓATÚN 3, FO-100 TÓRSHAVN, FAROE ISLANDS E-mail address: [email protected]
1812.08819	1	1812	2018-12-20T19:48:22	A Bezout ring of stable range 2 which has square stable range 1	[ "math.RA", "math.AC", "math.KT" ]	In this paper we introduced the concept of a ring of stable range 2 which has square stable range 1. We proved that a Hermitian ring $R$ which has (right) square stable range 1 is an elementary divisor ring if and only if $R$ is a duo ring of neat range 1. And we proved that a commutative Hermitian ring $R$ is a Toeplitz ring if and only if $R$ is a ring of (right) square range 1.	math.RA	math	A Bezout ring of stable range 2 which has square stable range 1 Bohdan Zabavsky, Oleh Romaniv Department of Mechanics and Mathematics Ivan Franko National University of Lviv, Ukraine [email protected], [email protected] November, 2018 Abstract: In this paper we introduced the concept of a ring of stable range 2 which has square stable range 1. We proved that a Hermitian ring R which has (right) square stable range 1 is an elementary divisor ring if and only if R is a duo ring of neat range 1. And we proved that a commutative Hermitian ring R is a Toeplitz ring if and only if R is a ring of (right) square range 1. We proved that if R be a commutative elementary divisor ring of (right) square stable range 1, then for any matrix A ∈ M2(R) one can find invertible Toeplitz matrices P and Q such that P AQ = (cid:0) e1 0 0 e2(cid:1) , where ei is a divisor of e2. Key words and phrases: Hermitian ring, elementary divisor ring, stable range 1, stable range 2, square stable range 1, Toeplitz matrix, duo ring, quasi-duo ring. Mathematics Subject Classification: 06F20,13F99. 1 Introduction The notion of a stable range of a ring was introduced by H. Bass, and became especially popular because of its various applications to the problem of can- cellation and substitution of modules. Let us say that a module A satisfies the power-cancellation property if for all modules B and C, A ⊕ B ∼= A ⊕ C implies that B n ∼= C n for some positive integer n (here B n denotes the direct sum of n copies of B). Let us say that a right R-module A has the power-substitution property if given any right R-module decomposition M = A1 ⊕ B1 = A2 ⊕ B2 which each Ai ∼= A, there exist a positive integer n and a submodule C ⊆ M n such that M n = C ⊕ B n 1 = C ⊕ B n 2 . Prof. K. Goodearl pointed out that a commutative rind R has the power- substitution property if and only if R is of (right) power stable range 1, i.e. if aR + bR = R than (an + bx)R = R for some x ∈ R and some integer n ≥ 2 depending on a, b ∈ R [1]. Recall that a ring R is said to have 1 in the stable range provided that whenever ax + b = 1 in R, there exists y ∈ R such that a + by is a unit in R. The following Warfield's theorem shows that 1 in the stable range is equivalent to a substitution property. Theorem 1. [1] Let A be a right R-module, and set E = EndR(A). Then E has 1 in the stable range if and only if for any right R-module decomposition M = A1 ⊕ B1 = A2 ⊕ B2 with each Ai ∼= A, there exists a submodule C ⊆ M such that M = C ⊕ B1 = C ⊕ B2. A ring R is said to have 2 in the stable range if for any a1, . . . , ar ∈ R where r ≥ 3 such that a1R+· · ·+arR = R, there exist elements b1, . . . , br−1 ∈ R such that (a1 + arb1)R + (a2 + arb2)R + · · · + (ar−1 + arbr−1)R = R. K. Goodearl pointed out to us the following result. Proposition 1. [1] Let R be a commutative ring which has 2 in the stable range. If R satisfies right power-substitution, then so does Mn(R), for all n. Our goal this paper is to study certain algebraic versions of the notion of stable range 1. In this paper we study a Bezout ring which has 2 in the stable range and which is a ring square stable range 1. A ring R is said to have (right) square stable range 1 (written ssr(R) = 1) if aR + bR = R for any a, b ∈ R implies that a2 + bx is an invertible element of R for some x ∈ R. Considering the problem of factorizing the matrix ( a 0 b 0 ) into a product of two Toeplitz matrices. D. Khurana, T.Y. Lam and Zhou Wang were led to ask go units of the form a2 + bx given that aR + bR = R. Obviously, a commutative ring which has 1 in the stable range is a ring which has (right) square stable range 1, but not vice versa in general. Exam- ples of rings which have (right) square stable range 1 are rings of continuous real-valued functions on topological spaces and real holomorphy rings in for- mally real fields [2]. Proposition 2. [2] For any ring R with ssr(R) = 1, we have that R is right quasi-duo (i.e. R is a ring in which every maximal right ideal is an ideal). We say that matrices A and B over a ring R are equivalent if there exist invertible matrices P and Q of appropriate sizes such that B = P AQ. If for a matrix A there exists a diagonal matrix D = diag(ε1, ε2, . . . , εr, 0, . . . , 0) such that A and D are equivalent and Rεi+1R ⊆ εiR ∩ Rεi for every i then we say that the matrix A has a canonical diagonal reduction. A ring R is called an elementary divisor ring if every matrix over R has a canonical diagonal reduction. If every (1 × 2)-matrix ((2 × 1)-matrix) over a ring R has a canonical diagonal reduction then R is called a right (left) Hermitian ring. A ring which is both right and left Hermitian is called an Hermitian ring. Obviously, a commutative right (left) Hermitian ring is an Hermitian ring. We note that a right Hermitian ring is a ring in which every finitely generated right ideal is principal. Theorem 2. [3] Let R be a right quasi-duo elementary divisor ring. Then for any a ∈ R there exists an element b ∈ R such that RaR = bR = Rb. If in addition all zero-divisors of R lie in the Jacobson radical, then R is a duo ring. Recall that a right (left) duo ring is a ring in which every right (left) ideal is two-sided. A duo ring is a ring which is both left and right duo ring. We have proved the next result. Theorem 3. Let R be an elementary divisor ring which has (right) square stable range 1 and which all zero-divisors of R lie in Jacobson radical of R, then R is a duo ring. Proof. By Proposition 2 we have that R is a right quasi-duo ring. By Theo- rem 2 we have that R is a duo ring. Proposition 3. Let R be a Hermitian duo ring. For every a, b, c ∈ R such that aR + bR + cR = R the following conditions are equivalent: 1) there exist elements p, q ∈ R such that paR + (pb + qc)R = R; 2) there exist elements λ, u, v ∈ R such that b+λc = vu, where uR+aR = R and vR + cR = R. Proof. 1)⇒2) Since paR + (pb + qc)R = R we have pR + qcR = R and since R is a duo ring we have pR + cR = R. Than Rp + Rc = R, i.e. vp + jc = 1 for some elements v, j ∈ R. Then vpb + jcb = b and b − vpb = jcb = cj ′b = ct where t = j ′b and jc = cj ′. Element j ′ exist, since R is a duo ring. Then v(pb + qc) = vpb + vqc = b + ct + vqc = b + ct + ck, where vqc = ck for some element k ∈ R. That is, we have v(pb + qc) − b = cλ for some element λ ∈ R. We have b + cλ = v(pb + qc). Let u = pb + qc. We have b + cλ = vu, where vR+cR = R, since vp+cj ′ = 1 and uR+aR = R, since paR+(pb+qc)R = R. 2)⇒1) Since vR + cR = R then Rv + Rc = R. Let pv + jc = 1 for some elements p, j ∈ R. Then pR + cR = R. Since b + λc = vu, we have pb = p(vu − λc) = (pv)u − pλc = (1 − jc)u + pλc = u − ju′c + pλc = u + qc for some element q = pλ − ju′, where cu = u′c for some element u′ ∈ R. Since u = pb + qc, therefore (pb + qc)R + aR = R. Since R is an Hermitian duo ring then we have pR+qR = dR where p = dp1, q = dq1 and p1R+q1R = R. Then p1R + (p1b + q1c)R = R since pR ⊂ p1R and pR + cR = R, p1R + q1R = R, i.e. we have p1R + (p1b + q1c)R = R. Hence, aR + (p1b + q1c)R we have p1aR + (p1b + q1c)R = R. Remark 1. In Proposition 3 we can choose the elements u and v such that uR + vR = R. Proposition 4. Let R be an Hermitian duo ring. Then the following condi- tions are equivalent: 1) R is an elementary divisor duo ring; 2) for every x, y, z, t ∈ R such that xR + yR = R and zR + tR = R there exists an element λ ∈ R such that x + λy = vu, where vR + zR = R and uR + tR = R. Proof. 1)⇒2) Let R be an elementary divisor ring. By [4] for any a, b, c such that aR + bR + cR = R there exist elements p, q ∈ R such that paR + (pb + qc)R = R. Since xR + yR = R, zR + tR = R and the fact that R is a Hermitian duo ring we have zR + xR + ytR = R. By Proposition 3 we have x + λyt = uv where uR + zR = R, vR + ytR = R. Since x + (λt)y = x + µy = uv where µ = λt, we have uR + zR = R, vR + yR = R. 2)⇒1) Let aR + bR + cR = R and Rb + Rc = Rd and b = b1d, c = c1d, where Rb1 = Rc1 = R. Since R is a duo ring then b1R + c1R = R. So now dR = Rd and aR + bR + cR = R, Rb + Rc = Rd we have aR + dR = R, i.e. dd1 + ax = 1 for some elements d1, x ∈ R. Then 1 − dd1 ∈ aR. Since b1R + c1R = R, by Conditions 2 of Proposition 3 there exists an element λ1 ∈ R such that b1 + c1λ = vu1 where u1R + (1 − dd1)R = R and vR + dd1R = R. Since (1 − dd1) ∈ aR and u1R + (1 − dd1)R = R. We have uR + aR = R. Let u = u1d. Since u1R + aR = R and dR + aR = R we have uR + aR = R. Since b1 + c1λ = vu1, we have b + cµ + vu, where λd = dµ. Recall that vR + dd1R = R then vR + dR = R. Since vR + cR = vR + c1dR = vR + c1R. So b1 + c1λ = vu1 and b1R + c1R = R then vR + c1R = R. Therefore, vR+cR = R. This means that the Condition 2 of Proposition 3 is true. By Proposition 3 we conclude that for every a, b, c ∈ R with aR + bR + cR = R there exist elements p, q ∈ R such that paR + (pb + qc)R = R, i.e. according to [4], R is an elementary divisor ring. Definition 1. Let R be a duo ring. We say that an element a ∈ R\{0} is neat if for any elements b, c ∈ R such that bR + cR = R there exist elements r, s ∈ R such that a = rs, where rR + bR = R, sR + cR = R, rR + sR = R. Definition 2. We say that a duo ring R has neat range 1 if for every a, b ∈ R such that aR + bR = R there exists an element t ∈ R such that a + bt is a neat element. According to Propositions 3, 4 and Remark 1 we have the following result. Theorem 4. A Hermitian duo ring R is an elementary divisor ring if and only if R has neat range 1. The term "neat range 1" substantiates the following theorem. Theorem 5. Let R be a Hermitian duo ring. If c is a neat element of R then R/cR is a clean ring. Proof. Let c = rs, where rR + aR = R, sR + (1 − a)R = R for any element a ∈ R. Let ¯r = r + cR, ¯s = s + aR. From the equality rR + sR = R we have ru + sv = 1 for some elements u, v ∈ R. Hence r2u + srv = r and rsu + s2v = s we have ¯r2¯u = ¯r, ¯s2¯v = ¯s. Let ¯s¯v = ¯e. It is obvious that ¯e2 = ¯e and ¯1 − ¯e = ¯u¯r. Since rR + aR = R, we have rx + ay = 1 for elements x, y ∈ R. Hence rxsv + aysv = sv we have rsx′v + aysv = sv where xs = sx′ for some element x′ ∈ R. Then ¯a¯y¯e = ¯e, i.e. ¯e ∈ ¯aR. Similarly from the equality sR + (1 − a)R = R, it follows ¯1 − ¯e ∈ (¯1 − ¯a)R. According to [5] R/cR is an exchange ring. Since R is a duo ring, R/cR is a clean ring. Taking into account the Theorem 3 and Theorem 4 we have the following result. Theorem 6. A Hermitian ring R which has (right) square stable range 1 is an elementary divisor ring if and only if R is a duo ring of neat range 1. Let R be a commutative Bezout ring. The matrix A of order 2 over R is said to be a Toeplitz matrix if it is of the form (cid:18)a b c a(cid:19) where a, b, c ∈ R. Notice that if A is an invertible Toeplitz matrix, then A−1 is also an invertible Toeplitz matrix. Definition 3. A commutative Hermitian ring R is called a Toeplitz ring if for any a, b ∈ R there exist an invertible Toeplitz matrix T such that (a, b)T = (d, 0) for some element d ∈ R. Theorem 7. A commutative Hermitian ring R is a Toeplitz ring if and only if R is a ring of (right) square range 1. Proof. Let R be a commutative Hermitian ring of (right) square stable range 1 and aR + bR = R for some elements a, b ∈ R. Then a2 + bt = u, where u is an invertible element of R. Let Then i.e. we have Since S = (cid:18)a −b a (cid:19) , K = (cid:18)u−1 0 t 0 u−1(cid:19) . (a, b)S = (u, 0), (u, 0)K = (1, 0), (a, b)SK = (1, 0). (cid:18)a −b a (cid:19)(cid:18)u−1 0 t 0 u−1(cid:19) = (cid:18) au−1 −bu−1 au−1 (cid:19) = T −tu−1 we have that T = SK is a Toeplitz matrix. So (a, b)T = (1, 0). If a, b ∈ R and aR + bR = dR then by a = da0, b = db0 and a0R + b0R = R [4]. Then there exists an element t ∈ R such that a0 + b0t = u, where a is an invertible element of R. Let (cid:18)a0 −b0 a0 (cid:19)(cid:18)u−1 0 t 0 u−1(cid:19) . Note that T is an invertible Toeplitz matrix. Then (a, b)T = (d, 0), i.e. R is a Toeplitz ring. Let R be a Toeplitz ring and aR + bR = R. The exists an invertible Toeplitz matrix T such that (a, b)T = (1, 0). Let S = T −1 = (cid:18)x t y x(cid:19), where x, y, t ∈ R. So det T −1 = z2 + ty = u is an invertible element of R. Since (a, b) = (1, 0)T −1, we have a = x, b = t. By equality x2 + ty = u we have a2 + by = u, i.e. R is a ring of (right) square stable range 1. Theorem 8. Let R be a commutative ring of square stable range 1. Then for any row (a, b), where aR + bR = R, there exists an invertible Toeplitz matrix T = (cid:18)a b x a(cid:19) , where x ∈ R. Proof. By Theorem 7 we have (a, b) = (1, 0)T for some invertible Toeplitz matrix T . Let T = (cid:18)x t y x(cid:19). Then a = x, b = t and T = (cid:18)a b y a(cid:19) is an invertible Toeplitz matrix. Recall that GEn(R) denotes a group of n × n elementary matrices over ring R. The following theorem demonstrated that it is sufficient to consider only the case of matrices of order 2 in Theorem 7. Theorem 9. [4] Let R be a commutative elementary divisor ring. Then for any n × m matrix A (n > 2, m > 2) one can find matrices P ∈ GEn(R) and Q ∈ GEm(R) such that P AQ = e1 0 . . . 0 0   0 e2 . . . 0 0 . . . . . . . . . . . . . . . 0 0 0 0 . . . . . . es 0 0 A0   where ei is a divisor of ei+1, 1 ≤ i ≤ s − 1, and A0 is a 2 × k or k × 2 matrix for some k ∈ N. Theorem 10. Let R be a commutative elementary divisor ring of (right) square stable range 1. Then for any 2 × 2 matrix A one can find invertible Toeplitz matrices P and Q such that P AQ = (cid:18)e1 0 0 e2(cid:19) , where ei is a divisor of e2. Proof. Since R is a Toeplitz ring it is enough to consider matrices of the form A = (cid:18)a b 0 c(cid:19) , where aR + bR + cR = R. Since R is an elementary divisor ring by [4] there exist elements p, q ∈ R such that paR+(pb+qc)R = R, i.e. par+(pb+qc)s = 1 for some elements r, s ∈ R. Since pR + qR = R and rR + sR = R, by Theorem 8 we have the invertible Toeplitz matrices P = (cid:18)p q (cid:18)r ∗ s ∗(cid:19) such that ∗ ∗(cid:19), Q = P AQ = (cid:18)1 x y z(cid:19) = A1. Then where S = (cid:18) 1 0 0 0 (cid:18) 1 −y 1(cid:19) A1(cid:18)1 −x −y 1(cid:19) and T = (cid:18)1 −x 1 (cid:19) = (cid:18)1 1 (cid:19) are invertible Toeplitz matrices. So 0 ac(cid:19) , 0 0 SP AQT = (cid:18)1 0 ac(cid:19) . 0 Theorem is proved. Open Question. Is it true that every commutative Bezout domain of stable range 2 which has (right) square stable range 1 is an elementary divisor ring? References [1] K. R. Goodearl, Power-cancellation of groups and modules, Pacific J. Math. 64 (1976) 487 -- 411. [2] D. Khurana, T. Y. Lam and Zh. Wang, Rings of square stable range one, J. Algebra 338 (2011) 122 -- 143. [3] B. V. Zabavsky and M. Ya. Komarnytskii, Distributive elementary di- visor domains, Ukr. Math. J. 42 (1990) 890 -- 892. [4] B. V. Zabavsky Diagonal reduction of matrices over rings (Mathemat- ical Studies, Monograph Series, volume XVI, VNTL Publishers, Lviv, 2012). [5] W. K. Nicholson, Lifting idempotents and exchange rings, Trans. Amer. Math. Soc. 229 (1977) 269 -- 278.
1512.08452	1	1512	2015-12-28T16:44:31	Typical ranks for 3-tensors, nonsingular bilinear maps and determinantal ideals	[ "math.RA", "math.AC" ]	Let $m,n\geq 3$, $(m-1)(n-1)+2\leq p\leq mn$, and $u=mn-p$. The set $\mathbb{R}^{u\times n\times m}$ of all real tensors with size $u\times n\times m$ is one to one corresponding to the set of bilinear maps $\mathbb{R}^m\times \mathbb{R}^n\to \mathbb{R}^u$. We show that $\mathbb{R}^{m\times n\times p}$ has plural typical ranks $p$ and $p+1$ if and only if there exists a nonsingular bilinear map $\mathbb{R}^m\times\mathbb{R}^n\to\mathbb{R}^{u}$. We show that there is a dense open subset $\mathscr{O}$ of $\mathbb{R}^{u\times n\times m}$ such that for any $Y\in\mathscr{O}$, the ideal of maximal minors of a matrix defined by $Y$ in a certain way is a prime ideal and the real radical of that is the irrelevant maximal ideal if that is not a real prime ideal. Further, we show that there is a dense open subset $\mathscr{T}$ of $\mathbb{R}^{ n\times p \times m}$ and continuous surjective open maps $\nu\colon\mathscr{O}\to\mathbb{R}^{u\times p}$ and $\sigma\colon\mathscr{T}\to\mathbb{R}^{u\times p}$, where $\mathbb{R}^{u \times p}$ is the set of $u\times p$ matrices with entries in $\mathbb{R}$, such that if $\nu(Y)=\sigma(T)$, then $\mathrm{rank} T=p$ if and only if the ideal of maximal minors of the matrix defined by $Y$ is a real prime ideal.	math.RA	math	Typical ranks for 3-tensors, nonsingular bilinear maps and determinantal ideals Toshio Sumi, Mitsuhiro Miyazaki and Toshio Sakata Abstract Let m, n ≥ 3, (m − 1)(n − 1) + 2 ≤ p ≤ mn, and u = mn − p. The set Ru×n×m of all real tensors with size u × n × m is one to one corresponding to the set of bilinear maps Rm × Rn → Ru. We show that Rm×n×p has plural typical ranks p and p + 1 if and only if there exists a nonsingular bilinear map Rm × Rn → Ru. We show that there is a dense open subset O of Ru×n×m such that for any Y ∈ O, the ideal of maximal minors of a matrix defined by Y in a certain way is a prime ideal and the real radical of that is the irrelevant maximal ideal if that is not a real prime ideal. Further, we show that there is a dense open subset T of Rn×p×m and continuous surjective open maps ν : O → Ru×p and σ : T → Ru×p, where Ru×p is the set of u × p matrices with entries in R, such that if ν(Y ) = σ(T ), then rank T = p if and only if the ideal of maximal minors of the matrix defined by Y is a real prime ideal. 1 Introduction For positive integers m, n, and p, we consider an m × n × p tensor which is an element of the tensor product of Rm, Rn, and Rp with standard basis. This tensor can be identified with a 3-way array (aijk) where 1 ≤ i ≤ m, 1 ≤ j ≤ n and 1 ≤ k ≤ p. We denote by Rm×n×p the set of all m × n × p tensors. This set is a topological space with Euclidean topology. Hitchcock [15] defined the rank of a tensor. An integer r is called a typical rank of Rm×n×p if the set of tensors with rank r is a semi-algebraic set of dimension mnp. In the other words, r is a typical rank of Rm×n×p if the set of tensors with rank r contains an open set of Rm×n×p. In this paper we discuss the typical ranks of 3-tensors and connect between plurality of typical ranks and existence of a nonsingular bilinear map. Let n ≤ p. A typical rank of R1×n×p is equal to an n × p matrix full rank, that is, n. If n ≥ 2, then the set of typical ranks of R2×n×p is equal to {n, n + 1} if n = p and otherwise min{p, 2n} [36]. This is also obtained from the equivalent class: almost all 2 × n × p tensors are equivalent to ((En, On×(p−n)); (On×(p−n), En)) which has rank min{p, 2n} if n < p (see [18] or [32]), see Section 2 for notation. Suppose that n ≥ m ≥ 3. The set of typical ranks of Rm×n×p is equal to min{p, mn} if (m − 1)n < p [35]. If p = (m − 1)n then the set of typical ranks of Rm×n×p depends on the existence of a nonsingular bilinear map Rm × Rn → Rn: It is equal to {p} if there is no nonsingular bilinear map Rm × Rn → Rn and {p, p + 1} otherwise [33]. Here, a bilinear map f : Rm × Rn → Rr is called nonsingular if f (x, y) = 0 implies x = 0 or y = 0. Suppose that (m − 1)(n − 1) + 1 ≤ p ≤ (m − 1)n. A typical rank of Rm×n×p is unknown except a few cases. First, p is a minimal typical rank, since p is a generic rank of Cm×n×p [5]. The authors [34] showed that the Hurwitz-Radon function gives a condition that Rm×n×(m−1)n has 1 plural typical ranks. We [24] also showed that Rm×n×p has plural typical ranks for some (m, n, p) by using the concept of absolutely full column rank tensors. We let m#n be the minimal integer r such that there is a nonsingular bilinear map Rm × Rn → Rr. Then m#n ≤ m + n − 1 (see Section 2). The set Rr×m×n of r × m × n tensors is one to one corresponding to the set of bilinear maps Rm × Rn → Rr. By this map the set of absolutely full column rank tensors is one to one corresponding to the set of nonsingular bilinear maps. Theorem 1.1 Let m, n ≥ 3 and (m − 1)(n − 1) + 1 ≤ p ≤ mn. (1) If there exists a nonsingular bilinear map Rm ×Rn → Rmn−p, then Rm×n×p has plural typical ranks. (2) If p ≥ (m − 1)(n − 1) + 2 and Rm×n×p has plural typical ranks, then there exists a nonsingular bilinear map Rm × Rn → Rmn−p. (1) of Theorem 1.1 is an extension of one of [24]. Furthermore, we completely determine the set trank(m, n, p) of typical ranks of Rm×n×p for p≥(m − 1)(n − 1) + 2 by the number m#n. Theorem 1.2 Let m, n ≥ 3, k ≥ 2, and p = (m − 1)(n − 1) + k. The set of typical ranks of Rm×n×p is given as follows. trank(m, n, p) =  {p, p + 1}, {p}, {mn}, 2 ≤ k ≤ m + n − 1 − (m#n) max{2, (m + n) − (m#n)} ≤ k ≤ m + n − 2 k ≥ m + n − 1. Consider the case where p = (m − 1)(n − 1) + 1, Friedland [12] showed that Rn×n×((n−1)2+1) has plural typical ranks. We extend this result. Theorem 1.3 Let m, n ≥ 3 and p = (m − 1)(n − 1) + 1. Rm×n×p has plural typical ranks if m − 1 and n − 1 are not bit-disjoint. This article is organized as follows. Sections 2 -- 7 are preparation to show the above theorems. In Section 2, we set notations and discuss the number m#n. In Section 3, we study absolutely full column rank tensors. Since the set of absolutely full column rank tensors is an open set, there exists a special form of an absolutely full column rank tensor if an absolutely full column rank tensor exists. In Section 4, we state the other notions and deal with ideals of minors of matrices. Theorem 4.31 in Section 4 which corresponds with the real radical ideals is quite interesting in its own right. We show that for integers with 0 < t ≤ min{u, n} and m ≥ (u−t+1)(n−t+1)+2, there exist open subsets O1 and O2 of Ru×n×m such that the union of them is dense, I(V(It(M (x, Y )))) = It(M (x, Y )) for Y ∈ O1 and I(V(It(M (x, Y )))) = (x1, . . . , xm) for Y ∈ O2, where It(M (x, Y )) k=1 xkYk given by the indeterminates x1, . . . , xm and Y = (Y1; . . . ; Ym) ∈ Ru×n×m. From this, we can give a subset of m × n × p tensors with rank p for 3 ≤ m ≤ n and (m − 1)(n − 1) + 2 ≤ p ≤ (m − 1)n. In Section 5 we discuss a property for the determinantal ideals by using monomial preorder. This property plays an important role for proving Theorem 1.1. We characterize m × n × p tensors with rank p in Section 6. In Section 7, we show that the existence of an absolutely full column rank tensor with suitable size implies that p + 1 is a typical rank of Rm×n×p. Moreover there exist a nonempty is the ideal generated by all t-minors of the u × n matrix M (x, Y ) = Pm 2 open subset T1 consisting of tensors with rank p and a possibly empty open subset T2 consisting of tensors with rank greater than p, corresponding O1 and O2 respectively, such that the union of them is a dense subset of Rm×n×p (see Theorem 7.14). Finally, in Section 8, we show that p + 2 is not a typical rank of Rm×n×p and complete proofs of the above theorems. 2 Nonsingular bilinear maps We first recall some basic facts and establish terminology. Notation (1) We denote by Rn (resp. R1×n) the set of n-dimensional column (resp. row) real vectors and by En the n × n identity matrix. Let ej be the j-th column vector of an identity matrix. (2) For a tensor x ∈ Rn ⊗ Rp ⊗ Rm with x = Pijk aijkei ⊗ ej ⊗ ek, we identify x with T = (aijk)1≤i≤n,1≤j≤p,1≤k≤m and denote it by (A1; . . . ; Am), where Ak = (aijk)1≤i≤n,1≤j≤p for k = 1, . . . , m is an n × p matrix, and call (A1; . . . ; Am) a tensor. (3) We denote the set of n × p × m tensors by Rn×p×m and the set of typical ranks by trank(n, p, m). (4) For an n × p × m tensor T = (T1; . . . ; Tm), an l × n matrix P and an k × p matrix Q, we denote by P T the l × p × m tensor (P T1; . . . ; P Tp) and by T Q⊤ the n × k × m tensor (T1Q⊤; . . . ; TpQ⊤). (5) For n × p matrices A1, . . . , Am, we denote by (A1, . . . , Am) the n × mp matrix obtained by aligning A1, . . . , Am horizontally. (6) We set Diag(A1, A2, . . . , At) =  O At   A1 A2 . . . O for matrices A1, A2, . . . , At. (7) For an m × n matrix M , we denote by M≤j (resp. j<M ) the m × j matrix consisting of the first j (resp. last n − j) columns of M . We denote by M ≤i (resp. i<M ) the i × n (resp. (m−i)×n) matrix consisting of the first i (resp. last m−i) rows of M . We put M <i = M ≤i−1 M<i = M≤i−1, and M =i = i−1<(M ≤i) which is the i-th row vector of M . (8) We set fl1(T ) = (T1, . . . , Tm) and fl2(T ) =  T1 ... Tm   for a tensor T = (T1; . . . ; Tm). Definition 2.1 A bilinear map f : Rm × Rn → Rl is called nonsingular if f (x, y) = 0 implies x = 0 or y = 0. For positive integers m and n, we set m#n := min{l there exists a nonsingular bilinear map Rm × Rn → Rl}. 3 Let g : R1×u × R1×v → R1×(u#v) be a nonsingular bilinear map. For positive integers m and n, let f : R1×mu × R1×nv → R1×(m+n−1)(u#v) be a map defined by f ((a1, . . . , am), (b1, . . . , bn)) = (g(a1, b1), g(a1, b2) + g(a2, b1), . . . ,Pi+j=k g(ai, bj), . . . , g(am, bn)). It is easily verified that f is a nonsingular bilinear map. Thus we have the following: Lemma 2.2 (mu)#(nv) ≤ (m + n − 1)(u#v). By applying this lemma to nonsingular bilinear maps obtained by multiplications of R, C, quaternions and octanions respectively, we have the following: Proposition 2.3 (cf. [30, Proposition 12.12 (3)]) For k = 1, 2, 4 and 8, km#kn ≤ k(m + n − 1). it holds that Let H (r, s, n) be the condition on the binomial coefficients, called the Stiefel-Hopf criterion, that the binomial coefficient (cid:0)n nonsingular, biskew map Rr × Rs → Rn then the Stiefel-Hopf criterion H (r, s, n) holds. Put k(cid:1) is even whenever n − s < k < r. If there exists a continuous, We have r ◦ s = min{n H (r, s, n) holds}. max{r, s} ≤ r ◦ s ≤ r#s ≤ r + s − 1. Putting n∗ = ⌈ n 2 ⌉ for n ∈ Z, the number r ◦ s is easily obtained by the formula r ◦ s =(2(r∗ ◦ s∗) − 1 2(r∗ ◦ s∗) if r, s are both odd and r∗ ◦ s∗ = r∗ + s∗ − 1, otherwise (cf. [30, Proposition 12.9]). For a positive integer n, we put integers αj(n) = 0, 1, j ≥ 0 such that n =P∞ dyadic expansion of n and let α(n) :=P∞ j=0 αj (n)2j is the j=0 αj(n) be the number of ones in the dyadic expansion of n. Two integers m and n are bit-disjoint if {j αj(m) = 1} and {j αj (n) = 1} are disjoint. For k > h, let τ (k, h) be a nonnegative number defined as τ (k, h) = #{j ≥ 0 αj(k − h) = 0, αj(k) 6= αj(h)}. Proposition 2.4 r#s = r + s − 1 if and only if r − 1 and s − 1 are bit-disjoint. If r − 1 and s − 1 are bit-disjoint, then r ◦ s = r#s = r + s − 1 (cf. Proof [30, p. 257]). Moreover, τ (k, h) = 0 if and only if h and k − h are bit-disjoint. There is a nonsingular bilinear map Rh+1 × Rk−h+τ (k,h) → Rk for k > h ≥ 0 [20] and thus (h + 1)#(k − h + τ (k, h)) ≤ k. Putting r = h + 1 and k = r + s − 2, we have r#(s − 1 + τ (r + s − 2, r − 1)) ≤ r + s − 2. In particular, if r − 1 and s − 1 are not bit-disjoint then r#s ≤ r + s − 2. Let ρ be the Hurwitz-Radon function defined as ρ(n) = 2b + 8c for nonnegative integers a, b, c such that n = (2a + 1)2b+4c and 0 ≤ b < 4. There is a nonsingular bilinear map Rn × Rρ(n) → Rn [17, 26] and is no nonsingular bilinear map Rn × Rρ(n)+1 → Rn for any n ≥ 1 [1]. Therefore, n#ρ(n) ≤ n and n#(ρ(n) + 1) > n. Corollary 2.5 n#n ≤ 2n − 2. In particular, the equality n#n = 2n − 2 holds for n = 2a + 1. 4 Proof The inequality n#n ≤ 2n − 2 is clear by Proposition 2.4 since n − 1 and n − 1 are not bit-disjoint. There is an immersion RPn → Rn+k if and only if there is a nonsingular biskew map Rn+1 × Rn+1 → Rn+1+k (cf. [2, 30]). Note that ρ(2) = 2 and ρ(4) = 4. Then 2#2 = 2 and 3#3 = 4 which follows from 4#4 = 4. Suppose that a ≥ 2. Put m = 2a−1. Since there is no immersion RP2m→R4m−2 (cf. [21]), we have 4m = (2m + 1)#(2m + 1). Many estimations for m#n are known from immersion problem for manifolds, as projective spaces. For example, the existence of a nonsingular bilinear map Rn+1 × Rn+1 → Rn+1+k implies that RPn immerses in Rn+k [13]. Proposition 2.6 (1) (n + 1)#(n + 1) ≤ 2n − α(n) + 1 [7]. (2) (2n + α(n))#(2n + α(n)) ≥ 4n − 2α(n) + 2 [8]. (3) (8n + 9)#(8n + 9) ≥ 16n + 6 and (16n + 12)#(16n + 12) ≥ 32n + 14 if α(n) = 2 [10, 31]. (4) (8n + 10)#(8n + 10) ≥ 16n + 1 and (8n + 11)#(8n + 11) ≥ 16n + 4 if α(n) = 3 [9, 10]. (5) (n + 1)#(m + 1) ≤ n + m + 1 − (α(n) + α(n − m) + min{k(n), k(m)}) if m, n are odd and n ≥ m, where k(n) is a nonnegative function depending only in the mod 8 residue class of n with k(8a + 1) = 0, k(8a + 3) = k(8a + 5) = 1 and k(8a + 7) = 4 [23]. (6) d(h + 1)#(d(k − h) + τ (k, h)) ≤ dk for k > h ≥ 0 and d = 1, 2, 4, 8 [20]. (7) (n + 1)#(n + τ (2n, n)) ≤ 2n. 3 Absolutely full column rank tensors For a tensor T of Rn ⊗ Rp ⊗ Rm, we define the rank of T , denoted by rank T , the minimal number r so that there exist ai ∈ Rn, bi ∈ Rp, and ci ∈ Rm for i = 1, . . . , r such that The set Rn×p×m has an action of GL(m) × GL(p) × GL(n) as T = r Xi=1 ai ⊗ bi ⊗ ci. (A, B, C) · r Xi=1 ai ⊗ bi ⊗ ci = r Xi=1 Aai ⊗ Bbi ⊗ Cci. For tensors T1, T2 ∈ Rm×n×p, T1 and T2 are said to be equivalent if T1 = (A, B, C) · T2 for some (A, B, C) ∈ GL(n) × GL(p) × GL(m). The equivalence relation preserves the rank. For a subset U and an open semi-algebraic subset S of Rm×n×p, we say that almost all tensors in S are equivalent to tensors in U if there exists a semi-algebraic subset S0 of S with dim S0 < mnp such that any tensor of S \ S0 is equivalent to a tensor of U . In particular, for a given tensor T0, if almost all tensors in Rm×n×p are equivalent to {T0}, then we say that any tensor is generically equivalent to T0. 5 An integer r is called a typical rank of n × p × m-tensors if there is a nonempty open subset O of Rn×p×m such that rank X = r for X ∈ O. Over the complex number field C, it is known that there is a unique typical rank, called the generic rank, of n × p × m-tensors for any n, p and m. The set of typical ranks of n × p × m-tensors over R is denoted by trank(n, p, m) and the generic rank of n × p × m-tensors over C is denoted by grank(n, p, m). We recall the following facts. Theorem 3.1 ([12, Theorem 7.1]) The space Rm1×m2×m3, m1, m2, m3 ∈ N, contains a fi- nite number of open connected disjoint semi-algebraic sets O1, . . . , OM satisfying the following properties. (1) Rm1×m2×m3 \SM than m1m2m3. i=1 Oi is a closed semi-algebraic set Rm1×m2×m3 of dimension strictly less (2) Each T ∈ Oi has rank ri for i = 1, . . . , M . (3) min{r1, . . . , rM } = grank(m1, m2, m3). (4) trank(m1, m2, m3) = {r ∈ Z min{r1, . . . , rM } ≤ r ≤ max{r1, . . . , rM }}. Let T = (A1; . . . ; Ap) be an m × n × p tensor over R. The tensor T is called an absolutely full column rank tensor if p rank( Xj=1 yjAj) = n for any (y1, . . . , yp)⊤ ∈ Rp \ {0}. From the definition of the absolutely full column rank property, we see the following fact. Lemma 3.2 Let T be an m × n × p tensor over R and P ∈ GL(m, R). Then T is absolutely full column rank if and only if so is P T . Lemma 3.3 (see Corollary 4.20 or [24, Theorem 3.6]) The set of m × n × p absolutely full column rank tensors is an open subset of Rm×n×p. Let T be an m × n × p-tensor. We define fT : Rn × Rp → Rm as fT (x, y) = yjAj x, p Xj=1 where y = (y1, . . . , yp)⊤. Then fT is a bilinear map. This assignment T 7→ fT induces a bijection from Rm×n×p to the set of all bilinear maps Rn ×Rp → Rm. It is easily verified that fT : Rn ×Rp → Rm is nonsingular if and only if T is absolutely full column rank. Therefore Corollary 3.4 There is an m × n × p absolutely full column rank tensor if and only if there is a nonsingular bilinear map Rn × Rp → Rm, i.e., n#p ≤ m. Lemma 3.5 Let n, m, and u be positive integers with u ≤ mn. Set p = mn − u. Then the following conditions are equivalent. 6 (1) n#m ≤ u. (2) There is a u × n × m absolutely full column rank tensor. (3) There is a u × n × m absolutely full column rank tensor Y such that p<fl1(Y ) = −Eu. Proof (1) ⇔ (2) follows from Corollary 3.4. It is clear that (3) ⇒ (2). (2) ⇒ (3): Let X = (X1; . . . ; Xm) be a u × n × m absolutely full column rank tensor. By Lemma 3.3, we may assume that p<fl1(X) is nonsingular. Set Y = −p<fl1(X)X. Then Y satisfies the required conditions. 4 Ideals of minors In this section, we state some results on ideals of minors, which we use in the following of this paper and interesting in its own right. First we recall the definition of normality of a ring. Definition 4.1 (see [22, Section 9]) Let R be a commutative ring. We say that R is normal if RP is an integrally closed integral domain for any prime ideal P of R. Remark 4.2 (1) A Noetherian integral domain is normal if and only if it is integrally closed. (2) If R is a Noetherian normal ring, then R ≃ R/P1 × · · · × R/Pr, where P1, . . . , Pr are associated prime ideals of R. We recall a criterion of normality in terms of Serre's condition. Definition 4.3 ([22, page 183]) Let R be a Noetherian ring and i a nonnegative integer. (1) We say that R satisfies (Ri) if RP is regular for any prime ideal P of R with htP ≤ i. (2) We say that R satisfies (Si) if depthRP ≥ min{i, htP } for any prime ideal P of R. Lemma 4.4 ([22, Theorem 23.8]) Let R be a Noetherian ring. Then R is normal if and only if R satisfies (R1)+(S2). The condition (R1)+(S2) is restated as follows. Lemma 4.5 Let R be a Noetherian ring. Then R satisfies (R1)+(S2) if and only if the following condition is satisfied: if P is a prime ideal of R with depthRP ≤ 1, then RP is regular. Proof First assume that R satisfies (R1)+(S2). Let P be a prime ideal of R with depthRP ≤ 1. Since R satisfies (S2), we see that depthRP ≥ min{htP, 2}. Therefore, htP ≤ 1. Thus by (R1), we see that RP is regular. Conversely, assume that RP is regular for any prime ideal P of R with depthRP ≤ 1. First we show that R satisfies (R1). If P is a prime ideal with htP ≤ 1, then depthRP ≤ htP ≤ 1. 7 Thus by assumption, we see that RP is regular. Next we show that R satisfies (S2). Let P If depthRP ≤ 1, then by assumption, RP is regular. Thus be an arbitrary prime ideal of R. depthRP = htP = min{htP, 2}. If depthRP ≥ 2, then depthRP ≥ min{htP, 2} holds trivially. Next we state notations and definitions used in this section. Definition 4.6 We denote by u, n, m, and t positive integers with t ≤ min{u, n} and set v = (u − t+ 1)(n− t+ 1). Let M = (mij ) be a u × n matrix with entries in a commutative ring A. We denote by It(M )A, or simply It(M ), the ideal of A generated by t-minors of M . For α(1), . . . , α(t) ∈ {1, . . . , u} and β(1), . . . , β(t) ∈ {1, . . . , n}, we set [α(1), . . . , α(t) β(1), . . . , β(t)]M := det(mα(i)β(j)), and if u ≥ n and α(1), . . . , α(n) ∈ {1, . . . , u}, we set [α(1), . . . , α(n)]M := det(mα(i)j). For i=1 aiTi and we define Γ(u × n) = {[a1, . . . , an] 1 ≤ a1 < · · · < an ≤ u, ai ∈ Z}. For γ = [a1, . . . , an] ∈ Γ(u × n), we set suppγ = {a1, . . . , an}. If B is a ring, A is a subring of B and T is a tensor (resp. matrix, vector) with entries in B, we denote by A[T ] the subring of B generated by the entries of T over A. If moreover, B is a field, we denote by A(T ) the subfield of B generated by the entries of T over A. If the entries of a tensor (resp. matrix, vector) T are independent indeterminates, we say that T is a tensor (resp. matrix, vector) of indeterminates. a tensor T = (T1; . . . ; Tm) and a = (a1, . . . , am) we set M (a, T ) := Pm Here we note the following fact, which is verified by using [3, Chapter 1 Exercise 2] or [25, (6.13)]. Lemma 4.7 Let A be a commutative ring, X a square matrix of indeterminates. Then det X is a non-zerodivisor of A[X]. Next we recall the following fact. Lemma 4.8 ([16, Theorem 1 and Corollaries 3 and 4] see also [4, (6.3) Theorem]) Let A be a Noetherian ring and X a u × n matrix of indeterminates. (1) ht(It(X)A[X]) = grade(It(X)A[X]) = v. (2) If A is a domain, then It(X)A[X] is a prime ideal of A[X]. (3) If A is a normal domain, then so is A[X]/It(X)A[X]. We also recall the following fact. Lemma 4.9 ([16, Theorem 1 and Corollaries 2 and 4] see also [4, (2.1) Theorem]) Let A be a Noetherian commutative ring and M a u × n matrix with entries in A. If It(M ) 6= A, then htIt(M ) ≤ v. Moreover, if A is Cohen-Macaulay and htIt(M ) = v, then It(M ) is height unmixed. The following Lemma is a generalization of [4, (12.4) Lemma]. Lemma 4.10 Let u, n, m, t and v be as in Definition 4.6, A a commutative Noetherian ring, T = (tijk) a u × n × m tensor of indeterminates and f1, . . . , fm elements of A. Suppose that (f1, . . . , fm) 6= A. Set g = grade(f1, . . . , fm)A, f = (f1, . . . , fm) and M = M (f , T ) = (mij ). (1) grade It(M )A[T ] = min{g, v}. 8 (2) If g ≥ v + 1 and A is a domain, then It(M )A[T ] is a prime ideal. (3) If g ≥ v + 2 and A is a Cohen-Macaulay normal domain, then A[T ]/It(M )A[T ] is a normal domain. Remark 4.11 If g ≥ v, then grade It(M ) = ht It(M ) = v by Lemma 4.10 (1) and [16, Theorem 1 and Corollary 4]. Proof of Lemma 4.10 First we prove (1). Set R = A[T ]. Set v′ = min{g, v}. Since It(M ) ⊂ (f1, . . . , fm)R, we see that gradeIt(M )R ≤ grade(f1, . . . , fm)R = g. Thus we see by Lemma 4.9, gradeIt(M )R ≤ v′. To prove the converse inequality, it is enough to show that if P be a prime ideal of R with P ⊃ It(M ), then depthRP ≥ v′. Since if P ⊃ (f1, . . . , fm)R, then depthRP ≥ g ≥ v′, we may assume that P 6⊃ (f1, . . . , fm)R. Take l with fl 6∈ P . Then M is essentially a matrix of indeterminates over A[f −1 ]) = v by Lemma 4.8. Since RP l is a localization of R[f −1 ][tijk k 6= l]. Thus grade(It(M )R[f −1 ], we see that depthRP ≥ v ≥ v′. l Next we prove (2). We may assume f1, . . . , fm 6= 0. Set B = R/It(M )R. Since It(M )R is grade unmixed by (1) and [16, Theorem 1 Corollaries 2 and 4] (see also [27, Corollary of Theorem 1.2] or [22, Exercise 16.3]), we see that every associated prime ideal of It(M )R is of grade v. In particular any associated prime ideal of It(M )R does not contain (f1, . . . , fm)R, since g > v by assumption. Thus ( ¯f1, . . . , ¯fm)B has grade at least 1, where ¯fk denote the natural image of fk in B for 1 ≤ k ≤ m. l Since A[f −1 l l ][tijk k 6= l], we see that B[ ¯fl ][tijk k 6= l] is an integral domain and M is essentially a matrix of indeterminates over A[f −1 ] is an integral domain for any l. Thus we see that ¯fl is contained in all associated prime ideals of B but one. We denote this prime ideal by Pl. Since B[( ¯fl ¯fl′)−1] = R[(flfl′)−1]/It(M )R[(flfl′ )−1] is not a zero ring by the same reason as above, we see that Pl = Pl′ for any l and l′. In particular, Pl = P1 for any l with 1 ≤ l ≤ m. Since grade( ¯f1, . . . , ¯fm)B ≥ 1 and any associated prime of B other than P1 contains ( ¯f1, . . . , ¯fm)B, we see that P1 is the only associated prime ideal of B. ]/It(M )R[f −1 ] = R[f −1 −1 l l −1 Therefore, B ⊂ B[ ¯f1 Finally we prove (3). Assume that P is a prime ideal of B with depthBP ≤ 1. Since B is Cohen-Macaulay by (1) and [16, Theorem 1 and Corollary 4] and ht( ¯f1, . . . , ¯fm)B ≥ 2, we see that P 6⊃ ( ¯f1, . . . , ¯fm)B. ] and we see that B is a domain. Take l with ¯fl 6∈ P . Then B[ ¯f −1 l as above. Since BP is a localization of B[ ¯f −1 Thus B is normal by Lemmas 4.4 and 4.5. l ] is a normal domain by Lemma 4.8 and the same argument ], we see that BP is regular by Lemmas 4.4 and 4.5. Here we note the following fact, which can be verified by considering the associated prime ideals of I and using [22, Theorems 15.5, 15.6]. Lemma 4.12 Let K be a field, x = (x1, . . . , xm) a vector of indeterminates and I a proper ideal of K[x]. Then dim K[x]/I = max{r ∃i1, . . . , ir; ¯xi1 , . . . , ¯xir are algebraically independent over K}, where ¯xi denote the natural image of xi in K[x]/I. 9 Lemma 4.13 Let K be a field, T = (tijk) a u × n × m-tensor of indeterminates, and x = Set R = K[T ], L = K(T ), M = M (x, T ) and (x1, . . . , xm) a vector of indeterminates. v′ = min{m, v}. Then L[x1, . . . , xm−v′ ] ∩ It(M )L[x] = (0), R[x1, . . . , xm−v′ ] ∩ It(M )R[x] = (0), L[x]/It(M )L[x] is algebraic over the natural image of L[x1, . . . , xm−v′ ] in L[x]/It(M )L[x] and R[x]/It(M )R[x] is algebraic over the natural image of R[x1, . . . , xm−v′ ] in R[x]/It(M )R[x]. Proof to x1, . . . , xm, we see that L ∩ It(M )L[x] = (0). Since It(M )L[x] is generated by homogeneous polynomials of positive degree with respect By Lemma 4.10, we see that It(M ) is an ideal of height v′. Thus by Lemma 4.12, we see tr.degL L[x]/It(M )L[x] = m − v′. Thus there is a permutation i1, . . . , in of 1, . . . , n such that ¯xi1 , . . . , ¯xim−v′ are algebraically independent over L and L[x]/It(M )L[x] is algebraic over L[¯xi1 , . . . , ¯xim−v′ ], where ¯xi denote the natural image of xi in L[x]/It(M )L[x]. By symmetry, we see that ¯x1, . . . , ¯xm−v′ are algebraically independent over L and L[x]/It(M )L[x] is algebraic over L[¯x1, . . . , ¯xm−v′ ]. We also see that R[x]/It(M )R[x] is algebraic over R[¯x1, . . . , ¯xm−v′ ]. Since ¯x1, . . . , ¯xm−v′ are algebraically independent over L, we see that and therefore R[x1, . . . , xm−v′ ] ∩ It(M )R[x] = (0). L[x1, . . . , xm−v′ ] ∩ It(M )L[x] = (0) Lemma 4.14 Let L/K be a field extension with charK = 0, x = (x1, . . . , xm) a vector of indeter- minates and Y a u × n × m tensor with entries in L. Set M = M (x, Y ). Suppose that the entries of Y are algebraically independent over K. Then the followings hold. (1) If m ≥ v + 1, then L[x1, . . . , xm−v] ∩ It(M )L[x] = (0) and L[x]/It(M )L[x] is algebraic over the natural image of L[x1, . . . , xm−v]. (2) If m ≥ v + 2, then L[x]/It(M )L[x] is a normal domain. In particular, It(M )L[x] is a prime ideal of L[x] of height v. Since the entries of Y are algebraically independent over K, we see by Lemma 4.13 that Proof K(Y )[x]/It(M )K(Y )[x] is algebraic over K(Y )[x1, . . . , xm−v]. Thus L[x]/It(M )L[x] is algebraic over L[x1, . . . , xm−v] since L[x]/It(M )L[x] = (K(Y )[x]/It(M )K(Y )[x]) ⊗K(Y ) L. On the other hand, since tr.degLL[x]/It(M )L[x] = dim L[x]/It(M )L[x] ≥ m − v, by Lemmas 4.9 and 4.12, we see that ¯x1, . . . , ¯xm−v are algebraically independent over L, where ¯xi denote the natural image of xi in L[x]/It(M )L[x]. Thus L[x1, . . . , xm−v] ∩ It(M )L[x] = (0). This proves (1). Next we prove (2). Take a transcendence basis S of L/K(Y ) and put A = K(Y )[x]/It(M )K(Y )[x], C = K(Y )(S)[x]/It(M )K(Y )(S)[x] and B = L[x]/It(M )L[x]. 10 By Lemma 4.10 (3), we see that A is a normal domain. Since K(Y )[S][x]/It(M )K(Y )[S][x] = (K(Y )[x]/It(M )K(Y )[x]) ⊗K(Y ) K(Y )[S] = A[S] is a polynomial ring (with possibly infinitely many variables) over A, it is an integrally closed integral domain. Since C is a localization of the above ring, C is a normal domain. Now let P be a prime ideal of B with depthBP ≤ 1. By Lemmas 4.4 and 4.5, it is enough to show that BP is regular. Set Q = C ∩ P . Then since B = C ⊗K(Y )(S) L is flat over C, we see that depthCQ ≤ 1 by [22, Theorem 23.3 Corollary]. Thus CQ is regular since C is normal. The fiber ring CQ/QCQ ⊗C BP is a localization of CQ/QCQ ⊗C B = CQ/QCQ ⊗K(Y )(S) L which is a 0-dimensional reduced ring, thus regular, since L is separably algebraic over K(Y )(S). (Note that L is an inductive limit of finitely generated algebraic extension fields of K(Y )(S). Or see [25, Theorem 3.2.6 and Theorem 3.2.8 (i)] and note the assumption of the existence of a field containing M and N is not used in the proof of [25, Theorem 3.2.8 (i)].) Thus by [22, Theorem 23.7], we see that BP is regular. Thus, B is a normal ring. Since B is a nonnegatively graded ring whose degree 0 component is a field, B is not a direct product of 2 or more rings. Therefore, B is a domain by Remark 4.2. Moreover, we see by (1), that htIt(M )L[x] = v. Definition 4.15 Let u, n, m, t and v be as in Definition 4.6. We set t A u×n×m C u×n×m I := {Y ∈ Ru×n×m the entries of Y are algebraically independent over Q}, := {Y ∈ Ru×n×m It(M (a, Y )) 6= (0) for any a ∈ R1×m \ {0}}, := Ru×n×m \ A u×n×m , t t and for Y = (Y1; . . . ; Ym) ∈ Ru×n×m and for integers k, k′ with t ≤ k ≤ u and t ≤ k′ ≤ n, we set µk,k′ (x, Y ) := [1, . . . , t − 1, k 1, . . . , t − 1, k′]M(x,Y ), where x is a vector of indeterminates. We also define Jt(x, Y ) = ∂(µtt, µt,t+1, . . . , µtn, µt+1,t, . . . , µt+1,n, . . . , µu,t, . . . , µun) (x, Y ), ∂(xm−v+1, . . . , xm) St(Y ) :=(cid:8)a ∈ R1×m det M (a, Y )<t Pt := {Y ∈ Ru×n×m St(Y ) 6= ∅}. <t 6= 0, Jt(a, Y ) 6= 0 and It(M (a, Y )) = (0) (cid:9) , Remark 4.16 (1) A u×n×m 1 ⊃ A u×n×m 2 ⊃ · · · ⊃ A u×n×m min{u,n}. (2) A u×n×m t is stabel under the action of GL(u, R) for any t. (3) C u×n×m t 6= ∅ for any t. (4) Pt is a subset of C u×n×m t and St(Y ) :=  a ∈ R1×m(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) <t 6= 0, Jt(a, Y ) 6= 0 det M (a, Y )<t and there exist linearly independent vectors bt, . . . , bn ∈ Rn such that M (a, Y )bj = 0 for any t ≤ j ≤ n   11 Lemma 4.17 Let u, n and t be as in Definition 4.6, A an integral domain and M a u × n matrix with entries in A. Suppose that det M <t <t 6= 0 and [1, . . . , t − 1, k 1, . . . , t − 1, k′]M = 0 for any integer with t ≤ k ≤ u and t ≤ k′ ≤ n. Then It(M ) = (0). In particular, <t 6= 0, Jt(a, Y ) 6= 0 det M (a, Y )<t and [1, . . . , t − 1, k 1, . . . , t − 1, k′]M = 0 for any integer with t ≤ k ≤ u and t ≤ k′ ≤ n.   (−1)t+1[1, . . . , t − 1 2, . . . , t − 1, l]M (−1)t+2[1, . . . , t − 1 1, 3, . . . , t − 1, l]M ... (−1)2t−1[1, . . . , t − 1 1, . . . , t − 2, l]M 0 (−1)2t[1, . . . , t − 1 1, . . . , t − 2, t − 1]M 0 ∈ An   St(Y ) :=  Set Proof a ∈ R1×m(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)   ξl := for each l with t ≤ l ≤ n ((−1)2t[1, . . . , t − 1 1, . . . , t − 2, t − 1]M in the l-th position). Then, since [1, . . . , t − 1 1 . . . , t − 1]M = det M <t <t 6= 0, we see that ξt, . . . , ξn are linearly independent over A. Since the k-th entry of M ξl is [1, . . . , t − 1, k 1, . . . , t − 1, l]M , we see, by assumption, that M ξl = 0 for t ≤ l ≤ n. Thus, rank M < t and we see that It(M ) = (0). It is verified the following fact, since Q is a countable field. Lemma 4.18 I is a dense subset of Ru×n×m. We also see that A u×n×m t is an open subset of Ru×n×m. First note the following fact, which is easily verified. Lemma 4.19 Let X and Y be topological spaces with X compact and f : X×Y → R is a continuous map. Set g : Y → R by g(y) := minx∈X f (x, y). Then, g is a continuous map. Corollary 4.20 A u×n×m t is an open subset of Ru×n×m. Proof Since A u×n×m t is the set consisting of Y ∈ Ru×n×m such that min a∈Sm−1 (the maximum of the absolute values of t-minors of M (a, Y )) > 0, we see the result by the previous lemma. Lemma 4.21 If v < m, then Pt is a dense subset of C u×n×m t . In particular, Pt 6= ∅. Let Y ∈ C u×n×m Proof first assertion, it suffices to show that Pt ∩ U 6= ∅. t and U an open neighborhood of Y in Ru×n×m. In order to prove the There exist a nonzero vector a ∈ R1×m and linearly independent vectors bt, . . . , bn ∈ Rn such that M (a, Y )bj = 0 for t ≤ j ≤ n. Let g3 ∈ GL(m) and g2 ∈ GL(n) such that the first entry of g⊤ 3 a is nonzero, t≤(g⊤ 2 bn) is nonsingular and sufficiently close to Em and En respectively so that (1, g−1 2 bt, 3 ) · Y ∈ U . By replacing Y , a and bt, . . . , bn by (1, g−1 2 bt, . . . , g⊤ 3 a and g⊤ 3 ) · Y , g⊤ 2 , g−1 2 , g−1 12 2 bn respectively, we may assume that the first entry of a is nonzero and t≤(bt, . . . , bn) is . . . , g⊤ nonsingular. Let e ∈ R. We take a tensor P (e) = (pijk) ∈ Ru×n×m as follows. For any i, j, k with j < t or k 6= 1, we put eij e (k = 1, i < t, j < t, ) ((i, j, k) = (t + l, t + l′, m − v + 1 + l + l′(u − t + 1)), 0 ≤ l ≤ u − t, 0 ≤ l′ ≤ n − t) 0 (otherwise) pijk =  and take pij1 for 1 ≤ i ≤ u and t ≤ j ≤ n so that M (a, P (e))bj = 0 for t ≤ j ≤ n. Note that we can take such pij1 since the first entry of a is nonzero and t≤(bt, . . . , bn) is nonsingular. Then we have for e ≫ 0. det M (a, Y + P (e))<t <t 6= 0 and Jt(a, Y + P (e)) 6= 0 Therefore, since the entries of P (e) are polynomials of e, we see that for a real number e0 6= 0 which is sufficiently closed to 0, det M (a, Y + P (e0))<n <n 6= 0, Jt(a, Y + P (e0)) 6= 0, Y + P (e0) ∈ U . (4.1) (4.2) (4.3) (4.1), (4.2) and the fact M (a, Y + P (e0))bj = M (a, Y )bj + M (a, P (e0))bj = 0 for t ≤ j ≤ n imply that a ∈ S(Y + P (e0)). Thus we have Y + P (e0) ∈ Pt and we see that Pt ∩ U 6= ∅. The latter assertion follows from Remark 4.16. Lemma 4.22 Suppose that v < m. Then the set Pt is an open subset of Ru×n×m and for any Y ∈ Pt and a = (a1, a2) ∈ St(Y ), where a1 ∈ R1×(m−v) and a2 ∈ R1×v, there exists an open neighborhood O(a, Y ) of a1 ∈ R1×(m−v) such that for any b1 ∈ O(a, Y ), there exists b2 ∈ R1×v such that (b1, b2) ∈ St(Y ). Proof Assume that Y ∈ Pt and a = (a1, a2) ∈ St(Y ). Then µkk′ (a, Y ) = 0 for any t ≤ k ≤ u and t ≤ k′ ≤ n. Thus by implicit function theorem, we see that there is an open neighborhood U of (a1, Y ) in R1×(m−v) × Ru×n×m and a continuous map ν : U → R1×v such that ν(a1, Y ) = a2, and µkk′ (b, Z) = 0 for any (b1, Z) ∈ U and any k, k′ with t ≤ k ≤ u and t ≤ k′ ≤ n, where b := (b1, ν(b1, Z)). By replacing U by a smaller neighborhood if necessary, we may assume that det M (b, Z)<t <t 6= 0 and Jt(b, Z) 6= 0 for any (b1, Z) ∈ U . Assume (b1, Z) ∈ U . Put b = (b1, ν(b1, Z)). By Lemma 4.17, we see that b ∈ St(Z). Thus it suffices to set O(a, Y ) := {b1 ∈ R1×(m−v) (b1, Y ) ∈ U }. Moreover, since {Z ∈ Ru×n×m (a1, Z) ∈ U } is an open subset of Pt containing Y , we see that Pt is an open subset of Ru×n×m. By Corollary 4.20 and Lemmas 4.21 and 4.22, we see the following: Corollary 4.23 If v < m, then Pt ⊂ int C u×n×m t and P t = int C u×n×m t = C u×n×m t . 13 Definition 4.24 We set Pt := {P Y P ∈ GL(u, R), Y ∈ Pt}. Lemma 4.25 Pt is an open subset of Ru×n×m, stable under the action of GL(u, R) and Pt = C u×n×m . t Since Pt =SP ∈GL(u,R) P Pt and P Pt is an open subset of Ru×n×m for any P ∈ GL(u, R) Proof by Lemma 4.22 and the fact that multiplication of a nonsingular matrix is a homeomorphism on Ru×n×m. Therefore, Pt is an open subset of Ru×n×m. The fact that Pt is stable under the action of GL(u, R) is clear from the definition of Pt. Finally, since A u×n×m is stable under the action . Therefore, we see that Pt = C u×n×m of GL(u, R), we see, by Remark 4.16, that Pt ⊂ C u×n×m by Lemma 4.21. t t t Lemma 4.26 Let L be an infinite field and x = (x1, . . . , xm) a vector of indeterminates. Set v′ = min{m, v} and v′′ = min{m, (u − t + 2)(n − t + 2)}. Then there is a Zariski dense open subset Q1 of Lu×n×m such that if Y ∈ Q1, then L[x]/It(M (x, Y ))L[x] is alge- braic over the natural image of L[x1, . . . , xm−v′ ], L[x1, . . . , xm−v′ ] ∩ It(M (x, Y ))L[x] = (0), htIt(M )L[x] = v′, L[x]/It−1(M (x, Y ))L[x] is algebraic over the natural image of L[x1, . . . , xm−v′′ ], L[x1, . . . , xm−v′′ ] ∩ It−1(M (x, Y ))L[x] = (0) and htIt−1(M (x, Y ))L[x] = v′′. Let T = (tijk) be the u × n × m tensor of indeterminates. Then by Lemma 4.13, we see Proof that L[T ][x]/It(M (x, T ))L[T ][x] is algebraic over the natural image of L[T ][x1, . . . , xm−v′ ]. Denote the natural image of xl in L[T ][x]/It(M (x, T ))L[T ][x] by ¯xl. Take a nonzero polynomial fl(t) with coefficient in L[T ][x1, . . . , xm−v′ ] such that fl(¯xl) = 0 for each l with m − v′ + 1 ≤ l ≤ m. Let g be the product of all nonzero elements of L[T ] appearing as the coefficient of at least one of fl and set Q′ 1 is a Zariski dense open subset of Lu×n×m. 1 = Lu×n×m \ V(g). Then Q′ Suppose that Y ∈ Q′ 1. And let xi be the natural image of xi in L[x]/It(M (x, Y ))L[x] and fl be an element of L[x1, . . . , xm−v′ ] obtained by substituting Y to T . Then fl is a nonzero element of L[x1, . . . , xm−v′ ] and fl(xl) = 0 for m − v′ + 1 ≤ l ≤ m. Therefore, L[x]/It(M (x, Y ))L[x] image of L[x1, . . . , xm−v′ ]. Thus htIt(M (x, Y ))L[x] = m − is algebraic over the natural dim L[x]/It(M )L[x] = m − tr.degLL[x]/It(M )L[x] ≥ v′. Thus, htIt(M (x, Y ))L[x] = v′ by Lemma 4.9 and we see that tr.degLL[x]/It(M (x, Y ))L[x] = m − v′. Therefore x1, . . . , xm−v′ are alge- braically independent over L, that is, L[x1, . . . , xm−v′ ] ∩ It(M (x, Y ))L[x] = (0). We see by the same way that there is a Zariski dense open subset Q′′ 1 of Lu×n×m such that if Y ∈ Q′′ 1 , then L[x]/It−1(M (x, Y ))L[x] is algebraic over the natural image of L[x1, . . . , xm−v′′ ], L[x1, . . . , xm−v′′ ] ∩ It−1(M (x, Y ))L[x] = (0) and htIt−1(M (x, Y ))L[x] = v′′. Thus it is enough to set Q1 = Q′ 1 ∩ Q′′ 1 . Let L be a field, x = (x1, . . . , xm) a vector of indeterminates and M a u × n matrix with entries <t )−1] is <t )−1] and M is equivalent to the matrix of the following form in <t ) 6∈ pIt(M ). Then It(M )L[x][det(M <t in L[x]. Suppose that htIt(M ) = v and det(M <t a proper ideal of L[x][det(M <t L[x][(det(M <t <t )−1]. (cid:18)Et−1 O ∗(cid:19) O In particular, It(M ) is a complete intersection ideal in L[x][(det(M <t that if htIt−1(M ) > v, then It(M ) is a generically complete intersection ideal. <t )−1]. By symmetry, we see 14 x′ 1 ... x′ m     = U  x1 ... xm 1, . . . , x′   . We use the notation of [11, p. 219]. Let L be a field of characteristic 0, T a u × n × m tensor of indeterminates and x = (x1, . . . , xm) a vector of indeterminates. Set M = L(T ). Suppose that m > v. Then It(M (x, T ))M[x] is a prime ideal and htIt−1(M (x, T ))M[x] > v by Lemma 4.10. Thus It(M (x, T )) : Jm−v(It(M (x, T ))) = It(M (x, T )) Thus if we set I ′ = It(M (x, T )) + by [11, Theorem 2.1] and the argument above. Jm−v(It(M (x, T ))), then htI ′ > v. Therefore the natural images of x1, . . . , xm−v in M[x]/I ′ are algebraically dependent over M by Lemma 4.12. Take a transcendence basis xi1 , . . . xid of M[x]/I ′ over M. By symmetry, we may assume that ik = k for 1 ≤ k ≤ d. Since htI ′ > v, we see that d < m − v. Take a nonzero polynomial f (t) with coefficients in L[T ][x1, . . . , xd] such that f (xm−v) ∈ I ′ and let g be the product of all nonzero elements of L[T ] which appear in some nonzero coefficient of f . Set Q2 = Lu×n×m \V(g). Then Q2 is a Zariski dense open subset of Lu×n×m and if Y ∈ Q1 ∩Q2, where Q1 is the one in Lemma 4.26, then ht(It(M (x, Y ))+Jm−v(It(M (x, Y )))) > v since tr.degLL[x]/(It(M (x, Y )) + Jm−v(It(M (x, Y )))) < m − v. Until the end of this section, assume that m ≥ v + 2 and let U be the m × m matrix of indeter- minates, T the u × n × m tensor of indeterminates, x = (x1, . . . , xm) the vector of indeterminates and L the algebraic closure of R(U ). Set 1, . . . , x′ Then L(T )[x′ m−v+1] ∩ It(M (x, T )) is a principal ideal generated by a polynomial F called the ground form of It(M (x, T )), since It(M (x, T )) is a prime ideal therefore is unmixed of height v. See [29, parts 28 and 29]. m] = L(T )[x] and L(T )[x′ Since L(T )[x′ 1, . . . , x′ m−v+1] ∩ It(M (x, T )) is the elimination ideal, F is obtained by the Buch- berger's algorithm. Let g3 be the products of all elements of L[T ] which appear as a numerator or a denominator of a nonzero coefficient of at least one polynomial in the process of Buchberger's al- gorithm to obtain the reduced Grobner basis of It(M (x, T )) in L[T ][x]. Set Q3 = Lu×n×m \ V(g3). Then Q3 is a Zariski dense open subset of Lu×n×m and if Y ∈ Q3, then the Buchberger's algo- rithm to obtain the reduced Grobner basis of It(M (x, Y ))L[x] in L[x] is identical with that of It(M (x, T ))L(T )[x] in L(T )[x]. In particular, L[x′ m−v+1] ∩ It(M (x, Y )) is a principal ideal generated by FY , the polynomial obtained by substituting Y in T in the coefficients of F . 1, . . . , x′ Let d = deg F and let PL(d, m − v + 1) (resp. PR(d, m − v + 1)) be the set of homogeneous m−v+1 and degree d. Since polynomials with coefficients in L (resp. R) with variables x′ m − v + 1 ≥ 3 and L is an algebraically closed field containing R, we see by [14] that 1, . . . , x′ {G ∈ PR(d, m − v + 1) G is absolutely irreducible} = PR(d, m − v + 1) ∩ {G ∈ PL(d, m − v + 1) G is irreducible} is a Zariski dense open subset of PR(d, m − v + 1). Definition 4.27 Set Q = {Y ∈ Q1 ∩ Q2 ∩ Q3 ∩ Ru×n×m FY is absolutely irreducible}, 15 where Q1 is the one in Lemma 4.26 and Q2, Q3 and FY are the ones defined after the proof of Lemma 4.26. Remark 4.28 Q is a Zariski open subset of Ru×n×m, since the correspondence Y to FY is a rational map whose domain contains Q3. Moreover, we see the following fact. Lemma 4.29 Q ⊃ I ∩ Q1 ∩ Q2 ∩ Q3. In particular, Q is not an empty set. Suppose that Y ∈ I ∩ Q1 ∩ Q2 ∩ Q3. Then we see, by applying Lemma 4.14 to L/Q, Proof that It(M (x, Y ))L[x] is a prime ideal. Thus the elimination ideal is also prime and therefore the generator FY of the elimination ideal is an irreducible polynomial in L[x′ m−v+1]. Therefore, Y ∈ Q. Since I is a dense subset of Ru×n×m by Lemma 4.18, we see that I ∩ Q1 ∩ Q2 ∩ Q3 6= ∅. Thus, Q 6= ∅. 1, . . . , x′ Thus we see that Q is a non-empty Zariski open subset of Ru×n×m. In particular, Q is dense. Lemma 4.30 If Y ∈ Q, then It(M (x, Y ))R[x] is a prime ideal of height v. 1, . . . , x′ m−v+1]. Thus R(U )[x′ 1, . . . , x′ Since Y ∈ Q, htIt(M (x, Y ))R[x] = v and L[x′ m−v+1] ∩ It(M (x, Y ))L[x] = Proof (FY )L[x′ 1, . . . , x′ since L is faithfully flat over R(U ). Thus we see that FY is the ground form of It(M (x, Y ))R[x] m−v+1] and therefore in [29, part 28]. R(U )[x′ 1, . . . , x′ m−v+1], we see by [29, part 31], that It(M (x, Y ))R[x] is a primary ideal. m−v+1]∩It(M (x, Y ))R(U )[x] = (FY )R(U )[x′ 1, . . . , x′ m−v+1] Since FY is an irreducible polynomial in L[x′ 1, . . . , x′ On the other hand, since Y ∈ Q1 ∩Q2, we see that ht(It(M (x, Y ))+Jm−v(It(M (x, Y )))) > v. Since It(M (x, Y )) is a primary ideal of height v, we see that It(M (x, Y )) : Jm−v(It(M (x, Y ))) = It(M (x, Y )). Therefore, by [11, Theorem 2.1], we see that It(M (x, Y )) is a radical ideal. Thus It(M (x, Y )) is a prime ideal. Now we show the following result. Theorem 4.31 Suppose that m ≥ v + 2. Set O1 = Q ∩ Pt and O2 = Q ∩ A u×n×m followings hold. t . Then the (1) O1 and O2 are disjoint open subsets of Ru×n×m and O1 6= ∅. (2) O1 ∪ O2 is a dense subset of Ru×n×m. (3) O 1 = C u×n×m t = Ru×n×m \ A u×n×m t . (4) If Y ∈ O1 ∪ O2, then It(M (x, Y ))R[x] is a prime ideal of height v. (5) If Y ∈ O1, then I(V(It(M (x, Y )))) = It(M (x, Y )). (6) If Y ∈ O2, then I(V(It(M (x, Y )))) = (x1, . . . , xm). 16 Proof The set A u×n×m open subset of Ru×n×m with A u×n×m open subset. Thus (1) holds. t t is an open subset of Ru×n×m by Corollary 4.20 and Pt is a nonempty ∩ Pt = ∅ by Lemmas 4.21 and 4.25. Further Q is a Zariski (2): Since Q and A u×n×m Lemma 4.25, we see that O1 ∪ O2 ⊃ Q ∩ (A u×n×m ∪ Pt are dense open subsets of Ru×n×m by Corollary 4.20 and ∪ Pt) is also a dense open subset of Ru×n×m. (3): Since Q is a dense subset of Ru×n×m, and Pt is an open set, we see that O 1 = Q ∩ Pt = t t Pt = C u×n×m t by Lemma 4.25. (4) follows from Lemma 4.30. (5): Assume the contrary and take g ∈ I(V(It(M (x, Y )))) with g 6∈ It(M (x, Y )). Set J = (g)R[x] + It(M (x, Y )). Then J ) It(M (x, Y ))R[x]. Since It(M (x, Y ))R[x] is a prime ideal of height v by Lemma 4.30, we see that htJ > v and therefore R[x1, . . . , xm−v] ∩ J 6= (0). Take 0 6= f ∈ J ∩ R[x1, . . . , xm−v]. Since Y ∈ Pt, we can take P ∈ GL(u, R) such that P Y ∈ Pt. Take b ∈ St(P Y ). Since O(b, P Y ) defined in Lemma 4.22 is an open set and f is a non- zero polynomial, we can take (a1, . . . , am−v) ∈ O(b, P Y ) with f (a1, . . . , am−v) 6= 0. On the other hand, we see that there are am−v+1, . . . , am ∈ R such that It(M (a, Y )) = It(P M (a, Y )) = It(M (a, P Y )) = (0) by Lemma 4.22, where a = (a1, . . . , am). Thus by assumption, we see that g(a) = 0. This contradicts to the fact that f ∈ J = (g)R[x] + It(M (x, Y ))R[x] and f (a1, . . . , am−v) 6= 0. Finally, (6) is clear from the definition of A u×n×m . t 5 Monomial preorder In this section, we introduce the notion of monomial preorder and prove a result about ideals of minors by using it. First we recall the notion of preorder. Definition 5.1 Let S be a nonempty set and (cid:22) a binary relation on S. We say that (cid:22) is a preorder on S or (S, (cid:22)) is a preordered set if the following two conditions are satisfied. (1) a (cid:22) a for any a ∈ S. (2) a (cid:22) b, b (cid:22) c ⇒ a (cid:22) c. If moreover, (3) a (cid:22) b or b (cid:22) a for any a, b ∈ S. is satisfied, then we say that (S, (cid:22)) is a totally preordered set or (cid:22) is a total preorder. Notation Let (S, (cid:22)) be a preordered set. We denote by b (cid:23) a the fact a (cid:22) b. We denote by a ≺ b or by b ≻ a the fact that a (cid:22) b and b (cid:14) a. We also denote by a ∼ b the fact that a (cid:22) b and b (cid:22) a. Remark 5.2 The binary relation ∼ defined above is an equivalence relation and if a ∼ a′ and b ∼ b′, then a (cid:22) b ⇐⇒ a′ (cid:22) b′. def⇐⇒ a (cid:22) b, In particular, we can define a binary relation ≤ on the quotient set P = S/ ∼ by ¯a ≤ ¯b where ¯a is the equivalence class which a belongs to. It is easily verified that (P, ≤) is a partially 17 ordered set and (S, (cid:22)) is a totally preordered set if and only if (P, ≤) is a totally ordered set. As usual, we denote ¯a > ¯b the fact ¯a ≥ ¯b and ¯a 6= ¯b. Definition 5.3 Let x1, . . . , xr be indeterminates. We denote the set of monomials or power products of x1, . . . , xr by PP (x1, . . . , xr). A monomial preorder on x1, . . . , xr is a total preorder (cid:22) on PP (x1, . . . , xr) satisfying the following conditions. (1) 1 (cid:22) m for any m ∈ PP (x1, . . . , xr). (2) For m1, m2, m ∈ PP (x1, . . . , xr), m1 (cid:22) m2 ⇐⇒ m1m (cid:22) m2m. Let ∼ be the equivalence relation on PP (x1, . . . , xr) defined by the monomial preorder (cid:22). We denote by P (x1, . . . , xr) the quotient set PP (x1, . . . , xr)/ ∼ and by qdeg m the class of m in P (x1, . . . , xr) and call it the quasi-degree of m, where m ∈ PP (x1, . . . , xr). Remark 5.4 Our definition of monomial preorder may seem to be different from that of [19], but it is identical except we allow m ∼ 1 for a monomial m 6= 1. Example 5.5 (c.f. [19, Example 3.1]) Let x1, . . . , xr be indeterminates, W = (w1, . . . , ws) an r × s matrix whose entries are real numbers such that the first nonzero entry of each row is positive. If one defines xa (cid:22) xb def⇐⇒ (a · w1, . . . , a · ws) ≤lex (b · w1, . . . , b · ws), where · denotes the inner product and ≤lex denotes the lexicographic order, then (cid:22) is a monomial preorder. In fact, one can prove by the same way as [28] that every monomial preorder is of this type. Definition 5.6 Let K be a field and x1, . . . , xr indeterminates. If a monomial preorder on x1, . . . , xr is defined, we say that K[x1, . . . , xr] is a polynomial ring with monomial preorder. Let f be a nonzero element of K[x1, . . . , xr]. We say that f is a form if all the monomials appearing in f have the same quasi-degree. We denote by qdeg f the quasi-degree of the monomials appearing in f . Let g be a nonzero element of K[x1, . . . , xr]. Then there is a unique expression g = g1 + g2 + · · · + gt of g, where gi is a form for 1 ≤ i ≤ t and qdeg g1 > qdeg g2 > · · · > qdeg gt. We define the leading form of g, denoted lf(g) as g1. Remark 5.7 Let K[x1, . . . , xr] be a polynomial ring with monomial preorder and f , g nonzero elements of K[x1, . . . , xr]. Then lf(f g) = lf(f )lf(g). Remark 5.8 It is essential to assume both implications in (2) of Definition 5.3. For example, let x and y be indeterminates. We define total preorder on PP (x, y) by 1 ≺ y ≺ x and m1 ≺ m2 if the total degree of m1 is less than that of m2. Then it is easily verified that (1) 1 ≺ m for any m ∈ PP (x, y) \ {1}. 18 (2) m1 (cid:22) m2 ⇒ m1m (cid:22) m2m. Let f = x + y. Then lf(f ) = x while lf(f 2) = x2 + 2xy + y2 6= x2 = (lf(f ))2. Definition 5.9 Let x1, Suppose that a total preorder on {x1, . . . , xr} is defined. {x1, . . . , xr} = {y11, . . . , y1s1, y21, . . . , y2ss, . . . , yt1, . . . , ytst }, s1 + · · · + st = r, y11 ∼ · · · ∼ y1s1 ≻ y21 ∼ · · · ∼ y2s2 ≻ · · · ≻ yt1 ∼ · · · ∼ ytst . . . . , xr be indeterminates. set {x1, . . . , xr} as Rewrite follows. the j=1 yaij ij (cid:22) i=1Qsi j=1 bt−2,j and j=1 bt−1,j and j=1 b1j. j=1 ybij ij j=1 a2j <Ps2 if and only if one of the following conditions is satisfied. The lexicographic monomial preorder on PP (x1, . . . , xr) is defined as follows. Qt i=1Qsi Qt • Ps1 j=1 a1j <Ps1 • Ps1 j=1 a1j =Ps1 • Ps1 j=1 a1j =Ps1 ... • Ps1 j=1 a1j = Ps1 Pst−1 j=1 at−1,j <Pst−1 • Ps1 j=1 a1j = Ps1 Pst j=1 at,j ≤Pst j=1 a2j =Ps2 j=1 a2j = Ps2 j=1 a2j = Ps2 j=1 b1j and Ps2 j=1 b1j, Ps2 j=1 b1j, Ps2 j=1 b1j, Ps2 j=1 a3j <Ps3 . . ., Pst−2 . . ., Pst−1 j=1 at−2,j = Pst−2 j=1 at−1,j = Pst−1 j=1 b2j and Ps3 j=1 bt−1,j. j=1 bt,j. j=1 b2j, j=1 b2j, j=1 b2j. j=1 b3j. Remark 5.10 Suppose that x1, . . . , xr are indeterminates and total preorder (cid:22) on {x1, . . . , xr} is defined. Suppose also that x1 ∼ · · · ∼ xm1 ≻ xm1+1 ∼ · · · ∼ xm2 ≻ · · · ≻ xmt−1+1 ∼ · · · ∼ xmt , mt = r. Then the lexicographic monomial preorder induced by this preorder on {x1, . . . , xr} is the one defined as in Example 5.5 by the r × t matrix whose j-th column has 1 in mj−1 + 1 through mj-th position and 0 in others, where we set m0 = 0. Definition 5.11 We set [a1, . . . , k ai, . . . , an] := [a1, . . . , ai−1, k, ai+1, . . . , an] and [a1, . . . , k ai, . . . , l aj, . . . , an] := [a1, . . . , ai−1, k, ai+1, . . . , aj−1, l, aj+1, . . . , an]. Lemma 5.12 Let K be a field, K[x1, . . . , xr] a polynomial ring with monomial preorder, S a subset of {x1, . . . , xr} and g1, . . . , gt ∈ K[x1, . . . , xr]. Set L = K[S]. If lf(g1), . . . , lf(gt) are linearly independent over L, then g1, . . . , gt are linearly independent over L. Proof Assume the contrary and suppose that cigi = 0 Xi 19 is a non-trivial relation where ci ∈ L and ci 6= 0 for any i which appears in the above sum. Then whereP′ runs through i's with qdeg lf(cigi) are maximal. Since lf(ci) ∈ L for any i, it contradicts the assumption. lf(ci)lf(gi) = 0, X′ Lemma 5.13 (Plucker relations, see e.g. [4, (4,4) Lemma]) For every u × n-matrix M, u ≥ n, with entries in a commutative ring and all indices a1, . . . , ak, bl, . . . , bn, c1, . . . , cs ∈ {1, . . . , u} such that s = n − k + l − 1 > n, t = n − k > 0 one has Xi1<···<it it+1 <···<is {1,...,s}={i1,...,is } sgn(i1, . . . , is)[a1, . . . , ak, ci1 , . . . , cit ]M [cit+1 , . . . , cis, bl, . . . , bn]M = 0, where sgn(σ) is the signature of a permutation σ and the notations are defined in Definition 4.6. An element a is called a non-zerodivisor if ab = 0 implies b = 0. Lemma 5.14 Let A = A0 ⊕ A1 ⊕ · · · be a graded ring, X = (xij ) a u × n matrix with u > n and entries in A1 and y = (y1, . . . , yn) ∈ A1×n . Set X =(cid:18)X y(cid:19) and Γ = Γ(u × n). Suppose that 1 • δ = [1, 2, . . . , n]X is a non-zerodivisor of A, • for any k1 and k2 with 1 ≤ k1 < k2 ≤ n, δX γX (γ ∈ Γ), [1, . . . , [1, . . . , n+1 k2 , . . . , n]X γX (γ ∈ Γ \ {δ}) and n+1 k1 , . . . , n]X γX (γ ∈ Γ, suppγ 6⊃ {1, . . . , k2, . . . , n}) are linearly independent over A0 and • In( X) = In(X), where the notations are defined in Definition 4.6. Then, y is an A0-linear combination of rows of X. Proof We denote γ X as γ and Xγ∈Γ as Xγ for simplicity. Set and [1, . . . , u+1 k , . . . , n] =Xγ a(k) γ γ [1, . . . , n+1 k1 , . . . , u+1 k2 , . . . , n] =Xγ a(k1,k2) γ γ 20 where a(k) γ , a(k1,k2) γ ∈ A. By considering the degree, we may assume that a(k) γ , a(k1,k2) γ ∈ A0. [1, . . . , n+1 k1 , . . . , u+1 k2 , . . . , n] XCof(X ≤n) = det(Cof(X ≤n))[1, . . . , n+1 k1 , . . . , u+1 k2 , . . . , n] = δn−1Xγ a(k1,k2) γ γ, where Cof(X ≤n) denotes the matrix of cofactors of X ≤n. On the other hand, since XCof(X ≤n) = δ δ n+1 1 , 2 . . . , n] n+2 1 , 2 . . . , n] [ [ n+1 2 , . . . , n] n+2 2 , . . . , n] [1, [1, ... ... u+1 1 , 2 . . . , n] [ u+1 2 , . . . , n] [1,   . . . · · · · · · ... · · · δ [1, 2, . . . , [1, 2, . . . , ... [1, 2, . . . , n+1 n ] n+2 n ] u+1 n ] ,   [1, . . . , n+1 k1 , . . . , u+1 k2 , . . . , n] XCof(X ≤n) we see that = δn−2 det  [1, . . . , [1, . . . , n+1 k1 , . . . , n] u+1 k1 , . . . , n] [1, . . . , [1, . . . , Since δ is a non-zerodivisor, we see that n+1 k2 , . . . , n] u+1 k2 , . . . , n]  . a(k1,k2) γ δγ Xγ = [1, . . . , n+1 k1 , . . . , n][1, . . . , u+1 k2 , . . . , n] − [1, . . . , a(k2) γ [1, . . . , = Xγ = −a(k1) δ δ[1, . . . , +a(k2) δ δ[1, . . . , n+1 k1 , . . . , n]γ −Xγ k2 , . . . , n] − Xγ∈Γ\{δ} n+1 n+1 k1 , . . . , n] + n+1 k2 , . . . , n][1, . . . , n+1 k2 , . . . , n]γ u+1 k1 , . . . , n] a(k1) γ [1, . . . , a(k1) γ [1, . . . , n+1 k2 , . . . , n]γ suppγ∩{1,...,n}={1,...,k2,...,n} X a(k2) γ [1, . . . , n+1 k1 , . . . , n]γ + Xsuppγ6⊃{1,...,k2,...,n} a(k2) γ [1, . . . , n+1 k1 , . . . , n]γ 21 Suppose that suppγ ∩ {1, . . . , n} = {1, . . . , k2, . . . , n}. Then γ = [1, . . . , k2, . . . , n, l] for some l with n + 1 ≤ l ≤ u. Thus γ = (−1)n−k2 [1, . . . , l k2, . . . , n]. By applying Lemma 5.13 to by substituting u, n, k, s, l by u, n, k1 − 1, n + 1, k1 + 1 respectively, we see that [1, . . . , l k2, . . . , n][1, . . . , n+1 k1 , . . . , n], [1, . . . , = δ[1, . . . , = δ[1, . . . , n+1 k1 , . . . , n] l k2, . . . , n][1, . . . , n+1 k1 . . . , n+1 k1 . . . , l k2, . . . , n] + [1, . . . , l k2, . . . , n] + (−1)n−k1 [1, . . . , n+1 k2 , . . . , n][1, . . . , l k1, . . . , n] n+1 k2 , . . . , n][1, . . . , k1, . . . , n, l] Therefore, if we set b(k1) γ =( a(k1) γ − (−1)k2−k1 a(k2) a(k1) γ [1,...,k2,...,n,l] if γ = [1, . . . , k1 . . . , n, l] otherwise for γ ∈ Γ \ {δ}, we see that there are bγ ∈ A0 such that Xγ + b(k1) γ bγδγ − Xγ∈Γ\{δ} Xsuppγ6⊃{1,...,k2,...,n} [1, . . . , n+1 k2 , . . . , n]γ a(k2) γ [1, . . . , n+1 k1 , . . . , n]γ = 0. Thus we see, by the assumption, that and b(k1) γ = 0 if γ ∈ Γ \ {δ} a(k2) γ = 0 if suppγ 6⊃ {1, . . . , k2, . . . , n}. Therefore, by the definition of b(k1) γ , we see that and (−1)n−k1a(k1) [1,...,k1,...,n,l] = (−1)n−k2a(k2) [1,...,k2,...,n,l] a(k1) γ = 0 if suppγ 6⊃ {1, . . . , k1, . . . , n}. Since these hold for any k1, k2 with 1 ≤ k1 < k2 ≤ n, we see that if we set cl = a(n) [1,...,n−1,l] for l with n + 1 ≤ l ≤ u, [1, . . . , u+1 k , . . . , n] = a(k) δ δ + (−1)n−k cl[1, . . . , k, . . . , n, l] u Xl=n+1 cl[1, . . . , l k, . . . , n] = a(k) δ δ + u Xl=n+1 22 for any k with 1 ≤ k ≤ n. Set and Then z = y − clX =l u Xl=n+1 n a(s) δ X =s − Xs=1 Z =(cid:18)X z(cid:19) . [1, . . . , u+1 k , . . . , n]Z = [1, . . . , u+1 k , . . . , n] X − = 0 a(s) δ [1, . . . s k, . . . n]X − n Xs=1 cl[1, . . . l k, . . . n]X u Xl=n+1 for any k with 1 ≤ k ≤ n. Thus zCof(X ≤n) = 0 and we see that z = 0 since det Cof(X ≤n) = δn−1 is a non-zerodivisor. Lemma 5.15 Let L be a field, n, m integers with n ≥ m ≥ 3, x1, . . . , xm indeterminates and α, β ∈ L with 0 6= α 6= β 6= 0. Then the following polynomials are linearly independent over L. (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2) (1 ≤ b1 ≤ · · · ≤ bn ≤ m − 1, bn−1 ≥ 3) (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2, 3 ≤ s ≤ m − 1) (x2 − βxm) xs (3 ≤ s ≤ m − 1) x2n 1 x2n−1 1 x2n−1 1 xn 1 (x2 − αxm)(x2 − βxm)xb1 · · · xbn−2 xn 1 (x2 − αxm)xb1 · · · xbn−2 xs xn 1 xb1 · · · xbn x2n−2 1 x2n−2 1 xn−1 1 xn−1 1 xn−1 1 xn−2 1 xn−2 1 xn−2 1 Set deg x1 = deg x2 = deg xm = 0 and deg x3 = · · · = deg xm−1 = 1. Then the Proof polynomials under consideration are homogeneous. Thus it is enough to show that for each integer d, the polynomials of degree d in the above list are linearly independent over L. (x2 − βxm)2 (x2 − βxm)xs (x2 − αxm)(x2 − βxm)2xb1 · · · xbn−2 (x2 − αxm)(x2 − βxm)xb1 · · · xbn−2xs (x2 − βxm)xb1 · · · xbn (x2 − αxm)2(x2 − βxm)2xb1 · · · xbn−2 (x2 − αxm)2(x2 − βxm)xb1 · · · xbn−2 xs (x2 − αxm)(x2 − βxm)xb1 · · · xbn (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2) (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2, 3 ≤ s ≤ m − 1) (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2, 3 ≤ s ≤ m − 1) (1 ≤ b1 ≤ · · · ≤ bn ≤ m − 1, bn−1 ≥ 3) (3 ≤ s ≤ m − 1) (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2) (1 ≤ b1 ≤ · · · ≤ bn ≤ m − 1, bn−1 ≥ 3) First consider the polynomials with degree more than 1. They are (1 ≤ b1 ≤ · · · ≤ bn ≤ m − 1, bn−1 ≥ 3) xn 1 xb1 · · · xbn xn−1 1 xn−2 1 (x2 − βxm)xb1 · · · xbn (x2 − αxm)(x2 − βxm)xb1 · · · xbn (1 ≤ b1 ≤ · · · ≤ bn ≤ m − 1, bn−1 ≥ 3) (1 ≤ b1 ≤ · · · ≤ bn ≤ m − 1, bn−1 ≥ 3). 23 By first substituting β−1x2 to xm and next by substituting α−1x2 to xm one sees that these polynomials are linearly independent. Next consider the polynomials with degree 1. They are xs (3 ≤ s ≤ m − 1) x2n−1 1 xn 1 (x2 − αxm)xb1 · · · xbn−2 xs x2n−2 1 xn−1 1 (x2 − βxm)xs (x2 − αxm)(x2 − βxm)xb1 · · · xbn−2 xs (3 ≤ s ≤ m − 1) (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2, 3 ≤ s ≤ m − 1) (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2, 3 ≤ s ≤ m − 1) xn−2 1 (x2 − αxm)2(x2 − βxm)xb1 · · · xbn−2 xs (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2, 3 ≤ s ≤ m − 1). By a similar but more subtle argument as above, one sees that these polynomials are linearly independent. Finally consider the polynomials with degree 0. They are (x2 − βxm) x2n 1 x2n−1 1 xn 1 (x2 − αxm)(x2 − βxm)xb1 · · · xbn−2 x2n−2 1 xn−1 1 xn−2 1 (x2 − βxm)2 (x2 − αxm)(x2 − βxm)2xb1 · · · xbn−2 (x2 − αxm)2(x2 − βxm)2xb1 · · · xbn−2 (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2) (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2) (1 ≤ b1 ≤ · · · ≤ bn−2 ≤ 2) By a similar but more subtle argument, one sees that these polynomials are linearly independent. Lemma 5.16 Let K be a field, T = (T1; · · · ; Tm) = (tijk) a u × n × m-tensor of indeterminates with u > n ≥ m ≥ u − n + 2, x = (x1, . . . , xm) a vector of indeterminates. Set X = M (x, T ) (c.f. Definition 4.6), Γ = Γ(u × n), δ = δ0 = [1, . . . , n] and A = K[T ][x]. Then δX is a non-zerodivisor of A and for any k1, k2 with 1 ≤ k1, k2 ≤ n, k1 6= k2, δX γX (γ ∈ Γ) [1, . . . , [1, . . . , n+1 k2 , . . . , n]XγX (γ ∈ Γ \ {δ}) n+1 k1 , . . . , n]XγX (γ ∈ Γ, suppγ 6⊃ {1, . . . , k2, . . . , n}) are linearly independent over K[T ]. Proof The first assertion is clear since A is an integral domain and δX 6= 0. Next we prove the second assertion. By symmetry, we may assume that k1 = n − 1 and k2 = n. Set δ1 = [1, 2, . . . , n − 1, n + 1] and δ2 = [1, 2, . . . , n − 2, n, n + 1]. We introduce the lexicographic monomial preorder induced by the preorder on the indeterminates defined as follows. If one of the following is satisfied, we define tijk ≻ ti′j′k′ . 24 • i < i′. • i = i′ and j < j′. • i = i′, j = j′, k < k′ and "i < j or i > j + u − n". In case 0 ≤ i − j ≤ u − n and (i, j) 6= (n, n − 1), (n + 1, n), we define ti,j,i−j+1 ≻ ti,j,i−j+2 ≻ · · · ≻ ti,j,m ≻ ti,j,1 ≻ · · · ≻ ti,j,i−j . In case (i, j) = (n, n − 1) or (n + 1, n), we define And for any i, j and k. Set Then ti,j,2 ∼ ti,j,m ≻ ti,j,1 ≻ ti,j,3 ≻ · · · ≻ ti,j,m−1. tijk ≻ x1 ∼ x2 ∼ · · · ∼ xm Γ0 = {[a1, . . . , an] ∈ Γ an−1 = n − 1}, Γ1 = {[a1, . . . , an] ∈ Γ an−1 = n}, Γ2 = {[a1, . . . , an] ∈ Γ an−1 ≥ n + 1}, Γ00 = {[a1, . . . , an] ∈ Γ0 an = n} = {δ0}, Γ01 = {[a1, . . . , an] ∈ Γ0 an = n + 1} = {δ1}, Γ02 = {[a1, . . . , an] ∈ Γ0 an ≥ n + 2}, Γ11 = {[a1, . . . , an] ∈ Γ1 an = n + 1} = {δ2}, Γ12 = {[a1, . . . , an] ∈ Γ1 an ≥ n + 2}. Γ = Γ0 ⊔ Γ1 ⊔ Γ2 Γ0 = Γ00 ⊔ Γ01 ⊔ Γ02 Γ1 = Γ11 ⊔ Γ12. Set α1 = tn,n−1,2, α2 = tn,n−1,m, β1 = tn+1,n,2 and β2 = tn+1,n,m. Then for γ = [a1, . . . , an] ∈ Γ, lf(γX ) is, up to multiplication of nonzero element of K[T ], as follows. if γ ∈ Γ01 if γ ∈ Γ00 (β1x2 + β2xm) xan−n+1 xn 1 xn−1 1 xn−1 1 xa1 xa2−1 · · · xan−2−n+3(α1x2 + α2xm)(β1x2+β2xm) xa1 xa2−1 · · · xan−2−n+3(α1x2 + α2xm)xan−n+1 xa1 xa2−1 · · · xan−n+1 if γ ∈ Γ02 if γ ∈ Γ11 if γ ∈ Γ12 if γ ∈ Γ2. 25 Therefore, for γ = [a1, . . . , an] ∈ Γ, lf(δX γX ) is, up to multiplication of nonzero element of K[T ], if γ ∈ Γ01 if γ ∈ Γ00 (β1x2 + β2xm) xan−n+1 x2n 1 x2n−1 1 x2n−1 1 xn 1 xa1 xa2−1 · · · xan−2−n+3(α1x2 + α2xm)(β1x2+β2xm) xn 1 xa1 xa2−1 · · · xan−2−n+3(α1x2 + α2xm)xan−n+1 xn 1 xa1 xa2−1 · · · xan−n+1 if γ ∈ Γ02 if γ ∈ Γ2. if γ ∈ Γ11 if γ ∈ Γ12 For γ = [a1, . . . , an] ∈ Γ \ {δ}, lf((δ1)X γX ) is, up to multiplication of nonzero element of K[T ], x2n−2 1 x2n−2 1 xn−1 1 xn−1 1 xn−1 1 if γ ∈ Γ01 (β1x2 + β2xm)2 (β1x2 + β2xm)xan−n+1 xa1 xa2−1 · · · xan−2−n+3(α1x2 + α2xm)(β1x2+β2xm)2 xa1 xa2−1 · · · xan−2−n+3(α1x2 + α2xm)(β1x2+β2xm)xan−n+1 (β1x2+β2xm)xa1 xa2−1 · · · xan−n+1 if γ ∈ Γ02 if γ ∈ Γ2. if γ ∈ Γ11 if γ ∈ Γ12 Finally, consider the leading form of (δ2)X γX , where γ = [a1, . . . , an] ∈ Γ and suppγ 6⊃ {1, . . . , n − 1}. It is easily verified that suppγ 6⊃ {1, . . . , n − 1} if and only if γ ∈ Γ1 ⊔ Γ2. Thus the leading form of (δ2)X γX is, up to multiplication of nonzero element of K[T ], xa1 xa2−1 · · · xan−2−n+3(α1x2 + α2xm)2(β1x2+β2xm)2 xa1 xa2−1 · · · xan−2−n+3(α1x2 + α2xm)2(β1x2+β2xm)xan−n+1 (α1x2 + α2xm)(β1x2+β2xm)xa1 xa2−1 · · · xan−n+1 if γ ∈ Γ2. if γ ∈ Γ11 if γ ∈ Γ12 xn−2 1 xn−2 1 xn−2 1 Since 1 ≤ a1 ≤ a2 − 1 ≤ · · · ≤ an − n + 1 ≤ u − n + 1 ≤ m − 1 an−2 − n + 3 ≤ 2 if γ ∈ Γ1 an−1 − n + 2 ≥ 3 if γ ∈ Γ2 and an − n + 1 ≥ 3 if γ ∈ Γ02 ⊔ Γ12, we see by Lemma 5.15 that (γ ∈ Γ) lf(δX γX ) lf((δ1)X γX ) lf((δ2)X γX ) (γ ∈ Γ \ {δ}) (γ ∈ Γ, suppγ 6⊃ {1, . . . , n − 1}) are linearly independent over K[T ]. The assertion follows by Lemma 5.12. Corollary 5.17 Let K be a field, T a u × n × m tensor of indeterminates with u > n ≥ m ≥ u − n + 2, R a commutative ring containing K(T ), x = (x1, . . . , xm) a vector of indeterminates. 26 Set M = M (x, T ) (c.f. Definition 4.6) and B = R[x]. Then δM is a non-zerodivisor of B and for any k1, k2 with 1 ≤ k1, k2 ≤ n, k1 6= k2, δM γM (γ ∈ Γ) [1, . . . , [1, . . . , n+1 k2 , . . . , n]M γM (γ ∈ Γ \ {δ}) n+1 k1 , . . . , n]M γM (γ ∈ Γ, suppγ 6⊃ {1, . . . , k2, . . . , n}) are linearly independent over R. Proof Set A = K[T ][x]. Then B = A ⊗K[T ] R. Since R is flat over K[T ], we see that B is flat over A. By Lemma 5.16, we see that δM is a non-zerodivisor of A and for any k1, k2 with 1 ≤ k1, k2 ≤ n, k1 6= k2, δM γM (γ ∈ Γ) [1, . . . , [1, . . . , n+1 k2 , . . . , n]M γM (γ ∈ Γ \ {δ}) n+1 k1 , . . . , n]M γM (γ ∈ Γ, suppγ 6⊃ {1, . . . , k2, . . . , n}) are linearly independent over K[T ]. Since R (resp. B) is flat over K[T ] (resp. A), we see that δM is a non-zerodivisor of B and for any k1, k2 with 1 ≤ k1, k2 ≤ n, k1 6= k2, δM γM (γ ∈ Γ) [1, . . . , [1, . . . , n+1 k2 , . . . , n]M γM (γ ∈ Γ \ {δ}) n+1 k1 , . . . , n]M γM (γ ∈ Γ, suppγ 6⊃ {1, . . . , k2, . . . , n}) are linearly independent over R. Corollary 5.18 Let x = (x1, . . . , xm) be a vector of indeterminates. Suppose that u > n and Y ∈ I (c.f. Definition 4.15) and set M = M (x, Y ). Then δM is a non-zerodivisor of R[x] and for any k1, k2 with 1 ≤ k1, k2 ≤ n, k1 6= k2, δM γM (γ ∈ Γ) [1, . . . , [1, . . . , n+1 k2 , . . . , n]M γM (γ ∈ Γ \ {δ}) n+1 k1 , . . . , n]M γM (γ ∈ Γ, suppγ 6⊃ {1, . . . , k2, . . . , n}) are linearly independent over R. Definition 5.19 Let x = (x1, . . . , xm) be a vector of {Y ∈ Ru×n×m n+1 k2 , . . . , n]M γM (γ ∈ Γ \ {δ}), n+1 k1 , . . . , n]M γM (γ ∈ Γ, suppγ 6⊃ {1, . . . , k2, . . . , n}) are linearly independent over R for indeterminates. We set Q′ := δM 6= 0 and δM γM (γ ∈ Γ), [1, . . . , [1, . . . , any k1, k2 with 1 ≤ k1 < k2 ≤ n, where M = M (x, Y )}. 27 Remark 5.20 Q′ is a Zariski open set of Ru×n×m and by Corollary 5.18, we see that Q′ ⊃ I . In particular, Q′ is a Zariski dense open subset of Ru×n×m. By Lemma 5.14, we see the following fact. Proposition 5.21 Let u, n and m be integers with u > n ≥ m ≥ u−n+2 and let x = (x1, . . . , xm) be a vector of indeterminates. Suppose that Y ∈ Q′ and y ∈ R1×n×m. Set Y =(cid:18)Y y(cid:19) . If In(M (x, Y )) = In(M (x, Y )), then fl1(y) is an R-linear combination of rows of fl1(Y ). Proof that δM is a non-zerodivisor of R[x]. Moreover, Set M = M (x, Y ). Since R[x] is a domain and δM 6= 0 by the definition of Q′, we see δM γM (γ ∈ Γ) [1, . . . , [1, . . . , n+1 k2 , . . . , n]M γM (γ ∈ Γ \ {δ}) n+1 k1 , . . . , n]M γM (γ ∈ Γ, suppγ 6⊃ {1, . . . , k2, . . . , n}) are linearly independent over R for any k1, k2 with 1 ≤ k1 < k2 ≤ n by the definition of Q′. Thus by Lemma 5.14, we see that M (x, y) is an R-linear combination of rows of M = M (x, Y ). Since x1, . . . , xm are indeterminates, we see that fl1(y) is an R-linear combination of rows of fl1(Y ). 6 Tensor with rank p Let 3 ≤ m ≤ n, (m − 1)(n − 1) + 1 ≤ p ≤ (m − 1)n and set l = (m − 1)n − p and u = n + l. In the following of this paper, we use the results of the previous sections by setting t = n. See Definition 4.6. Then v = l + 1 and it follows that v < m since l ≤ m − 2. Note also u + p = nm. We make bunch of definitions used in the sequel of this paper. Definition 6.1 We put V = V n×p×m := {T ∈ Rn×p×m fl2(T )≤p is nonsingular} and define σ : V → Ru×p be a map defined as σ(T ) = (p<fl2(T ))(fl2(T )≤p)−1. We denote by A u×n×m the set of all u × n × m absolutely full column rank tensors and put in the C u×n×m = Ru×n×m \ A u×n×m. Note that A u×n×m = A u×n×m notation of Definition 4.15. n and C u×n×m = C u×n×m n Let M be the subset of Ru×nm consisting of all matrices W = (W1, . . . , Wm) satisfying that there are A = (a1, . . . , ap) ∈ Rn×p and p × p diagonal matrices D1, . . . , Dm such that Dk = 28 Diag(d1k, . . . , dpk) for 1 ≤ k ≤ m, (dj1W1 + · · · + djmWm)aj = 0 for 1 ≤ j ≤ p, and AD1 ... ADm−2 A≤n−lDm−1     is nonsingular. Let ι : Ru×p → Ru×nm be a map which sends A to (A, −Eu). Moreover put (6.1) C := {W ∈ Ru×nm W /∈ fl1(A u×n×m)} = {W ∈ Ru×n×m W ∈ fl1(C u×n×m)}. We define φ : R1×m × Rn → Rp a map defined as φ(a, b) =  a1b a2b ... am−1b≤n−l   ∈ Rp, where a = (a1, . . . , am−1, am). Recall that the set A u×n×m is an open subset of Ru×n×m by Lemma 3.3 or Corollary 4.20. Proposition 6.2 σ is an open, surjective and continuous map. Proof Clearly σ is surjective and continuous. Let O be an open subset of V and let h : Rnm×p → Rp×p ×Ru×p be a homeomorphism defined as h(M ) = (M ≤p, p<M ). Then h(fl2(O)) can be written by for some open subsets O1,λ ⊂ Rp×p and O2,λ ⊂ Ru×p and thus O1,λ × O2,λ [λ σ(O) =[λ [A∈O1,λ O2,λA−1. The set O2,λA−1 is open and then σ(O) is open. The following fact follows from the definition. Lemma 6.3 M is stable under the action of GL(u, R). Lemma 6.4 M ⊂ C. Proof By observing the first column of (6.1), we see that a1 6= 0 and at least one of d11, d12, . . . , d1,m−1 is nonzero, since   AD1 ... ADm−2 A≤n−lDm−1   1 ≤ k ≤ m − 1. Since d11W1 + · · · + d1mWm is singular, (W1; . . . ; Wm) is not absolutely full column rank, i.e., M ⊂ C. is nonsingular, where Dk = Diag(d1k, . . . , dpk) for 29 Theorem 6.5 Let X ∈ V ⊂ Rn×p×m. Put (W1, . . . , Wm−1, Wm) = ι(σ(X)), where W1, . . . , Wm ∈ Ru×n. The following four statements are equivalent. (1) rank X = p. (2) There are an n × p matrix A, and diagonal p × p matrices D1, . . . , Dm such that W1AD1 + W2AD2 + · · · + Wm−1ADm−1 + WmADm = O (6.2) and is nonsingular. (3) ι(σ(X)) ∈ M. AD1 ... ADm−2 A≤n−lDm−1 N =    (6.3) Proof Then rank X = rank(S1; . . . ; Sm) and It holds that rank X ≥ p, since fl2(X)≤p has rank p. Put (S1; . . . ; Sm) = X(fl2(X)≤p)−1 (cid:18)n−l<Sm−1 Sm (cid:19) = (W1, W2, . . . , Wm−2, (Wm−1)≤n−l). (1) ⇒ (2): Since rank(S1; . . . ; Sm) = p, there are an n × p-matrix A, a p × p-matrix Q, and p × p diagonal matrices D1, . . . , Dm such that ADkQ = Sk for k = 1, . . . , m. Since Sm and ADk = SkN for k = m − 1, m, we see that Ou×p = (cid:18)n−l<Sm−1 ADm = W1AD1 + · · · + Wm−2ADm−2 (cid:19) N −(cid:18)n−l<ADm−1 Sm (cid:19) +(Wm−1)≤n−lA≤n−lDm−1 +(cid:18)−El O (cid:19) n−l<ADm−1 +(cid:18) O −En(cid:19) ADm = W1AD1 + · · · + Wm−2ADm−2 + Wm−1ADm−1 + WmADm. 30 we see that N and Q are nonsingular and Q−1 = N . Since S1 ... Sm−2 = Ep, ADm−2 AD1 ...   Q =  N Q =  (cid:18)n−l<Sm−1 (cid:19) = σ(X) = (W1, W2, . . . , Wm−2, (Wm−1)≤n−l), A≤n−lDm−1 (Sm−1)≤n−l   Therefore the equation (6.2) holds. (2) ⇒ (1): Set Q = N −1. Then, since N Q = Ep, we see that and Furthermore, since ADkQ = Sk 1 ≤ k ≤ m − 2 A≤n−lDm−1Q = S≤n−l m−1 . W1AD1 + · · · + Wm−2ADm−2 +(Wm−1)≤n−lA≤n−lDm−1 +(cid:18)−El O (cid:19) n−l<ADm−1 +(cid:18) O −En(cid:19) ADm = W1AD1 + · · · + WmADm = Ou×p, we see that Thus (cid:18)n−l<Sm−1 Sm (cid:19) N =(cid:18)El (cid:18)n−l<ADm−1 En(cid:19) ADm. O(cid:19) n−l<ADm−1 +(cid:18) O (cid:19) Q =(cid:18)n−l<Sm−1 (cid:19) ADm Sm and we see that ADkQ = Sk for k = m − 1, m. Therefore, rank X = rank(S1; . . . ; Sm) ≤ p and we see (1). Finally it is easy to see that (2) ⇔ (3). Since M ⊂ C by Lemma 6.4, we see the following: Proposition 6.6 For X ∈ V , if rank X = p, then ι(σ(X)) 6∈ fl1(A u×n×m). 7 Contribution of absolutely full column rank property Let m, n and p be integers with 3 ≤ m ≤ n and (m − 1)(n − 1) + 1 ≤ p ≤ (m − 1)n. We set u = nm − p and t = n and use the results of Sections 4 and 5. Note v = u − n + 1 = (m − 1)n − p + 1 in the notation of Definition 4.6. It is known that the generic rank grank(n, p, m) of n × p × m tensors over C is equal to p ([5, Theorem 3.1] or [6, Theorem 2.4 and Remark 2.5]) and it is also equal to the minimal typical rank of n × p × m tensors over R. Thus if we discuss the plurality of typical ranks, it is enough to consider whether there exists a typical rank that is greater than p or not. Definition 7.1 We set A := ι−1(fl1(A u×n×m)) ⊂ Ru×p, where ι and A u×n×m are defined in Definition 6.1. Lemma 7.2 If Y ∈ V n×p×m and σ(Y ) ∈ A, then rank Y > p. Proof This follows from the fact that rank Y ≥ p if Y ∈ V n×p×m and Proposition 6.6. 31 Theorem 7.3 If A u×n×m 6= ∅, then there are plural typical ranks of n × p × m tensors over R. Proof By Lemma 3.5, we see that A 6= ∅. Since A u×n×m is an open subset of Ru×n×m, we see that A is an open subset of Ru×p. Moreover, since σ : V n×p×m → Ru×p is a surjective continuous map, we see that σ−1(A) is a nonempty open subset of V n×p×m. Thus, there is a typical rank greater than p by Lemma 7.2. Since p is a typical rank of n × p × m tensors over R, we see that there are plural typical ranks of n × p × m tensors over R. From now on until the end of this section, we assume that p ≥ (m − 1)(n − 1) + 2. Thus, m ≥ v + 2. Definition 7.4 Let Y ∈ Ru×n×m and let x = (x1, . . . , xm) be a vector of indeterminates. For i1, . . . , in−1 ∈ {1, . . . , u}, we set ψi1,...,in−1(x, Y ) :=  For the definition [a1, . . . , at Ru×n×m → R[x]p by (−1)n+1[i1, . . . , in−1 2, . . . , n − 1, n]M(x,Y ) (−1)n+2[i1, . . . , in−1 1, 3, . . . , n − 1, n]M(x,Y ) ... (−1)2n[i1, . . . , in−1 1, . . . , n − 2, n − 1]M(x,Y ) ∈ R[x]n.   b1, . . . , bt], see Definition 4.6. We define ψi1,...,in−1 : R1×m × ψi1,...,in−1(x, Y ) :=  x1ψi1,...,in−1 (x, Y ) x2ψi1,...,in−1 (x, Y ) ... xm−1ψi1,...,in−1(x, Y ) ≤p ∈ R[x]p.   We also define the R-vector space U (Y ) by U (Y ) := h ψi1 ,...,in−1 (u, Y ) u ∈ V(In(M (x, Y ))), i1, . . . , in−1 ∈ {1, . . . , u} i ⊂ Rp. For c = (c11, . . . , cn1, c12, . . . , cn2, . . . , c1m, . . . , cnm) ∈ R1×nm, we set Zk = (cid:18) 1 ≤ k ≤ m, Z = (Z1; . . . ; Zm) ∈ R(u+1)×n×m and Yk c1k · · · cnk(cid:19) for gi1,...,in−1(x, Y, c) = [i1, . . . , in−1, u + 1]M(x,Z) for any i1, . . . , in−1 ∈ {1, . . . , u}. For the definition [i1, . . . , in−1, u + 1]M(x,Z), see Definition 4.6. Lemma 7.5 Suppose that Y ∈ Ru×n×m. Then the following claims are equivalent. (1) dim U (Y ) = p. (2) If c ∈ R1×nm satisfies the following conditions, then c = 0. (∗) p<c = 0 and 32 (∗∗) gi1,...in−1 (u, Y, c) = 0 for any u ∈ V(In(M (x, Y ))) and any i1, . . . , in−1 ∈ {1, . . . , u}. The vector d ∈ Rp is perpendicular to U (Y ) if and only if d is perpendicular to Proof ψi1,...,in−1(u, Y ) for any u ∈ V(In(M (x, Y ))) and any i1, . . . , in−1. Since the inner product of ψi1,...,in (u, Y ) with d is gi1,...,in−1(u, Y, (d⊤, 0)), the result follows. Next we show the following result. For the definition of M, see Definition 6.1. Lemma 7.6 If dim U (Y ) = p, then fl1(Y ) ∈ M. Proof V(In(M (x, Y ))) and t11, . . . , t1,n−1, . . . , tp1, . . . , tp,n−1 such that Set Y = (Y1; . . . ; Ym). Suppose that dim U (Y ) = p. Then there are u1, . . . , up ∈ ψt11,...,t1,n−1(u1, Y ), . . . , ψtp1,...,tp,n−1(up, Y ) are linearly independent over R. Set uj = (uj1, . . . , ujm) for 1 ≤ j ≤ p, Dk = Diag(u1k, . . . , upk) for 1 ≤ k ≤ m and A = (ψt11,...,t1,n−1(u1, Y ), . . . , ψtp1,...,tp,n−1(up, Y )). Then,     =  AD1 ... ADm−2 ADm−1 ≤p   = ( ψt11,...,t1,n−1(u1, Y ), . . . , ψtp1,...,tp,n−1(up, Y )) AD1 ... ADm−2 A≤n−lDm−1 is a nonsingular matrix and (uj1Y1 + · · · + ujmYm)ψtj1 ,...,tj,n−1(uj, Y ) =  since In(M (uj, Y )) = (0) for 1 ≤ j ≤ p. [1, tj1, . . . , tj,n−1]M(uj ,Y ) [2, tj1, . . . , tj,n−1]M(uj ,Y ) ... [u, tj1, . . . , tj,n−1]M(uj ,Y ) = 0   Definition 7.7 Set U := {Y ∈ Ru×n×m p<fl1(Y ) is nonsingular}, O3 := U ∩ Q ∩ Q′ ∩ Pn = O1 ∩ U ∩ Q′ and O4 := U ∩ Q ∩ Q′ ∩ A u×n×m = O2 ∩ U ∩ Q′, where Q, Q′ and Pn are the ones defined in Definitions 4.27, 5.19, and 4.24 and O1 and O2 are the ones in Theorem 4.31 under t = n. Define ν : U → Ru×p as ν(Y ) := −(p<fl1(Y ))−1fl1(Y )≤p for i = 1, 2, where σ is the one defined in Definition 6.1. Set Oi = ν(Oi+2) ⊂ Ru×p and Ti = σ−1(Oi)⊂ V n×p×m for i = 1, 2. The following fact is immediately verified. Lemma 7.8 ι(ν(Y )) = fl1(−(p<fl1(Y ))−1Y ). By the same way as Theorem 4.31 (1), (2) and (3), we see the following fact. Lemma 7.9 Then the followings hold. 33 (1) O3 and O4 are disjoint open subsets of Ru×n×m and O3 is nonempty. (2) O3 ∪ O4 is a dense subset of Ru×n×m. (3) O3 = C u×n×m = Ru×n×m \ A u×n×m. Lemma 7.10 fl1(O3) ⊂ M. Proof and (∗∗), then c = 0. Let Y ∈ O3. By Lemmas 7.6 and 7.5, it is enough to show that if c ∈ R1×nm satisfies (∗) Set c = (c1, . . . , cm), where cj ∈ R1×n, c′ = (c1; · · · ; cm) ∈ R1×n×m and Y =(cid:18)Y (∗∗), gi1,...,in−1 (x, Y, c) ∈ I(V(In(M (x, Y )))) for any i1, . . . , in−1 ∈ {1, . . . , u}. Therefore, by the definition of O3 and Theorem 4.31 (5), we see that gi1,...,in−1(x, Y, c) ∈ In(M (x, Y )) for any i1, . . . , in−1 ∈ {1, . . . , u}. Thus we see that In(M (x, Y )) = In(M (x, Y )). Thus, by Proposition 5.21, we see that fl1(c) is an R-linear combination of rows of fl1(Y ). Since p<c = 0 and Y ∈ U , we see that c = 0. c′(cid:19). Then by By the same way as Proposition 6.2, we see the following: Proposition 7.11 ν is an open, surjective and continuous map. We see the following fact. Lemma 7.12 Then the followings hold. (1) Y ∈ A u×n×m if and only if ν(Y ) ∈ A for Y ∈ U . (2) O1 and O2 are disjoint open subsets of Ru×p and O1 6= ∅. (3) O1 ∪ O2 is a dense subset of Ru×p. (4) O1 = Ru×p \ A. (5) O2 ⊂ A and O2 = A. (1): Suppose that Y ∈ U . Since A u×n×m and U are stable under the action Proof of GL(u, R), we see that Y ∈ A u×n×m if and only if −(p<fl1(Y ))−1Y ∈ A u×n×m. Since ι(ν(Y )) = fl1(−(p<fl1(Y ))−1Y ) and fl1 is a bijection, we see (1). We see (2) by the facts that Pn and A u×n×m are stable under the action of GL(u, R), Lemma 7.9, and Proposition 7.11. (3) also follows from Lemma 7.9 and Proposition 7.11. We see by (1) that if Y ∈ O3, then ν(Y ) 6∈ A. Thus O1 = ν(O3) ⊂ Ru×p \ A. Since O3 = Ru×n×m \ A u×n×m by Lemmas 7.9 (3), we see that O1 ⊃ ν(O3 ∩ U ) = ν((Ru×n×m \ A u×n×m) ∩ U ) = Ru×p \ A by (1) and the surjectivity of ν. Thus we see (4). Therefore O2 ⊂ A by (2). Further, we see that O2 ⊃ A by (3) and (4). Thus we see (5). Lemma 7.13 Let X and Y be topological spaces, f : X → Y a mapping and B a subset of Y . (1) If f is continuous, then f −1(B) ⊃ f −1(B). 34 (2) If f is an open map, then f −1(B) ⊂ f −1(B). Proof f −1(B). (1): Since f −1(B) is a closed subset of X containing f −1(B), we see that f −1(B) ⊃ (2): Suppose that x ∈ f −1(B) and let U be an open neighborhood of x. We show that U ∩ f −1(B) 6= ∅. Since f (x) ∈ B, f (x) ∈ f (U ) and f (U ) is an open subset of Y , we see that f (U ) ∩ B 6= ∅. Take b ∈ f (U ) ∩ B and a ∈ U such that f (a) = b. Then, since f (a) ∈ B, we see that a ∈ f −1(B). Thus, a ∈ U ∩ f −1(B) and we see that U ∩ f −1(B) 6= ∅. Theorem 7.14 Let m, n and p be integers with 3 ≤ m ≤ n and (m − 1)(n − 1) + 2 ≤ p ≤ (m − 1)n. The followings hold. (1) T1 and T2 are disjoint open subsets of V n×p×m and T1 is nonempty. (2) T1 ∪ T2 is a dense subset of Rn×p×m. (3) T1 ∩ V n×p×m = V n×p×m \ σ−1(A) and T2 ∩ V n×p×m = σ−1(A) ∩ V n×p×m. (4) If T ∈ T1, then rank T = p. (5) If T ∈ T2, then rank T > p. Proof since σ is an open continuous map. First note that σ−1(X ) ∩ V n×p×m = σ−1(X ) for any subset X of Ru×p by Lemma 7.13, (1) and (2) follow from Lemma 7.12 and the facts that σ is surjective and V n×p×m is a dense subset of Rn×p×m. (3): We see by Lemma 7.12 that T 1∩V n×p×m = σ−1(O1) = σ−1(Ru×p\A) = V n×p×m\σ−1(A) and T 2 ∩ V n×p×m = σ−1(O2) = σ−1(A) ∩ V n×p×m. (4): Suppose that T ∈ T1. Then σ(T ) ∈ O1. Thus there exists Y ∈ O3 such that ν(Y ) = σ(T ). By Lemma 7.10, we see that fl1(Y ) ∈ M. Hence ι(σ(T )) = ι(ν(Y )) = −(p<fl1(Y ))−1fl1(Y ) ∈ M, since M is stable under the action of GL(u, R). Therefore rank T = p by Theorem 6.5. (5): If T ∈ T2, then σ(T ) ∈ O2 ⊂ A by Lemma 7.12. Thus rank T > p by Lemma 7.2. 8 Upper bound for typical ranks In Lemma 7.2, we see a class of tensors with rank greater than p. To complete the proof of Theorem 1.2, we give an upper bound of the set of typical ranks of Rn×p×m: Theorem 8.1 Let 3 ≤ m ≤ n and (m − 1)(n − 1) + 1 ≤ p ≤ (m − 1)n. Any typical rank of Rn×p×m is less than or equal to p + 1. We prepare the proof. Let 3 ≤ m ≤ n, (m − 1)(n − 1) + 1 ≤ p < (m − 1)n and u = mn − p. Let σ′ : V n×(p+1)×m → R(u−1)×(p+1) be the counterpart of σ : V n×p×m → Ru×p. Also, let A′ ⊂ R(u−1)×(p+1) and T ′ 1 ⊂ V n×(p+1)×m be the counterparts of A ⊂ Ru×p and T1 ⊂ V n×p×m re- spectively. Let π : Rn×(p+1)×m → Rn×p×m be a canonical projection defined as π(Y1; . . . ; Ym) = ((Y1)≤p; . . . ; (Ym)≤p). Clearly π is a continuous, surjective and open map. 35 Lemma 8.2 π(T ′ 1 ) is an open dense subset of Rn×p×m. Since T ′ Proof We show that π(T ′ defined as 1 is an open set and π is an open map, π(T ′ 1 ) is an open subset of Rn×p×m. 1 ) is dense. Let X ∈ V n×p×m. Consider the map f : V n×p×m → V n×(p+1)×m f (X1; . . . ; Xm−2; Xm−1; Xm) = ((X1, 0); . . . ; (Xm−2, 0); (Xm−1, e); (Xm, 0)) where e is the (2n − u + 1)th column vector of the identity matrix En. Since the (p + 1)th column vector of the matrix σ′(f (X)) is zero, f (X) /∈ σ′−1(A′) holds and by Theorem 7.14 (3), f (X) ∈ T ′ 1 ) holds. Therefore V n×p×m ⊂ π(T ′ 1 . Since π ◦ f is the identity map and π is continuous, X ∈ π(T ′ 1 ) and thus Rn×p×m = π(T ′ 1 ). 1 ) ⊂ π(T ′ By Theorem 7.14 (5), and Lemma 8.2, we have immediately the following corollary. Corollary 8.3 Let 3 ≤ m ≤ n and (m − 1)(n − 1) + 1 ≤ p < (m − 1)n. T2 6= ∅ if and only if T2 ∩ π(T ′ 1 ) 6= ∅, and rank T = p + 1 for any T ∈ T2 ∩ π(T ′ 1 ). Note that arbitrary tensor of π(T ′ 1 ) has rank less than or equal to p + 1 by Theorem 7.14 (4). Proof of Theorem 8.1 The assertion for p = (m − 1)n holds by [33]. Suppose that (m − 1)(n − 1) + 1 ≤ p < (m − 1)n. Then rank(T ) ≤ p + 1 for T ∈ π(T ′ 1 ) is dense, arbitrary integer greater than p + 1 is not a typical rank. 1 ). Since π(T ′ Recall that trank(m, n, p) = trank(n, p, m). We are ready to prove main theorems. Proof of Theorem 1.1 (1) follows from Theorem 7.3 and Corollary 3.4. (2): We may assume that 3 ≤ m ≤ n without the loss of generality. Ten Berge [35] showed that Rm×n×p has a unique typical rank for p ≥ (m − 1)n + 1. Therefore, we see that p ≤ (m − 1)n. Set u = mn − p. By Theorem 7.14 (2) and (4), we see that T2 6= ∅. Furthermore, T2 6= ∅ ⇒ O4 6= ∅ by definitions and the surjectivity of σ. Since O4 ⊂ A u×n×m, we see that there exists an absolutely full column rank u × n × m tensor. The result follows from Corollary 3.4. Proof of Theorem 1.2 We may assume that 3 ≤ m ≤ n. Note that trank(m, n, p) = {min{p, mn}} for k ≥ m [35]. Suppose that 2 ≤ k ≤ m−1. By Theorem 8.1, the maximal typical rank of Rm×n×p is less than or equal to p + 1. Since p is the minimal typical rank of Rm×n×p, trank(m, n, p) is {p} or {p, p+ 1}. By Theorem 1.1, Rm×n×p has a unique typical rank if and only if m#n≥ mn − p + 1, equivalently, k≥ m + n − (m#n). This completes the proof. We immediately have Theorem 1.3 by Proposition 2.4 and Theorem 7.3. In the case where p = (m − 1)(n − 1) + 1, we have many examples for having plural typical ranks. Corollary 8.4 Let m, n ≥ 3 and a ≥ 1. If m ≡ 2a−1 + s (mod 2a) and n ≡ 2a−1 + t (mod 2a) for some integers s and t with 1 ≤ s, t ≤ 2a−1 then Rm×n×((m−1)(n−1)+1) has plural typical ranks. Proposition 8.5 Let a = 4, 8. Rm×n×((m−1)(n−1)+k) has plural typical ranks. If m and n are divisible by a, then for each 1 ≤ k < a, 36 Proof and thus m + n − 1 − (m#n) ≥ a − 1. Then the assertion follows by Theorem 1.2. For a = 4, 8, if m and n are divisible by a, then m#n ≤ m + n − a by Proposition 2.3 Corollary 8.6 (1) R4×4×k has plural typical rank whenever 10 ≤ k ≤ 12. (2) R8×8×k has plural typical rank whenever 50 ≤ k ≤ 56. Proposition 8.7 Let m, n ≥ 3. If Rm×n×((m−1)(n−1)+1) has a unique typical rank, i.e., m#n = m + n − 1, then trank(m, n, (m − 1)(n − 1) − k) = {(m − 1)(n − 1) + 1} holds whenever 0 ≤ k < (m−1)(n−1) m+n−1 . m+n−1 Let 0 ≤ k < (m−1)(n−1) , q = (m − 1)(n − 1) − k and p = (m − 1)(n − 1) + 1. Suppose Proof that Rm×n×p has a unique typical rank. Then trank(n, p, m) = {p}. Since the set of all n × p × m tensors with rank p is a dense subset of Rn×p×m, the image of this set by a canonical projection Rn×p×m → Rn×q×m is also a dense subset of Rn×q×m. Thus any typical rank of Rn×q×m is less than or equal to p. On the other hand, by elementary calculation, we see that (m − 1)(n − 1) < m+n+q−2 ⇔ k < (m−1)(n−1) . Thus the minimal typical rank of Rn×q×m is greater than or equal to p. Therefore Rn×q×m has a unique typical rank p. m+n−1 mnq Corollary 8.8 Let 3 ≤ m ≤ n. Suppose that Rm×n×((m−1)(n−1)+1) has a unique typical rank, i.e., m#n = m + n − 1. If 0 ≤ k ≤ ⌊ m 2 ⌋ − 1 then trank(m, n, (m − 1)(n − 1) − k) = {(m − 1)(n − 1) + 1}. Proof and thus (m + n − 1)k < (m − 1)(n − 1). Therefore the assertion follows from Proposition 8.7. 2 ⌋ − 1. Then (m + n − 1)(k + 1) ≤ (m + n − 1) m 2 ≤ (n + n − 1) m Let 0 ≤ k ≤ ⌊ m 2 < mn References [1] J. F. Adams. Vector fields on spheres. Ann. of Math. (2), 75:603 -- 632, 1962. [2] J. Adem. Some immersions associated with bilinear maps. Bol. Soc. Mat. Mexicana (2), 13:95 -- 104, 1968. [3] M. F. Atiyah and I. G. Macdonald. Introduction to commutative algebra. Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills, Ont., 1969. [4] W. Bruns and U. Vetter. Determinantal rings, volume 1327 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 1988. [5] M. V. Catalisano, A. V. Geramita, and A. Gimigliano. Ranks of tensors, secant varieties of Segre varieties and fat points. Linear Algebra Appl., 355:263 -- 285, 2002; erratum: 367:347 -- 348, 2003. [6] M. V. Catalisano, A. V. Geramita, and A. Gimigliano. On the ideals of secant varieties to certain rational varieties. J. Algebra, 319(5):1913 -- 1931, 2008. [7] R. L. Cohen. The immersion conjecture for differentiable manifolds. Ann. of Math. (2), 122(2):237 -- 328, 1985. 37 [8] D. M. Davis. A strong non-immersion theorem for real projective spaces. Ann. of Math., 120(3):517 -- 528, 1984. [9] D. M. Davis. Some new nonimmersion results for real projective spaces. Bol. Soc. Mat. Mexicana (3), 17(2):159 -- 166, 2011. [10] D. M. Davis and M. Mahowald. Nonimmersions of RPn implied by tmf, revisited. Homology, Homotopy Appl., 10(3):151 -- 179, 2008. [11] D. Eisenbud, C. Huneke, and W. Vasconcelos. Direct methods for primary decomposition. Invent. Math., 110(2):207 -- 235, 1992. [12] S. Friedland. On the generic and typical ranks of 3-tensors. Linear Algebra Appl., 436(3):478 -- 497, 2012. [13] M. Ginsburg. Some immersions of projective space in Euclidean space. Topology, 2:69 -- 71, 1963. [14] J. Heintz and M. Sieveking. Absolute primality of polynomials is decidable in random polyno- mial time in the number of variables. In Automata, languages and programming (Akko, 1981), volume 115 of Lecture Notes in Comput. Sci., pages 16 -- 28. Springer, Berlin-New York, 1981. [15] F. L. Hitchcock. The expression of a tensor or a polyadic as a sum of products. J. Math. Phys., 6(1):164 -- 189, 1927. [16] M. Hochster and J. A. Eagon. Cohen-Macaulay rings, invariant theory, and the generic perfection of determinantal loci. Amer. J. Math., 93:1020 -- 1058, 1971. [17] A. Hurwitz. Uber die Komposition der quadratischen Formen. Math. Ann., 88(1-2):1 -- 25, 1922. [18] J. Ja'Ja'. Optimal evaluation of pairs of bilinear forms. SIAM J. Comput., 8(3):443 -- 462, 1979. [19] G. Kemper and N. Viet Trung. Krull dimension and monomial orders. J. Algebra, 399:782 -- 800, 2014. [20] K. Y. Lam. Construction of some nonsingular bilinear maps. Bol. Soc. Mat. Mexicana (2), 13:88 -- 94, 1968. [21] J. Levine. Imbedding and immersion of real projective spaces. Proc. Amer. Math. Soc., 14:801 -- 803, 1963. [22] H. Matsumura. Commutative ring theory, volume 8 of Cambridge Studies in Advanced Math- ematics. Cambridge University Press, Cambridge, second edition, 1989. Translated from the Japanese by M. Reid. [23] R. J. Milgram. Immersing projective spaces. Annals of Mathematics, 85(3):473 -- 482, 1967. [24] M. Miyazaki, T. Sumi, and T. Sakata. Typical ranks of certain 3-tensors and absolutely full column rank tensors. preprint, arXiv:1103.0154v2, Dec. 2012. 38 [25] M. Nagata. Theory of commutative fields, volume 125 of Translations of Mathematical Mono- graphs. American Mathematical Society, Providence, RI, 1993. Translated from the 1985 Japanese edition by the author. [26] J. Radon. Lineare scharen orthogonaler matrizen. Abh. Math. Sem. Univ. Hamburg, 1(1):1 -- 14, 1922. [27] D. Rees. The grade of an ideal or module. Proc. Cambridge Philos. Soc., 53:28 -- 42, 1957. [28] L. Robbiano. Term orderings on the polynomial ring. In EUROCAL '85, Vol. 2 (Linz, 1985), volume 204 of Lecture Notes in Comput. Sci., pages 513 -- 517. Springer, Berlin, 1985. [29] A. Seidenberg. Constructions in algebra. Trans. Amer. Math. Soc., 197:273 -- 313, 1974. [30] D. B. Shapiro. Compositions of quadratic forms, volume 33 of de Gruyter Expositions in Mathematics. Walter de Gruyter & Co., Berlin, 2000. [31] N. Singh. On nonimmersion of real projective spaces. Topology and its Applications, 136:233 -- 238, 2004. [32] T. Sumi, M. Miyazaki, and T. Sakata. Rank of 3-tensors with 2 slices and Kronecker canonical forms. Linear Algebra Appl., 431(10):1858 -- 1868, 2009. [33] T. Sumi, M. Miyazaki, and T. Sakata. Typical ranks of m × n × (m − 1)n tensors with 3 ≤ m ≤ n over the real number field. Linear Multilinear Algebra, 63(5):940 -- 955, 2015. [34] T. Sumi, T. Sakata, and M. Miyazaki. Typical ranks for m × n × (m − 1)n tensors with m ≤ n. Linear Algebra Appl., 438(2):953 -- 958, 2013. [35] J. M. F. ten Berge. The typical rank of tall three-way arrays. Psychometrika, 65(4):525 -- 532, December 2000. [36] J. M. F. ten Berge and H. A. L. Kiers. Simplicity of core arrays in three-way principal component analysis and the typical rank of p × q × 2 arrays. Linear Algebra Appl., 294(1- 3):169 -- 179, 1999. 39
1710.07338	1	1710	2017-10-19T19:59:14	Minimal Reversible Nonsymmetric Rings	[ "math.RA" ]	Marks showed that $\mathbb{F}_2Q_8$, the $\mathbb{F}_2$ group algebra over the quaternion group, is a reversible nonsymmetric ring, then questioned whether or not this ring is minimal with respect to cardinality. In this work, it is shown that the cardinality of a minimal reversible nonsymmetric ring is indeed 256. Furthermore, it is shown that although $\mathbb{F}_2Q_8$ is a duo ring, there are also examples of minimal reversible nonsymmetric rings which are nonduo.	math.RA	math	Minimal Reversible Nonsymmetric Rings Steve Szabo Department of Mathematics and Statistic Eastern Kentucky University Richmond, KY 40475 [email protected] Abstract Marks showed that F2Q8, the F2 group algebra over the quaternion group, is a reversible nonsymmetric ring, then questioned whether or not this ring is minimal with respect to cardinality. In this work, it is shown that the cardinality of a minimal reversible nonsymmetric ring is indeed 256. Furthermore, it is shown that although F2Q8 is a duo ring, there are also examples of minimal reversible nonsymmetric rings which are nonduo. keywords: finite rings, reversible rings, symmetric rings, minimal rings, 2-primal rings 1 Introduction and Overview In [1], Cohn introduced reversible rings. A ring R is reversible if for any a, b ∈ R, ab = 0 implies ba = 0. Since Cohn's introduction of reversible rings there has been many works connecting them to other ring properties (see [5, 7, 8, 10]). In [9], Lambek studied what are now called symmetric rings. A ring R is symmetric if for any a, b, c ∈ R, abc = 0 implies bac = 0. Although Lambek's introduction was some time ago, more recently, the connections between symmetric rings and rings with related properties have been of some interest. In particular, Marks, in [10], studied the relationship between reversible and symmetric rings with and without identity. He showed that when considering rings without identity, these two properties are independent. However, it is easy to see that a symmetric ring with identity is reversible. He went on to show that F2Q8, the F2 group algebra over the quaternion group, is a reversible nonsymmetric ring. He then asked whether or not a minimal reversible nonsymmetric ring with respect to cardinality is of order 256. The sole purpose of this paper is to answer this question. A ring is called right (resp. left) ideal is two-sided. Marks showed that this property even for rings with identity is independent of being reversible nonsymmetric by proving that F2Q8 is right left) duo if every right (resp. 1 duo as well as showing the ring k hx, y, zi hx2, y2, z2, xyz, yzx, zxyi for a field k is reversible nonsymmetric and neither left nor right duo. This begs the question, "Does minimality of reversible nonsymmetric rings depend on the duo property?" With k = F2, this ring is of order 8192. We have found a ring of order 256, F2 hu, vi hu3, v3, u2 + v2 + vu, vu2 + uvu + vuvi which is reversible nonsymmetric and neither left nor right duo (see Example 4.1). In analyzing indecomposable reversible nonsymmetric rings, which we know must be local (see Proposition 2.3), for minimality the nilpotency of the jacobson radical must be at least 4 (Proposition 2.4) and at most 6 (Proposition 3.2). Examples having nilpotency 4 or 5 are given in Section 4. As of yet we have not identified an example with nilpotency 6. There has been some work on specific ring types and minimality. In [14], Xue showed that a minimal noncommutative duo ring is of order 16 and that there are only three such rings. In [7], Kim and Lee showed that a minimal reversible noncommutative ring is of order 16. In another paper [13], Xu and Xue showed that a minimal zero-insertive ring (also known as semicommutative or having S I condition) also has order 16 and that there are only 5 such rings. From [4] we know there are only 13 noncommutative rings of order 16. From their list and the fact that there is only one noncommutative ring or order 8, with some effort, the minimal rings just mentioned could be found. Such work becomes difficult when rings of larger orders are involve. With the classification of rings of order p5 in [2] and [3], we see that such work becomes cumbersome as the classification of finite rings is difficult and that even for small orders, there is an abundance of rings. Hence, the techniques used in the papers mentioned are important to find minimal rings with various properties. The connections between various ring properties related to commutativity such as reduced, symmetric, reversible, semicommutative and 2-primal are stud- ied in [11] where the strict inclusion of these ring classes is presented. Adding the fact that a ring is reversible if and only if it is semicommutative and reflexive to their ring class inclusions we have the following diagram. reduced symmetric 3 reversible ❥❥❥❥❥❥❥❥ ❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚ ❚❚❚❚❚❚❚❚❚ reflexive + commutative semicommutative 3 2-primal Having minimal examples showing these strict inclusions can be instructive when studying them and our result here provides such an example. In a companion paper, [12], we list minimal rings of these classes. Interestingly, many of the 2 0 8 + k s K S % - + examples come from the set of noncommutative rings of order 16. It is shown there for instance that F4[x;σ] hx2i where σ is the Frobenius automorphism on F4, is F2hu,vi the minimal noncommutative nonreduced symmetric ring and that hu2,v2,uvi is a minimal semicommutative nonreversible ring. Both are rings of order 16. As we will see here, minimal reversible nonsymmetric rings are much larger than min- imal rings of related types. Notice that reversible nonsymmetric rings in some sense, live between noncommutative nonreduced symmetric rings and semicom- mutative nonreversible rings. It is curious then that minimal nonsymmetric rings are significantly larger than these other two types. In Section 2 preliminaries are presented. Section 3 contains the main result of the paper, namely that a minimal reversible nonsymmetric ring has order 256. Finally, in Section 4 examples are provided. 2 Preliminaries We assume all rings have unity and that rings are finite with the exception of a polynomial ring. Given a ring R, J(R) is the Jacobson radical of R and S(R) is the socle of R. A ring R is symmetric if for all a, b, c ∈ R, abc = 0 implies bac = 0. A ring R is reversible if for all a, b ∈ R, ab = 0 implies ba = 0. The following is a well-known result. Lemma 2.1 (see [6]). Let R be a finite local ring and F = R/J(R). Then F is a field and J(R)k/J(R)k+1 is a finite dimensional F -vector space. The following results on reversible rings are needed throughout. Lemma 2.2. Let R be a ring which contains an idempotent e such that eR(1 − e) 6= 0. Then R is not reversible. Proof. Let a ∈ eR(1 − e) 6= 0. Then ea = a 6= 0 and ae = 0 showing R is not reversible. Proposition 2.3. Let R be a reversible ring. Then R is local if and only if it is indecomposable. Proof. Local rings are indecomposable rings. central idempotents other than 1. The result follows from Lemma 2.2. Indecomposable rings have no Proposition 2.4. Let R be a local reversible ring. symmetric. If J(R)3 = 0 then R is Proof. Assume J(R)3 = 0. Let a, b, c ∈ R \ 0 and assume abc = 0. Since R is local, at least two of a, b and c are in J(R). Assume b is a unit. Since R is reversible, we have that cab = 0, ca = 0, ac = 0 and bac = 0. This logic also shows if either a or c is a unit, bac = 0. Now assume a, b, c ∈ J(R). Since J(R)3 = 0, bac = 0. Hence, R is symmetric. 3 Gutan in [5] characterized reversible group rings, provided examples of re- versible nonsymmetric group rings and showed that F2Q8 is a minimal reversible nonsymmetric group ring. This result is easy to see in light of Lemma 2.2. The only noncommutative group ring smaller than F2Q8 is F2S3. Since F2S3 has a subring isomorphic to M2(F2), by Lemma 2.2, it is not reversible. This verifies Gutan's result. Gutan left open the question of F2Q8 being a minimal reversible nonsymmetric ring or not. 3 Minimal Reversible Nonsymmetric Rings Theorem 3.1. A minimal reversible nonsymmetric ring has order 256. The proof of Theorem 3.1 will be given at the end of this section. As Marks showed that F2Q8 is a reversible nonsymmetric ring of order 256, we show that no ring of order less than 256 is reversible but nonsymmetric. Since the direct sum of a set of reversible nonsymmetric rings is reversible nonsymmetric and by proposition 2.3, a reversible ring is indecomposable if and only if it is local, we need only consider local reversible rings. Throughout this section, let R be a finite local reversible ring of order less than 256, J = J(R), S = S(R) and F = R/J. Since R is finite and local, F is a field. From Lemma 2.1 we know J i/J i+1 is an F -vector space. Let di = dimF (J i/J i+1) and D = (d1, d2, . . . , dK) where K is such that dK 6= 0 If J 3 = 0, by Proposition 2.4 R is symmetric. So, assume and dK+1 = 0. J 3 6= 0. Then R ≥ F 4. If F ≥ 4, R ≥ 256 but we have assumed R < 256. Therefore, F ∈ {F2, F3}. Now, R = pm for a prime p and some positive integer m. Since R < 256, R ∈ {24, 25, 26, 27, 34, 35}. Proposition 3.2. Assume d2 = 1. If d1 ≤ 2 or F ∼= F2 then R is commutative. Proof. From the assumption, J = F a1 + F a2 + · · · + F an + J 2 for a1, . . . , an ∈ J \ J 2, d2 = · · · = dK = 1, J K−1 = F t + S(R) for some t ∈ J K−1 and S(R) = JJ K−1 = F a1t+ F a2t+ · · ·+ F ant. Since dK = 1, we may assume S(R) = F a1t after renumbering. For some λi ∈ F , ait = λia1t and so (ai − λia1)t = 0. Let n + J 2 and since R is a′ i = (ai − λia1) for i > 1. Then J = F a1 + F a′ reversible, a′ i for i > 1. For notational purposes replace ai with a′ i. 2 + · · · + F a′ it = 0 = ta′ There must be a product of K − 1 elements from {a1, a2, . . . , an} in J K−1 \ S(R). Let x1x2 . . . xK−1 be such a product. Then a1x1 . . . xK−1 6= 0 since a1 does not annihilate J K−1. By reversibility, any cyclic shift of the product is also nonzero. Since ai for i > 1 annihilates J K−1, xi = a1 for 1 ≤ i ≤ K − 1. Hence, aK 1 6= 0 and J j/J j+1 = F aj j=2 αj aj Let 2 ≤ i ≤ n, aia1 = PK 1 + J j+1 for 2 ≤ j ≤ K. 1 and a1ai = PK j=2 βjaJ 1 for some αj, βj ∈ F . Since F is a prime field, K X j=2 βjaj+1 1 = a1aia1 = K X j=2 αjaj+1 1 4 So, αj = βj showing aia1 = a1ai. Then a1 ∈ Z(R) and if n = 2 then R is commutative. So, now we further assume F ∼= F2. Now let 2 ≤ i, l ≤ n. If ala1 = 0 then aiala1 = 0. This implies that aial ∈ S. Since F ∼= F2, by reversibility, aial = alai. So, assume ala1 6= 0. Then aial = PK 1 for some γj, δj, σ ∈ F . Since F ∼= F2 and a1 ∈ Z(R), 1 and ala1 = a1al = PK k=2 σkak j=2 γjaj 1, alai = PK j=2 δjaj K X j,k=2 δjσkaj+k−1 1 = alaial = K X j,k=2 γjσkaj+k−1 1 From this, since ala1 6= 0, it can be shown that γj = δj. In any case, aial = alai and R is commutative. If m ∈ {4, 5} the only possibilities for D are (1, 1, 1), (1, 1, 1, 1) or (2, 1, 1). Proposition 3.2 shows in these cases R is symmetric. At this point, if R is nonsymmetric then F ∼= F2 and R ∈ {26, 27}. In light of Proposition 3.2 the only possibilities for D would be (2, 2, 1), (2, 2, 1, 1), (2, 2, 2), (2, 3, 1) or (3, 2, 1). The next proposition rules out D ∈ {(2, 2, 1), (2, 2, 2), (2, 3, 1)}. Proposition 3.3. If R is nonsymmetric, F ∼= F2 and D = (2, l, n) then l = 3 and n ≥ 2. Proof. Assume R is non-symmetric, F ∼= F2 and D = (2, l, n) for some l and n. Then there exists a, b, c ∈ J \J 2 such that abc = 0 and bac 6= 0. If a+J 2 = b+J 2, since J 4 = 0, bac = a2c = abc = 0. So, a + J 2 6= b + J 2. It can be similarly shown that a + J 2 6= c + J 2 and b + J 2 6= c + J 2. So, a + J 2 and b + J 2 are linearly independent in J/J 2 showing J = F a + F b + J 2 and c = a + b + t for some t ∈ J 2. Since R is reversible, 0 = abc = aba + ab2, 0 = cab = a2b + bab and 0 = bca = ba2 + b2a. So, aba = ab2, a2b = bab and ba2 = b2a. Again by reversibility, if any degree two monomial involving a and b is in J 3, all degree 3 monomials involving a and b other than a3 and b3 are 0. This would mean bac = 0. So, a2, ab, ba, b2 ∈ J 2 \ J 3. Assume αa2 + βab + γba + δb2 ∈ J 3 for some α, β, γ, δ ∈ F . Considering left and right multiples of a and b we have that 0 = αa3 + βa2b + γaba + δab2 0 = αa3 + βaba + γba2 + δb2a 0 = αba2 + βbab + γb2a + δb3 0 = αa2b + βab2 + γbab + δb3 and with aba = ab2, a2b = bab and ba2 = b2a this becomes 0 = αa3 + βbab + γaba + δaba 0 = αa3 + βaba + γba2 + δba2 0 = αba2 + βbab + γba2 + δb3 0 = αbab + βaba + γbab + δb3. 5 Adding the first and second equations, 0 = (β + γ + δ)aba + βbab + (γ + δ)ba2. If β = 1 then either aba = bab or ba2 = bab. In either case, bac = 0. So, β = 0. Then 0 = (γ + δ)aba + (γ + δ)ba2 and adding the third and fourth equations above, 0 = (α + γ)bab + (α + γ)ba2. If α 6= γ, bab = ba2. Similarly, if δ 6= γ, aba = ba2. In either case, bac = 0. So, α = γ = δ. Then α = γ = δ = 1 gives a non-zero solution and a2 + J 3, ba + J 3 and b2 + J 3 are linearly dependent in J 2/J 3. From these calculations we see then that ab + J 3, ba + J 3 and b2 + J 3 are linearly independent in J 2/J 3. We know that J 2 = F a2 + F ab + F ba + F b2 + J 3 so l = 3. It was shown earlier that a2b = bab and ba2 = b2a. So, by reversibility, either a2b = bab = b2a = ba2 = 0 or a2b = bab 6= 0 and ba2 = b2a 6= 0. Since ba2 + bab = bac 6= 0, ba2 and bab are both nonzero and hence linearly independent in J 3. So, n ≥ 2. In Section 4, a few examples of reversible nonsymmetric rings are discussed. The proof of Proposition 3.3 led to Example 4.1 which turns out to be a minimal reversible nonsymmetric non-duo ring. The next proposition takes care of the case when F ∼= F2 and D = (3, 2, 1). Proposition 3.4. If F ∼= F2 and D = (n, 2, 1) for some n ≥ 2 then R is symmetric. Proof. Assume F ∼= F2, D = (n, 2, 1) and R is non-symmetric. Since J 3 6= 0 and J 4 = 0, there exists a, b, c ∈ J \ J 2 such that abc = 0 and bac 6= 0. Since R is reversible we can deduce the following: S = {0, bac}, acb = cba = bac 6= 0, abc = cab = abc = 0, ac, cb ∈ J 2 \ J 3, ca, bc 6= 0, a2b = aba = ba2, b2a = bab = ab2 and b2c = bcb = cb2. We know that J 2 = F ab + F ba + J 3 because otherwise abc = bac. Since 1 − a is a unit, (1 − a)cb 6= 0 which shows c(1 − a)b 6= 0 and bc 6= bac. Then bc ∈ J 2 \ J 3. Now, cb = αab + βba + s for some α, β ∈ F not both 0 and s ∈ J 3. Then 0 6= acb = αa2b + βaba = (α + β)a2b so a2b 6= 0 and α 6= β. In any case, b2c = cb2 = αab2 + βbab = (α + β)b2a = b2a. Next, ac = ǫab + ζba + u for some ǫ, ζ ∈ F not both 0 and u ∈ J 3. Then 0 6= acb = ǫab2 + ζbab = (ǫ + ζ)b2a so b2a 6= 0. Also, bc = γab + δba + t for some γ, δ ∈ F not both 0 and t ∈ J 3. Then 0 = abc = γa2b + δaba = (γ + δ)a2b. Since a2b 6= 0, γ = δ = 1 showing bc = ab + ba + t. Then 0 6= b2a = b2c = bab + b2a = 0. Hence, if F ∼= F2 and D = (n, 2, 1) then R is symmetric. 6 We have one final case to rule out, that being when R = 27 with D = (2, 2, 1, 1). Proposition 3.5. If F ∼= F2 and D = (2, 2, 1, 1) then R is symmetric. Proof. Assume F ∼= F2 and D = (2, 2, 1, 1). Then J = F a1 + F a2 + J 2 for a1, a2 ∈ J \ J 2, J 3 = F t + J 4 for some t ∈ J 3 and J 4 = JJ 3 = F a1t + F a2t. Since d4 = 1, we may assume J 4 = F a1t after renumbering. For some λ ∈ F , a2t = λa1t and so (a2 − λa1)t = 0. Replace a2 with a2 − λa1. Since R is reversible, a2t = 0 = ta2. There must be a product of 3 elements from {a1, a2} in J 3 \ J 4. Let x1x2x3 be such a product. Then a1x1x2x3 6= 0 since a1 does not annihilate J 3. By reversibility, any cyclic shift of the product is also nonzero. Since a2 annihilates J 3, x1 = x2 = x3 = a1. Hence, a4 1 + J 4 and J 4 = F a4 1. Further more there exists u ∈ {a1a2, a2a1, a2 1 +F u+J 3. Note, a1u, a2u, ua1, ua2 ∈ J 4 since a1ua1, a2ua1, ua1a1 and ua2a1 are all 0. 2} such that J 2/J 3 = F a2 1 6= 0, J 3/J 4 = F a3 Let a, b, c ∈ R such that abc = 0. If a, b or c is a unit, bac = 0. So assume a, b, c ∈ J. Then a = α1a1+β1a2+γ1a2 1+δ2u+t2 and c = α3a1 + β3a2 + γ3a2 1 + δ3u + t3 for αi, βi, γi, δi ∈ F and ti ∈ J 3. First, if α1 = α2 = α3 = 1 then abc 6= 0. So, by reversibility we may assume α3 = 0. If β3 = 0, bac = 0. So assume β3 = 1. Since R is reversible and a1a2 2 ∈ J 4, a1a2 1+δ1u+t1, b = α2a1+β2a2+γ2a2 2 = a2a1a2. Then bac = bac − abc = (α2β1 − α1β2)a1a2 2 + (α1β2 − α2β1)a2a1a2 = (α2β1 − α1β2 + α1β2 − α2β1)a1a2 2 = 0 Hence, R is symmetric. We can now prove Theorem 3.1. Proof to Theorem 3.1. In [10] it was shown that F2Q8 is reversible nonsym- metric ring of order 256. Let T be a local reversible ring of order less than If J(T )3 = 0, by Lemma 2.4 T is symmetric. So, assume J(T )3 6= 0. 256. By Lemma 2.1, T /J(T ) is a field and T = T /J(T )l for some l. This im- plies T /J(T ) ∈ {2, 3} and 4 ≤ l ≤ 7. This is precisely what was assumed about R in this section. We reiterate: R is a finite local reversible ring where R = pm ∈ {24, 25, 26, 27, 34, 35}, J(R)3 6= 0, J = J(R), F = R/J, F is a field, F ∈ {2, 3}, di = dimF (J i/J i+1), D = (d1, d2, . . . , dK) where K is such that dK 6= 0 and dK+1 = 0. If m = 4, D = (1, 1, 1) and if m = 5, D ∈ {(1, 1, 1, 1), (2, 1, 1)} then, by Proposition 3.2, R is symmetric. If m ≥ 6, then F = 2. If m = 6, D ∈ {(1, 1, 1, 1, 1), (2, 1, 1, 1), (2, 2, 1), (3, 1, 1)} and if m = 7, D ∈ {(1, 1, 1, 1, 1, 1), (2, 1, 1, 1, 1), (2, 2, 1, 1), (2, 2, 2), (2, 3, 1), (3, 2, 1), (4, 1, 1)}. 7 If D ∈ {(1, 1, 1, 1, 1), (2, 1, 1, 1), (3, 1, 1), (1, 1, 1, 1, 1, 1), (2, 1, 1, 1, 1), (4, 1, 1)}, by Proposition 3.2 R is symmetric. If D ∈ {(2, 2, 1), (2, 2, 2), (2, 3, 1)}, by Proposition 3.3 R is symmetric. If D = (3, 2, 1), by Proposition 3.4 R is symmetric. If D = (2, 2, 1, 1), by Proposition 3.5 R is symmetric. So, a local reversible ring of order less than 256 is symmetric. Again, for reversible rings local and indecomposable are equivalent conditions. In a ring direct sum, if a direct summand is nonsymmetric, the ring itself is nonsymmetric. Hence, a minimal reversible nonsymmetric ring is of order 256. 4 Examples of Minimal Reversible Nonsymmet- ric Rings In this section we provide two examples of minimal reversible nonsymmetric rings, one which is duo and one that is not. The first example is a byproduct of the proof to Proposition 3.3. Example 4.1. Let R = F2 hu, vi hu3, v3, u2 + v2 + vu, vu2 + uvu + vuvi . We will show that this is a minimal reversible nonsymmetric ring that is neither left nor right duo. First, R is an 8-dimensional algebra over R/J(R) ∼= F2 with basis {1, v1 = u, v2 = v, v3 = u2, v4 = uv, v5 = v2, v6 = uvu, v7 = vuv} showing R = 256. Second, since uv(u + v) = u3 + uv(u + v) = u3 + uvu + uv2 = u(u2 + v2 + vu) = 0 and vu(u + v) = vu2 + vuv = uvu 6= 0, R is nonsymmetric. Third, since uv /∈ vR and uv /∈ Ru, R is neither right nor left duo. Finally, we show R is reversible. Let a, b ∈ J(R) with a = P7 i=1 αivi and b = P7 i=1 βivi where αi, βi ∈ R/J(R). Then ab = (α1β1 + α2β1)u2 + (α1β2)uv + (α2β2 + α2β1)v2 + (α1β4 + α2β3 + α2β4 + α5β1 + α3β2)vuv + (α1β5 + α2β3 + α4β1 + α5β1 + α4β2)uvu. 8 Assume ab = 0. From the coefficient of uv, α1 = 0 or β2 = 0. Then, considering the coefficients of u2 and v2 there are three restrictions: (1) α2 = 0 or β1 = 0, (2) α1 = 0 or β1 = 0 and (3) α2 = 0 or β2 = 0. This produces 7 possibilities which we list along with their implications. α1 β2 α2 β1 0 0 1 0 0 0 0 0 1 0 0 1 1 0 0 0 0 1 1 0 0 0 0 1 0 0 0 1 ab = 0 ab = α5vuv + (α4 + α5)uvu α4 = α5 = 0 ab = (β3 + β4)vuv + β3uvu β3 = β4 = 0 ab = α3vuv + α4uvu α3 = α4 = 0 ab = (α3 + α5)vuv + α5uvu α3 = α5 = 0 β4 = β5 = 0 ab = β4vuv + β5uvu ab = β3vuv + (β3 + β5)uvu β3 = β5 = 0 Then for any of these situations, ba = (β1α1 + β2α1)u2 + (β1α2)uv + (β2α2 + β2α1)v2 + (β1α4 + β2α3 + β2α4 + β5α1 + β3α2)vuv + (β1α5 + β2α3 + β4α1 + β5α1 + β4α2)uvu = 0. Hence, R is reversible. As with the previous example, the next example, which was originally given in [10], may have been discovered from the proof of Proposition 3.5. Example 4.2. Let Q8 = {±1, ±i, ±j, ±k} be the quaternions, {xg : g ∈ Q8} and R = F2Q8 where we write elements of R as F2-linear combinations of {xg : g ∈ Q8}. Let u = 1 + xi and v = 1 + xj . Since uv(uv) 6= 0 and vu(uv) = 0, R is nonsymmetric. We first verify these key relations, u4 = 0, v4 = 0, u2 + v2 = 0, and u2 + uv + vu + uvu = 0. Then one can see that J(R) = F2u+F2v+J(R)2, J(R)2 = F2uv+F2vu+J(R)3, J(R)3 = F2uv2+F2vu2+J(R)4 and J(R)4 = F2uv3. Then we deduce that R ∼= F2 hu, vi hu4, v4, u2 + v2, u2 + uv + vu + uvui . These facts fit with the proof of Proposition 3.5 if the restriction of J(R)3 being principal is removed. Had we found this ring we could have verified reversibility similar to what was done for Example 4.1 although the proof in [10] is more elegant. Though their proof is predicated on R being right duo for which their proof is not straight forward. From our view of R we deduce the duo property very simply. We need to show that sr ∈ rR for any r, s ∈ R. From the generators of J(R)i, we see we need only verify that uv ∈ vR and that vu ∈ uR. First, note that vuv = uvu. Then we have that vu = u2 + uv + uvu = u(u + v + vu) and uv = u2 + vu + uvu = v2 + vu + vuv = v(v + u + uv). 9 To close we point out that we have two examples of minimal reversible non- symmetric rings, one which is duo and the other which is not. Furthermore, one has a jacobson radical of nilpotency of 4 and the other of 5. From Propositions 2.4 and 3.2 we see that the nilpotency of jacobson radical must be between 4 and 6. Can we find a minimal reversible nonsymmetric ring that has a ja- cobson radical of nilpotency 6? From the results in Section 3 we know that D = (2, 2, 1, 1, 1) (as defined in Section 3) if such an example exists. Also, in viewing the proof to Proposition 3.4, is there a minimal reversible nonsymmet- ric ring with D = (3, 2, 2). Finally, can we identify all the minimal reversible nonsymmetric rings? References [1] P. M. Cohn. Reversible rings. Bull. London Math. Soc., 31(6):641 -- 648, 1999. [2] B. Corbas and G. D. Williams. Rings of order p5. I. Nonlocal rings. J. Algebra, 231(2):677 -- 690, 2000. [3] B. Corbas and G. D. Williams. Rings of order p5. II. Local rings. J. Algebra, 231(2):691 -- 704, 2000. [4] J. B. Derr, G. F. Orr, and Paul S. Peck. Noncommutative rings of order p4. J. Pure Appl. Algebra, 97(2):109 -- 116, 1994. [5] Marin Gutan and Andrzej Kisielewicz. Reversible group rings. J. Algebra, 279(1):280 -- 291, 2004. [6] Nathan Jacobson. Structure of rings. American Mathematical Society Colloquium Publications, Vol. 37. Revised edition. American Mathematical Society, Providence, R.I., 1964. [7] Byung-Ok Kim and Yang Lee. Minimal noncommutative reversible and reflexive rings. Bull. Korean Math. Soc., 48(3):611 -- 616, 2011. [8] Nam Kyun Kim and Yang Lee. Extensions of reversible rings. J. Pure Appl. Algebra, 185(1-3):207 -- 223, 2003. [9] J. Lambek. On the representation of modules by sheaves of factor modules. Canad. Math. Bull., 14:359 -- 368, 1971. [10] Greg Marks. Reversible and symmetric rings. J. Pure Appl. Algebra, 174(3):311 -- 318, 2002. [11] Greg Marks. A taxonomy of 2-primal rings. J. Algebra, 266(2):494 -- 520, 2003. [12] Steve Szabo. Some minimal rings related to 2-primal rings. in preparation, 2017. 10 [13] Liqiong Xu and Weimin Xue. Structure of minimal non-commutative zero- insertive rings. Math. J. Okayama Univ., 40:69 -- 76 (2000), 1998. [14] Weimin Xue. Structure of minimal noncommutative duo rings and minimal strongly bounded nonduo rings. Comm. Algebra, 20(9):2777 -- 2788, 1992. 11
1206.2973	1	1206	2012-06-14T00:22:41	Symmetric Matrices over F_2 and the Lights Out Problem	[ "math.RA", "cs.DM" ]	We prove that the range of a symmetric matrix over F_2 contains the vector of its diagonal elements. We apply the theorem to a generalization of the "Lights Out" problem on graphs.	math.RA	math	Symmetric Matrices over F2 and the Lights Out Problem Igor Minevich May 1, 2014 Abstract We prove that the range of a symmetric matrix over F2 contains the vector of its diagonal elements. We apply the theorem to a gener- alization of the “Lights Out” problem on graphs. 1 Introduction The purpose of this note is to prove that the range of a symmetric matrix over F2 contains the vector of its diagonal elements; we do this in section 2. We then apply this theorem to a generalization of the “Lights Out” problem on graphs: we have a simple graph and the vertices all have a state of either ‘on’ (represented by 0 ∈ F2) or ‘off’ (represented by 1 ∈ F2). When a vertex is ‘clicked,’ the states of it and its adjacent vertices are toggled. Starting with arbitrary states for all vertices, the problem is to determine whether or not we can click a subset of vertices and get to the state where they are all off, hence the name “Lights Out”. It is known that if we start with all vertices on, then this is possible (see [Car96] or [Sut88]). The theorem we prove here generalizes this result to the case where clicking some of the vertices does not actually affect the state of those vertices, but only the states of those adjacent to them. 2 Linear Algebra Theorem Theorem. Any symmetric N × N matrix A = (aij) over F2 has its diagonal vector ~d = (a11, a22, . . . , aN N )T in its range. 1 Proof. We show that if A~x = ~0 then ~x · ~d = 0. This implies ~d is perpendicular to every vector in Nul A, i.e. ~d ∈ Nul(A)⊥ = Col(AT ) = Col(A), as desired. If ~x = ~0, then this is obvious. Without loss of generality, we may assume ~x = (1, . . . , 1, 0, . . . , 0), where the first n + 1 entries are 1 and n ≥ 0 (otherwise, we can just permute the entries of ~x and also the corresponding columns and rows of A so that this is true). The linear dependence relation on the columns of A given by A~x = ~0 says that the (n + 1)st column is the sum of the first n columns, so we have aj,n+1 = ajj + n X i=1 i6=j aji, j = 1, . . . , N an+1,n+1 = n X j=1 an+1,j = n X j=1 aj,n+1 = = n X j=1 n n n ajj + X X aji = X ajj + 2 X aij = j=1 i=1 i6=j j=1 n≥i>j≥1 n X j=1 ajj so we see that ~x · ~d = Pn+1 j=1 ajj = 0. 3 “Lights Out” Problem The original motivation for this theorem came from the “Lights Out” problem as explained in the introduction. Corollary. Let G be a finite undirected graph, where there can be at most one edge between any two vertices and a vertex can be connected to itself. Let v1, . . . , vn be the vertices connected to themselves. Assign to each vertex a state of either 1 or 0, and assume that clicking a vertex makes those vertices adjacent to it (including itself if it is connected to itself) switch their state. Then, starting with all vertices at the state of 0, we can choose vertices to click so that, as a result, precisely v1, . . . , vn are 1 and the rest are 0. Proof. Define the N ×N matrix A over F2, where N is the number of vertices of G, by aij = 1 if and only if the vertex vi is connected to the vertex vj. Since the graph is undirected, A is symmetric. Furthermore, the first n diagonal elements of A are 1 and the rest are 0, since the first n vertices are the ones 2 connected to themselves. It is easy to see that if A~x = ~y, then ~y is precisely the vector of states obtained by starting with all vertices being 0 and clicking the vertices corresponding to 1’s in ~x, because multiplication by ~x simply adds up the columns corresponding to 1’s in ~x. Since the diagonal vector ~d is in the range of A, there is a vector ~x such that A~x = ~d. Thus we can click the vertices corresponding to 1’s in ~x and, as a result, the vertices that are connected to themselves will be 1 and the rest will be 0, as desired. This is the generalization of the “Lights Out” problem on graphs to the case where not all vertices affect themselves. To make this more concrete, we first list a few examples where all vertices are connected to themselves and then an example where they are not. The original motivator for the problem was the case where the vertices of the graph are squares in a fi- nite two-dimensional grid and clicking a square toggles its color (say between white and black) and also the colors of the squares that share an edge with the square clicked. The problem is to find out which squares to click so that, if we start with an all-white grid, we end up with all squares black. This is not easy to do by inspection for grids 5 × 5 and bigger, but the corollary shows it can always be done. We could generalize this scenario to be n-dimensional, where an n-dimensional cube affects itself and those with which it shares an (n − 1)-dimensional hy- perplane. Or we could start with an arbitrary subset of a 2-dimensional or 3-dimensional grid, perhaps in a pleasing shape such as a star or a flower, and have vertices affect themselves and the ones next to them as before. We could also change the rules and have squares affect each other diagonally, or in a torus-like fashion, where the top squares affect the bottom ones and/or the leftmost ones affect the rightmost ones (this particular case has been studied in more detail in [GY09]). Or we could change ‘squares’ to triangles in a triangular lattice or hexagons in a hexagonal lattice, etc. Now for an example of the generalization, where some vertices do affect themselves and some don’t, we can have squares with green and red lamps; the green lamps affect themselves and also those around them, but the red lamps only affect those around them and not themselves. Starting with all lamps off, the goal is to turn on all the green lamps and turn off all the red ones, and by the corollary this can always be achieved. Of course, the amount of such scenarios is limited only by the imagination. 3 References [Car96] Yair Caro. Simple proofs to three parity theorems. Ars Combin., 42:175–180, 1996. [GY09] Masato Goshima and Masakazu Yamagishi. Two remarks on torus lights out puzzle. Adv. Appl. Discrete Math., 4(2):115–126, 2009. [Sut88] Klaus Sutner. Additive automata on graphs. Complex Systems, 2(6):649–661, 1988. 4
1502.01746	1	1502	2015-02-05T22:15:36	Non isomorphic pure Galois-Eisenstein rings	[ "math.RA" ]	Let $n; r; e; s$ be are positive integers and the prime p; the finite local principal ideals ring of parameters $p; n; r; e; s)$ $GR(p^n;r)[x]/(x^e - pu ; x^s),$ is defined by an invertible element u of the Galois ring $GR(p^n; r)$ of characteristic $p^n$ of order $p^{nr}.$ It is called Galois-Eisenstein ring of parameters $(p; n; r; e; s)$. A basic problem, which seems to be very difficult is to determine all non-isomorphism pure Galois-Eisenstein rings of parameters $(p; n; r; e; s).$ In this paper, this isomorphism problem for pure Galois-Eisenstein rings of parameters $(p; n; r; e; s)$ is investigated.	math.RA	math	NON-ISOMORPHIC PURE GALOIS-EISENSTEIN RINGS ALEXANDRE FOTUE TABUE AND CHRISTOPHE MOUAHA Abstract. Let n, r, e, s be are positive integers and the prime p, the finite local principal ideals ring of parameters (p, n, r, e, s) GR(pn, r)[x]/(xe − pu , xs), is defined by an invertible element u of the Galois ring GR(pn, r) of char- acteristic pn of order pnr. It is called Galois-Eisenstein ring of parameters (p, n, r, e, s). A basic problem, which seems to be very difficult is to de- termine all non-isomorphism pure Galois-Eisenstein rings of parameters (p, n, r, e, s). In this paper, this isomorphism problem for pure Galois- Eisenstein rings of parameters (p, n, r, e, s) is investigated. Keywords: Galois field, Galois ring. AMS Subject Classification: 13Exx, 13M05, 13M10, 13Hxx, 13B05 1. Introduction Throughout this paper, all rings are finite, associative, commutative with 1(, 0). For a ring R, we denote by R× be the set of invertible elements of R, and J(R) the Jacobson radical of R. Let r, n, and p denote positive integers, p a prime, the residue class ring Z/pnZ of integers modulo pn with p prime and n > 1 and let GR(pn, r) denote the (unique up to isomorphism) Galois extension of degree r of the ring Z/pnZ of integers mod pn. To begin, let f ∈ (Z/pZ)[x] be a primitive irreducible polynomial of degree r. Then there is a unique monic polynomial fn ∈ (Z/pnZ)[x] of degree r such that f ≡ fn(mod p), and fn divides xpr −1 − 1 in (Z/pZ)[x]. Let ξ be a root of fn, so that ξ pr−1 = 1. The Galois ring GR(pn, r) is defined to be (Z/pnZ)[ξ]. Moreover, GR(pn, r) is local ring with J(GR(pn, r)) = (p) and the order of multiplicative subgroup Γp(r)∗ := hξi of GR(pn, r)× is pr −1. The set Γp(r) := Γp(r)∗ ∪ {0} is called Teichmüller set of GR(pn, r). The Galois ring GR(pn, r) depends only on p, n and r. A ring R is a Galois-Eisenstein ring of parameters (p, n, r, e, s) if GR(pn, r) is the largest Galois ring contained in R and all ideals form the chain 1.1 {0} = (θs) ( (θs−1) ( · · · ( (θ) = J(R) ( R. (1.1) Since J(GR(pn, r)) = (p), there exists an integer e such that (θe) = pR. The integer s the nilpotency index of J(R), the Jacobson radical of R, the integer 1 2 ALEXANDRE FOTUE TABUE AND CHRISTOPHE MOUAHA e ramification index of R. According to the Theorem 17.5 of [4], a GE-ring of parameters (p, n, r, e, s) is isomorphic to the ring GR(pn, r)[x]/(xe − pu(x) ; xs) (1.2) where u(x) := ue−1xe−1 + · · · + u1x + u0 ∈ GR(pn, r)[x], with u0 ∈ GR(pn, r)×. The polynomial xe − pu(x) is an associated Eisenstein polynomial to the GE-ring 1.2. W. Clark and J. Liang shown in Lemma 2. of [1] that when n > 1, an element θ in R is a root of an Eisenstein polynomial of degree e over GR(pn, r) if and only if J(R) = (θ). We say that a GE-ring of the form (1.2) is pure if u(x) = u0 ∈ GR(pn, r)×. Clark and Liang shown in [1] that, if p ∤ e then pure GE-rings in the form (1.2) are pure and they enumerate all non-isomorphic pure GE-rings and when p e, Clark and Liang shown in [1] that there are GE-rings which are not pure. Moreover, Xiang-Dong Hou in [2] gave the number of all non- isomorphic pure GE-rings, when n = 2 or p e, p2 ∤ e and (p −1) ∤ e. In this paper, the main goal is the determination all non-isomorphic pure GE-ring of parameters (p, n, r, e, s), when (p − 1) ∤ e. The paper is organized as follows. In Section 2, we review some basic facts about Galois rings and pure GE-rings of parameters (p, n, r, e, s) to be used in sequel. In Section 3, we determine all non-isomorphic pure GE- rings of parameters (p, n, r, e, s). Let R be a pure GE-ring of parameters (p, n, r, e, s). Let 2. Preliminaries Le(R) := (cid:0)R×(cid:1)·e ∩ (GR(pn, r))× be a multiplicative subgroup of GR(pn, r)×, where (R×)·e := {ue : u ∈ R×} . The aim of this section is the determination of the integer ♭(e) ∈ {1; 2; · · · ; n} such that Le(R) = (Γp(r)∗)·e ∩ (1 + p♭(e)GR(pn, r)). (2.1) 2.1. Galois Rings. The theory of Galois Rings was firstly developed by W. Krull (1924) and the reader will find in the monograph [7], more informa- tion about on Galois rings quoted. For this subsection, we gather the results on Galois rings allowing to determine the integer ♭(e). A finite local ring of characteristic pn is called Galois ring GR(pn, r) of characteristic pn of rank r, if its Jacobson radical is generated by p. It is obvious that GR(pn, 1) = Z/pnZ and GR(p, r) = GF(pr), where GF(pr) is the Galois field of the size pr. Let fn ∈ (Z/pnZ)[x] be a monic polynomial of degree r such that fn mod p ∈ (Z/pZ)[x] NON-ISOMORPHIC PURE GALOIS-EISENSTEIN RINGS 3 is irreducible over Z/pZ and fn divides xpr−1 − 1(mod pn). In [5], an algo- rithm allows to compute the polynomial fn is given. Let ξ be a root of fn. Then GR(pn, r) = (Z/pnZ)[ξ] and Γp(r)∗ := hξi is the unique cyclic sub- group of order pr −1 of multiplicative group GR(pn, r)× isomorphic to cyclic multiplicative group GF(pr)∗. The Teichmüller set Γp(r) of GR(pn, r) forms a complete system of representatives modulo p in GR(pn, r). The following proposition gives the immediate proprieties of the Teich- müller set of the Galois ring GR(pn, r). Proposition 1. Let ξ be a generator of Γp(r)∗. Then (1) GR(pn, r) ⊆ GR(pn, r′) if and only if Γp(r) ⊆ Γp(r′) if and only if r devises r′; (2) Γp(r) ∩ Γp(r′) = Γp(gcd(r; r′)); = hξei and the order of hξei is pr−1 gcd(pr −1;e) . (3) (cid:16)Γp(r)∗(cid:17)·e := x2 + x + 2 is irreducible over Example 1. The monic polynomials f GF(3). We denote by α the root of f. Then GF(3)(α) is the Galois field with 9 elements. Moreover, the monic polynomial f2 := x2 − 5x − 1 ∈ (Z/9Z)[x] such f2 mod = f and f2 devises x8 − 1 in ∈ (Z/9Z)[x]. We then construct the Galois ring GR(9, 2) and GR(9, 2) = (Z/9Z)[ξ], where ξ2 = 5ξ + 1. Thus the Teichmüller set of the Galois ring GR(9, 2) is Γ3(2) = {0, 1, ξ, ξ2(= 5ξ+1), ξ3(= 8ξ+5), ξ4(= 8), ξ5(= 8ξ), ξ6(= 4ξ+8), ξ7(= ξ+4)}, and (Γ3(2)∗)·18 = (Γ3(2)∗)·2 = hξ2i = {1, 5ξ + 1, 8}. The following lemma gives the immediate proprieties of a complete residue system modulo pℓ, where ℓ ∈ {0, 1, · · · , n}. Lemma 2.1. Let ξ be a generator of Γp(r)∗ and ℓ ∈ {0, 1, · · · , n}, we con- sider the set ℓ−1 Xi=0 Rr(ℓ) :=  ξipi ∈ GR(pn, r) : ξi ∈ Γp(r) and by convention, we adopt Rr(0) := {0}. Then (2.2) (1) Rr(1) = Γp(r) and Rr(n) = GR(pn, r); (2) Rr(ℓ) forms a complete residue system modulo pℓ in GR(pn, r); (3) for each α ∈ GR(pn, r), for each ℓ ∈ {0; 1; 2; · · · ; n}, there exists a unique (γ; β) in Rr(ℓ) × Rr(n − ℓ), such that α = γ + pℓβ. We remark that Rr(0) ⊂ Rr(1) ⊂ · · · ⊂ Rr(n). (2.3) 4 ALEXANDRE FOTUE TABUE AND CHRISTOPHE MOUAHA Proposition 2. The automorphism group of the ring GR(pn, r) is Aut(GR(pn, r)) := h{σp : ξ 7→ ξ p}i, and for all ξi ∈ Γp(r), Moreover, the groups Aut(GR(pn, r)) and {0; 1; · · · ; r −1}, are isomorphic and for all j ∈ {0; 1; · · · ; r − 1}, n−1 n−1 = Xi=0 σ(ξi)pi. σ ξipi Xi=0 nx ∈ Γp(r) : σ j p(x) = xo = Γp (gcd(r, j)) . We remark that if gcd(r, j) = 1 then nx ∈ Γp(r) : σ j Example 2. Consider the Galois ring p(x) = xo = Z/pnZ. where ξ2 = 5ξ + 1. Thus the Teichmüller set of the Galois ring GR(9, 2) is GR(9, 2) = (Z/9Z)[ξ], Γ3(2) = {0, 1, ξ, 5ξ + 1, 8ξ + 5, 8, 8ξ, 4ξ + 8, ξ + 4}. The Galois group of of the ring GR(pn, r) is Aut(GR(9, 2)) = {Id, σ3 : ξ 7→ ξ3} and {x ∈ Γ3(2) : σ3(x) = x} = Γ3 (gcd(2, 1)) = Z/9Z. The structure of group of invertible elements of GR(pn, r) is given by the following theorem. Theorem 2.2. ([[7] Theorem 14.11]) Let GR(pn, r)× be the group of invert- ible elements of GR(pn, r). Then GR(pn, r)× is the internal direct product of subgroups Γp(r)∗ and 1 + pGR(pn, r). Moreover, (1) If p is odd or if p = 2 and n ≤ 2, then 1 + pGR(pn, r) (cid:27) (Z/(pn))r , (2.4) (2) if p = 2 and n ≥ 3, then 1 + pGR(pn, r) (cid:27) GF(2) ×(cid:16)Z/(2n−2)(cid:17) ×(cid:16)Z/(2n−1)(cid:17)r−1 . (2.5) The Theorem 2.2 has as a consequence the following corollary and this corollary gives a simple expression of the subgroup (1 + pGR(pn, r))·pi of 1 + pGR(pn, r). NON-ISOMORPHIC PURE GALOIS-EISENSTEIN RINGS 5 Corollary 1. Let i be an integer and p is a prime. If p odd or if p = 2 and n ≤ 2, then (1 + pGR(pn, r))·pi = 1 + pi+1GR(pn, r). (2.6) 2.2. Pure Galois-Eisenstein Rings. Let R be a pure Galois-Eisenstein ring (short: pure GE-ring) of parameters (p, n, r, e, s). Then there exists u ∈ GR(pn, r)× such that R = GR(pn, r)[x]/(xe − pu ; xs). The writing GE(u) := GR(pn, r)[x]/(xe − pu ; xs) (2.7) means that GE(u) is a GE-ring of parameters (p, n, r, e, s) defines by the in- vertible element u ∈ GR(pn, r)×. In the sequel, we write GE(u) = GR(pn, r)[θ], such that θe = pu and θs = 0, but θs−1 , 0, for denote the pure Galois- Eisenstein ring of parameters (p, n, r, e, s). We obvious that the Jacobson radical of GE(u) is (θ). Lemma 2.3. Let R be a pure Galois-Eisenstein ring of parameters (p, n, r, e, s). Then the group of invertible elements R× of R is the internal direct product of subgroups Γp(r)∗ and 1 + J(R). The following theorem gives the structure of subset (2.8) of group of invertible elements GR(pn, r)×. This is the main result in this section. Le(R) := (cid:0)R×(cid:1)·e ∩ (GR(pn, r))× Theorem 2.4. Let R be a pure Galois-Eisenstein ring of parameters (p, n, r, e, s) such that s = (n − 1)e + t, 0 ≤ t ≤ e and (p − 1) ∤ e. Then there exists an integer ♭(e) ∈ {1; 2; · · · ; n} such that Le(R) = hηi · (1 + p♭(e)GR(pn, r)), where ♭(e) = ( 1, min{ω + 1; n}, if t ≤ e if t > e p and n = 2; p and n = 2 or n , 2,, where ω := max{i ∈ N∗ : pi e}. Proof 1. The size of the multiplicative group 1 + J(R) is a power of p, therefore (1 + J(R))·pe = 1 + J(R))·pω and ♭(e) = ♭(ω). By Theorem 2.2 and Lemma 2.3., Le(R) = (cid:16)(Γp(r) · (1 + J(R))·pω(cid:17) ∩(cid:16)Γp(r) · (1 + pGR(pn, r)(cid:17) , 6 ALEXANDRE FOTUE TABUE AND CHRISTOPHE MOUAHA It follows that Le(R) = hηi ·(cid:16)(1 + J(R))·pω ∩ (1 + pGR(pn, r))(cid:17) , in according by Lemma 2.1., we have (cid:16)Γp(r)∗(cid:17)·e mine the integer ♭(e) such that = hηi . It suffices to deter- (1 + J(R))·pω ∩ (1 + pGR(pn, r)) = 1 + p♭(e)GR(pn, r)). We write where (1 + J(R))·pω ∩ (1 + pGR(pn, r)) = L1 ∪ L2, L1 = n(1 + θbε)pω ∈ 1 + pGR(pn, r) : ≥ e ; ε ∈ R×o , L2 = n(1 + θbε)pω ∈ 1 + pGR(pn, r) : 1 ≤ b < e ; ε ∈ R×o . Let b be an integer and ε ∈ R×. On the one hand, suppose that b ≥ e. We have and Thus 1 + θbε ∈ 1 + pR (1 + θbε)pω ∈ (1 + pR)·pω ⊆ 1 + pω+1R. (1 + θbε)pω ∈ (1 + pω+1R) ∩ (1 + pGR(pn, r)) = 1 + pω+1GR(pn, r) = (1 + pGR(pn, r))·pω . Since we have (1 + pGR(pn, r))·pω ⊆ L1, L1 = (1 + pGR(pn, r))·pω , NON-ISOMORPHIC PURE GALOIS-EISENSTEIN RINGS 7 on the other hand, suppose that b < e, we develop the following expres- sion (1 + θbε)pω and we obtain: (1 + θbε)pω = 1 + = 1 + since and thus pω−1 l Xl=1 Xi=1 Xjpi≤pω−1 pω ! (θbε)l + (θbε)pω jpi pω ! (θbε) jpi ω−1 p ∤ j , since l = jpi and p ∤ j; + (θbε)pω , pω−i jpi pω ! pω−i+1 ∤ jpi pω ! ; (1 + θbε)pω = 1 + ω−1 Xi=1 θe(ω−i)+bpi εi + θbpω εpω , where εi ∈ R×. We write h(b, i) := e(ω − i) + bpi and h(b) := min{e(ω − i) + bpi : 0 ≤ i < ω}. If (p − 1) ∤ e, then h(b, i) , h(b, i + 1). Thus there exists a unique ı ∈ {0; · · · ; ω − 1} such that h(b) = h(b, ı). Therefore, there exists εb ∈ R× such that (1 + θbε)pω = 1 + θh(b)εb + θbpω εpω . This forces that h(b) ≥ s, and bpω ≡ 0 mod e. We have a ∈ N∗ such that (pu)aεpω ∈ pGR(pn, r). We obtain, L2 = ( 1 + pGR(pn, r), {1}, if n = 2 and t ≤ e p; if n , 2 or n = 2 and t > e p . (cid:4) Example 3. Let R be a GE-ring of parameters (3, 2, 2, 18, 25). Then by Theorem 2.2, ♭(2) = 2. Thus L9(R) = hξ2i · (1 + 38GR(9, 2)) = {1, ξ2, ξ4}, where Γ3(2)∗ = hξi and ξ2 = 5ξ + 1. 8 ALEXANDRE FOTUE TABUE AND CHRISTOPHE MOUAHA 3. Main result In this section, we determine the pure GE-rings of parameters (p, n, r, e, s). Note when n = 1, R = GF(pr)/(xe). Such rings need no classification, so we will suppose that n ≥ 2. The isomorphism problem for pure GE-rings with given parameters (p, n, r, e, s) mentions by A. A. Nechaev in [6] and T. G. Gazaryan in [3], shown that pure GE-rings GE(1) and GE(2) of param- eters (3, 2, 1, 2, 3) are non-isomorphic rings with isomorphic additive and multiplicative groups. Let ∼ be the equivalence relation defined on GR(pn, r)× by u ∼ v ⇔ GE(u) (cid:27) GE(v), (3.1) where u, v ∈ GR(pn, r)×. Let σ be a generator of Aut(GR(pn, r)), and the multiplicative subgroup Le(R) = hηi · (1 + p♭(e)GR(pn, r)) . Consider of GR(pn, r)× and η a generator of (cid:16)Γp(r)∗(cid:17)·e Ur(♭(e)) := hχi + pRr(∂(e) − 1), (3.2) where ∂(e) = min{∂(e); n − 1}. We write χ := ξ then hχi · hηi = Γp(r)∗. pr−1 d and d := gcd(pr − 1; e). Since gcd(cid:16)d; ( pr−1 d )(cid:17) = 1, The isomorphic GE-rings have the same parameters, but there are non- isomorphic GE-ring with the same parameters. The following example il- lustrates the isomorphism problem for the GE-rings. Example 4. T. G. Gazaryan, given in [3], two GE-rings of parameters (3, 2, 1, 2, 3) GE(1) = (Z/9Z)[θ] and GE(2) = (Z/9Z)[δ] as an example of non-isomorphic commutative chain rings. We have θ2 = 3, δ2 = 6, and θ3 = δ3 = 0. Then the radical Jacobson are: J(GE(1)) = (θ) = {0; 3; 6; θ; 3θ} and J(GE(1)) = (δ) = {0; 3; 6; δ; 3δ}. The Teichmüller set Γ3(1) = {0; 1; 8}, allows to write GE(1) = {a + bθ : (a; b) ∈ Γ3(1)} and GE(2) = {a + bδ : (a; b) ∈ Γ3(1)}. NON-ISOMORPHIC PURE GALOIS-EISENSTEIN RINGS 9 But J(GE(1)) , J(GE(2)). Indeed, if J(GE(1)) = J(GE(2)), then there exists z ∈ GE(1)× such that θ = zδ. As 3 = θ2 = z2δ2 = 6z2. It follows, z2 = (a + θb)2 = a2 + 2θab. Therefore, the equation 3 = 6z2 is equivalent to 3 = 6a2. Now, for all a ∈ (Z/9Z)× = {1; 2; 4; 5; 7; 8}, we have 6a2 = 6. It is absurd. Lemma 3.1. Let R be the GE-ring of parameters (p, n, r, e, s). Then there exists v in Ur(♭(e)) such that R = GE(v). Proof 2. Let R be the GE-ring of parameters (p, n, r, e, s). Then there exists u ∈ GR(pn, r)× such that R = GE(u). By the item3 of Lemma 2.1., there exists (γ; β) ∈ Rr(♭(e)) × Rr(n − ♭(e)), such that u = γ + p♭(e)β and γ ∈ GR(pn, r)×. Since hδi · hηi = Γp(r)∗, there exists (α, v) ∈ hηi × Ur(♭(e)) such that γ = αv. Now, u = v(α + p♭(e)β), where β = βv−1. We have g := α + p♭(e)β ∈ Le(R). By Theorem 2.4, there exists z ∈ R× such that g = ze. Since GE(u) = GR(pn, r)[θ] with θe = pv and s = min{i ∈ N : θi = 0}. Thus, J(GE(u)) = (δ) where δ := zθ. We have δe = pv and s = min{i ∈ N : δi = 0}. So GE(u) = GE(v). (cid:4) The following lemma gives a fundamental condition for non-isomorphic pure GE-rings. Lemma 3.2. For each u and v in Ur(♭(e)). Then u ∼ v ⇔ u = σi(v), for some i ∈ {0; 1; · · · ; r − 1}. Proof 3. Let θ := x + (xe − pu ; xs) be a generator of J(GE(u)). Then pure GE-rings GE(u) and GE(v) are isomorphic means by Lemma. 4 of [1], the existence of (i; z) ∈ {0; 1; · · · ; r − 1} × GE(u)× such that (θz)e = pσi(v) and θe = pu. Since uze = σi(v) and u, σi(v) ∈ Ur(♭(e)), we have ze ∈ 1 + p♭(e)GR(pn, r). Therefore, u ∼ v ⇔ ∃(i; γ) ∈ {0; 1; · · · ; r − 1} × (1 + p♭(e)GR(pn, r)) : uγ = σi(v), where γ := ze; ⇔ ∃(i; γ) ∈ {0; 1; · · · ; r − 1} × (1 + p♭(e)GR(pn, r)) : uγ = σi(v). Now, γ ∈ (1 + p♭(e)GR(pn, r)) ∩ Ur(♭(e)) = {1}. (cid:4) 10 ALEXANDRE FOTUE TABUE AND CHRISTOPHE MOUAHA Let u ∈ Ur(♭(e)), we write C(u) = {σi(u) : i ∈ {0; 1; · · · ; r − 1}}, the Frobenius class of u and Cr(♭(e)) a complete set of Frobenius representative of Ur(♭(e)). Theorem 3.3. Let Ξ∗(p, n, r, e, s) be the set of pure GE-rings of parameters (p, n, r, e, s) up to isomorphism. Suppose that (p − 1) ∤ e. Then the mapping GE : Cr(♭(e)) → Ξ∗(p, n, r, e, s) GE(u) u 7→ (3.3) is bijective. Moreover, Ξ∗(p, n, r, e, s) = 1 r r Xi=1 gcd(pgcd(r,i) − 1, e)p(∂(e)−1)gcd(i,r). (3.4) Proof 4. By Lemma 3.1, the mapping GE is bijection. u ∈ Ur(♭(e)), there exists a unique v in Cr(♭(e)) such that GE(u) = GE(v). Indeed, For each Now, we determine the size of Ξ∗(p, n, r, e, s). Then we use the action of σ on Ur(♭(e)) is given by: σ : Ur(♭(e)) → Ur(♭(e)) 7→ σ(ε). ε (3.5) For each 1 ≤ i ≤ r, we have: f ix(σi) = nε ∈ Ur(♭(e)) : σi(ε) = εo ; = Uri(♭(e)), where ri := gcd(r, i). Thus by the Burnside's Lemma, the number of hσi−orbits in Ur(♭(e)) is 1 hσi r Xi=1 f ix(σi) = 1 r r Xi=1 gcd(pri − 1, e)p(∂(e)−1)ri. (cid:4) Thus, the mapping GE allows to determine up to isomorphism all pure GE-rings. Example 5. Consider the pure GE-rings of parameters (3, 2, 1, 2, 3). In ac- cording by Theorem Theorem 2.3, ♭(2) = 1. Then C1(1) = U1(1) = {1; 8}. By formula of Theorem 3.3, there are two non-isomorphism pure GE-rings of parameters (3, 2, 1, 2, 3). Therefore, in the article [3], the GE-rings r(1) et r(8) are only non-isomorphism pure GE-rings of parameters (3, 2, 1, 2, 3). Example 6. Consider the GE-rings GR(9, 2)[θ] of parameters (3, 2, 2, 18, 25) and GR(9, 2) = (Z/9Z)[ξ], with ξ2 = 5ξ + 1. In according by Theorem Theorem 2.3, ♭(18) = 2. Then C2(18) = U2(2) = hξ4i = {1; 8}. NON-ISOMORPHIC PURE GALOIS-EISENSTEIN RINGS 11 By formula of Theorem 3.3, there are 2 non-isomorphism pure GE-rings of parameters (3, 2, 2, 18, 25), and the GE-rings of parameters (3, 2, 2, 18, 25) up to isomorphism are r(1), and r(8) Conclusion A pure Galois-Eisenstein ring of (p, n, r, e, s) is constructed from an el- ement of the complete set of Frobenius representative of Ur(♭(e)) and this construction are unique up to isomorphism. In general, the isomorphism problem for Galois-Eisenstein Rings stays open. References [1] W. Clark and J. Liang, Enumeration of Finite Communtative Chain Rings, J. Algebra 27,(1973) 445-453 . [2] X. Hou, Finite Communtative Chain Rings, Finite Fields Appl. 7 (2001) 382 - 396 [3] T. G. Gazaryan, An example of non-isomorphic commutative chain rings, Uspekhi Mat. Nauk, 47:3(285) (1992), 155-156 [4] B. R. McDonald , Finite Rings with Identity, Marcel Dekker, New York, 1974. [5] Gary McGuire, An Approach to Hensel's Lemma, Irish Math. Soc. Bullentin 47 (2001), 15-21. [6] A. A. Nechaev, Finite Rings with Applications, Handbook of Algebra, Vol. 5,(2008) 213-320. [7] Zhe-Xian Wan, Lectures on Finite Fields and Galois Rings, World Scientific, (2003) 342 pages. Alexandre Fotue Tabue Department of mathematics, Faculty of Sciences, University of Yaounde 1, Cameroon E-mail address: [email protected] Christophe Mouaha Department of mathematics, Higher Teachers Training College of Yaounde, University of Yaounde 1, Cameroon E-mail address: [email protected]
1810.07881	1	1810	2018-10-18T02:43:16	Hom-Lie groups of a class of Hom-Lie algebra	[ "math.RA", "math-ph", "math-ph" ]	In this paper, the definition of Hom-Lie groups is given and one conntected component of Lie group $GL(V)$, which is not a subgroup of $GL(V)$, is a Hom-Lie group. More, we proved that there is a one-to-one relationship between Hom-Lie groups and Hom-Lie algebras $(\gl(V),[\cdot,\cdot]_\beta,\rm{Ad}_\beta)$. Next, we also proved that if there is a Hom-Lie group homomorphism, then, there is a morphism between their Hom-Lie algebras. Last, as an application, we use these results on Toda lattice equation.	math.RA	math	Hom-Lie groups of a class of Hom-Lie algebra Department of Mathematics and Computer, Yichun University, Jiangxi, 336000, China Zhen Xiong Abstract: In this paper, the definition of Hom-Lie groups is given and one conntected com- ponent of Lie group GL(V ), which is not a subgroup of GL(V ), is a Hom-Lie group. More, we proved that there is a one-to-one relationship between Hom-Lie groups and Hom-Lie algebras (gl(V ), [·, ·]β, Adβ). Next, we also proved that if there is a Hom-Lie group homomorphism, then, there is a morphism between their Hom-Lie algebras. Last, as an application, we use these results on Toda lattice equation. Keyword: Hom-Lie groups; Hom-Lie algebras; homomorphism; Toda hierarchy. MR(2010) Subject Classification: 17B99, 22E99. 1 Introduction The notion of Hom-Lie algebras was introduced by Hartwig, Larsson, and Silvestrov in [1] as part of a study of deformations of the Witt and the Virasoro algebras. Some q-deformations of the Witt and the Virasoro algebras have the structure of a Hom-Lie algebra [1, 2]. Because of close relation to discrete and deformed vector fields and differential calculus [1, 3, 4], more people pay special attention to this algebraic structure [5, 6, 7, 8, 11]. Its geometric generalization is given in [9] and [10]. In [7],the authors give a Hom-Lie algebra (gl(V ), [·, ·]β , Adβ). This Hom-Lie algebra play an im- portant role of studying structures of Hom-Lie algebras. Base on Hom-Lie algebra (gl(V ), [·, ·]β, Adβ), there are many results are given: Omni-Hom-Lie algebra is given in [7]; Hom-big brackets are given in [8]; a Hom-Lie algebroid structure on ϕ!T M is given in [9], and have the following results: there is a purely Hom-Poisson algebra structure on C∞(M ); a Hom-Lie algebra structure on the set of (σ, σ)− derivations of an associative algebra is givn in [11], and so on. As we know, from a Lie group, we get a Lie algebra; on the other hands, from a Lie algebra, we can have a Lie group. Hence, are there Hom-Lie groups, have similar results like Lie groups? In this paper, first, we study some properties of Hom-Lie algebra (gl(V ), [·, ·]β, Adβ), then, give the definition of Hom-Lie groups, similar to Hom-Lie algebras, a Hom-Lie group twisted by a diffeomorphism is a Lie group. Next, on matrix space, we give some examples of Hom-Lie groups, and proved that one connected component of GL(V ), which is not a subgroup, is a Hom-Lie group. More, we proved that there is a one-to-one relationship a Hom-Lie group and a Hom-Lie algebra (gl(V ), [·, ·]β , Adβ). Then, we give definitions of homomorphism on Hom-Lie groups, and 0Supported by the NSF of China (No.11771382) and The Science and Technology Project(GJJ161029)of Depart- ment of Education, Jiangxi Province. 0E-mail address: [email protected] 1 proved that a homomorphism of Hom-Lie groups induce a morphism of Hom-Lie algebras. As an application, we study a deformation of Toda lattice equation. The paper is organized as follow. In Section 2, we recall some necessary background knowledge, including Hom-Lie algebras and morphism, Hom-Lie algebra (gl(V ), [·, ·]β, Adβ). In Section 3, we study cohomology of Hom-Lie algebra (gl(V ), [·, ·]β , Adβ), and have: cohomologies of Hom-Lie algebra (gl(V ), [·, ·]β, Adβ) and Lie algebra gl(V ) are isomorphic. In Section 4, we give definitions of Hom-Lie groups, homomorphism of Hom-Lie groups, and have mainly results: Theorem 4.7 and Theorem 4.13. In Section 5, for a class of deformations of Toda lattice equation, it is integrable. 2 Preliminaries The notion of a Hom-Lie algebra was introduced in [1], see also [6] for more information. Definition 2.1. (1.) A Hom-Lie algebra is a triple (g, [·, ·], α) consisting of a vector space g, a skewsymmetric bilinear map (bracket) [·, ·] : ∧2g −→ g and a linear transformation α : g → g satisfying α[x, y] = [α(x), α(y)], and the following hom-Jacobi identity: [α(x), [y, z]] + [α(y), [z, x]] + [α(z), [x, y]] = 0, ∀x, y, z ∈ g. A Hom-Lie algebra is called a regular Hom-Lie algebra if α is a linear automorphism. When α = id, Hom-Lie algebra (g, [·, ·], α) is just Lie algebra (g, [·, ·]). (2.) A subspace h ⊂ g is a Hom-Lie sub-algebra of (g, [·, ·], α) if α(h) ⊂ h and h is closed under the bracket operation [·, ·], i.e. for all x, y ∈ h, [x, y] ∈ h. (3.) A morphism from the Hom-Lie algebra (g, [·, ·]g, α) to the Hom-Lie algebra (h, [·, ·]h, δ) is a linear map ψ : g −→ h such that ψ([x, y]g) = [ψ(x), ψ(y)]h and ψ ◦ α = δ ◦ ψ. When ψ is invertible, then ψ is an isomorphism. The set of k-cochains on Hom-Lie algebra (g, [·, ·], α) with values in V , which we denote by Ck(g; V ), is the set of skewsymmetric k-linear maps from g × · · · × g(k-times) to V : Ck(g; V ) := {η : ∧kg −→ V is a linear map}. Associated to the trivial representation, the set of k-cochains is ∧kg∗. The corresponding cobound- ary operator d : ∧kg∗ −→ ∧k+1g∗ is given by (see [7, 12]) dξ(x1, · · · , xk+1) =Xi<j (−1)i+jξ([xi, xj], α(x1), · · · , xi, · · · , xj, · · · , α(xk+1)). Theorem 2.2. [7] Let V be a vector space, and β ∈ GL(V ). Define a skew-symmetric bilinear bracket operation [·, ·]β : gl(V ) × gl(V ) −→ gl(V ) by [A, B]β = βAβ−1Bβ−1 − βBβ−1Aβ−1, ∀ A, B ∈ gl(V ), where β−1 is the inverse of β. Denote by Adβ : gl(V ) −→ gl(V ) the adjoint action on gl(V ), i.e. Adβ(A) = βAβ−1. Then (gl(V ), [·, ·]β , Adβ) is a regular Hom-Lie algebra. Obviously, when β = id, (gl(V ), [·, ·]β , Adβ) is just Lie algebra (gl(V ), [·, ·]). 2 3 Properties of Hom-Lie algebra (gl(V ), [·, ·]β, Adβ) In this paper, we just study β ◦ β = id, i.e. β−1 = β. Proposition 3.1. For any C ∈ GL(V ), let γ = CβC −1, then there is an isomorphism from (gl(V ), [·, ·]β, Adβ) to (gl(V ), [·, ·]γ , Adγ). Proof. Define map F : (gl(V ), [·, ·]β , Adβ) −→ (gl(V ), [·, ·]γ, Adγ) by F (x) = CxC −1, then F is an isomorphism. Remark 3.2. When β 6= ±id, there is not an isomorphism from (gl(V ), [·, ·]) to (gl(V ), [·, ·]β , Adβ). For Hom-Lie algebra (gl(V ), [·, ·]β , Adβ), coboundary operator d : ∧kgl(V )∗ −→ ∧k+1gl(V )∗ is given by dξ(x1, · · · , xk+1) =Xi<j (−1)i+jξ([xi, xj]β, Adβ(x1), ·,dxi,j , ·, Adβ(xk+1)). For Lie algebra (gl(V ), [·, ·]), coboundary operator d : ∧kgl(V )∗ −→ ∧k+1gl(V )∗ is given by dξ(x1, · · · , xk+1) =Xi<j (−1)i+jξ([xi, xj], x1, ·,dxi,j, ·, xk+1). Then, we have two chomology complexes: (⊕k ∧k gl(V )∗, d) of Hom-Lie algebra (gl(V ), [·, ·]β , Adβ) and (⊕k ∧k gl(V )∗, d) of Lie algebra (gl(V ), [·, ·]). β induce map βr : ∧kgl(V )∗ −→ ∧kgl(V )∗ by And β also induce map βl : ∧kgl(V )∗ −→ ∧kgl(V )∗ by βr(ξ)(x1, · · · , xk) = ξ(x1β, · · · , xkβ). βl(ξ)(x1, · · · , xk) = ξ(βx1, · · · , βxk). Proposition 3.3. For ξ ∈ ∧kgl(V )∗, we have: βr ◦ dξ = dβl(ξ); dβr(ξ) = βl ◦ dξ. (1) (2) Proof. For ξ ∈ ∧kgl(V )∗, βr ◦ dξ(x1, · · · , xk+1) = dξ(x1β, · · · , xk+1β) = Xi<j = Xi<j (−1)i+jξ(βxixj − βxjxi, βx1, ·,dxi,j , ·, βxk+1) (−1)i+jβl(ξ)(xixj − xjxi, x1, ·,dxi,j, ·, xk+1) = dβl(ξ)(x1, · · · , xk+1) The proof of the rest of the conclusions is similar. Remark 3.4. In fact, we also have: βl ◦ d 6= d ◦ βl, βr ◦ d 6= d ◦ βr. 3 Denote the set of closed k-cochains of complex (⊕k ∧k gl(V )∗, d) by Z k(HL; R) and the set of exact k-cochains of complex (⊕k∧k gl(V )∗, d) by Bk(HL; R). Denote the corresponding cohomology by H k(HL; R) = Z k(HL; R)/Bk(HL; R). Similarly, denote the set of closed k-cochains of complex (⊕k ∧k gl(V )∗, d) by Z k(L; R) and the set of exact k-cochains of complex (⊕k ∧k gl(V )∗, d) by Bk(L; R). Denote the corresponding cohomology by H k(L; R) = Z k(L; R)/Bk(L; R). Theorem 3.5. With the above notations, we have: H k(HL; R) = H k(L; R). Proof. For ξ1 ∈ Z k(HL; R), by 0 = βr ◦ dξ1 = dβl(ξ1), we have: βl(ξ1) ∈ Z k(L; R). For ξ2 ∈ Z k(L; R), by βr ◦dβl(ξ2) = dξ2 = 0, we have: βl(ξ2) ∈ Z k(HL; R). So, we have: Z k(HL; R) = Z k(L; R). For η1 ∈ Bk(HL; R), there is a η2 ∈ ∧k−1gl(V )∗, and such that dη2 = η1. By dβl(η2) = βr ◦ dη2 = βr(η1), we have: βr(η1) ∈ Bk(L; R). On the other hand, for η ∈ Bk(L; R), there is a η3 ∈ ∧k−1gl(V )∗, and such that dη3 = η. By dβr(η3) = βl dη3 = βl(η), then, βl(η) ∈ Bk(HL; R). So, we have: Bk(HL; R) = Bk(L; R). 4 Hom-Lie groups Definition 4.1. Let G is a Lie group, S is a submanifold of G. If there is a diffeomorphism F : G −→ G, such that F (S) is a subgroup of G. Then (S, F ) is called a Hom-Lie group. For ∀x, y ∈ S, if F (y) is the inverse of F (x) in Lie group F (S), then we called y is the Hom-inverse of x in Hom-Lie group (S, F ). When F = id, Hom-Lie group (S, F ) is the Lie group S. Example 4.2. Matrix Lie group GL(n; R), let S1 = {A ∈ GL(n; R)A < 0}, S2 = {A ∈ GL(n; R)A > 0}. S1 and S2 are connected components of GL(n; R), S2 is a subgroup of GL(n; R) and S1 is not a subgroup of GL(n; R). We define the map F : GL(n; R) −→ GL(n; R) by F (A) = APi,j , where Pi,j is given by exchanging line i and line j of id, P 2 i,j = id. F is a diffeomorphism and F 2 = id. So (S1, F ) is a Hom-Lie group. ∀A ∈ S1, Pij A−1Pij is the Hom-inverse of A. Example 4.3. Matrix Lie group O(n), let S3 = {A ∈ O(n)A = −1}, S4 = {A ∈ O(n)A = 1}. S3 and S4 are connected components of O(n), S4 is a subgroup of O(n) and S3 is not a subgroup of O(n). We define the map F : O(n) −→ O(n) by F (A) = APi,j, where Pi,j is given by exchanging line i and line j of id, P 2 i,j = id. F is a diffeomorphism and F 2 = id. So (S3, F ) is a Hom-Lie group. ∀A ∈ S1, Pi,j A−1Pi,j is the Hom-inverse of A. Definition 4.4. A function p : R −→ (S, F ) is called a one-parameter subgroup of Hom-Lie group (S, F ) if 1. p is continuous, 2. F (p(0)) = id, 3. F (p(t + s)) = F (p(t))F (p(s)). 4 Example 4.5. For matrix Lie group O(1; 1), let sinht S1 = {(cid:18) cosht S3 = {(cid:18) cosht −sinht sinht cosht (cid:19) t ∈ R}; S2 = {(cid:18) −cosht sinht −cosht (cid:19) t ∈ R}; S4 = {(cid:18) −cosht −sinht sinht sinht sinht −cosht (cid:19) t ∈ R}; cosht (cid:19) t ∈ R}. S1, S2, S3, S4 are connected components of O(1; 1). S2, S3, S4 are not subgroups of O(1; 1), S1 is a subgroup of O(1; 1), hence S1 is a Lie group. We define the map F2 : O(1; 1) −→ O(1; 1) by F2(A) = A(−id), then (S2, F2) is a Hom-Lie group and F 2 2 = id. The map F3 : O(1; 1) −→ O(1; 1) is given by F3(A) = A(cid:18) 1 0 We define the map F4 : O(1; 1) −→ O(1; 1) by F4(A) = A(cid:18) −1 0 0 −1 (cid:19), then (S3, F3) is a Hom- 1 (cid:19), then (S4, F4) is a 3 = id. 0 Lie group and F 2 Hom-Lie group and F 2 4 = id. Element of Si is a one-parameter subgroup of (Si, Fi), i = 2, 3, 4. GL(V ) is a matrix Lie group, (gl(V ), [·, ·]) its Lie algebra, there is a map exp : gl(V ) −→ GL(V ), for any X ∈ gl(V ), exp(X) = eX . More about the map exp, please see [13]. For β ∈ GL(V ), let Mβ = {eβXβX ∈ gl(V )}, then Mβ ⊂ GL(V ), for X ∈ gl(V ), let p(t) = etβX β, then p(t) ⊂ Mβ. Proposition 4.6. For β ∈ GL(V ), β2 = id, let Mβ = {eβXβX ∈ gl(V )}, then (Mβ, Rβ) is a Hom-Lie group. Proof. When β = id, Mβ = GL(V ) is a Lie group. When β 6= id, if X0 ∈ gl(V ) and such that eβX0 β = id, we have eβX0 = β, and e2βX0 = β2 = id, then X0 = 0, we have β = id, but β 6= id. So, Mβ is not a subgroup of GL(V ). connected component of GL(V ). Then Mβ is a submanifold of GL(V ). By p(t) ⊂ Mβ ⊂ SX ∈gl(V ) p(t), we have Mβ = SX ∈gl(V ) p(t). Because of β ∈ T p(t), Mβ is a Rβ(cid:0)Mβ(cid:1) = {eβXX ∈ gl(V )}. So(Mβ, Rβ) is a Hom-Lie group, and Hom-inverse of eβX β is e−βXβ. We define the map Rβ : GL(V ) −→ GL(V ) by Rβ(A) = Aβ, Rβ is a diffeomorphism. Then For X ∈ gl(V ), let p(t) = etβXβ, then p(t) is a one-parameter subgroup of Hom-Lie group (Mβ, Rβ). For Y ∈ gl(V ), we have d dt (etβX βY e−tβXβ)t=0 = βXβY e0β − e0βY βXβ = βXβY β − βY βXAβ = [X, Y ]β. (3) Actually, we proved the following results. Theorem 4.7. For β ∈ GL(V ), β2 = id, there is a one-to-one relationship between Hom-Lie group (Mβ, Rβ) and regular Hom-Lie algebra (gl(V ), [·, ·]β, Adβ). When β = id, Hom-Lie group (Mβ, Rβ) is the Lie group GL(V ) and Hom-Lie algebra (gl(V ), [·, ·]β , Adβ) is the Lie algebra gl(V ). 5 Example 4.8. Let M be a manifold and ϕ : M −→ M a diffeomorphism. Define Adϕ∗ : Γ(ϕ!TM) −→ Γ(ϕ!TM) by Adϕ∗(X) = ϕ∗ ◦ X ◦ (ϕ∗)−1, ∀X ∈ Γ(ϕ!TM). Define a skew-symmetric bilinear operation [·, ·]ϕ∗ : ∧2Γ(ϕ!T M ) −→ Γ(ϕ!T M ) by [X, Y ]ϕ∗ = ϕ∗ ◦ X ◦ (ϕ∗)−1 ◦ Y ◦ (ϕ∗)−1 − ϕ∗ ◦ Y ◦ (ϕ∗)−1 ◦ X ◦ (ϕ∗)−1. Then, (ϕ!T M, [·, ·]ϕ∗, Adϕ∗ , id) is a Hom-Lie algebroid and is called tangent Hom-Lie algebroid ([9]) . In this example, now, we suppose ϕ2 = id and ϕ 6= id, at a point m ∈ M , x is a vector field for any f ∈ C∞(M ), on M , if θt(m) is a one parameter Lie subgroup with respect to xm, i.e. df = xmf = limt→0 1 t [f (θt(m)) − f (m)]. For tangent Hom-Lie algebroid, Xm ∈ Γ(ϕ!TmM ) and Xm = xϕ(m), by representation of Hom- Lie algebroids on C∞(M ) ([14]), for any f ∈ C∞(M ) d1f (Xm) = (ϕ∗)2 ◦ Xm ◦ ϕ∗(f ) = Xm ◦ (f ◦ ϕ) = xϕ(m)(f ◦ ϕ) = lim t→0 f(cid:0)ϕ ◦ θt(ϕ(m))(cid:1) − f (m) t . Hence, Xm with respect to one parameter Hom-Lie subgroup is ϕ ◦ θt(ϕ(m)). On the other hands, xm ∈ TmM , d dt etxmf t=0 = xmf. Xm ∈ Γ(ϕ!TmM ), then p(t) = etϕ∗ ◦Xm ϕ∗ is one parameter Hom-Lie subgroup, and d dt etϕ∗ ◦Xm ϕ∗ ◦ f t=0 = ϕ∗ ◦ Xm ◦ ϕ∗ ◦ f = xϕ(m)(f ◦ ϕ). So, we got the same result through different paths. Actually, we can get the following results: Theorem 4.9. Let M be a manifold, ϕ : M −→ M is a diffeomorphism and ϕ2 = id, ϕ 6= id. If x is a vector field on M , for any m ∈ M , then Dmϕ(xm) 6= xϕ(m). Proof. Let θt(m) is a one parameter Lie subgroup with respect to xm. If Dmϕ(xm) = xϕ(m), then we have ϕ ◦ θt(m) = θt(ϕ(m)), i.e., ϕ ◦ θt(ϕ(m)) = θt(m). But, by Example 4.8, ϕ ◦ θt(ϕ(m)) is a one parameter Hom-Lie subgroup. Hence, Dmϕ(xm) 6= xϕ(m). 6 Definition 4.10. Let G1 and G2 are Lie groups, Si is a submanifold of Gi, i = 1, 2. (S1, F1) and (S2, F2) are Hom-Lie groups. Φ : G1 −→ G2 is a Lie group homomorphism. We called Φ is a Hom-Lie group homomorphism from (S1, F1) to (S2, F2), if 1.) Φ(S1) ⊂ S2; 2.) ∀x, y ∈ S1, F2 ◦ Φ(xy) = Φ ◦ F1(xy). Example 4.11. In Example 4.2 and Example 4.3, we define map I : O(n) −→ GL(n; R) by I(A) = A, i.e., I is an inclusion. I is a homomorphism from (S3, F ) to (S1, F ). Example 4.12. The map g : GL(n; R) −→ R is given by g(A) = A, then g is a Lie group homomorphism. In Example 4.2, (S1, F ) is a Hom-Lie group, let S2 = {y ∈ Ry < 0} and define the map F2 : R −→ R by F2(y) = −y, then (S2, F2) is a Hom-Lie group. We have: g(S1) ⊂ S2 and ∀A, B ∈ S1, g ◦ F (AB) = g(ABPi,j ) = ABPi,j = −AB = F2 ◦ g(AB), so g is a Hom-Lie group homomorphism. Theorem 4.13. Let γ 6= β, γ 6= id, β 6= id, γ2 = id = β2, (Mγ, Rγ) and (Mβ, Rβ) are Hom-Lie groups, with Hom-Lie algebras (gl(V ), [·, ·]γ , Adγ) and (gl(V ), [·, ·]β, Adβ), respectively. Suppose that Φ : (Mγ, Rγ) −→ (Mβ, Rβ) is a Hom-Lie group homomorphism. Then, there exists a real linear map φ : (gl(V ), [·, ·]γ, Adγ) −→ (gl(V), [·, ·]β , Adβ) such that Φ(eγX) = eβφ(X) for all X ∈ gl(V ). The map φ has following additional properties: (1.) φ ◦ Adγ(X) = Adβ ◦ φ(X), for all X ∈ gl(V ); (2.) φ(eγX γY e−γXγ) = Φ(eγX γ)φ(Y )Φ(e−γX γ), for all X, Y ∈ gl(V ); (3.) φ([X, Y ]γ) = [φ(X), φ(Y )]β, for all X, Y ∈ gl(V ); (4.) βφ(X) = d dt Φ(eγX)t=0, for all X ∈ gl(V ). Proof. Since Φ is a Lie group homomorphism, then Rβ ◦ Φ(eγX γ) = Φ ◦ Rγ(eγXγ), we have Φ(γ) = β. And Φ is a continuous group homomorphism, for each X ∈ gl(V ), so Φ(etγX ) will be a one-parameter subgroup of Lie group GL(V ). Thus, there is a unique Z ∈ gl(V ) such that Φ(etγX) = etβZ. (4) We define φ(X) = Z and check in several steps that φ has the required properties. Step 1, Φ(eγX) = eβφ(X). This follows from (4) and our definition of φ, by putting t = 1. Step 2, φ(sX) = sφ(X), for all s ∈ R. This is obviously. Step 3, φ(X + Y ) = φ(X) + φ(Y ). By Steps 1 and 2, etβφ(X+Y ) = eβφ(t(X+Y )) = Φ(et(γX+γY )). By the Lie product formula and the fact that Φ is a continuous homomorphism, we have etβφ(X + Y ) = Φ(cid:0) lim = lim m→∞(cid:0)etγX/metγY /m(cid:1)m(cid:1) m→∞(cid:0)Φ(etγX/m)Φ(etγY /m)(cid:1)m . 7 Then, we have etβφ(X + Y ) = lim m→∞(cid:0)etβφ(X)/metβφ(Y )/m(cid:1)m Differentiating this result at t = 0 gives the desired result. = etβ(φ(X)+φ(Y )). Step 4, φ ◦ Adγ(X) = Adβ ◦ φ(X). By Step 1, Φ(eγAdγ (X)γ) = Φ(eγAdγ (X))Φ(γ) = eβφ(Adγ (X))β. And eγAdγ (X)γ = eXγγ = γeγX, then Φ(eγAdγ (X)γ) = Φ(γeγX) = βΦ(eγX ) = βeβφ(X). By (5) and (6), we have: eβφ(Adγ (X)) = βeβφ(X)β = eφ(X)β. Hence, φ(Adγ(X)) = βφ(X)β = Adβ ◦ φ(X). Step 5, φ(eγX γY e−γXγ) = Φ(eγXγ)φ(Y )Φ(e−γXγ). eβφ(cid:0)etγX γY e−tγX γ(cid:1) = Φ(cid:0)eγetγX γY e−tγX γ(cid:1) = Φ(cid:0)eetXγ Y γe−tXγ(cid:1) = Φ(cid:0)etXγγγeY γe−tXγ(cid:1) = Φ(cid:0)γetγXeγY e−tγX γ(cid:1) = βΦ(etγX )Φ(eγY )Φ(e−tγX )β = βΦ(etγX )eβφ(Y )Φ(e−tγX)β = eβΦ(etγX )βφ(Y )Φ(e−tγX )β. (5) (6) Hence, we have φ(cid:16)etγXγY e−tγXγ(cid:17) = Φ(etγX γ)φ(Y )Φ(e−tγX γ). By putting t = 1, we have the desired result. Step 6, φ([X, Y ]γ) = [φ(X), φ(Y )]β. By Step 5, we have βφ(cid:16)etγX γY e−tγXγ(cid:17) = βΦ(etγX )βφ(Y )Φ(e−tγX )β = βetβφ(X)βφ(Y )e−tβφ(X)β. On the other hands, φ([X, Y ]γ) = φ(cid:16) d etγXγY e−tγXγt=0(cid:17) dt φ(etγX γY e−tγX γ)t=0 etβφ(X)βφ(Y )e−tβφ(X)βt=0 = = d dt d dt = [φ(X), φ(Y )]β. 8 Step 6, βφ(X) = d dt Φ(eγX)t=0. This follows our definition of φ. 5 Applications Consider the following equation d dt L = BL − LB = [B, L], (7) where L is an n × n symmetric real tridiagonal matrix, and B is the skew symmetric matrix obtained from L by B = L>0 − L<0, where L>0(<0) denotes the strictly upper (lower) triangular part of L. Based on Lie algebras, Kodama,Y. and Ye, J.([15, 16]) study equation (7), and give an explicit formula for the solution to the initial value problem. Now we consider the following system: d dt L = βBβLβ − βLβBβ = [B, L]β, (8) where β2 = id, then, from results of Section 4, we know that equation (8) is also integrable. References [1] J. Hartwig, D. Larsson and S. Silvestrov, Deformations of Lie algebras using σ-derivations. J. Algebra 295 (2006), 314-361. [2] N. Hu, q-Witt algebras, q-Lie algebras, q-holomorph structure and representations. Algebra Colloq. 6 (1999), no. 1, 51-70. [3] D. Larsson and S. Silvestrov, Quasi-hom-Lie algebras, central extensions and 2-cocycle-like identities. J. Algebra 288 (2005), 321-344. [4] D. Larsson and S. Silvestrov, Quasi-Lie algebras. Contemp. Math. 391 (2005), 241-248. [5] Y. Sheng, Representations of Hom-Lie Algebras. Algebr. Represent. Theor. 15 (2012), 1081- 1098. [6] A. Makhlouf and S. Silvestrov, Hom-algebra structures. J. Gen. Lie Theory Appl. Vol. 2 (2008), No. 2, 51-64. [7] Y. Sheng and Z. Xiong, On Hom-Lie algebras. Linear and Multilinear Algebra 12 (63) (2014), 2379-2395. [8] L. Cai and Y. Sheng, Hom-big brackets: theories and applications. SIGMA 14(2016), 18pp. [9] L. Cai, J. Liu and Y. Sheng, Hom-Lie algebroids, Hom-Lie bialgebroids and Hom-Courant algebroids. J. Geom. Phys. 121 (2017), 15-32. 9 [10] C. Laurent-Gengoux and J. Teles, Hom-Lie algebroids, J. Geom. Phys. 68 (2013), 69-75. [11] L. Cai, Y. Sheng, Purely Hom-Lie bialgebras. Sci. China Math., 61 (9) (2018), 1553-1566. [12] Z. Xiong, A complement on representations of Hom-Lie algebras. ADVANCES IN MATHE- MATICS(CHINA), in press, arXiv:1703.01400 [13] C. Hall Brian, Lie Groups, Lie Algebras, and Representations: An Elementary Introduction. Springer-Verlag, 2003. [14] Z. Xiong, Equivalent description of Hom-Lie algebroids. Adv. Math. Phy., 2018, 8pp. [15] Y. Kodama, K. McLaughlin, Explicit integration of the full symmetric Toda hierarchy and the sorting property. Lett. Math. Phys., 1996, 37, 37-47. [16] Y. Kodama, J. Ye, Iso-Spectral deformations of general matrix and their reductions on Lie algebras. Commun. Math. Phys., 1996, 178, 765-788. 10
1708.09237	1	1708	2017-08-30T12:32:06	Bordering for spectrally arbitrary sign patterns	[ "math.RA" ]	We develop a matrix bordering technique that can be applied to an irreducible spectrally arbitrary sign pattern to construct a higher order spectrally arbitrary sign pattern. This technique generalizes a recently developed triangle extension method. We describe recursive constructions of spectrally arbitrary patterns using our bordering technique, and show that a slight variation of this technique can be used to construct inertially arbitrary sign patterns.	math.RA	math	Bordering for spectrally arbitrary sign patterns 7 1 0 2 D.D. Oleskya, P. van den Driesscheb, K. N. Vander Meulenc,∗ aDepartment of Computer Science, University of Victoria, BC, Canada, V8W 2Y2 bDepartment of Mathematics and Statistics, University of Victoria, BC, Canada, V8W 2Y2 cDepartment of Mathematics, Redeemer University College, ON, Canada, L9K 1J4 g u A 0 3 ] . A R h t a m [ 1 v 7 3 2 9 0 . 8 0 7 1 : v i X r a Abstract We develop a matrix bordering technique that can be applied to an irreducible spectrally arbitrary sign pattern to construct a higher order spectrally arbitrary sign pattern. This technique generalizes a recently developed triangle extension method. We describe recursive constructions of spectrally arbitrary patterns using our bordering technique, and show that a slight variation of this technique can be used to construct inertially arbitrary sign patterns. Keywords: nilpotent matrix, spectrally arbitrary pattern, nilpotent-Jacobian method, inertially arbitrary pattern. 2010 Mathematics Subject Classification. 15A18, 15B35. 1. Introduction A number of methods have been developed to check that a specific pattern is spectrally or inertially arbitrary, such as the analytic nilpotent-Jacobian method and the algebraic nilpotent-centralizer method (see e.g. [4, 7, 8, 9]), and these have been applied to various classes of patterns (see e.g. [1, 5, 13]). Recently in [11], a digraph method called triangle extension has been developed for constructing higher order spectrally or inertially arbitrary patterns from lower order patterns. In this paper, we generalize the triangle extension method by formulating it as a matrix bordering technique (see Remark 2.2). With this bor- dering technique, we construct higher order patterns (some of which cannot be obtained by triangle extension) that are spectrally or inertially arbitrary from lower order patterns. We give examples of new spectrally and inertially arbitrary sign patterns obtained by bordering. ∗Corresponding author Email addresses: [email protected] (D.D. Olesky), [email protected] (P. van den Driessche), [email protected] (K. N. Vander Meulen) c(cid:13) 2017. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/ Preprint submitted to Elsevier August 3, 2017 1.1. Definitions and the nilpotent Jacobian method. Given an order n matrix A = [aij], denote the characteristic polynomial of A by pA(z) = det(zI − A). A sign pattern is a matrix A = [αij] of order n with entries in {0, +, −}. Let Q(A) = {A aij = 0 if αij = 0, aij > 0 if αij = + and aij < 0 if αij = −}. If A ∈ Q(A) for some pattern A, then A is a realization of A and we sometimes refer to A as sgn(A). A pattern A is spectrally arbitrary if for every degree n monic polynomial p(z) over R, there is some real matrix A such that A ∈ Q(A) and pA(z) = p(z). A pattern B = [βij] is a superpattern of A if αij 6= 0 implies βij = αij, and A is a subpattern of B. Two patterns A and B are equivalent if B can be obtained from A via any combination of negation, transposition, permutation similarity and signature similarity. A matrix A is nilpotent if Ak = 0 for some positive integer k and the smallest posi- tive integer k such that Ak = 0 is the index of A. An order n nilpotent matrix A has characteristic polynomial pA(z) = zn. Suppose A is an order n sign pattern with a nilpotent matrix A ∈ Q(A) with m ≥ n nonzero entries ai1j1, ai2j2 , . . . , aimjm. Let X = XA(x1, x2, . . . , xm) denote the matrix obtained from A by replacing aikjk with the variable xk for k = 1, . . . , m. Writing pX(z) = zn + f1zn−1 + · · · + fn−1z + fn for some fi = fi(x1, x2, . . . , xm), let J = JX be the n × m Jacobian matrix with (i, j) entry equal to ∂fi for 1 ≤ i ≤ n, and 1 ≤ j ≤ m. Let ∂xj JX=A denote the Jacobian matrix evaluated at the nilpotent realization, that is JX=A = J(x1,x2,...,xm)=(ai1 j1 ,ai2j2 ,...,aimjm ). A nilpotent matrix A allows a full-rank Jacobian if the rank of JX=A is n. Finding a nilpotent matrix A ∈ Q(A) that allows a full-rank Jacobian is known as the nilpotent-Jacobian method. As noted in part (c) of Theorem 1.1, this method guarantees that every superpattern of A is spectrally arbitrary. A matrix A (or pattern A) is reducible if there is a permutation matrix P such that P AP T (resp. P AP T ) is block triangular with more than one nonempty diagonal block. Otherwise it is irreducible. A matrix A is nonderogatory if the dimension of the eigenspace of every eigenvalue is equal to one. The following theorem combines known results from [4] and [7]. Theorem 1.1. Let A be a sign pattern of order n. If a nilpotent matrix A ∈ Q(A) allows a full-rank Jacobian, then (a) A is irreducible, (b) A is nonderogatory, and (c) every superpattern of A is spectrally arbitrary. Proof. Suppose A ∈ Q(A) is a nilpotent matrix of order n that allows a full-rank Jaco- bian. Part (c) is [4, Theorem 3.1], which is a reframing of the nilpotent-Jacobian method introduced in [7]. Part (b) is [4, Corollary 4.5]. If A is a reducible nilpotent matrix and P AP T is block triangular for some permutation matrix P , then the index of A is at most the index of the largest order diagonal block of 2 P AP T . Thus the index is bounded above by the order of the largest diagonal block. Since the index of A is n, it follows that A is irreducible, proving part (a). Because of part (c) of Theorem 1.1, minimal spectrally arbitrary patterns (that is, spectrally arbitrary patterns for which no proper subpattern is spectrally arbitrary), are of special interest. For n = 2 and n = 3, the minimal spectrally arbitrary patterns are well-known (see, e.g., [1, 5]) and, up to equivalence, are: T2 =(cid:20) + − + − (cid:21) , , V3 =    + − + + − 0 + 0 − , + − 0 + 0 − 0 + − T3 =    , and W3 =    + − 0 + 0 − + 0 − + + − + 0 − + 0 − .   U3 =  1.2. Bordering Let A = [aij] be an order n matrix, x, z ∈ Rn, and let B be the bordered matrix of order n + 1: B =(cid:20) In 0 1 (cid:21)(cid:20) A z xT 0 (cid:21)(cid:20) In −xT 0T 0 1 (cid:21) =(cid:20) A − zxT xT (A − zxT ) xT z (cid:21) . z (1) Since this is a similarity transformation, it follows that pB(z) = zpA(z), and thus B is nilpotent if A is nilpotent. Note that (1) is a special case of a construction introduced in [10, Theorem 3.1]. Let ei = [0, . . . , 0, 1, 0, . . . , 0]T with a 1 in position i. In this paper, we focus on the special cases z = ej and xT = bek for some b 6= 0, which we call standard unit bordering. In the next two sections we use bordering to construct higher order spectrally arbitrary patterns out of lower order patterns without having to recalculate a Jacobian matrix. In addition, at each stage, the construction provides an explicit nilpotent realization of the spectrally arbitrary pattern. 2. Standard unit bordering with equal indices Let A = [aij], and denote the kth row of A by rk(A). Suppose x = akkz = akkek for some akk 6= 0. Then xT A = akkrk(A) and A − zxT = A − akkPkk where Pkk has a 1 in entry (k, k) and zeros elsewhere. In this case, the matrix B in (1) is B =(cid:20) A − akkPkk akkrk(A − akkPkk) akk (cid:21) . ek (2) Let A(u, v) denote the matrix obtained from A by deleting row u and column v. Theorem 2.1. Let A be a sign pattern of order n. Suppose A = [aij] ∈ Q(A) is a nilpotent matrix and A allows a full-rank Jacobian. Suppose akk 6= 0 and akv 6= 0 for some v 6= k. If det A(k, v) 6= 0, then B in (2) is a nilpotent matrix that allows a full-rank Jacobian and hence every superpattern of B = sgn(B) is spectrally arbitrary. 3 Proof. Let A ∈ Q(A) be a nilpotent matrix and XA be a matrix with the nonzero pattern of A having variable entries such that the Jacobian JXA=A has rank n. For convenience, assume k = n and the last row of XA is [xn1, xn2, . . . , xnn], recognizing that some of these entries may be zero. Note that by assumption xnv and xnn are nonzero. Let B be as in (2) and (3) XB =  XA − xnnPnn y 0 1 0 xnn   with y = [y1, y2, . . . , yn−1] such that yi 6= 0 if and only if xni 6= 0. (Note that, other than the placement of the variables in y, the nonzero entries of y are independent of the variables in XA.) Then XB has the nonzero pattern of B. Using cofactor expansion along the last row of XB gives pXB (z) = det(zIn+1 − XB) = (z − xnn) det(zIn − XA + xnnPnn) + n−1 (−1)n+ℓyℓ det ([zIn − XA](n, ℓ)) . Xℓ=1 However, applying cofactor expansion along the last row of the first summand gives det(zIn − XA + xnnPnn) = z det ([zIn − XA](n, n)) n−1 (−1)n+ℓxnℓ det ([zIn − XA](n, ℓ)) . + Xℓ=1 Thus pXB (z) = z det(zIn − XA + xnnPnn) − xnnz det([zIn − XA](n, n)) n−1 (−1)n+ℓ(yℓ − xnnxnℓ) det ([zIn − XA](n, ℓ)) . + Xℓ=1 Since the determinant is linear in the rows (or using a rank 1 perturbation of a determinant), it follows that n−1 pXB (z) = zpXA(z) + Xℓ=1 (−1)n+ℓ(yℓ − xnnxnℓ) det ([zIn − XA](n, ℓ)) . (4) Focusing on the coefficients of pXB (z), the second summand can be rewritten as n+1 Xr=3"n−1 Xℓ=1 Sr,ℓ(yℓ − xnnxnℓ)# zn−r+1 for some polynomials Sr,ℓ of the variable entries in XA. To consider the Jacobian of XB, we assume the last columns of JXB are indexed by the nonzeros of xn1, . . . , xnn, y1, . . . , yn−1. 4 Let m be the number of nonzero entries of y and w be the number of variables in XA. Then the (n + 1) × (w + m) Jacobian matrix JXB is JXB =(cid:20) JXA O 0T (cid:21) + 0T n−1 Xℓ=1 (yℓ − xnnxnℓ)Mℓ +(cid:2) O N (cid:3) (5) for some matrices Mℓ and (n + 1) × (2m + 1) matrix N with columns indexed by the nonzeros of xn1, . . . , xnn, y1, . . . , yn−1. Note that by (1) and (3), yℓ = annanℓ = xnnxnℓ in the nilpotent realization, so that we can ignore each matrix Mℓ in (5), since its coefficient vanishes at the nilpotent realization. Further the column of N corresponding to yℓ is −→ N yℓ = [0, 0, S3,ℓ, S4,ℓ, . . . , Sn+1,ℓ]T for 1 ≤ ℓ ≤ n, and in addition, the column corresponding −→ to xnℓ is N yℓ. It follows that N is column equivalent to [ O −→ N yℓ for 1 ≤ ℓ ≤ n − 1 and −→ N yn−1]. From (4), with z = 0, −→ N xnℓ = −xnn −→ N y2 · · · ℓ=1 −xnℓ −→ N xnn = Pn−1 −→ N y1 giving Sn+1,ℓ = (−1)ℓ−1 det(XA(n, ℓ)), Sn+1,ℓXB =B = (−1)ℓ−1 det(A(n, ℓ)). Thus, the condition that there exists an index v 6= n such that anv 6= 0 and det A(n, v) 6= 0 implies that Sn+1,ℓXB =B 6= 0 for some ℓ, 1 ≤ ℓ ≤ n − 1. It follows that JXB =B is equivalent to (cid:20) JXA=A 0T ∗ sT (cid:21) for some s 6= 0. Hence B allows a full-rank Jacobian. Thus by Theorem 1.1, every super- pattern of B is spectrally arbitrary. Remark 2.2. A method in [11] called triangle extension on arc (u, v) (in the digraph associated with A) is equivalent to a special case of applying Theorem 2.1 to row u of A and entry (u, v), namely in the situation that auu and auv are the only nonzero entries in row u of A. Example 2.3. If then A is nilpotent, A ∈ Q(A), and A allows a full-rank Jacobian. Hence A is spectrally arbitrary (A is equivalent to the second matrix in Appendix A of [6]). Further, a44 6= 0, A =  0 + 0 0 0 − + 0 + 0 0 + + 0 − +   0 0 0 1 1 0 0 −1 0 1 1 0 0 −1 1 1 A =  ,   and 5 a41 6= 0 and det(A(4, 1)) 6= 0. Applying Theorem 2.1 to row 4 and entry (4, 1) gives a spectrally arbitrary pattern B5 with nilpotent matrix B ∈ Q(B5) for B5 =  0 0 0 + 0 0 0 − + 0 + 0 0 + 0 + 0 − 0 + + 0 − 0 +   and 0 0 0 1 0 1 0 0 0 −1 0 1 0 1 0 0 −1 0 1 1 1 0 −1 0 1 B =  .   Note that since row 4 of A has more than one off-diagonal entry, triangle extension as de- scribed [11] is not possible on the arc (4, 1) in the digraph associated with A, demonstrating that Theorem 2.1 provides a more general technique than triangle extension in [11]. Remark 2.4. Theorem 2.1 can be applied recursively. In particular, suppose det(A(n, v)) 6= 0 and Theorem 2.1 was applied to row n and entry (n, v) of A to obtain B. It follows that det(B(n + 1, v)) = (−1)n det(A(n, v)) 6= 0 since there is only one nonzero entry in the last column of B(n + 1, v), namely 1 in the last row. Thus Theorem 2.1 can now be applied to row n + 1 and entry (n + 1, v) of B. Example 2.5. The sign pattern B5 in Example 2.3 can be recursively bordered using The- orem 2.1, starting with row 5 and entry (5, 1), to obtain a spectrally arbitrary pattern of order n ≥ 6, with 3n − 4 nonzero entries, of the form Bn =   0 + 0 + 0 − + + 0 + 0 − 0 + ... ... . . . · · · + 0 − 0 + 0 − 0 · · · ... ... O . . . 0 + 0 + .   Note that each nonzero entry of the nilpotent realization of Bn has magnitude 1. It can be shown that B5 and B4 = A in Example 2.3 are minimally spectrally arbitrary. 3. Standard unit bordering with unequal indices Referring to (1), suppose x = bek for some b 6= 0 and z = ej for j 6= k; thus xT z = 0. With the kth row of A denoted by rk(A), xT A = brk(A) and A − zxT = A − bPjk where Pjk has a 1 in entry (j, k) and zeros elsewhere. In this case, the matrix B in (1) is B =(cid:20) A − bPjk brk(A − bPjk) ej 0 (cid:21) . (6) Recall that XA is obtained from A be replacing some of the nonzero entries with vari- ables. In the case that JX=A has rank n, we call a nonzero entry of A Jacobian in XA if 6 it is replaced by a variable in XA, otherwise the entry is non-Jacobian in XA. Note that a non-Jacobian entry may be zero. To simplify the next proof, for U, V ⊆ {1, 2, . . . , n}, let A(U, V ) denote the matrix obtained from A by deleting the rows in U and the columns in V . Theorem 3.1. Let A be a sign pattern of order n. Suppose A = [aij] ∈ Q(A) is a nilpotent matrix and A allows a full-rank Jacobian. Suppose ajk, j 6= k, is non-Jacobian for some choice of XA. If akv 6= 0 and det A(j, v) 6= 0, for some v, then B in (6) is a nilpotent matrix that allows a full-rank Jacobian and hence every superpattern of B = sgn(B) is spectrally arbitrary. Proof. Let A ∈ Q(A) be a nilpotent matrix and XA be a matrix with the nonzero pattern of A having variable entries such that the Jacobian JXA=A has rank n with no variable placed in position (j, k). For convenience, assume that j = 1, k = n, and the last row of XA is [xn1, xn2, . . . , xnn], recognizing that some of these entries may be zero. Let B be as in (6) and XB =  XA − bP1n y 1 0 0   (7) with y = [y1, y2, . . . , yn] such that yi 6= 0 if and only if xni 6= 0. (Note that, other than the placement of the variables in y, the nonzero entries of y are independent of the variables in XA.) Then XB has the nonzero pattern of B. Using cofactor expansion along the last row of XB gives pXB (z) = det(zIn+1 − XB) n = z det(zIn − XA + bP1n) + (−1)ℓyℓ det ([zIn − XA](1, ℓ)) Xℓ=1 = zpXA(z) + (−1)n+1zb det ([zIn − XA](1, n)) (8) n + (−1)ℓyℓ det ([zIn − XA](1, ℓ)) . Xℓ=1 Let Wℓ = z det ([zIn − XA]({1, n}, {ℓ, n})). Applying cofactor expansion on the determinant in the second summand of (8) gives z det ([zI − XA](1, n)) = n−1 (−1)n+ℓxnℓWℓ. Xℓ=1 (9) Using the fact that the last row of zIn − XA is [0, · · · , 0, z] − [xn1, xn2, . . . , xnn], and that a determinant is linear in the last row, n (−1)ℓyℓ det ([zIn − XA](1, ℓ)) = Xℓ=1 n−1 n (−1)ℓyℓWℓ + Xℓ=1 (−1)ℓyℓUℓ Xℓ=1 (10) 7 for Uℓ = det(cid:18)[zIn − XA](1, ℓ) − z(cid:20) O 0 1 (cid:21)(cid:19) , 0T and Un = det([zIn − XA](1, n)). Using (9) and (10) in (8) gives with 1 ≤ ℓ ≤ n − 1, pXB (z) = zpXA(z) + n−1 Xℓ=1(cid:16)(−1)ℓ+1bxnℓWℓ + (−1)ℓyℓWℓ(cid:17) + n (−1)ℓyℓUℓ. Xℓ=1 However, using cofactor expansion along the last row of the matrix in Uℓ gives ℓ−1 Uℓ = (−1)n+ixni det ([zIn − xA]({1, n}, {i, ℓ})) (−1)n+i−1xni det ([zIn − xA]({1, n}, {ℓ, i})) . Thus n−1 pXB = zpXA + (−1)ℓ(yℓ − bxnℓ)Wℓ Xi=1 Xi=ℓ+1 + n Xℓ=1 + X1≤i<ℓ≤n (yℓxni − yixnℓ)(−1)n+i+ℓ det ([zIn − XA]({1, n}, {ℓ, i})) . Focusing on the coefficients of pXB (z), we can rewrite pXB (z) as zpXA(z)+ n+1 Xr=3"n−1 Xℓ=1 Sr,ℓ(yℓ − bxnℓ)# zn−r+1+ n+1 Xr=5   X1≤i<ℓ≤n Tr,i,ℓ(yℓxni − yixnℓ)  zn−r+1 (11) for some polynomials Sr,ℓ and Tr,i,ℓ in the variable entries of XA. Note that the variables in the last row of XA do not appear in Sr,ℓ or Tr,i,ℓ. To consider the Jacobian of XB, we assume the last columns of JXB are indexed by the nonzeros of xn1, . . . , xnn, and y1, . . . , yn. Let m be the number of nonzero entries of y and w be the number of variables in XA. Since x1n is non-Jacobian in XA, the (n + 1) × (w + m) Jacobian matrix JXB is JXB =(cid:20) JXA O 0T (cid:21) + 0T n−1 Xℓ=1 (yℓ − bxnℓ)Mℓ + X1≤i<ℓ≤n (yℓxni − yixnℓ)Hi,ℓ +(cid:2) O N (cid:3) (12) for some matrices Mℓ, Hi,ℓ and (n + 1) × (2m) matrix N with columns indexed by the nonzeros of xn1, . . . , xnn, y1, . . . , yn. Note that by (6) and (7), yℓ = banℓ = bxnℓ in the nilpotent realization, so that we can ignore each matrix Mℓ and Hi,ℓ in (12), since their coefficients vanish at the nilpotent realization. Let sℓ = [0, 0, S3,ℓ, S4,ℓ, . . . , Sn+1,ℓ]T and tiℓ = [0, 0, 0, 0, T5,i,ℓ, T6,i,ℓ, . . . , Tn+1,i,ℓ]T . By (11), the column of N corresponding to yℓ is −→ N yℓ = sℓ + xnitiℓ − ℓ−1 Xi=1 n Xi=ℓ+1 xnitiℓ, 8 and the column corresponding to xnℓ is −→ N xnℓ = −bsℓ − yitiℓ ℓ−1 n yitiℓ + Xi=ℓ+1 Xi=1 N yn =Pn−1 −→ −→ N yℓ for 1 ≤ ℓ ≤ n − 1. Further for 1 ≤ ℓ ≤ n − 1. Thus, evaluated at the nilpotent realization with yi = bani = bxni, −→ N xnℓ = −b i=1 (−yi)tin. It follows that N XB =B is column equivalent to [ O −→ N yℓ is From (8), with z = 0, the (n + 1) entry of −→ −→ N yn]. N xnn =Pn−1 i=1 xnitin with −→ −→ N y2 · · · N y1 which evaluated at XB = B is (−1)ℓ det XA(1, ℓ), (−1)ℓ det A(1, ℓ). Thus, the hypothesis that there exists an index v such that akv 6= 0 and det A(j, v) 6= 0 implies that the (n + 1) entry of −→ N yv is nonzero. It follows that JXB =B is equivalent to (cid:20) JXA=A 0T ∗ rT (cid:21) (13) for some r 6= 0. Hence B allows a full-rank Jacobian. Thus by Theorem 1.1, every super- pattern of B is spectrally arbitrary. Example 3.2. Starting with T2, the unique spectrally arbitrary pattern of order 2 up to equivalence [7], the bordering technique of Theorem 3.1 gives spectrally arbitrary patterns of order 3. In particular, consider the nilpotent matrix A =(cid:20) 1 −1 1 −1 (cid:21) ∈ Q(T2) and let XA =(cid:20) x1 −1 x2 −1 (cid:21) . Then JX=A has full rank and entry a12 is non-Jacobian in XA. Thus, the bordering tech- nique of Theorem 3.1 gives the matrix 1 −1 − b 1 0 1 0 b −1 −b   ,   providing different spectrally arbitrary patterns depending on the chosen value of b 6= 0. Taking b = 1 2 gives a sign pattern equivalent to W3 (see [1]) with a full-rank Jacobian. Taking b = − 1 2 gives a pattern equivalent to a superpattern of V3 (see [1]) with a full-rank Jacobian. Taking b = −1 gives a pattern equivalent to V3 with a full-rank Jacobian. This last option, using b = a12 maintains sparsity (i.e, it gives a minimal spectrally arbitrary pattern.) 9 Remark 3.3. With a well-chosen example, Theorem 3.1 can be applied recursively. In particular, note that in (13), the n variables of XA are used to show that B allows a full- rank Jacobian. Thus, at most one of the nonzero entries in row n + 1 of B needs to be Jacobian in XB. Further, an entry in row n + 1 that is Jacobian in XB can be chosen to be any nonzero position (n + 1, v) for which det A(j, v) is nonzero. Note that, since the last column of B has only one nonzero entry, det B(n + 1, v) = (−1)j+n det A(j, v). Thus, if there is more than one v with akv 6= 0, and det A(j, v) 6= 0, then bordering can be repeated recursively, applying it to (j, k) = (n + 1, k) in B. Example 3.4. Consider the nilpotent realization 1 −1 1 1 0 0 −1 0 −1     of the spectrally arbitrary pattern V3. By applying Theorem 3.1 with b = k = 1, j = 3 and v = 2, and repeating recursively, increasing j but keeping b = k = 1 and v = 2, a spectrally arbitrary pattern Kn is obtained for n ≥ 4, with Kn = + − + 0 − 0 0 − 0 ... ... 0 − 0 + − 0 ...   O 0 − + 0 + ... . . . · · · 0 0 · · · . . . 0 + 0 0 .   The nonzero entries in a nilpotent realization of Kn have magnitude 1. As far as we know, the spectrally arbitrary sign patterns Bn in Example 2.5 and Kn in Example 3.4 have not previously appeared in the literature. 4. General bordering for n = 3 In Theorems 2.1 and 3.1, we restricted to bordering with standard unit vectors in the place of x and z in (1). We next illustrate the more general bordering (1) with a couple of examples. Example 4.1. Starting with the nilpotent realization A of T2 given in Example 3.2, a nilpotent realization of T3 can be obtained as follows: B =  1 0 0 0 1 0 1 1 − 1 2     1 −1 0 1 −1 −1 0 0 0     0 0 1 1 0 0 1 2 −1 1   =   1 −1 1 2 0 0 0 −1 1 2 −1   ∈ Q(T3). Matrix B allows a full-rank Jacobian and hence (as is well-known [7]) every superpattern of T3 is spectrally arbitrary by Theorem 1.1. 10 Example 4.2. Starting with the nilpotent realization A of T2 given in Example 3.2, a nilpotent realization of U3 (see [1]) can be obtained as follows: 1 0 0 0 1 0 −1 2 1 B =      1 −1 1 −1 0 0 1 2 0 0 1 0 0 0 1 0 1 −2 1       =   3 2 −2 1 −1 1 2 1 2 0 0 − 1 2 .   In particular, matrix B is a nilpotent realization of U3 that allows a full-rank Jacobian and hence every superpattern of U3 is spectrally arbitrary by Theorem 1.1. As demonstrated in [1], every spectrally arbitrary sign pattern of order 3 is a superpat- tern of one of the four patterns T3, U3, V3 and W3. From Example 3.2, every superpattern of V3 and W3 is spectrally arbitrary by Theorem 3.1 using a standard unit bordering of T2. The other two order 3 patterns can be obtained by using a general bordering of T2 as demonstrated in Examples 4.1 and 4.2. Corollary 4.3. Every spectrally arbitrary sign pattern of order 3 is a superpattern of a pattern obtained from T2 by bordering as in (1). 5. Inertially arbitrary borderings We conclude by extending the main results in Sections 2 and 3 to obtain inertially arbitrary sign patterns. The inertia of a matrix A is the ordered triple i(A) = (a, b, c) for which a is the number of eigenvalues of A with positive real parts, b is the number with negative real parts, and c is the number of eigenvalues with real parts zero. The refined inertia of a matrix A is the ordered 4-tuple ri(A) = (a, b, c1, c2) for which c1 is the algebraic multiplicity of zero as an eigenvalue for A and c1 + c2 = c. Then c2 is the number of nonzero imaginary eigenvalues of A. A sign pattern A of order n is inertially arbitrary if, for every non-negative integer choice of (a, b, c) with a + b + c = n, there is some matrix A ∈ Q(A) with i(A) = (a, b, c). As with nilpotent matrices, a matrix A of order n with refined inertia (0, 0, c1, c2) allows a full-rank Jacobian if the Jacobian matrix JX=A has rank n. The next theorem combines [3, Theorem 2.13] and [4, Corollary 4.5]. Theorem 5.1. Let A be a sign pattern and A ∈ Q(A) be a matrix with ri(A) = (0, 0, c1, c2) for some c1 ≥ 2. If A allows a full-rank Jacobian, then (a) A is nonderogatory, and (b) every superpattern of A is inertially arbitrary. Note that, unlike the context of Theorem 1.1, A is not necessarily irreducible if A allows a full-rank Jacobian in Theorem 5.1. Example 5.2. Let 1 −1 0 1 −1 0 0 0 0 0 0 1 −2 0 1 −1 A =    XA =  1 x1 0 1 x2 0 0 0 0 0 0 1 x3 0 1 x4 .   and 11 This matrix A ∈ Q(T2 ⊕ T2) is a nonderogatory reducible matrix with refined inertia (0, 0, 2, 2) and JXA=A has rank 4. Therefore, by Theorem 5.1 every superpattern of + − 0 + − 0 0 0 0 0 0 + − 0 + − A =    is inertially arbitrary. Note that while it is known [2] that T2⊕T2 is spectrally arbitrary (and hence inertially arbitrary), it is not yet known if every superpattern of T2 ⊕ T2 is spectrally arbitrary. The proof of the next theorem is the same as that for Theorem 2.1 except it uses Theorem 5.1 instead of Theorem 1.1. Theorem 5.3. Let A be a sign pattern. Suppose A = [aij] ∈ Q(A) is a matrix having refined inertia (0, 0, c1, c2) with c1 ≥ 2, and A allows a full-rank Jacobian. Suppose akk 6= 0 and akv 6= 0 for some v 6= k. If det A(k, v) 6= 0, then B in (2) has refined inertia (0, 0, c1 + 1, c2), B allows a full-rank Jacobian and every superpattern of B = sgn(B) is inertially arbitrary. Example 5.4. Let A be the matrix in Example 5.2. With k = 2 and v = 1, Theorem 5.3 implies that every superpattern of B =   + − 0 + 0 0 0 0 − 0 0 0 0 + 0 + − 0 0 + − 0 0 − 0   is inertially arbitrary. Note that B is spectrally arbitrary since B is equivalent to T2 ⊕ V3, but it is not known if every superpattern of B is spectrally arbitrary. The proof of the next theorem is the same as that for Theorem 3.1 except it uses Theorem 5.1 instead of Theorem 1.1. Theorem 5.5. Let A is a sign pattern. Suppose A = [aij] ∈ Q(A) is a matrix with refined inertia (0, 0, c1, c2) for some c1 ≥ 2 and A allows a full-rank Jacobian. Suppose ajk, j 6= k, is non-Jacobian for some choice of XA. If akv 6= 0 and det A(j, v) 6= 0, for some v, then B in (6) has refined inertia (0, 0, c1 +1, c2), B allows a full-rank Jacobian, and every superpattern of B = sgn(B) is inertially arbitrary. Example 5.6. Consider the matrix 0 0 0 −1 −1 −1 1 0 0 −1 −1 0 −1 0 0 0 0 2 1 0 0 0 −1 0 −1 A =     , and XA = 12   0 −1 x1 −1 0 x3 0 −1 0 0 2 x2 0 0 0 x5 0 0 0 x4 0 −1 0 0 −1 .   Matrix A has sign pattern G5 from [12] (see also Section 5.3 of [4]), A has refined inertia (0, 0, 3, 2), JX=A has full rank and entry (1, 3) is non-Jacobian. Thus with j = 1, k = 3, v = 4, and b = −1 in Theorem 5.5, we obtain the inertially arbitrary pattern B = 0 0 + − − 0 0 0 0 + + + 0 0 − − 0 0 0 − 0 0 − 0 0 − 0 0 0 0 + + 0 0 0 0   .   Since G5 has no nilpotent realization, it follows that B has no nilpotent realization; thus B is not spectrally arbitrary. Note that, for n ≥ 2, using the sign pattern G2n+1 with matrix A2n+1 as listed in Section 5.3 of [4], then A2n+1 has refined inertia (0, 0, 2n − 1, 2), det( A(1, 4)) 6= 0, and entry (1, 3) is non-Jacobian. Thus, using Theorem 5.5 with j = 1, k = 3, v = 4, and b = −1 applied to A2n+1, we can construct an even order inertially arbitrary sign pattern with no nilpotent realization for each even order 2n + 2 ≥ 6. In [12], only odd order sign patterns were provided with these conditions. Acknowledgements. This research was initiated when the third author visited the University of Victoria with support from the Pacific Institute for Mathematical Sciences (PIMS). The research is partially supported by the authors' NSERC Discovery Grants. The authors thank an anonymous referee for comments that helped to clarify parts of this paper. References References [1] T. Britz, J.J. McDonald, D.D. Olesky, and P. van den Driessche. Minimal spectrally arbitrary sign patterns. SIAM J. Matrix Anal. Appl. 26 (2004) 257 -- 271. [2] M.S. Cavers. On reducible matrix patterns. Linear and Multilinear Algebra 58.2 (2010) 257 -- 267. [3] M.S. Cavers and S.M. Fallat. Allow problems concerning spectral properties of patterns. Electron. J. Linear Algebra 23:1 (2012) 731 -- 754. [4] M. Cavers, C. Garnett, I.-J. Kim, D.D. Olesky, P. van den Driessche, and K. Vander Meulen. Techniques for identifying inertially arbitrary patterns. Electron. J. Linear Algebra 26 (2013) 71 -- 89. [5] M.S. Cavers and K.N. Vander Meulen. Spectrally and inertially arbitrary patterns. Linear Algebra Appl. 394 (2005) 53-72. 13 [6] L. Corpuz and J.J. McDonald. Spectrally arbitrary nonzero patterns of order 4. Linear and Multilinear Algebra 55 (2007) 249 -- 273. [7] J.H. Drew, C.R. Johnson, D.D. Olesky and P. van den Driessche. Spectrally arbitrary patterns. Linear Algebra Appl. 308 (2000) 121 -- 137. [8] C. Garnett, B.L. Shader. A proof of the Tn conjecture: centralizers, Jacobians and spectrally arbitrary sign patterns. Linear Algebra Appl. 436 (2012) 4451 -- 4458. [9] C. Garnett, B.L. Shader. The nilpotent-centralizer method for spectrally arbitrary patterns. Linear Algebra Appl. 438 (2013) 3836 -- 3850. [10] I.-J. Kim, D.D. Olesky, B.L. Shader, P. van den Driessche, H. van der Holst and K.N. Vander Meulen. Generating potentially nilpotent full sign patterns. Electron. J. Linear Algebra 18 (2009) 162 -- 175. [11] I.-J. Kim, B. Shader, K.N. Vander Meulen and M. West. Spectrally arbitrary pattern extensions. Linear Algebra Appl. 517 (2017) 120 -- 128. [12] I.-J. Kim, D.D. Olesky and P. van den Driessche. Inertially arbitrary sign patterns with no nilpotent realization. Linear Algebra Appl. 421 (2007) 264 -- 283. [13] R.J. Pereira. Nilpotent matrices and spectrally arbitrary sign patterns. Electron. J. Linear Algebra 16 (2007) 232 -- 236. 14
1706.04929	2	1706	2018-02-28T17:29:52	Triple Linkage of Quadratic Pfister Forms	[ "math.RA", "math.AC" ]	Given a field $F$ of characteristic 2, we prove that if every three quadratic $n$-fold Pfister forms have a common quadratic $(n-1)$-fold Pfister factor then $I_q^{n+1} F=0$. As a result, we obtain that if every three quaternion algebras over $F$ share a common maximal subfield then $u(F)$ is either $0,2$ or $4$. We also prove that if $F$ is a nonreal field with $\operatorname{char}(F) \neq 2$ and $u(F)=4$, then every three quaternion algebras share a common maximal subfield.	math.RA	math	Triple Linkage of Quadratic Pfister Forms Department of Computer Science, Tel-Hai Academic College, Upper Galilee, 12208 Israel Adam Chapman Andrew Dolphin Department of Mathematics, Ghent University, Ghent, Belgium David B. Leep Department of Mathematics, University of Kentucky, Lexington, KY 40506 Abstract Given a field F of characteristic 2, we prove that if every three quadratic n-fold Pfister forms have a common quadratic (n − 1)-fold Pfister factor then In+1 q F = 0. As a result, we obtain that if every three quaternion algebras over F share a common maximal subfield then u(F) is either 0, 2 or 4. We also prove that if F is a nonreal field with char(F) , 2 and u(F) = 4, then every three quaternion algebras share a common maximal subfield. Keywords: Quadratic Forms, Quaternion Algebras, Linked Fields, u-Invariant 2010 MSC: 11E81 (primary); 11E04, 16K20, 11R52 (secondary) 1. Introduction We say that a set of quadratic n-fold Pfister forms is linked if there exists a quadratic or bilinear (n − 1)-fold Pfister form which is a common factor to all the forms in this set. By the natural identification of quaternion algebras with their norm forms which are quadratic 2-fold Pfister forms, a set of quaternion algebras is linked if they share a common maximal subfield. A maximal subfield K of a quaternion algebra over F is a quadratic field exten- sion of F. When char(F) = 2, K/F can be either separable or inseparable, and one can refine the definition of linkage accordingly: a set of quaternion algebras is sep- arably (inseparably) linked if they share a common separable (inseparable) quadratic field extension of F. It was observed in [Dra75] that inseparable linkage for pairs of quaternion algebras implies separable linkage, and a counter example for the converse was provided in [Lam02]. This observation was extended to pairs of Hurwitz algebras in [EV05] and pairs of cyclic p-algebras of any prime degree in [Cha15]. Email addresses: [email protected] (Adam Chapman), [email protected] (Andrew Dolphin), [email protected] (David B. Leep) 1 We extend the notion of separable and inseparable linkage of quaternion algebras to arbitrary n-fold Pfister forms in the following way: a set of quadratic n-fold Pfister forms are separably (inseparably) linked if there exists a quadratic (bilinear) (n−1)-fold Pfister form which is a common factor to all the forms in the set. It can be easily con- cluded from the fact mentioned above (as can be seen in [Fai06]) that if two quadratic n-fold Pfister forms ϕ1 and ϕ2 satisfy ϕ1 = B ⊗ π1 and ϕ2 = B ⊗ π2 for some bilinear (n − 1)-fold Pfister form B and quadratic 1-fold Pfister forms π1 and π2, then ϕ1 and ϕ2 are separably linked. It was proven in [EL73, Main Theorem] that if F is a field of char(F) , 2 and every two quaternion algebras over F are linked then the possible values u(F) can take are 0, 1, 2, 4 and 8, where u(F) is defined to be the supremum of the dimensions of nonsingular anisotropic quadratic forms over F of a finite order in WqF. For fields F of characteristic 2, it was shown in [Bae82, Theorem 3.1] that every two quaternion algebras over F share a quadratic inseparable field extension of F if and only if u(F) 6 4, and in [CD, Corollary 5.2] that if every two quaternion algebras over F share a maximal subfield then u(F) is either 0, 2, 4 or 8. In [Bec], the case of linkage of three bilinear n-fold Pfister forms in any character- istic was studied. It was shown in [Bec, Theorem 5.1] that if F is a nonreal field and every three bilinear n-fold Pfister forms are linked then In+1F = 0. When char(F) , 2, by the natural identification of quadratic forms with their underlying symmetric bilin- ear forms, this implies that if every three n-fold Pfister forms are linked then In+1 q F = 0. This was used to study the Hasse number of F. The Hasse number, denoted u(F), is the supremum of the dimensions of anisotropic totally indefinite quadratic forms over F. Note that if F is nonreal, then u(F) = u(F). It was shown in [Bec, Corollary 5.8] that if F is a field with char(F) , 2 and every three quaternion algebras over F are linked then u(F) 6 4. The question of whether the converse held, that is whether u(F) 6 4 im- plies that all triples of quaternion algebras over F are linked, appeared in a preliminary version of of [Bec]. In this paper we prove that if char(F) = 2 and every three quadratic n-fold Pfister forms in F are linked then In+1 q F = 0. We conclude that if every three quaternion algebras over F are linked then u(F) 6 4. In the last section we show that if F is a nonreal field with char(F) , 2 and u(F) = 4 then every three quaternion algebras over F are linked. Note that we shared this last result with the author of [Bec] who included a similar proof in the final version of that paper, acknowledging our contribution. 2. Bilinear and Quadratic Pfister Forms Let F be a field with char(F) = 2. We recall what we need from the algebraic theory of quadratic forms. For general reference see [EKM08, Chapters 1 and 2]. Let V be an n-dimensional F-vector space. A symmetric bilinear form on V is a map B : V × V → F satisfying B(v, w) = B(w, v), B(cv, w) = cB(v, w) and B(v + w, t) = B(v, t) + B(w, t) for all v, w, t ∈ V and c ∈ F. A symmetric bilinear form B is degenerate if there exists a vector v ∈ V \ {0} such that B(v, w) = 0 for all w ∈ V. If such a vector does not exist, we say that B is nondegenerate. Two symmetric bilinear forms B : V × V → F and B′ : W × W → F are isometric if there exists an isomorphism M : V → W such that B(v, v′) = B′(Mv, Mv′) for all v, v′ ∈ V. 2 A quadratic form over F is a map ϕ : V → F such that ϕ(av) = a2ϕ(v) for all a ∈ F and v ∈ V and the map defined by Bϕ(v, w) = ϕ(v + w) − ϕ(v) − ϕ(w) for all v, w ∈ V is a bilinear form on V. The bilinear form Bϕ is called the polar form of ϕ and is clearly symmetric. Two quadratic forms ϕ : V → F and ψ : W → F are isometric if there exists an isomorphism M : V → W such that ϕ(v) = ψ(Mv) for all v ∈ V. We are interested in the isometry classes of quadratic forms, so when we write ϕ = ψ we actually mean that they are isometric. We say that ϕ is singular if Bϕ is degenerate, and that ϕ is nonsingular if Bϕ is nondegenerate. Every nonsingular form ϕ is even dimensional and can be written as ϕ = [α1, β1] ⊥ · · · ⊥ [αn, βn] for some α1, . . . , βn ∈ F, where [α, β] denotes the two-dimensional quadratic form ψ(x, y) = αx2 + xy + βy2 and ⊥ denotes the orthogonal sum of quadratic forms. We say that a quadratic form ϕ : V → F is isotropic if there exists a vector v ∈ V \ {0} such that ϕ(v) = 0. If no such vector exists, we say that ϕ is anisotropic. The unique nonsingular two-dimensional isotropic quadratic form is = [0, 0], called "the hyperbolic plane". A hyperbolic form is an orthogonal sum of hyperbolic planes. We say that two nonsingular quadratic forms are Witt equivalent if their orthogonal sum is a hyperbolic form. We denote by hα1, . . . , αni the diagonal bilinear form given by (x, y) 7→ Pn i=1 αi xiyi. Given two symmetric bilinear forms B1 : V × V → F and B2 : W × W → F, the tensor product of B1 and B2 denoted B1 ⊗ B2 is the unique F-bilinear map B1 ⊗ B2 : (V ⊗F W) × (V ⊗F W) → F such that (B1 ⊗ B2) ((v1 ⊗ w1), (v2 ⊗ w2)) = B1(v1, v2) · B2(w1, w2) for all w1, w2 ∈ W, v1, v2 ∈ V. A bilinear n-fold Pfister form over F is a symmetric bilinear form isometric to a bilinear form h1, α1i ⊗ h1, α2i ⊗ · · · ⊗ h1, αni for some α1, α2, . . . , αn ∈ F ×. We denote such a form by hhα1, α2, . . . , αnii. By con- vention, the bilinear 0-fold Pfister form is h1i. Let B : V × V → F be a symmetric bilinear form over F and ϕ : W → F be a quadratic form over F. We may define a quadratic form B ⊗ ϕ : V ⊗F W → F determined by the rule that (B ⊗ ϕ)(v ⊗ w) = B(v, v) · ϕ(w) for all w ∈ W, v ∈ V. We call this quadratic form the tensor product of B and ϕ. A quadratic n-fold Pfister form over F is a tensor product of a bilinear (n − 1)-fold Pfister form hhα1, α2, . . . , αn−1ii and a two-dimensional quadratic form [1, β] for some β ∈ F. We denote such a form by hhα1, . . . , αn−1, β]]. Quadratic n-fold Pfister forms are isotropic if and only if they are hyperbolic (see [EKM08, (9.10)]). The set of Witt equivalence classes of nonsingular quadratic forms is an abelian group with ⊥ as the binary group operation and the class of as the zero element. We denote this group by IqF or I1 q F. This group is generated by scalar multiples of quadratic 1-fold Pfister forms. Let In q F denote the subgroup generated by scalar multi- ples of quadratic n-fold Pfister forms. A quaternion algebra over F is an F-algebra of the form [α, β)2,F = Fhx, y : x2 + x = α, y2 = β, yxy−1 = x + 1i for some α ∈ F and β ∈ F ×. Its norm form is the quadratic 2-fold Pfister form hhβ, α]]. This matching provides a 1-to-1 correspondence between 3 quaternion F-algebras and quadratic 2-fold Pfister forms over F. Therefore, the notions of separable and inseparable linkage translate naturally from quaternion algebras to 2- fold Pfister forms. We extend these notions of linkage to quadratic n-fold Pfister forms for any n > 2 in the following way: a set S = {ϕ1, . . . , ϕm} of m quadratic n-fold Pfister forms is separably linked if there exist a quadratic (n − 1)-fold Pfister form φ and bilinear 1-fold Pfister forms B1, . . . , Bm such that ϕi = Bi ⊗ φ for all i ∈ {1, . . . , m}. The set S is inseparably linked if there exists a bilinear (n − 1)-fold Pfister form B and quadratic 1-fold Pfister forms φ1, . . . , φm such that ϕi = B ⊗ φi for all i ∈ {1, . . . , m}. We make use of the following well-known results on quaternion algebras, and, in the case of (1), its reinterpretation in terms of 2-fold Pfister forms. Lemma 2.1 ([BO13, VII.1.9]). For α ∈ F and β, γ ∈ F × we have (1) [α, β)2,F (cid:27) [α2 + β, β)2,F . Further if α , 0, then [α, β)2,F (cid:27) [α, βα)2,F. (2) [α, β)2,F ⊗F [α, γ)2,F is Brauer equivalent to [α, βγ)2,F. 3. Triple Linkage for Quadratic Pfister Forms In this section we study properties of fields F with char(F) = 2 in which every three quadratic n-fold Pfister forms are linked for some given integer n > 3. The case n = 2 is somewhat different and will be handled in the next section. We make use of the following invariant defined in [CGV, Definition 4.1]: Definition 3.1. Let n be an integer > 2. Consider a set S of quadratic n-fold Pfister forms. We say S is tight if every element in the subgroup G of In q F generated by the Witt classes of the forms in S is represented by a quadratic n-fold Pfister form. For such a finite tight set, we define ΣS to be the Witt Class of the orthogonal sum of the quadratic n-fold Pfister representatives of the elements in G. By the identification of quaternion algebras with their norm forms, the notion of tightness and the associated invariant also apply for sets of quaternion algebras in the Brauer group. q F/In+1 A similar invariant was studied in [Siv14] when char(F) , 2 and S = 3. Lemma 3.2. Let n and k be integers > 2 and let S = {ϕ1, . . . , ϕk} be a tight set of k quadratic n-fold Pfister forms over a field F with char(F) = 2. 1. If S is separably linked then ΣS is the Witt class of a quadratic (n + k − 1)-fold Pfister form. More precisely, if ϕi = hhaiii ⊗ φ for each i ∈ {1, . . . , k}, then ΣS = hha1, . . . , akii ⊗ φ. 2. If S is inseparably linked then ΣS is the trivial Witt class. Proof. The first statement is a special case of [CGV, Proposition 4.3]. The second statement is an immediate result of [CGV, Corollary 3.6 and Proposition 4.2 (2)] (cid:3) Theorem 3.3. Let F be a field with char(F) = 2. Then for any integer n > 3, if every set of three quadratic n-fold Pfister forms over F is either separably or inseparably linked then In+1 q F = 0. 4 Proof. Let n be an integer > 3. Consider a quadratic (n + 1)-fold Pfister form Φ. Then Φ can be written as hha, b, cii⊗ψ where ψ is some quadratic (n−2)-fold Pfister form and a, b, c are some elements in F ×. Consider the forms ϕ1 = hha, bii ⊗ ψ, ϕ2 = hha, cii ⊗ ψ and ϕ3 = hhb, cii ⊗ ψ. Then in In q F we have q F/In+1 ϕ1 ⊥ ϕ2 = hb, ab, c, aci ⊗ ψ = hha, bcii ⊗ ψ mod In+1 q F . Similarly, and ϕ1 ⊥ ϕ3 = hhb, acii ⊗ ψ mod In+1 ϕ2 ⊥ ϕ3 = hhc, abii ⊗ ψ mod In+1 q F , q F , ϕ1 ⊥ ϕ2 ⊥ ϕ3 = hhab, acii ⊗ ψ mod In+1 q F . Therefore S = {ϕ1, ϕ2, ϕ3} is a tight triplet. A straight-forward computation then gives that ΣS is Witt equivalent to Φ. Since ϕ1, ϕ2, ϕ3 are quadratic n-fold Pfister forms, S is separably or inseparably linked. If they are separably linked then ΣS is a quadratic (n + 2)-fold Pfister form by Lemma 3.2 (1). By the Hauptstatz theorem [EKM08, 23.7], this can happen only when Φ is hyperbolic. If S is inseparably linked then ΣS is hyperbolic by Lemma 3.2, (2) and so Φ is hyperbolic as well. Consequently In+1 (cid:3) q F = 0. Corollary 3.4. For any integer n > 3, if every set of three quadratic n-fold Pfister forms over F is separably linked then every set of two quadratic n-fold Pfister forms over F is inseparably linked. Proof. By the previous theorem In+1 (F) = 0. By the assumption, every set of two quadratic n-fold Pfister forms over F is separably linked. By [CGV, Corollary 5.4], In+1 (F) = 0 implies that two quadratic n-fold Pfister forms over F are separably linked q if and only if they are inseparably linked. Hence every set of two quadratic n-fold Pfister forms over F are inseparably linked. (cid:3) q 4. Triple Linkage for Quaternion Algebras We again fix F to be a field with char(F) = 2. In this section we complete the picture for the case of quadratic 2-fold Pfister forms over F. These forms are in in 1-1 correspondence with quaternion algebras and our proofs are mainly written in terms of these algebras. The proof in this particular case is somewhat different from the case of higher-fold Pfister forms. Lemma 4.1. Let α, β, λ ∈ F with λ2 , β , 0. Then [α, β)2,F = [α + λ2αβ−1, β + λ2)2,F . Proof. Write [α, β)2,F = Fhx, y : x2 + x = α, y2 = β, yxy−1 = x + 1i. Let w = y + λ and z = x + λxy−1. Note that w2 = λ2 + β ∈ F ×, and in particular w is invertible. We have that wz − zw = y + λ = w, i.e. wzw−1 = z + 1 . Therefore [α, β)2,F = [z2 + z, w2)2,F = [α + λ2αβ−1, β + λ2)2,F , as required. (cid:3) 5 Lemma 4.2. Let α ∈ F, β, γ ∈ F × with β , α2. Then the quaternion algebras ψ = [α2 + β, γ)2,F and φ = [β + α4β−1, β + α2)2,F are separably linked and Σ{ψ,φ} is Witt equivalent to hhγ, β, α]] . Proof. Let φ′ = [α2 + β, β)2,F. Then by Lemma 4.1 (taking λ = α) we have φ′ = [α2 + β + α2(α2 + β)β−1, β + α2)2,F = [β + α4β−1, β + α2)2,F = φ . Hence ψ and φ are separably linked. Furthermore, by Lemma 2.1, (2) we have that [α2 + β, βγ)2,F is the quaternion representative of ψ ⊗F φ in the Brauer group. Hence S = {ψ, φ} is a tight pair, and by Lemma 3.2 (1) and Lemma 2.1 (1) we have ΣS = hhγ, β, α2 + β]] = hhγ, β, α2 + β]]. (cid:3) Proposition 4.3. Let α ∈ F, β, γ ∈ F × with β , α2. Then the quaternion algebras ψ = [β + α4β−1, γ)2,F, φ = [β + α4β−1, γ(α2 + β))2,F and π = [α2 + β, γ)2,F form a tight triplet and Σ{ψ,φ,π} is Witt equivalent to hhγ, β, α]]. Proof. The algebra π is evidently inseparably, and hence separably (by [Lam02]), linked to ψ and to φ using Lemma 2.1, (1). Further, ψ and φ are separably linked. By Lemma 2.1 (2), we have that ξ = [β + α4β−1, α2 + β)2,F is the quaternion repre- sentative of ψ ⊗F φ in the Brauer group. By Lemma 4.2, ξ is separably linked to π as well. Using Lemma 2.1 (2) we see that the tensor product of any two separably linked quaternion algebras is represented by a quaternion algebra in the Brauer group. We conclude that S = {ψ, φ, π} is a tight triplet of quaternion algebras. Since the pairs ψ, φ and ψ, π are separably linked, by [CGV, Lemma 11.2] we have that ΣS is Witt equivalent to Σ{ξ,π}. Finally by Lemma 4.2 we have that Σ{ξ,π} is Witt equivalent to hhγ, β, α]]. (cid:3) Theorem 4.4. If every set of three quaternion algebras over F is linked, then I3 q F = 0. Proof. Let ϕ = hhγ, β, α]] be an arbitrary quadratic 3-fold Pfister form. If β = α2 then ϕ is hyperbolic. Otherwise, by Proposition 4.3, ϕ is Witt equivalent to ΣS for some linked set S of three quaternion algebras. By the assumption, these three quaternion algebras share a common maximal subfield. If this subfield is a separable field extension of F then by Lemma 3.2, (1) we have that ΣG is Witt equivalent to a quadratic 4-fold Pfister form. If this subfield is an inseparable field extension of F then by Lemma 3.2, (2) we have that ΣG is hyperbolic. In both cases, ϕ must be hyperbolic, and so I3 (cid:3) q F = 0. Corollary 4.5. If F is linked and I3 three quaternion algebras over F is linked then u(F) 6 4. q (F) = 0 then u(F) 6 4. In particular, if every set of Proof. If I3 q F = 0, by [Cha17, Lemma 4.3] then there exists an anisotropic nonsingular quadratic form of dimension u(F) whose Witt class is in I2 q F. However, such a form is a direct sum of 2-fold Pfister forms (recall that when I3 q F = 0, 2-fold Pfister forms are isometric to their similar forms), and since every two quaternion algebras over F are linked, such a direct sum is Witt equivalent to one 2-fold Pfister form. Consequently, u(F) 6 4. The second statement follows from Theorem 4.4. (cid:3) 6 5. Characteristic different from 2 Let F be now a field of char(F) , 2. It is the case that u(F) = 4 implies that every three quaternion algebras over F are linked? In this section we prove that this is indeed the case if the field is nonreal, in which case u(F) = u(F). Note that this question is trivial if u(F) < 4, as then there are no division quaternion algebras over F. The main tool is the complex constructed in [Pey95] and studied further in [QMT15]. For a finite subset U of Br(F), the Brauer group of F, we let F × ·U denote the subgroup of H3(F, µ2) generated by classes λ · α with λ ∈ F × and α ∈ U. Proposition 5.1 ([QMT15, Theorem 3.13]). Let S = {Q1, Q2, Q3} be a tight set of quaternion algebras over F and U be the subgroup hQ1, Q2, Q3i of the Brauer group 2H2(F, Q/Z(2)). Further, let M be the function field of the Cartesian product 2Br(F) (cid:27) of the underlying Severi-Brauer varieties of Q1, Q2, Q3. Then the homology of the complex F × · U / H3(F, Q/Z(2)) res / H3(M, Q/Z(2)) has order 1 or 2. It is of order 1 if and only if S is linked. Proposition 5.2. Let F be a field with char(F) , 2 and I3 of three quaternion algebras over F is linked. q (F) = 0. Then every tight set Proof. Let S = {Q1, Q2, Q3} be a tight set of quaternion algebras over F, U be the subgroup of 2Br(F) generated by S and M be the function field of the Cartesian product of the Severi-Brauer varieties of Q1, Q2 and Q3. Since 2H3(F, Q/Z(2)) (cid:27) I3 q (F) = 0, the orders of elements in H3(F, Q/Z(2)) must be odd or infinite. By Proposition 5.1, the order of the homology of the complex F × · U / H3(F, Q/Z(2)) res / H3(M, Q/Z(2)) is either 1 or 2. In particular, the kernel of the restriction map cannot contain any elements of infinite order, as elements of F × · U are of finite order. Therefore the order of the homology of the above complex must be 1. By Proposition 5.1, it follows that S is linked. (cid:3) Corollary 5.3. Let F be a nonreal field with char(F) , 2 and u(F) = 4. Then every set of three quaternion algebras over F is linked. Proof. Since u(F) = 4 we have that F is linked by [EKM08, (39.1)]. Hence for all triples Q1, Q2 and Q3 of quaternion algebras over F, the set S = {Q1, Q2, Q3} is tight. That F is nonreal and u(F) = 4 implies that I3 q (F) = 0. The result therefore follows from Proposition 5.2. (cid:3) Question 5.4. Let F be a field with char(F) = 2 and u(F) = 4. Is every set of three quaternion algebras over F linked? Note that if a similar result to Proposition 5.1 can be shown for fields of character- istic 2, the same proof of Corollary 5.3 could be used for these fields to give a positive answer to Question 5.4. 7 / / / / Acknowledgements We thank the referee for useful suggestions that improved the clarity of the paper. The second author was supported by Automorphism groups of locally finite trees (G011012) with the Research Foundation, Flanders, Belgium (F.W.O. Vlaanderen). Bibliography References [Bae82] R. Baeza, Comparing u-invariants of fields of characteristic 2, Bol. Soc. Brasil. Mat. 13 (1982), no. 1, 105 -- 114. [Bec] K. J. Becher, Triple linkage, Ann. K-theory, to appear. [BO13] G. Berhuy and F. Oggier, An introduction to central simple algebras and their applications to wireless communications, Mathematical surveys and monographs, vol. 191, 2013. [CD] A. Chapman and A. Dolphin, Differential Forms, Linked Fields and the u- Invariant, Arch. Math. (Basel), to appear. [CGV] A. Chapman, S. Gilat, and U. Vishne, Linkage of quadratic Pfister forms, Comm. Algebra, to appear. [Cha15] A. Chapman, Common subfields of p-algebras of prime degree, Bull. Belg. Math. Soc. Simon Stevin 22 (2015), no. 4, 683 -- 686. [Cha17] A. Chapman, Symbol length of p-algebras of prime exponent, J. Algebra Appl. 16 (2017), no. 5, 1750136, 9. [Dra75] P. Draxl, Uber gemeinsame separabel-quadratische Zerfallungskorper von Quaternionenalgebren, Nachr. Akad. Wiss. Gottingen Math.-Phys. Kl. II (1975), no. 16, 251 -- 259. [EKM08] R. Elman, N. Karpenko, and A. Merkurjev, The algebraic and geomet- ric theory of quadratic forms, American Mathematical Society Colloquium Publications, vol. 56, Amer. Math. Soc., Providence, RI, 2008. [EL73] R. Elman and T. Y. Lam, Quadratic forms and the u-invariant II, Invent. Math. 21 (1973), 125 -- 137. [EV05] A. Elduque and O. Villa, A note on the linkage of Hurwitz algebras, Manuscripta Math. 117 (2005), no. 1, 105 -- 110. [Fai06] F. Faivre, Liaison des formes de Pfister et corps de fonctions de quadriques en caract´eristique 2, 2006, Thesis (Ph.D.) -- Universit´e de Franche-Comt´e. [Lam02] T. Y. Lam, On the linkage of quaternion algebras, Bull. Belg. Math. Soc. Simon Stevin 9 (2002), no. 3, 415 -- 418. 8 [Pey95] Emmanuel Peyre, Products of Severi-Brauer varieties and Galois cohomol- ogy, K-theory and algebraic geometry: connections with quadratic forms and division algebras (Santa Barbara, CA, 1992), Proc. Sympos. Pure Math., vol. 58, Amer. Math. Soc., Providence, RI, 1995, pp. 369 -- 401. [QMT15] Anne Qu´eguiner-Mathieu and Jean-Pierre Tignol, The Arason invariant of orthogonal involutions of degree 12 and 8, and quaternionic subgroups of the Brauer group, Doc. Math. (2015), no. Extra vol.: Alexander S. Merkur- jev's sixtieth birthday, 529 -- 576. [Siv14] A.S. Sivatski, Linked triples of quaternion algebras, Pacific J. Math. 268 (2014), no. 2, 465 -- 476. 9
1307.5501	1	1307	2013-07-21T07:35:56	Value Functions and Dubrovin Valuation Rings on Simple Algebras	[ "math.RA" ]	In this paper we prove relationships between two generalizations of commutative valuation theory for noncommutative central simple algebras: (1) Dubrovin valuation rings; and (2) the value functions called gauges introduced by Tignol and Wadsworth in [TW1] and [TW2]. We show that if v is a valuation on a field F with associated valuation ring V and v is defectless in a central simple F-algebra A, and C is a subring of A, then the following are equivalent: (a) C is the gauge ring of some minimal v-gauge on A, i.e., a gauge with the minimal number of simple components of C/J(C); (b) C is integral over V with C = B_1 \cap ... \cap B_xi$ where each B_i is a Dubrovin valuation ring of A with center V, and the B_i satisfy Graeter's Intersection Property. Along the way we prove the existence of minimal gauges whenever possible and we show how gauges on simple algebras are built from gauges on central simple algebras.	math.RA	math	VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Abstract. In this paper we prove relationships between two generalizations of commutative valu- ation theory for noncommutative central simple algebras: (1) Dubrovin valuation rings; and (2) the value functions called gauges introduced by Tignol and Wadsworth in [TW1] and [TW2]. We show that if v is a valuation on a field F with associated valuation ring V and v is defectless in a central simple F -algebra A, and C is a subring of A, then the following are equivalent: (a) C is the gauge ring of some minimal v-gauge on A, i.e., a gauge with the minimal number of simple components of C/J(C); (b) C is integral over V with C = B1 ∩ . . . ∩ Bξ where each Bi is a Dubrovin valuation ring of A with center V , and the Bi satisfy Grater's Intersection Property. Along the way we prove the existence of minimal gauges whenever possible and we show how gauges on simple algebras are built from gauges on central simple algebras. Introduction Valuation theory has been a very useful tool in the study of finite-dimensional division algebras, particularly in the construction of examples, such as noncrossed products and division algebras with nontrivial reduced Whitehead group SK1. (See [W2] for a survey of valuation theory on division algebras.) But there has been some difficulty in applying valuation theory in noncommutative settings because division algebras do not have many valuations and simple algebras that are not division algebras do not have valuations. This has led to efforts to find structures similar to but less restrictive than valuations that would exist more widely. One such approach was initiated by Dubrovin in [Du1] and [Du2]. By generalizing the idea of places in commutative valuation theory, he defined what are now called Dubrovin valuation rings. Such rings share many of the distinctive properties of commutative valuation rings, and it is known that for every central simple algebra A over a field F and every valuation ring V of F there is a Dubrovin valuation ring B of A with center V , and B is unique up to isomorphism. However, there is in general no valuation associated with such a B (except in the integral case, see below). The substantial theory of Dubrovin valuation rings is the topic of the book [MMU]. Another approach was initiated rather recently by Tignol and the second author in [TW1] and [TW2] by the introduction of gauges, which are a kind of value function on a semisimple algebra S Some of the results in this paper are based on the first author's doctoral dissertation written under supervi- sion of the second author and Antonio Jos´e Engler. The first author was partially supported by FAPESP, Brazil (Grant 06/00157-3) during his graduate studies. This author would like to thank the second author and UCSD for their hospitality during his visit in 2009. Some of the research for this paper was carried out during the second author's visit to the University of Campinas, Brazil, during July-August, 2010, which was made possible by a grant from FAPESP, Brazil. The second author would like thank Prof. Antonio Jos´e Engler and the first author for their hospitality during that visit. 1 2 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH finite-dimensional over a field F , but satisfying weaker axioms than for a valuation. (See §1 below for the definition of a gauge.) The theory of gauges is still developing, but it has already become a useful complement to the classical valuation theory of division algebras. A gauge α on S always extends some valuation v on F , so α is called a v-gauge. Just as the valuation v induces a filtration on F , so α induces a filtration on S yielding an associated graded ring grα(S), which is a finite-rank semisimple graded algebra over the graded field grv(F ). The associated graded algebra captures much of the essential information about α on S, but grα(S) is often much easier to work with than with S itself. Gauges are easy to construct in many cases, and they have good behavior with respect to tensor products and scalar extensions of algebras. The question naturally arises what kind of connections there may be between Dubrovin valuation rings and gauges. A limited answer was provided in [TW1]: Morandi had shown in [M2] that a Dubrovin valuation ring B of a central simple F -algebra A has a special kind of associated value function µB if and only if B is integral over its center Z(B); when this occurs, B determines µB and vice versa since B = {a ∈ A µB(a) ≥ 0}. Here, Z(B) is a valuation ring V of F , say with associated valuation v. In [TW1, Prop. 2.5] it was shown that the gauge ring Rα = {a ∈ A α(a) ≥ 0} of a v-gauge α on A is a Dubrovin valuation ring if and only if its residue ring Rα/J(Rα) is a simple ring. (J(•) denotes the Jacobson radical of •.) Moreover, when this occurs, Rα is integral over its center V and α = µRα. Furthermore, if v is defectless in A, then every Dubrovin valuation ring B of A with center V and integral over V is the gauge ring of the v-gauge µB on A. Since it is known that gauge rings Rα are always integral over their centers, and Dubrovin valuation rings B always satisfy B/J(B) is simple, this result says that Dubrovin valuation rings and gauge rings coincide "whenever possible." When V has rank (= Krull dimension) 1, then every Dubrovin valuation ring of A with center V is integral over V . But, when V has rank 2 or more, the Dubrovin valuation rings of A with center V are very often not integral over V . We show in this paper that there are still significant, and somewhat surprising, connections between the Dubrovin theory and gauges even when the Dubrovin valuation rings are not integral over their centers. We use the special intersections of Dubrovin valuation rings analyzed by Grater in [G1]. He showed that to every valuation ring V of a field F and every central simple F -algebra A there is a subring C of A, with C integral over V and determined uniquely up to isomorphism, such that C = B1 ∩ . . . ∩ Bξ, where the Bi are each Dubrovin valuation rings of A with center V , and the Bi are related by satisfying a special "Intersection Property." We dub such a C a Grater ring for V in A. Grater proved that among other nice properties C is a noncommutative B´ezout ring and that the Bi are determined from C as the localizations of C with respect to its maximal ideals. The number ξ of Bi in the intersection (= the number of maximal ideals of C) is an invariant ξ = ξV,[A] of V and the Brauer class [A] of A. Grater called this number the "extension number." The same number had appeared earlier in [W1] in the "Ostrowski Theorem" for Dubrovin valuation rings. The extension number equals 1 if and only if some (hence every) Dubrovin valuation ring of A with center V is integral over V . Let α be a v-gauge on the central simple F -algebra A. The degree zero piece of the associ- ated graded ring grα(A), denoted Aα 0 , coincides with Rα/J(Rα), and is a semisimple ring finite- v dimensional over the residue ring F of the valuation v. Let ω(α) = the number of simple components of the semisimple ring Aα 0 . We show in Th. 3.5 that ω(α) ≥ ξV,[A], where V is the valuation ring of v. We call α a minimal gauge if equality holds. We show in Th. 3.9 that if α is a minimal gauge on A then its gauge ring Rα is a VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 3 Grater ring. Conversely, we prove in Th. 3.11, that if v is defectless in A then a Grater ring of A with center V is the gauge ring of some minimal v-gauge. Defect in valuation theory refers to the failure of equality in the Fundamental Inequality. The background results we need on defect are given in §1.b. The defect is trivial for valuations of residue characteristic 0 or of prime residue characteristic not dividing the index of a central simple algebra. It is not hard to prove that if a semisimple F -algebra S has a v-gauge, then the valuation v must be defectless in S (see Prop. 1.9). In Th. 4.3 we prove the very nontrivial converse that if v is defectless in S, then S necessarily has a minimal v-gauge. This is of interest in itself, and is also essential for the proof of Th. 3.11. 0 , see Prop. 1.5. However, the semisimple F Our approach to proving results about gauges and Grater rings with respect to a central val- uation v often involves working back from corresponding objects for a coarser valuation w. A v-gauge α on a central simple F -algebra A has a coarsening to a w-gauge β on A with an induced v/w-gauge on the residue ring Aβ 0 need not be central simple. Therefore, it has been necessary to determine how gauges on semisimple algebras are related to gauges for central simple algebras. We do this in §2. Reduction from the semisimple case to the simple case is very easy. The simple case turns out to have a nice description that takes some work to prove: If S is a finite-dimensional simple F -algebra, and v is a valuation on F , let v1, . . . , vr be the extensions of v to the center Z(S). If α is a v-gauge on F , we show in Th. 2.2 that there are uniquely determined vi-gauges αi on S such that α = min(α1, . . . , αr). Moreover, in Th. 2.8 we give a necessary and sufficient compatibility condition on vi-gauges αi so that min(α1, . . . , αr) is a v-gauge. This is a notable result in its own right for the still-developing theory of gauges. w -algebra Aβ For valuations on fields, the valuation is determined (up to equivalence) by the valuation ring. This is likewise true for the Morandi value functions determined by Dubrovin valuation rings integral over their centers. We give an example in §5 to show that the corresponding property does not hold for minimal gauges. In our example there are infinitely many nonisomorphic minimal gauges on a central simple algebra all with the same gauge ring Rα. There is a book in preparation [TW3] that will give a more extensive treatment of gauges than is available in the currently published literature. We have adapted §1.b below on defect and the result in the Appendix from that book because the material is essential for this paper and there is no adequate published reference available. Also, the existence of gauges for defectless algebras was first proved in that book. What is needed here to complete the results in §3 below is the existence of minimal gauges for defectless algebras. This is proved in §4 by an argument that is substantially different from the one in [TW3]. 1. Value functions on semisimple algebras For the convenience of the reader, we recall a few basic properties of gauges proved in [TW1] and [TW2]. First, a few words of notation: If R is any ring (with 1), we write: R× for the group of units of R; Z(R) for the center of R; J(R) for the Jacobson radical of R; and Mk(R) for the k × k-matrix ring over R. Fix for now a divisible totally ordered abelian group Γ, chosen to be sufficiently large to contain the values of all valuations and the degrees of all gradings we consider. If v : F → Γ∪{∞} is a valuation on a field F , let V be the corresponding valuation ring, V = {x ∈ F v(x) ≥ 0}; then J(V ) = {x ∈ F v(x) > 0}, which is the unique maximal ideal of V . We write F for the residue field V /J(V ) and Γv for the value group v(F ×), which is a subgroup of Γ. We say that v 4 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH v and V are trivial if V = F . For basic assumed background on valuations on fields we refer to Bourbaki [B, Ch. 6] or Engler-Prestel [EP]. For valuations on division rings we use analogous notation and terminology to that for fields. A good reference for valuation theory on division rings is the paper [JW]. Let D be a division ring finite-dimensional over its center. A valuation v : D → Γ ∪ {∞} defines a filtration on D: for each γ ∈ Γ, we set D≥γ = {d ∈ D v(d) ≥ γ}, D>γ = {d ∈ D v(d) > γ}, and Dγ = D≥γ/D>γ. The associated graded ring for v on D is grv(D) = Lγ∈Γ Dγ. a division ring. Also, since Γv is torsion-free, D× =Sγ∈Γv It is a graded division ring because every nonzero homogeneous element in gr(D) is invertible. (If D is commutative, then gr(D) is also commutative and is then called a graded field.) The grade set Γgr(D) of gr(D) is {γ ∈ Γ Dγ 6= {0}}, which is a subgroup of Γ coinciding with Γv. Note that D0 is a graded right gr(D)-module, i.e., M is a right gr(D)-module with Mγ·Dδ ⊆ Mγ+δ for all γ, δ ∈ Γ. Then, M is a free gr(D)-module with well-defined rank and a homogeneous base. We therefore call M a right graded gr(D)-vector space, and write dimgr(D) M for the rank of M as a gr(D)-module. If N is another graded gr(D)-vector space, then a gr(D)-homomorphism ψ : M → N is called a graded homomorphism if ψ(Mγ) ⊆ Nγ for all γ ∈ Γ. We write M ∼=g N if there is a graded isomorphism M → N. If m ∈ M \ {0} is homogeneous, then there is a unique γ ∈ Γ with m ∈ Mγ; we call γ the degree of m, and write γ = deg m. For any δ ∈ Γ, we write M(δ) for the gr(D)-vector space obtained from M by shifting the grading by δ: Dγ \ {0}. Now, let M =Lγ∈Γ Mγ be M(δ) = Lγ∈Γ M(δ)γ where M(δ)γ = Mγ+δ. Now let M be a right D-vector space. A v-value function on M is a map α : M → Γ∪{∞} such (i) α(x) = ∞ if and only if x = 0; (iii) α(xd) = α(x) + v(d) for all x ∈ M and d ∈ D. (ii) α(x + y) ≥ min(cid:0)α(x), α(y)(cid:1) for x, y ∈ M ; We write Γα to denote the value set α(A \ {0}), which need not be a group, but it is a union of cosets of Γv. The v-value function α is called a v-norm if M is finite-dimensional and contains a D-vector space base (xi)n i=1 such that (1.1) that (1.2) α(cid:0) nPi=i xidi(cid:1) = min 1≤i≤n(cid:0)v(di) + α(xi)(cid:1) for d1, . . . , dn ∈ D. Such a base (xi)n defines a filtration on M : for each γ ∈ Γ, we set i=1 is called a splitting base of M for α. Just as with valuations, the value function α M≥γ = {x ∈ M α(x) ≥ γ}, M>γ = {x ∈ M α(x) > γ}, and Mγ = M≥γ/M>γ. (When we need to specify the value function, we write M α object ≥γ, M α >γ, and M α γ .) The associated graded grα(M ) = Lγ∈Γ Mγ VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 5 is a graded right gr(D)-vector space with dimgr(D) grα(M ) ≤ dimD M (see [RTW, Cor. 2.3]). We will write gr(M ) instead of grα(M ) when the value function α is clear. Every nonzero element i=1 of M is a splitting base of M for α if and only i=1 is a gr(D)-base of gr(M ) (see [RTW, Cor. 2.3]). Hence, α is a v-norm on M if and only x ∈ M has an imageex α in gr(M ), defined byex α = x + M>α(x) ∈ Mα(x). We will often write simply ex instead of ex α when α is clear. A D-base (xi)k if (exi)k A v-value function α on a finite-dimensional F -algebra A is surmultiplicative if α(1) = 0 and if dimgr(D) gr(M ) = dimD M . α(xy) ≥ α(x) + α(y) for x, y ∈ A. For such an α, set and (1.3) Rα = {x ∈ A α(x) ≥ 0} Jα = {x ∈ A α(x) > 0}. It is clear from the axioms for α that Rα is a subring of A and Jα is a two-sided ideal of Rα. One can check further that 1 + Jα ⊆ R× α ; hence Jα lies in the Jacobson radical J(Rα). Thus, if Rα/Jα is semisimple, then Jα = J(Rα). Also, if F = Z(A) and Γα lies in the divisible hull of Γv then Z(Rα) = Rα ∩ F = V , the valuation ring of v, and J(Rα)∩ F = J(V ). Moreover, if in addition α is a v-norm, then it follows from [TW1, Lemma 1.20 and Th. 3.1] that Rα is integral over V . Because α is surmultiplicative, the multiplication in A induces a multiplication on the graded gr(F )-vector space grα(A) such that for x, y ∈ A (1.4) if α(xy) = α(x) + α(y), if α(xy) > α(x) + α(y). exey = xy + A>α(x)+α(y) = (cid:26) fxy 0 Thus, grα(A) is a graded algebra over gr(F ). We write [grα(A): gr(F )] for dimgr(F ) grα(A). Let F be a field with a valuation v and valuation ring V . A surmultiplicative v-value function α on a finite-dimensional F -algebra A is called a v-gauge if α is a v-norm and grα(A) is a graded semisimple gr(F )-algebra, i.e., grα(A) has no nonzero homogeneous nilpotent ideal. In this case, Rα is called the gauge ring of α. If α is a v-gauge on a semisimple F -algebra A and α′ is a v-gauge on a semisimple F -algebra A′ we say that α and α′ are isomorphic v-gauges if there is an F -algebra isomorphism η : A → A′ such that α′ ◦ η = α. For finite-dimensional graded algebras over a graded field there are structure theorems analogous to the Wedderburn theorems in the ungraded context. The graded theory is developed in [HW, §1]. Example 1.1. One of the basic constructions of gauges is that of End-gauges on endomorphism rings determined by norms on vector spaces, which we now recall. Let D be a finite-dimensional division F -algebra and let M be a finite-dimensional right D-vector space. Suppose the valuation v on F extends to a valuation w on D and let α be a w-norm on M . Then there is a well-defined surmultiplicative v-value function End(α) on the endomorphism ring EndD(M ), given by (1.5) (see [TW1, Prop. 1.19]). Moreover, End(α)(f ) = min m ∈ M \{0}(cid:0)α(f (m)) − α(m)(cid:1) grEnd(α)(EndD(M )) ∼=g Endgrw(D)(cid:0) grα(M )(cid:1). (The endomorphism ring Endgrw(D)(cid:0) grα(M )(cid:1) has a natural grading in which a map g : grα(M ) → grα(M ) has degree γ if g(Mδ) ⊆ Mγ+δ for all δ ∈ Γ.) Thus grEnd(α)(EndD(M )) is graded simple, i.e., it has no proper nonzero homogeneous ideals. It follows by dimension count that End(α) is a v-gauge if and only if w on D is defectless over v, i.e., [grw(D): grv(F )] = [D:F ] (see [TW1, Prop. 1.19]). Since dimD M < ∞ the grade set Γα = α(M \ {0}) consists of finitely many cosets of Γw, say 6 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Γα = Fk decomposition i=1 γi + Γw, a disjoint union. Then grα(M ) has a canonical graded gr(D)-vector space grα(M ) = Mi, kLi=1 where Mi = Lδ ∈ γi+Γw Mδ ∼=g gr(D) ⊗D0 Mγi. An endomorphism of grα(M ) of degree 0 sends each Mδ to itself. Thus, the degree-0 component of grEnd(α)(EndD(M )) is (1.6) EndD(M )0 ∼= (cid:0) Endgr(D)(gr(M ))(cid:1)0 ∼= kQi=1(cid:0) Endgr(D)(Mi)(cid:1)0 ∼= kQi=1 Thus,(cid:0) Endgr(D)(gr(M ))(cid:1)0 is semisimple with exactly k simple components, each Brauer equivalent to the division ring D0. Note that the number k of simple components equals the number of cosets of Γw in Γα. EndD0(Mγi). The graded ring Endgr(D)(gr(M )) can be viewed as a matrix ring with shifted grading as fol- lows: Let (m1, . . . , mn) be any splitting base of M with respect to α, and let γi = α(mi), (1.7) i=1 D(γi). Then, as graded gr(F )-algebras, i=1fmiD ∼=gLn for each i. Then (fm1, . . . ,fmn) is a homogeneous base of the gr(D)-vector space gr(M ) with degfmi = γi. Let D = gr(D). Each fmi spans a 1-dimensional graded D-subspace of M, and fmiD ∼=g D(γi), which is D with its grading shifted by γi, as in (1.1). Thus, as graded D-vector spaces, gr(M ) =Ln Endgr(D)(gr(M )) ∼=g EndD(cid:0)D(γ1) ⊕ . . . ⊕ D(γn)(cid:1) ∼=g Mn(D)(γ1, . . . , γn), where Mn(D)(γ1, . . . , γn) is the matrix ring Mn(D) but with grading shifted so that its δ-component cosists of the matrices with each ij-entry Dγj −γi+δ, for all δ ∈ Γ. Since Dγj−γi+δ 6= {0} if and only if δ ∈ γi − γj + Γw, we have (1.8) ΓEnd(α) = ΓEndgr(D)(gr(M )) = γi − γj + Γw. nSi=1 nSj=1 We will need in §4 the following more general construction of End-gauges: Lemma 1.2. Let A be a semisimple F -algebra with a v-gauge α. Let M be a free right A-module with base (m1, . . . , mn). Take any γ1, . . . , γn ∈ Γ, and let η : M → Γ ∪ {∞} be the "α-v-norm" defined by η(cid:0) nPi=1 miai(cid:1) = min 1≤i≤n(cid:0)γi + α(ai)(cid:1), for all a1, . . . , an ∈ A. and grψ(E) ∼=g Endgrα(A)(grη(M )). and ψ is a v-gauge on E. Moreover, grη(M ) is a free right grα(A)-module with base (em1, . . . ,emn), (1.9) Then, (1.10) Let E = EndA(M ) and let ψ be the v-value function End(η) on E defined by ψ(f ) = min 1≤i≤n(cid:0)η(f (mi)) − η(mi)(cid:1). m∈M \{0}(cid:0)η(f (m)) − η(m)(cid:1), ψ(f ) = min VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 7 Proof. Note that for all m, m′ ∈ M and a ∈ A, η(ma) ≥ η(m) + α(a) Now, take any f ∈ E and nonzero m ∈ M , say m = and η(m + m′) ≥ min(η(m), η(m′)). miai with ai ∈ A. Then, nPi=1 η(f (m)) − η(m) = η(cid:0) nPi=1 = min 1≤i≤n(cid:0)η(f (mi)) + α(ai)(cid:1) − η(m) f (mi)ai(cid:1) − η(m) ≥ min 1≤i≤n(cid:2)(cid:0)η(f (mi)) − η(mi)(cid:1) +(cid:0)η(mi) + α(ai)(cid:1)(cid:3) − η(m) 1≤i≤n(cid:0)η(mi) + α(ai)(cid:1) − η(m) 1≤i≤n(cid:0)η(f (mi)) − η(mi)(cid:1) + min 1≤i≤n(cid:0)η(f (mi)) − η(mi)(cid:1) = ψ(f ). ≥ min = min Thus, η(f (m)) − η(m) ≥ ψ(f ). Hence, ψ(f ) = min 1≤i≤n(cid:0)η(f (mi)) − η(mi)(cid:1) ≥ min m ∈M \{0}(cid:0)η(f (m)) − η(m)(cid:1) ≥ ψ(f ), so equality holds throughout, proving (1.10). Note that for the identity map idM , clearly ψ(idM ) = 0. Also, let f, g ∈ E. Then, ψ(f ◦ g) = = min min m ∈M \{0}(cid:0)η(f (g(m))) − η(m)(cid:1) m ∈M \{0}(cid:2)(cid:0)η(f (g(m))) − η(g(m))(cid:1) +(cid:0)η(g(m)) − η(m)(cid:1)(cid:3) m ∈M \{0}(cid:0)η(f (g(m))) − η(g(m))(cid:1) + min m ∈M \{0}(cid:0)η(g(m)) − η(m)(cid:1) min ≥ ≥ ψ(f ) + ψ(g). Thus, ψ is surmultiplicative. Now consider the graded structures. Note that as α is a v-norm, η is a v-value function on M . So, grη(M ) is a graded grv(F )-vector space. Formula (1.9) shows η(ma) ≥ η(m) + α(a) for all m ∈ M and a ∈ A. Hence, the right A-module structure of M induces a well-defined grα(A)-module action on grη(M ) such that if η(ma) = η(m) + α(a); if η(ma) > η(m) + α(a). j=1 be a splitting base of A for α. By formula (1.9), (miaj)n,r i,j=1 is emea = ma + M>η(m)+α(a) = (cid:26) fma rLj=1fmi eaj grv(F ) = rLj=1 In particular,fmiea = gmia. Let (aj)r nLi=1 a splitting base of M for η. Hence, grη(M ) = 0 nLi=1 ]miaj grv(F ) = nLi=1fmi(cid:0) rLj=1eaj grv(F )(cid:1) = Therefore, grη(M ) is a free grα(A)-module of rank n with base (fm1, . . . ,fmn). map bf : grη(M ) → grη(M ) defined on homogeneous elements by Take any nonzero f ∈ E. Formula (1.10) above shows that η(f (m)) ≥ ψ(f ) + η(m), for all m ∈ M , with equality for some m ∈ M \ {0}. From this one can see that f induces a well-defined nLi=1fmi grα(A). if η(f (m)) = ψ(f ) + η(m); if η(f (m)) > ψ(f ) + η(m). bf (em) = ( ]f (m) 0 8 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Routine calculations show that bf is a grv(F )-vector space endomorphism of grη(M ) which maps each Mγ into Mγ+ψ(f ). The definition of ψ shows that bf (fmi) 6= 0 for some i; so bf 6= 0. Moreover, for any m ∈ M and a ∈ A, it is easy to check that if η(f (ma)) = ψ(f ) + η(m) + α(a) if η(f (ma)) > ψ(f ) + η(m) + α(a) = bf (em)ea, from which it follows that bf ∈ Endgrα(A)(grη(M )) and bf is homogeneous of degree ψ(f ). Moreover, for nonzero g ∈ E and any m ∈ M, if η(g(f (m))) = ψ(g) + ψ(f ) + η(m), if η(g(f (m))) > ψ(g) + ψ(f ) + η(m), 0 0 bf (emea) = ( ^f (ma) bg ◦ bf (em) = ( ^g(f (m)) bg ◦ bf = ( [g ◦ f 0 hence, (1.11) if ψ(g ◦ f ) = ψ(g) + ψ(f ); if ψ(g ◦ f ) > ψ(g) + ψ(f ). and ι is injective since it is injective on each homogeneous component of grψ(E). Thus, there is a well-defined map ι : grψ(E) → Endgrα(A)(grη(M )) given on homogeneous elements Also, if ψ(g) > ψ(f ), then [f + g = bf . Therefore, bf depends only on the image ef of f in grψ(E). by ι(ef ) = bf for f ∈ E. Using (1.11), one can check that ι is a graded grv(F )-algebra homomorphism, andbh(fmk) = gmja = fmjea whilebh(fmi) = 0 for i 6= k. Of course,bh = ι(eh) ∈ im(ι). Because maps such as bh generate Endgrα(A)(grη(M )) as an abelian group, the map ι is surjective, so a graded For surjectivity of ι, take any j, k ∈ {1, . . . , n} and any nonzero a ∈ A. Define h ∈ E by h(mk) = mja and h(mi) = 0 for i 6= k. Then, as η(mja) = γj + α(a), we have ψ(h) = γj − γk + α(a) isomorphism. Because grα(A) is graded semisimple (as α is a v-gauge) and grη(M ) is a free grα(A)- module, Endgrα(A)(grη(M )) is graded semisimple by the graded Wedderburn theory (see [HW, §1, especially Prop. 1.3]). Hence, grψ(E) is graded semisimple. Furthermore, [grψ(E) : grv(F )] = [Endgrα(A)(grη(M )) : grv(F )] = n2[grα(A) : grv(F )] = n2[A : F ] = [E : F ]. Thus, ψ is a surmultiplicative v-norm on E with grψ(E) graded semisimple, showing that ψ is a v-gauge on E. (cid:3) The following proposition shows how to construct gauges on direct products of algebras. Proposition 1.3. Let A1, . . . , Ak be finite-dimensional simple F -algebras with respective v-gauges α1, . . . , αk, and let A = A1 × . . . × Ak. The map α : A → Γ ∪ {∞} defined by (1.12) for x1 ∈ A1, . . . , xk ∈ Ak, α(x1, . . . , xk) = min(cid:0)α1(x1), . . . , αk(xk)(cid:1), is a v-gauge on A and there exists a canonical identification of gr(F )-algebras grα(A) = grα1(A1) × . . . × grαk (Ak). Proof. It is easily checked that α is a surmultiplicative v-value function on A. For γ ∈ Γ, we have α(x1, . . . , xk) ≥ γ if and only if αi(xi) ≥ γ for i = 1, . . . , k. Hence A≥γ = A1,≥γ × . . . × Ak,≥γ. Similarly, A>γ = A1,>γ × . . . × Ak,>γ. Therefore, (1.13) grα(A) = grα1(A1) × . . . × grαk (Ak). VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 9 By counting dimensions, we have [grα(A) : gr(F )] = [grαi(Ai) : gr(F )]. kXi=1 Since [grαi(Ai) : gr(F )] = [Ai : F ] for i = 1, . . . , k, it follows that [grα(A) : gr(F )] = [A : F ]. Finally, the projections of a homogeneous nilpotent two-sided ideal of grα(A) are homogeneous nilpotent two-sided ideals of each grαi(Ai), which is trivial since grαi(Ai) is assumed semisimple. Therefore, grα(A) is also semisimple. (cid:3) Every v-gauge on A as in Prop. 1.3 is given by the formula (1.12); more precisely, if β : A → Γ ∪ {∞} is a v-gauge, then for each i = 1, . . . , k the map βi : Ai → Γ ∪ {∞} defined by (x in the i-th position) (1.14) βi(x) = β(0, . . . , 0, x, 0, . . . , 0) is a v-gauge on Ai and β(x1, . . . , xk) = min(cid:0)β1(x1), . . . , βk(xk)(cid:1), (see [TW1, Prop. 1.6] for a proof). for x1 ∈ A1, . . . , xk ∈ Ak, w 1.a. Composition of gauges. Let v : F → Γ ∪ {∞} be a valuation on a field F , where Γ is a divisible totally ordered abelian group, and let V be the valuation ring of v. A valuation v′ on F is said to be equivalent to v if its valuation ring is also V . Recall that a valuation w on F is said to be a coarsening of v if its valuation ring W contains V . When this occurs, the maximal ideal J(W ) of W is a prime ideal of V , and W is the localization VJ(W ) of V at J(W ). As is well-known, the map w 7→ J(W ) gives a one-to-one correspondence between the equivalence classes of coarsenings of v and the set of prime ideals of V . Given v and w, the ring U = V /J(W ) is a valuation ring of F with ring U is called the residue valuation determined by v and w; we sometimes denote u by v/w. Its residue field is F . From the perspective of w, the valuation v is called a refinement of w, and v is determined up to equivalence by w and u, since V = π−1 is the canonical projection; we call v the composite of w and u, and write v = u ∗ w. w (U ) where πw : W → F = W/J(W ), and a valuation u on F We now look at coarsenings from the perspective of the valuation v : F → Γ∪{∞}. Let ∆ ⊆ Γ be any convex subgroup, i.e., a subgroup of Γ satisfying for all γ ∈ Γ, δ ∈ ∆, if 0 ≤ γ ≤ δ then γ ∈ ∆. So, ∆ is a divisible group. Let Λ = Γ/∆, which is a divisible totally ordered abelian group under the ordering induced from Γ. Let ε : Γ → Λ be the canonical map, which we extend to Γ ∪ {∞} by setting ε(∞) = ∞. By composing v with ε, we obtain a valuation w on F w w u v = F w w = ε ◦ v : F −→ Λ ∪ {∞}, which is a coarsening of v. Let W be the valuation ring of w. The residue valuation u = v/w is w u : F −→ ∆ ∪ {∞} given by u(c + J(W )) = (v(c) ∞ if w(c) = 0, if w(c) > 0, for all c ∈ W. Note that all coarsenings of v (up to equivalence) are obtainable this way: If w′ is a coarsening of v, let ∆0 = {v(c) w(c) = 0}, and let ∆′ be the convex hull of ∆0 in Γ. Then w′ is equivalent to the coarsening of v determined by ∆′. Let (1.15) H(Γv) = Γv ⊗Z Q, 10 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH which is the divisible hull of Γv. Recall that the ordering on Γv extends uniquely to H(Γv), making the latter into a divisible totally ordered abelian group. Moreover, we may view H(Γv) as a subgroup of Γ, since there is a unique monomorphism H(Γv) → Γ extending the inclusion Γv ֒→ Γ. Since valuations extending v on algebraic extensions of F all have value groups lying in H(Γv), there is generally no loss for us to assume Γ = H(Γv). Recall that the convex subgroups of H(Γv) are in one-to-one correspondence with the convex subgroups of Γv, which are in one-to-one correspondence with prime ideals of V , which are in turn in one-to-one correspondence with equivalence classes of coarsenings of v. Now let A be a finite-dimensional F -algebra and let α : A → Γ ∪ {∞} be a surmultiplicative v- value function. With ∆, Λ, ε, and w as above, the composition of α with ε yields a surmultiplicative w-value function If α is a v-gauge, then β is a w-gauge, by [TW2, Prop. 4.3]. In this case, β is called a coarsening of α. If Γα ⊆ H(Γv), then w = βF determines ∆ ∩ H(Γv), which determines β; we then call β the w-coarsening of α. β = ε ◦ α : A −→ Λ ∪ {∞}. Proposition 1.4. Let α be a v-gauge on a central simple F -algebra A such that Γα lies in the divisible hull H(Γv) of Γv. Let w be any valuation on F which is a coarsening of v, and let W be the valuation ring of w. Let β be the w-coarsening of α. Then the gauge ring Rβ is a central localization of Rα by P = J(W ), that is, Rβ = Rα·VP . Proof. For each x ∈ A, if α(x) ≥ 0, then β(x) = ε(α(x)) ≥ 0. Thus, Rα ⊆ Rβ. Since W = VP ⊆ Rβ, we have Rα·VP ⊆ Rβ. For the reverse inclusion, let b ∈ Rβ. Since β(x) > 0 implies α(x) > 0, In this case, α(x) ∈ ∆. Since Γα ⊆ H(Γv), there we only have to consider the case β(x) = 0. exists a positive integer n such that nα(x) ∈ Γv. Thus, nα(x) ∈ Γv ∩ ∆. Let c ∈ F such that v(c) = nα(x). Hence, we have w(c) = ε(v(c)) = 0. Since −nα(x) ≤ α(x) ≤ nα(x), it follows that α(cx) = v(c) + α(x) ≥ 0. Therefore, x = (xc)c−1 ∈ Rα·VP . (cid:3) For the coarsening β of the v-gauge α on A as above, let Aβ 0 be the degree zero part of grβ(A), which is a finite-dimensional semisimple F -algebra. Thus, w Aβ 0 = Aβ ≥0(cid:14) Aβ >0 = {x ∈ A α(x) ∈ ∆ or α(x) > δ for all δ ∈ ∆}(cid:14) {x ∈ A α(x) > δ for all δ ∈ ∆}. For u = v/w, we can define a u-value function on Aβ 0 : which is easily checked to be a graded ring isomorphism. Thus, we may view grα0(Aβ 0 ) as a graded subring of grα(A) =Lγ∈Γ Aα γ . This is well-defined by [TW2, Lemma 4.1]. Note that Γα0 = ∆ ∩ Γv. For any δ ∈ ∆ we have (1.16) α0 : Aβ 0 −→ ∆ ∪ {∞} by x + Aβ (Aβ 0 )δ = (Aβ 0 )≥δ(cid:14) (Aβ 0 )>δ = (cid:0)Aα grα0(Aβ Hence, (1.17) if β(x) = 0, ∞ if β(x) > 0. >0(cid:1) ∼= Aα ≥δ/Aα >δ = Aα δ . >0 7→ (α(x) >0(cid:1)(cid:14)(cid:0)Aα 0 ) ∼=g Lδ∈∆ >δ/Aβ Aα δ , ≥δ/Aβ VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 11 Proposition 1.5. The value function α0 is a u-gauge on Aβ 0 . (1.18) EndD(Vi)γ. Proof. We know from [TW2, Prop. 4.3] that α0 is a u-norm. Moreover, α0 is surmultiplicative since α is surmultiplicative. Thus, it remains only to prove that grα0(Aβ 0 ) is graded semisimple. Since grα(A) is graded semisimple, it suffices to consider the case when grα(A) is graded simple. In this case, by the graded version of Wedderburn's Theorem (see [HW, Prop. 1.3]), we can identify grα(A) with EndD(V), where D is a graded division ring and V is a finite-dimensional right graded D-vector space. (The grading on EndD(V) is given by: for η ∈ Γ, an f ∈ EndD(V) is homogeneous of degree η if and only if f (Vγ) ⊆ Vγ+η for all γ ∈ Γ.) Let {b1, . . . , bn} be a homogeneous D-base of V, and let γi = deg(bi). Since ΓV is a finite union of cosets of ΓD, we can write {b1, . . . , bn} ℓ=1 Sr, where bi and bj are in the same Sr if and only if γi − γj ∈ ΓD + ∆. Let Vi be the graded D-vector subspace of V generated by Si. We have a direct sum decomposition 0 ) is a homogeneous element of EndD(V) satisfying also deg(f ) ∈ ∆. Hence, f maps each Vi to itself. Thus, we can write as a disjoint union Fk V = Lk i=1 Vi. By (1.17), each nonzero homogeneous f ∈ grα0 (Aβ Bi = Lγ∈∆ grα0 (Aβ kQi=1 0 ) = where Bi The proof is completed by showing that each Bi is a simple graded algebra. Let which is a graded division subring of D. Now fix an index i ∈ {1, . . . , k}. The grade set of Vi lies in some coset λi + (ΓD + ∆). Let D′ = Lγ∈∆ Dγ, V′ i = Lγ ∈ λi+∆ Vγ ⊆ Lγ∈λi+ΓD+∆ Vγ = Vi, so V′ D′dj. Also, as V′ i is a graded D′-vector subspace of Vi. Let ΓD =Fj∈J δj + (∆ ∩ ΓD), a disjoint union of cosets of ∆ ∩ ΓD. Then we also have ΓD + ∆ =Fj∈J δj + ∆, which is again a disjoint union. For each j ∈ J, pick some dj ∈ Dδj \ {0}. Then, as ΓD′dj = δj + (∆ ∩ ΓD), we have i dj =Lγ∈λi+δj +∆ Vγ, Vi = Lj∈J D = Lj∈J i dj = Lj∈J Any map in Bi sends V′ i to itself, since ΓBi ⊆ ∆. Thus, there exists a homomorphism of graded rings . This map ψ has an inverse given by sending h ∈ EndD′(V′ ψ : Bi → EndD′(V ′ i) to h ⊗ idD ∈ EndD(V′ i ⊆ λi + ∆. Hence, h ⊗ idD is homogeneous with deg(h ⊗ idD) = deg(h) ∈ ∆, showing that h ⊗ idD ∈ Bi. Therefore, Bi ∼=g EndD′(V′ i ⊗D′ D). Note that if h is homogeneous, then deg(h) ∈ ∆, as ΓV′ i ⊗D′ D′dj = Vi ⊗D′ D. V′ i), which is a simple graded algebra. i ), given by g 7→ gV ′ (cid:3) V′ i 1.b. Defectlessness of valuations in semisimple algebras. We now develop the notion of defectlessness of a valuation v on F in a finite-dimensional semisimple F -algebra A. We will see in §4 that defectlessness is the condition required for the existence of v-gauges on A. First, we review the concept of defect on division algebras. A good reference for the division algebra case is [M1]. 12 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Let D be a finite-dimensional division algebra over a field F . For a valuation w on D, extending a valuation v on F , we have the "fundamental inequality" (1.19) [D:F ] ≥ [ D:F ] · Γw :Γv. When equality holds in (1.19), we say the valuation w on D is defectless over F . uniquely to Z(D), we define the defect ∂D/F of D over F by If v extends (1.20) ∂D/F = [D:F ] . [ D:F ] · Γw :Γv In particular, ∂D/Z(D) is always defined; we call it simply the defect of D and use the simpler notation ∂D for ∂D/Z(D). When ∂D/F is defined we have (1.21) ∂D/F = ∂D · ∂Z(D)/F by the transitivity formulas for residue degrees and and indices of value groups. In fact, for p = char F , we have ∂D/F = pℓ for some integer ℓ ≥ 0 if p 6= 0, and ∂D/F = 1 if p = 0. This result is known as Ostrowski's Theorem and was proved by Draxl in [D, Th. 2] for v Henselian and in general by Morandi in [M1, Th. 3]. Hence in particular, if p = 0 or if p 6= 0 and p ∤ [D:F ], then ∂D/F = 1. Let A be a (finite-dimensional) semisimple F -algebra, let (Fh, vh) be a Henselization of (F, v), and let Ah = A ⊗F Fh. Since Fh is a separable extension of F , the Fh-algebra Ah is semisimple; hence, it has a decomposition into simple components Ah ∼= Mn1(D1) × . . . × Mnr (Dr) for some integers n1, . . . , nr and some division algebras D1, . . . , Dr over Fh. We say that v is defectless in A if for each i = 1, . . . , r the unique valuation on Di extending vh is defectless over F , i.e., ∂Di/Fh = 1 for i = 1, . . . , r. It is clear from the definition that v is defectless in A if and only if v is defectless in each simple component of A, and that this condition holds if and only if vh is defectless in Ah. We single out two particular cases: • If K is a finite-degree field extension of F and v1, . . . , vn are all the extensions of v of F to K, then v is defectless in K if and only if equality holds in the Fundamental Inequality, i. e. This follows readily from Th. A.1 in the Appendix. [K :F ] = [ K vi : F v ] · Γvi : Γv. nPi=1 • If A is a central simple F -algebra, then v is defectless in A if and only if the valuation on the division algebra associated to Ah is defectless over Fh. In particular, if v is defectless in A, it is also defectless in every algebra Brauer-equivalent to A. In simple algebras that are not central, we have the following reduction to the central case: Proposition 1.6. Let A be a simple F -algebra, and let v1, . . . , vr be the valuations on Z(A) extending v. The following conditions are equivalent: (a) v is defectless in A; (b) v is defectless in Z(A) and v1, . . . , vr are each defectless in A. VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 13 Proof. For each i = 1, . . . , r, let (Zi, vi,h) be a Henselization of (Z(A), vi). From Th. A.1 in the Appendix, we have Z(A)h ∼= Z1 × . . . × Zr. Since Z(Ah) = Z(A)h, the number of prime components of Ah is r, and we have division algebras D1, . . . , Dr with centers Z1, . . . , Zr respectively such that Ah ∼= Mn1(D1) × . . . × Mnr (Dr) for some integers n1, . . . , nr. Now, we have ∂Di/Fh = ∂Di · ∂Zi/Fh (cf. (1.21)); hence ∂Di/Fh = 1 if and only if ∂Di = 1 and ∂Zi/Fh = 1. The equivalence of (a) and (b) follows. (cid:3) When ΓF ∼= Z, v is defectless in any central simple F -algebra and v is defectless in a semisimple F -algebra A if and only if it is defectless in Z(A). Also, by Ostrowski's Theorem, if char F = 0, then v is defectless in every semisimple F -algebra; if char F = p 6= 0, then v is defectless in every central simple F -algebra A whose index ind(A) is not divisible by p. Lemma 1.7. Let w be any coarsening of the valuation v on F , and let K be a finite-degree field extension of F . If v is defectless in K, then w is defectless in K. Proof. Let w1, . . . , wℓ be the extensions of w to K. For j ∈ {1, . . . , ℓ}, let v1j, v2j, . . . , vkjj be the extensions of v to K that are refinements of wj. Thus, the vij are all the extensions of v to K. Let wj induced v/w denote the valuation on F wj . For each i, j by vij. Note that v1j/wj, v2j /wj, . . . , vkj j/wj are all the extensions of v/w to K there is a commutative diagram of value groups with exact rows: induced by v and likewise vij/wj the valuation on K w 0 0 / Γv/w Γv Γw / Γvij /wj / Γvij / Γwj 0 / 0 Because the map Γw → Γwj is injective, the Snake Lemma yields a short exact sequence of cokernels of the columns: 0 Hence, (1.22) Since v is defectless in K, we have / 0 / Γvij(cid:14)Γv / Γwj(cid:14)Γw / Γvij /wj(cid:14)Γv/w (cid:12)(cid:12)Γvij : Γv(cid:12)(cid:12) = (cid:12)(cid:12)Γvij /wj : Γv/w(cid:12)(cid:12)(cid:12)(cid:12)Γwj : Γw(cid:12)(cid:12). v(cid:3)(cid:12)(cid:12)Γvij : Γv(cid:12)(cid:12). [K :F ] = ℓPj=1 kjPi=1(cid:2) K vij : F / / / / / / / / / / / / / / / 14 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Equation (1.22) together with the Fundamental Inequality for each K then yield, wj over F w and for K over F vij : F kjPi=1(cid:2) K ℓPj=1 ℓPj=1(cid:18) kjPi=1(cid:2) K ℓPj=1(cid:2) K wj : F wj : F vij /wj v(cid:3)(cid:12)(cid:12)Γvij /wj : Γv/w(cid:12)(cid:12)(cid:12)(cid:12)Γwj : Γw(cid:12)(cid:12) w(cid:3)(cid:12)(cid:12)Γwj : Γw(cid:12)(cid:12) ≤ [K :F ]. w v/w(cid:3)(cid:12)(cid:12)Γvij /wj : Γv/w(cid:12)(cid:12)(cid:19)(cid:12)(cid:12)Γwj : Γw(cid:12)(cid:12) [K :F ] = = ≤ The last inequality must therefore be an equality, showing that w is defectless in K. (cid:3) Proposition 1.8. Let w be any coarsening of the valuation v on F . Let A be a semisimple F - algebra. If v is defectless in A, then w is defectless in A. Proof. Assume first that A is central simple over F . Let (Fh,v, vh) be a Henselization of (F, v). Let w′ be the valuation on Fh,v with ring W ·Vh. So, w′ is the extension of w which is a coarsening of vh. Since vh is Henselian, its coarsening w′ is also Henselian, by [EP, Cor. 4.1.4, p. 90], so (Fh,v, w′) contains a Henselization (Fh,w, wh) of (F, w). It follows from [M1, Th. 2] that w′ is inertial (= unramified) over wh. Let Dh,v (resp. Dh,w) be the central division algebra over Fh,v (resp. Fh,w) associated to A ⊗F Fh,v (resp. A ⊗F Fh,w). Since v is defectless in A, Dh,v is defectless for vh; it is then also defectless for the coarser valuation w′ by [M4, Lemma 1]. Then, by [JW, Remark 3.4] applied to the inertial extension (Fh,v, w′) of (Fh,w, wh), Dh,w is defectless for wh. Hence, w is defectless in A, as desired. Now assume only that A is simple. Let K = Z(A), and let v1, . . . , vr be the extensions of v to K, and w1, . . . , wℓ the extensions of w to K. For i ∈ {1, . . . , r}, let j(i) ∈ {1, . . . , ℓ} be the index such that wj(i) is the w-coarsening of vi. Since v is defectless in A, v is defectless in K and each vi is defectless in A. Then, w is defectless in K by Lemma 1.7, and each wj(i) is defectless in A by the central simple case just considered. Since wj(1), . . . , wj(r) are all the extensions of w to K, it follows by Prop. 1.6 that w is defectless in A. This completes the proof for A simple, and the general case for A semisimple follows easily by considering the simple components of A. (cid:3) The following result says that defectlessness is a necessary condition for the existence of a v-gauge on an arbitrary semisimple F -algebra. We will see in Th. 4.3 below that this necessary condition is also sufficient. Proposition 1.9. Let A be any semisimple (finite-dimensional) F -algebra. If A has a v-gauge, then v is defectless in A. Proof. A has a v-gauge if and only if each simple component of A has a v-gauge, by Prop. 1.3. Also, by definition, v is defectless in A if and only if v is defectless in each simple component of A. Thus, we may assume that A is simple. Let K = Z(A), and let v1, . . . , vr be the extensions of v to the field K. Let (Fh, vh) be a Henselization of (F, v), and let (Kh,vi, vi,h) be a Henselization of (K, vi). Let α be a v-gauge on A. Then the restriction αK is a surmultiplicative v-norm on K with grαK (K) graded semisimple, since it is a central graded subalgebra of the graded semisimple algebra grα(A). Hence αK is a v-gauge on K, so v is defectless in K by [TW1, Cor. 1.9]. Moreover, VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 15 α ⊗ vh is a vh-gauge on A ⊗F Fh and A ⊗F Fh ∼= (A ⊗K Kh,v1) × . . . × (A ⊗K Kh,vr ). By Prop. 1.3 each A ⊗K Kh,vi carries a gauge, hence the corresponding central division algebra over Kh,vi is defectless by [TW1, Th. 3.1]. (cid:3) 2. Gauges on simple algebras Throughout this section, let v be an arbitrary valuation on a field F , and let A be a (finite- dimensional) simple F -algebra. Let K be the center of A, so K is a finite-degree field extension of F . Let v1, . . . , vr be all the extensions of v to K, and let Vi be the valuation ring of vi. Our goal is to characterize v-gauges on A in terms of vi-gauges for i = 1, . . . , r. This will be achieved in Th. 2.2 and Th. 2.8. Proposition 2.1. Let L be a field with F ⊆ L ⊆ K, and suppose v has a unique extension to a valuation vL of L. If A has a v-gauge α, then αL = vL, which is a v-gauge on L, and α is a vL-gauge. Thus, whenever vL is a v-gauge on L, the v-gauges on A are the same as the vL-gauges on A. Proof. The restriction αL of α to L is clearly surmultiplicative, and is a v-norm on L by [RTW, Prop. 2.5] since L is an F -subspace of A. Moreover, as L ⊆ Z(A), we have grαL(L) ⊆ Z(grα(A)). Because grα(A) is graded semisimple, it contains no nonzero central homogeneous nilpotent ele- ments. Therefore, the commutative gr(F )-algebra grαL(L) is semisimple, and hence αL is a v-gauge on L. Because vL is the unique extension of v to L, [TW1, Cor. 1.9] shows that αL = vL. Hence, for c ∈ L×, we have α(c−1) = vL(c−1) = −vL(c) = −α(c). So, a short computation (cf. [TW1, Lemma 1.3]) shows that α(ca) = α(c) + α(a) for all a ∈ A. This proves that the v-value function α on A is actually a vL-value function. Since vL and α are v-norms, we have [L:F ] [grα(A): grvL(L)] = [grvL(L): grv(F )] [grα(A): grvL(L)] = [grα(A): grv(F )] = [A:F ] = [L:F ] [A:L]. Hence, [grα(A): grvL(L)] = [A:L], showing that α is a vL-norm. The other conditions needed for α to be a vL-gauge hold because it is a v-gauge. Conversely, whenever vL is a v-gauge (hence a v-norm) and β is a vL-gauge on A, then [grβ(A): grv(F )] = [grβ(A): grvL(L)] [grvL(L): grv(F )] = [A:L] [L:F ] = [A:F ], so β is also a v-norm and hence a v-gauge on A. (cid:3) Theorem 2.2. Let α be a v-gauge on A. Then there exist vi-gauges αi on A for i = 1, . . . , r such that Furthermore, (2.1) α(a) = min(cid:0)α1(a), . . . , αr(a)(cid:1) for all a ∈ A. grα(A) ∼=g grα1(A) × . . . × grαr (A). 16 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Hence, the semisimple F -algebra grα(A)0 has at least r simple components. Moreover, the grαi(A) are the graded simple components of grα(A) and [grαi(A): gr(F )] = [A:K] [grvi(K): gr(F )]. We call the αi of the theorem the vi-component of α, for i = 1, . . . , r. We will see in Cor. 2.5 below that the αi are uniquely determined by α. Proof. Let (Fh, vh) be the Henselization of (F, v), and let B = A ⊗F Fh and L = Z(B) = K ⊗F Fh. Then, L is a direct product of finitely many fields. Let e1, . . . , er be the primitive idempotents of L, so and each Li is a field. The Li are indexed by the vi, as we will explain below. Since B ∼= A ⊗K L, the ring B is a product of algebras L = L1 × . . . × Lr where Li = eiL, B = B1 × . . . × Br where Bi = eiB ∼= A ⊗K Li. So, each Bi is a central simple Li-algebra. We identify K, A, Fh with their isomorphic copies K⊗1, A⊗1, 1⊗Fh in B. But, we do not identify them with their isomorphic copies eiK, eiA, eiFh in Bi. For each i we have canonical inclusions pi : A ֒→ Bi, a 7→ ei(a ⊗ 1) and qi : Fh ֒→ Bi, c 7→ ei(1 ⊗ c). Thus, Bi has subalgebras pi(A), pi(K), and Li, with Bi = pi(A) ⊗pi(K) Li ∼= A ⊗K Li, hence, [Bi:Li] = [A:K]. Each field Li is a compositum of fields, Li = pi(K)·qi(Fh). The Henselian valuation vh on Fh has an isomorphic (Henselian) valuation vh ◦ q−1 on qi(Fh), which extends uniquely to a Henselian valuation wi on Li. This pulls back to a valuation wi ◦ pi on K which extends v on F . The proof of Th. A.1 in the Appendix shows that the valuations w1 ◦ p1, . . . , wr ◦ pr are all distinct and are all the extensions of v to K. Thus, after renumbering the ei if necessary, we can assume wi ◦ pi = vi for i = 1, . . . , r. That is, i (2.2) for all d ∈ K. From Th. A.1, we have also that (Li, wi) is a Henselization of (K, vi). vi(d) = wi(ei(d ⊗ 1)) Let β = α ⊗ vh, which is a vh-gauge on B with (2.3) by [TW1, Cor. 1.26]. Let βi = βBi , which is a vh-gauge on Bi via the embedding qi : Fh → Bi. By Prop. 1.3, grβ(B) ∼=g grα(A) ⊗gr(F ) gr(Fh) ∼=g grα(A) Since wi is the unique extension of the Henselian valuation vh to Li, Prop. 2.1 above shows that each βi is a wi-gauge. The structure theorem [TW1, Th. 3.1] for gauges on simple algebras with (2.4) and (2.5) β(b) = min 1≤i≤r(cid:0)βi(eib)(cid:1) rQi=1 grβ(B) ∼=g for all b ∈ B, grβi(Bi). VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 17 respect to Henselian valuations shows that βi is an End-gauge as in Ex. 1.1; hence, grβi(Bi) is graded simple. Define v-value functions α1, . . . , αr on A by αi(a) = β(pi(a)) = βi(ei(a ⊗ 1)). Then, each αi is surmultiplicative as βi is surmultiplicative, and for all a ∈ A, (2.6) α(a) = β(a ⊗ 1) = min 1≤i≤r(cid:0)βi(ei(a ⊗ 1))(cid:1) = min αi(ca) = βi(ei(ca ⊗ 1)) = βi(cid:0)ei(c ⊗ 1) · ei(a ⊗ 1)(cid:1) = wi(ei(c ⊗ 1)) + βi(ei(a ⊗ 1)) = vi(c) + αi(a). 1≤i≤r(cid:0)αi(a)(cid:1). Furthermore, as βi is a wi-value function, for all c ∈ K and a ∈ A, Thus, αi is a vi-value function on A. The following diagram shows the algebras related to Bi and the associated value functions being considered here. (2.7) A, αi Bi, βi pi ;✈✈✈✈✈✈✈✈ ❍❍❍❍❍❍❍❍❍ ❍❍❍❍❍❍❍❍❍ :✈✈✈✈✈✈✈✈ pi Li, wi qi d■■■■■■■■■ K, vi Fh, vh Now, for all γ ∈ Γ, the definition of the βi and (2.4) and (2.6) above show that for each i we have a commutative diagram Aα ≥γ −−−−→ Aαi ≥γ y Bβ ≥γ y ei· −−−−→ B βi i, ≥γ given by a −−−−→ y a y . a ⊗ 1 −−−−→ ei(a ⊗ 1) There is a corresponding commutative diagram with > γ replacing ≥ γ, hence an induced com- mutative diagram of corresponding factor groups; these together yield a commutative diagram of graded gr(F )-algebra homomorphisms: (2.8) grα(A) −−−−→ ∼=y grβ(B) ∼=−−−−→ grαi(A) y grβi(Bi) rQi=1 rQi=1 Here, the top map is injective by (2.6); the left map is the isomorphism of (2.3); the right map is injective since for each i, the definition of αi shows that grαi(A) → grβi(B) is injective; and the bottom map is the isomorphism of (2.5). Therefore, all the maps in this diagram must be isomorphisms. Hence, for each i, grαi(A) ∼=g grβi(Bi), which is graded simple as we saw above after (2.5). Since the Henselization (Li, wi) (resp. (Fh, vh)) is an immediate extension of (K, vi) (resp. (F, v)), we have [Li :Fh] ≥ [grwi(Li): grvh(Fh)] = [grvi(K): gr(F )]. , ; , : 2 R d 18 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH So, as βi is a vh-norm, [grαi(A): grvi(K)] [grvi(K): gr(F )] = [grαi(A): gr(F )] (2.9) = [grβi(Bi): grvh(Fh)] = [Bi :Fh] = [Bi :Li] [Li :Fh] = [A:K] [Li :Fh] ≥ [A:K] [grvi(K): gr(F )], and hence [grαi(A): grvi(K)] ≥ [A:K]. Since the reverse inequality holds for any vi-value func- tion, we have [grαi(A): grvi(K)] = [A:K]. Hence, αi is a vi-norm on A; with the graded simplic- ity noted above, this yields that αi is a vi-gauge. Furthermore, equality holds in (2.9), yielding [grαi(A): gr(F )] = [A:K] [grvi(K): gr(F )]. The isomorphism for grα(A) in the theorem is the top iso- morphism in the commutative diagram (2.8). Since each grαi(A) is graded simple, the grαi(A) are the graded simple components of grα(A). Because grα(A) has r graded simple components, its degree zero part must have at least r simple components. (cid:3) we write α = min(cid:0)η1, . . . , ηr(cid:1) if α(z) = min(cid:0)η1(z), . . . , ηr(z)(cid:1) for all z ∈ M . It is easy to construct Let v be a valuation on some division algebra D, and let α, β, η1, . . . , ηr be v-value functions on a finite-dimensional D-vector space M . We write α ≤ β if α(z) ≤ β(z) for all z ∈ M . Likewise, examples of v-norms α, β on M with α ≤ β and α 6= β. By contrast, we will see in Cor. 2.6 below for gauges on semisimple algebras that α ≤ β implies α = β. This will be proved by showing a minimality property characterizing the components of a gauge on a simple algebra. Lemma 2.3. Let A and B be F -algebras with respective surmultiplicative v-value functions α and β. Suppose there is an F -algebra homomorphism f : A → B such that (2.10) for all a ∈ A. β(f (a)) ≥ α(a) Then f induces a graded gr(F )-algebra homomorphism bf : grα(A) −→ grβ(B) equality holds in (2.10). Proof. For any δ ∈ Γ, we have f (A≥δ) ⊆ B≥δ and f (A>δ) ⊆ B>δ, hence f induces a map such that bf (ex) = f (x) + B>α(x) ∈ Bα(x) for all x ∈ A. Moreover, bf is injective if and only if (bf )δ : Aδ → Bδ given by (bf )δ(x + A>δ) = f (x) + B>δ. Then set bf =Lδ∈Γ(bf )δ : grα(A) → grβ(B). Clearly, bf is a graded gr(F )-vector space homomorphism. To see that bf is multiplicative, it suffices to check this for homogeneous elements. That is, for any nonzero x, y ∈ A we need (2.11) When the left expression in (2.11) is nonzero, it equals ^f (xy), and when the right expression is bf (exey) = bf (ex)bf (ey). nonzero it equals gf (x)gf (y) = ^f (x)f (y) = ^f (xy). So, equality indeed holds in (2.11) when each side is nonzero. Now we have (2.12) β(f (xy)) ≥ α(xy) ≥ α(x) + α(y) and (2.13) β(f (xy)) = β(f (x)f (y)) ≥ β(f (x)) + β(f (y)) ≥ α(x) + α(y). β(f (xy)) = α(xy), i.e., equality holds throughout (2.12). The right expression in (2.11) is nonzero if The left expression in (2.11) is nonzero if and only if α(xy) = α(x) + α(y) (so exey = fxy 6= 0) and VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 19 and only if β(f (x)) = α(x) and β(f (y)) = α(y), and β(f (x)f (y)) = β(f (x)) + β(f (x)), i.e., equality holds throughout (2.13). Each of these conditions holds if and only if β(f (xy)) = α(x)+α(y). Thus, we have equality in (2.11) in all cases. Now note that bf (ex) = 0 if and only if β(f (x)) > α(x). Since ker(bf ) is a homogeneous two-sided ideal of grα(A), the stated condition for injectivity of bf holds. (cid:3) Theorem 2.4. Let α be any v-gauge on the simple F -algebra A, and, as in Th. 2.2 above, let αi be the vi-component of α for i = 1, . . . , r. Suppose η is a vk-gauge on A, for some k. If α ≤ η, then η = αk. j = 1, . . . , n and i = 1, . . . , r, Proof. Pick a homogeneous base (cid:0)b1, . . . , bn(cid:1) of grαk (A) as a graded grvk (K)-vector space, and α 7→ (0, . . . , 0, bj , 0, . . . , 0) (bj in the k- let γj = deg(bj). Then, pick a1, . . . , an ∈ A with each eaj th position) under the isomorphism grα(A) ∼=g Qr i=1 grαi(A) of Th. 2.2. This means that for all Since(cid:0)ea1 a =Pn is a splitting base of the K-vector space A for the vk-gauge αk (see the comments preceding (1.3) above). Furthermore, αk(aj) = α(aj) ≤ η(aj), for all j. Therefore, for any a ∈ A, writing αk(cid:1) is a homogeneous base of the graded grvk (K)-vector space grαk (A), (a1, . . . , an) for i 6= k, α(aj) = αk(aj) = γj, j=1 ajcj with cj ∈ K, we have αk , . . . ,fan αk(a) = min and αk = bj in grαk (A). αi(aj) > γj eaj 1≤j≤n(cid:0)αk(aj) + vk(cj )(cid:1) ≤ min ≤ η(cid:0) nPj=1 ajcj(cid:1) = η(a). 1≤j≤n(cid:0)η(aj ) + vk(cj)(cid:1) Thus, αk ≤ η as vk-gauges on A. Since αk ≤ η, Lemma 2.3 (with f = idA) shows that there is a well-defined graded grvk (K)- αk(a) for all a ∈ A. But, as αk is a vk-gauge on the central simple K-algebra A, [TW1, Cor. 3.7] shows that grαk (A) is a simple graded algebra. Hence ϕ must be injective. Again by Lemma 2.3 we have αk = η. (cid:3) algebra homomorphism ϕ : grαk (A) → grη(A) given byeaαk 7→ a + Aη αi be the vi-component of α for i = 1, . . . , r. Suppose α = min(cid:0)η1, . . . , ηr(cid:1) for some vi-gauges ηi. Corollary 2.5. Let α be any v-gauge on the simple F -algebra A, and, as in Th. 2.2 above, let >αk(a) ∈ Aη Then each ηi = αi. Proof. For each i, we have α ≤ ηi. Hence, ηi = αi by the preceding theorem. Corollary 2.6. Let α and η be v-gauges on a semisimple F -algebra C. If α ≤ η, then α = η. Proof. It suffices to check this for the restrictions of α and η on the simple components of C. Therefore, we may assume that C is a simple F -algebra. Then, let v1, . . . , vr be all the extensions of v to Z(C), and let αi (resp. ηi) be the vi-component of α (resp. η). Since, for each i we have α ≤ η ≤ ηi, Th. 2.4 shows that αi = ηi. Hence, (cid:3) α = min(cid:0)α1, . . . , αr(cid:1) = min(cid:0)η1, . . . , ηr(cid:1) = η. (cid:3) Corollary 2.7. Let α be a v-gauge on the simple F -algebra A, with vi-components αi, for i = 1, . . . , r, and suppose Γα lies in the divisible hull of Γv. Let w be any valuation on F which is a coarsening of v, and let w1, . . . , wℓ be all the extensions of w to K. Let β be the w-coarsening of α, and let βj 20 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH be the wj-component of β for j = 1, . . . , ℓ. For i ∈ {1, . . . , r}, let j(i) ∈ {1, . . . , ℓ} be the index such that wj(i) is the w-coarsening of vi (i.e., Wj(i) = W ·Vi). Then, βj(i) is the wj(i)-coarsening of αi. Proof. Let αi,w denote the wj(i)-coarsening of αi. Let ∆ be the convex subgroup of Γ associated to w. Recall that β = ε ◦ α, where ε : Γ → Γ/∆ is the canonical surjection. This ε is compatible with the orderings on Γ and Γ/∆. Likewise, wj(i) = ε ◦ vi and αi,w = ε ◦ αi. Now fix any i. Since α ≤ αi, β = ε ◦ α ≤ ε ◦ αi = αi,w, i.e., β ≤ αi,w. Since αi,w is a wj(i)-gauge, Th. 2.4 shows that αi,w = βj(i). We next determine when given vi-gauges η1, . . . , ηr yield a v-gauge as min(cid:0)η1, . . . , ηr(cid:1). For i, j ∈ {1, . . . , r} let vij denote the finest common coarsening of vi and vj. That is, vij is the valuation on K with associated valuation ring Vij = Vi·Vj. Theorem 2.8. Suppose v is defectless in the simple F -algebra A. For i = 1, . . . , r, let ηi be a vi-gauge on A with Γηi ⊆ H(Γv). Let α = min(cid:0)η1, . . . , ηr). Then, α is a v-gauge on A if and only if ηi and ηj have the same vij-coarsening for all pairs i, j. When this occurs, each ηi is the vi-component of α. (cid:3) Proof. ⇒ Suppose our α is a v-gauge on A. By Cor. 2.5, each ηi is the vi-component of α. Fix indices i, j, let w = vijF , and let β be the w-coarsening of α. The vij-coarsenings of ηi and ηj must be the same, since by Cor. 2.7 they coincide with the vij-component of β. ⇐ Suppose each ηi and ηj have the same vij-coarsening. Now, α = min(cid:0)η1, . . . , ηr(cid:1) is clearly a surmultiplicative v-value function on A. Consider the graded gr(F )-algebra homomorphism (2.14) Ψ : grα(A) −→ grηi(A) given by a + Aα >α(a), . . . , a + Aηr Then, Ψ is well-defined and injective because α = min(cid:0)η1, . . . , ηr(cid:1). We will show below that Ψ is an isomorphism. It then follows that grα(A) is semisimple, as each grηi(A) is semisimple. Moreover, v is defectless in K by Prop. 1.6 since it is defectless in A, whence >α(a) 7→ (cid:0)a + Aη1 >α(a)(cid:1). rQi=1 [grα(A): gr(F )] = = Hence, α is v-gauge. rPi=1 rPi=1 [grηi(A): gr(F )] = [grηi(A): grvi(K)] [grvi(K): gr(F )] [A:K] [grvi(K): gr(F )] = [A:K] [K :F ] = [A:F ]. rPi=1 To prove surjectivity of Ψ, we use the following approximation lemma, for which we use the notation: let ∆ij be the convex subgroup of the divisible hull Γ of Γv associated to vij; so vij is a map K → Γ/∆ij ∪ {∞}. For any δ1, . . . , δn ∈ Γ we say that the n-tuple (δ1, . . . , δn) is compatible in Γn if for all i, j we have δi − δj ∈ ∆ij. Lemma 2.9. With the notation just above, fix some k ∈ {1, . . . , n}, and let (δ1, . . . , δn) be a compatible n-tuple in Γn with δk = 0. Then, there is c ∈ K × with vi(c) > δi for each i 6= k, vk(c) = 0, and vk(c − 1) > 0. Proof. For each pair of indices i, j, we have 0 ≤ (cid:12)(cid:12)δi − δj(cid:12)(cid:12) ≤ δi − δj. VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 21 Since δi − δj ∈ ∆ij and ∆ij is convex, it follows that δi − δj ∈ ∆ij. Furthermore, as Γ/Γv is a torsion group, there is m ∈ N with each mδi ∈ Γv. Hence, each mδi ∈ Γvi and mδi−mδj ∈ ∆ij. These are the precise conditions needed for(cid:0)mδ1, . . . , mδn(cid:1) to be compatible in Γv1 × . . .× Γvn in the terminology of Ribenboim's paper [R]. Since this compatibility holds, the general approximation theorem for incomparable valuations on K [R, Th. 5] says that there exists d ∈ K × with vi(d) = mδi, for all i. In particular, vk(d) = mδk = 0. Now, let Vi be the valuation ring of vi, and let T = V1 ∩ . . . ∩ Vn ⊆ K. A weaker approximation theorem for incomparable valuations on K (see [EP, Th. 3.2.7(3), p. 64]) says that the canonical map ρ : T → K Then for i 6= k we have ρ(t) =(cid:0)0, 0, . . . , d −1, . . . , 0(cid:1), i.e., vi(t) > 0 for i 6= k and vk(t) = 0 with t = d −1 in K vn is surjective. Therefore, there is t ∈ T with vk . Let c = td. v1 × . . . × K vi(c) = vi(t) + vi(d) > 0 + mδi ≥ δi, vk , hence vi(c) > δi. Also, vk(c) = vk(t) + vk(d) = 0, and in K c = t · d = d −1d = 1. Thus, c has all the required properties. (cid:3) Proof of Th. 2.8 completed: It remains only to prove the surjectivity of the map Ψ in (2.14). Fix any k ∈ {1, . . . , r}, and take any b ∈ A \ {0}. Let δi = ηk(b) − ηi(b) for i = 1, . . . , r. For each pair of indices i, j, let ∆ij be the convex subgroup of H(ΓF ) associated to the finest common coarsening vij of vi and vj on K. Then, since ηi and ηj are assumed to have the same vij-coarsening, ηi(b) and ηj(b) have the same image in Γ/∆ij; hence, Thus, (δ1, . . . , δr) is a compatible r-tuple in H(ΓF )r. Since δk = 0, Lemma 2.9 yields c ∈ K × with vi(c) > δi for all i 6= k and vk(c) = 0 with c = 1 in K vk . Let a = cb ∈ A. Then, for i 6= k, δi − δj = ηj(b) − ηi(b) ∈ ∆ij. so ηi(a) > ηk(b). But ηi(a) = vi(c) + ηi(b) > δi + ηi(b) = ηk(b), ηk(a) = vk(c) + ηk(b) = ηk(b). >α(a) ∈ grα(A) andeb = b + Aηk ηk(a − b) = ηk(cb − b) = vk(c − 1) + ηk(b) > ηk(b). >ηk(b) ∈ grηk (A), we have Hence, α(a) = min(cid:0)η1(a), . . . , ηr(a)(cid:1) = ηk(b). Moreover, Thus, forea = a + Aα Since for arbitrary k and b such elements generate Qr pairwise independent. Take any vi-gauges ηi on A, i = 1, . . . , r. Then min(cid:0)η1, . . . , ηr(cid:1) is a v-gauge Corollary 2.10. Suppose v is defectless in A, and suppose the extensions v1, . . . , vr of v to K are >α(a)(cid:1) = (cid:0)0, . . . , 0, eb on A with components η1, . . . ηr. In particular, this holds whenever v has rank 1. i=1 grηi(A), the map Ψ is surjective. This (cid:3) Ψ(ea) = (cid:0)a + Aη1 >α(a), . . . , a + Aηr , 0, . . . , 0(cid:1) ∈ rQi=1 k grηi(A). completes the proof. 22 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Proof. This is immediate from the preceding theorem. For when vi and vj are independent valua- tions, their finest common coarsening vij is the trivial valuation, so the compatibility condition on ηi and ηj holds automatically. (cid:3) 3. Dubrovin valuation rings and Grater rings In this section we study the connection between gauges and Dubrovin valuation rings. The best general reference for Dubrovin theory is the book [MMU]. Let F be a field and let A be a central simple F -algebra. A subring B of A is called a Dubrovin valuation ring of A if there is an ideal J of B such that the following hold: (1) B/J is simple Artinian; (2) for each s ∈ A \ B there exist b, c ∈ B such that bs, sc ∈ B \ J. By [MMU, Lemma 5.2, p. 22], the ideal J is the only maximal ideal of B; therefore, J = J(B), the Jacobson ideal of B. Moreover, the center Z(B) = F ∩ B is a valuation ring of F and J ∩ F = J(Z(B)) (see [MMU, Lemma 7.1, p. 35]). We review a few of the nontrivial properties of Dubrovin valuation rings that we will use repeatedly below. Proofs of all of them can be found in Chapters 5 and 6 of [MMU]. Let S be an overring of B in A, i.e., a ring with B ⊆ S ⊆ A. Then, S is a Dubrovin valuation ring of A; J(S) is a prime ideal of B; S is the left (and right) localization of B with respect to the elements of B regular mod J(S); and B/J(S) is a Dubrovin valuation ring of S/J(S). Also, S is the central localization S = B ⊗Z(B) W = B·W , where the valuation ring W = Z(S) is the localization of Z(B) at its prime ideal J(S) ∩ F . Moreover, every prime ideal of B has the form J(S) for some such S. If R is a subring of B with J(B) ⊆ R, then R is a Dubrovin valuation ring of A if and only if R/J(B) is a Dubrovin valuation ring of B/J(B). Examples of Dubrovin valuation rings include • total valuation ring in division algebras D, i.e., subrings T of D such that d or d−1 lies in T for every d ∈ D×; • matrix rings Mn(B) for any Dubrovin valuation ring B; • rings eBe for any Dubrovin ring B of A and any nonzero idempotent e of A; • Azumaya algebras over commutative valuation rings. Associated to a Dubrovin valuation ring B there is its residue ring B = B/J(B), which is simple Artinian. This B also has a value group ΓB = st(B)/B×, where st(B) = {x ∈ A× xBx−1 = B}; the value group ΓB is a totally ordered abelian group. It was proved in [W1] that there is a strong connection between Dubrovin valuation rings and invariant valuation rings (i.e., the valuation rings associated to a valuation on a division algebra), which shows up with passage to the Henselization of the valuation of the center: Let (Fh, vh) be a Henselization of (F, v). The ring A ⊗F Fh is a central simple Fh-algebra; hence, (3.1) A ⊗F Fh ∼= MnB (Dh), where nB is a positive integer and Dh is a central division Fh-algebra. Thus, nB is the matrix size of the algebra A⊗F Fh. The Henselian valuation vh has a unique extension to a valuation w on Dh. By [W1, Th. B] or [MMU, Lemma 11.4, p. 59], (3.2) ΓB = Γw in H(Γv) VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 23 and (3.3) B ∼= MtB (Dh w ), where the positive integer tB is the matrix size of B. Moreover, nB/tB is always an integer that appears in the Ostrowski Theorem for Dubrovin valuation rings, which says that (3.4) [A : F ] = [B : F v ] · ΓB : Γv ·(cid:0)nB/tB(cid:1)2 · ∂B, v where the defect ∂B = pd, for p = char(F ) and d a non-negative integer, with pd = 1 if p = 0 (see [W1, Th. C]). It follows from (1.20) applied to Dh and (3.4) that ∂B = ∂Dh/Fh. Hence, v is defectless in A if and only if ∂B = 1 for any Dubrovin valuation ring B of A extending V . Dubrovin valuation rings have good properties with respect to extension from the center. More precisely, if V is a valuation ring of F then there always exists a Dubrovin valuation ring B of the central simple F -algebra A extending V , i.e., V = B ∩ F , see [MMU, Th. 9.4, p. 50]. Moreover, by [W1, Th. A] or [MMU, Th. 9.8, p. 52] there is as much uniqueness as possible, i.e., all Dubrovin ring extensions of V to A are conjugate in A. But the number of extensions of V to Dubrovin valuation rings of A is usually infinite; the exception occurs only when A is a division algebra and V can be extended to a total valuation ring T of A. In this special case, the number of total valuation rings of A extending V is given by nB/tB. In order to obtain a better understanding of the Ostrowski Theorem (equivalently, a better interpretation of the integer nB/tB for arbitrary Dubrovin valuation rings), Grater introduced in [G1] the Intersection Property for a finite number of Dubrovin valuation rings as follows: Let B1, . . . , Bn be Dubrovin valuation rings of A and let R = B1 ∩ . . . ∩ Bn. Let B(Bi) denote the set of all overrings of Bi in A. Then B1, . . . , Bn have the Intersection Property (IP) if the map (3.5) ϕ : B(B1) ∪ . . . ∪ B(Bn) → Spec(R) 7→ J(S) ∩ R S is a well-defined order-reversing bijection, where Spec(R) is the set of prime ideals of R. Actually, if one supposes only that ϕ is well-defined (i.e., each ideal J(S) ∩ R is a prime ideal of R), then in fact ϕ is an order-reversing bijection (see [Z]). Note that if Bj ⊇ Bi for some i, j, then we may delete Bj from the list of B's and the ring R and the domain and target of ϕ are unchanged. Thus, in working with the IP, we may delete all such redundant Bj and assume that the Bi are pairwise incomparable. At the same time that Intersection Property was introduced by Grater, Morandi was working independently on a general approximation theorem for a finite set of Dubrovin valuation rings (see [M3]). He found the following condition: Let B1, . . . , Bn be pairwise incomparable Dubrovin valuation rings. For each i 6= j, let Bij be the subring of A generated by Bi and Bj; so Bij is a which are Dubrovin valuation rings of Bij = Bij/J(Bij). The condition needed for the approxima- Dubrovin valuation ring, since it is an overring of Bi. Set fBi = Bi/J(Bij) and fBj = Bj/J(Bij), tion theorem to hold for B1, . . . , Bn is that the valuation rings Z(fBi) and Z(fBj) be independent in the field Z(Bij), i.e., Z(fBi)·Z(fBj) = Z(Bij). Grater proved that this condition is equivalent to the Intersection Property for B1, . . . , Bn (see [G1, Cor. 6.2, Prop. 6.3, Cor. 6.7] or [MMU, Cor. 16.9, p. 93]). It follows that (3.6) B1, . . . , Bn satisfy the IP if and only if each pair Bi, Bj satisfies the IP, for all i, j ∈ {1, . . . , n} with i 6= j. 24 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH We call a subring R of a central simple F -algebra A a Grater ring if R = B1 ∩ . . . ∩ Bn where B1, . . . , Bn is a family of incomparable Dubrovin valuation rings of A with the IP and R is integral over Z(R). This name is appropriate because of the remarkable properties Grater proved about such rings in [G1] and [G2]. (His results appear also in [MMU].) He showed that a Grater ring is a semilocal B´ezout ring whose center is a finite intersection of valuation rings of F . When R = B1 ∩ . . . ∩ Bn as above, then the localizations of the R with respect to its maximal ideals exist and coincide with the Bi. Thus in particular, (3.7) the number n of Bi = the number of maximal ideals of R. Moreover, if T is any finite intersection of valuation rings of F then there exists a Grater ring R of A with Z(R) = T . Further, Grater proved a very strong uniqueness property: If R′ is another Grater ring of A with Z(R′) = T = Z(R), then there is q ∈ A× with R′ = qRq−1. See [MMU, Th. 16.14, p. 94; Th. 16.15, p. 96] for proofs of these properties. 1, . . . , B′ k+1, . . . , B′ m each with center V so that R′ = B′ There are also striking properties when the center of a Grater ring is a valuation ring V : Let R = B1∩ . . .∩ Bn be a Grater ring with center V (and the Bi incomparable); then, each Bi has cen- ter V , by [MMU, Lemma 16.13, pp. 93 -- 94]. Moreover, if B′ k is another family of Dubrovin valuation rings of A each with center V and having the IP, then the B′ i are incomparable (since overrings of B′ i are central localizations) and this family can be enlarged with further Dubrovin valuation rings B′ m is a Grater ring with center V (see [MMU, Th. 16.14, p. 94]). The conjugacy result noted above shows that R′ ∼= R, hence m = n by (3.7). Thus, this number n depends only on V and A, and Grater defined it to be the extension number of V in A. Thus, the extension number is the number of maximal ideals in any Grater ring of A with center V , and it equals the number of incomparable Dubrovin valuation rings in any family whose intersection is a Grater ring with center V (see (3.7)); it is also the maximum number of Dubrovin rings with center V which have the IP. It turns out that the extension number also equals the quotient nB/tB of the integers nB and tB given in (3.1) and (3.3), for any Dubrovin valuation ring B of A, see [MMU, Prop. 19.2, p. 108]. From the conjugacy of such rings B, it is clear that nB and tB do not depend on the choice of B. Moreover, it follows from the definition of nB and tB that the extension number depends only on the valuation ring V of the center F of A and the Brauer equivalence class of A. Let 1 ∩ . . . ∩ B′ (3.8) ξV,[A] = the extension number of V to A = nB/tB. When the algebra A in question is clear, we write ξV for ξV,[A]. Now let W be any (valuation) ring with V ⊆ W ⊆ F , and let S = W ·B. Then S is a Dubrovin valuation ring of A containing B and W = S ∩ F . Let eV = V /J(W ), which is a valuation ring of W = W/J(W ), and let eB = B/J(S), which is a Dubrovin valuation ring of the simple W -algebra S = S/J(S). Let ℓV,W = the number of extensions of eV to valuation rings of Z(S). This number does not depend on the choice of the Dubrovin extensions B of V and S of W , by the conjugacy of B with center V and the property that overrings of B are obtained by central localization. The following fundamental relation was given in [W1, Th. E]: (3.9) (3.10) ξV,[A] = ξW,[A] (n eB/t eB) ℓV,W = ξW,[A] ξZ( eB),[S] ℓV,W . VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 25 Hence, (3.11) ξW,[A](cid:12)(cid:12) ξV,[A] for any valuation ring W with V ⊆ W ⊆ F . There is another interpretation of ℓV,W that will be needed later: Let w be the valuation on F associated to W , let (Fh,w, wh) be the Henselization of (F, w), let Dh be the associated division be the residue division algebra of Dh for the valuation w′ on Dh algebra of A⊗F Fh,w, and let Dh w′ w extending wh on Fh,w. Let u be the valuation of eV on F ℓV,W = the number of extensions of u to valuations of Z(Dh . Then, (3.12) This holds because S ∼= MtS (Dh centers. w′ ), see (3.3), so the F w -algebras S and Dh,w have isomorphic w′ ). w′ Note that the following conditions are equivalent: (a) Some (so every) Dubrovin valuation ring B of A extending V is integral over V . (b) ξV,[A] = 1. (c) Every Grater ring C of A with C ∩ F = V is a Dubrovin valuation ring. These conditions do not hold in general, but for given V and A, they hold for some nontrivial coarsening of V , see Prop. (3.3) below. Example 3.1. Let A be a central simple F -algebra, and let V be a rank 2 valuation ring of F . Let W be the rank 1 valuation ring with V $ W $ F . Then ξW,[A] = 1 by Prop. 3.3(iv) below. Let S be a Dubrovin valuation ring of A extending W . So, S is integral over W . Let S = S/J(S), which is a simple F = W/J(W )-algebra. Suppose the valuation ring U = V /J(W ) of F has exactly k different extensions U1, . . . , Uk to Z(S). The map B 7→ B/J(S) gives a one-to-one correspondence between the Dubrovin valuation rings B of A extending V and lying in S, and the Dubrovin valuation rings T of S with T ∩ F = U . For each such T , the intersection T ∩ Z(S) is a valuation ring of J(S) extending U , so it is one of the Uj. Moreover, T is integral over Uj since Uj has rank 1. Let ξV,[A] = k. B1, . . . , Bn be Dubrovin valuation rings of A extending V with each Bi ⊆ S. Let fBi = Bi/J(S), and letfBi∩Z(S) = Uj(i). Then, B1, . . . , Bn have the IP if and only if j(1), . . . , j(n) are all different. The ring C = Tn Tn i=1fBi is integral over U , if and only if n = k and {j(1), j(2), . . . , j(n)} = {1, . . . , n}. Thus, i=1 Bi is a Grater ring if and only if further C is integral over V , if and only if In general, there is no valuation associated to a Dubrovin valuation ring B of A. However, a sig- nificant exception occurs when B is integral over Z(B): Morandi introduced in [M2] a type of value function associated to any Dubrovin valuation ring integral over its center. Let α : A → Γ ∪ {∞} be a surmultiplicative v-value function. We say that α is a Morandi value function if (1) A0 is a simple ring; (2) Γα = α(st(α)), where st(α) = {a ∈ A× α(a−1) = −α(a)}. If v = αF , then v is a valuation on F and we call α a v-Morandi value function. Morandi showed in [M2, Th. 2.4] (or see [MMU, Th. 23.3, p. 135]) the following: If α is a Morandi value function on A, then its associated ring is a Dubrovin valuation ring integral over its center and Rα = {x ∈ A α(x) ≥ 0} (3.13) ΓRα = Γα. 26 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Conversely, to every Dubrovin valuation ring B of A integral over its center, there exists a Morandi value function α on A such that Rα = B; moreover, α is uniquely determined by B (see [M2, Th. 2.3, Prop. 2.6] or [MMU, Th. 23.2, p. 134]). (By contrast, we will see in Example 5.1 below that if β is a gauge on A, then its gauge ring Rβ does not always determine β.) The connection between gauges and Morandi value functions was shown in [TW1, Prop. 2.5]: If A is a central simple F -algebra and v is a valuation on F defectless in A, then a surmultiplicative v-value function α is a Morandi value function if and only if α is a gauge with A0 simple. The following result generalizes this for simple but not necessarily central simple algebras. If α is a gauge on a semisimple algebra A, then since grα(A) is graded semisimple, the degree zero part A0 of grα(A) must be semisimple (cf. [TW1, Prop. 2.1]). But A0 is not necessarily simple, even if A is simple, as is illustrated in Ex. 1.1. Let (3.14) ω(α) = the number of simple components of A0. Theorem 3.2. Suppose v is a valuation on F defectless in a simple F -algebra A. Let v1, . . . , vr be all the extensions of v to K = Z(A). only if B1, . . . , Br have the IP. When this occurs, ω(α) = r. (i) For i ∈ {1, . . . , r}, let αi be a vi-Morandi value function on A and let Bi be the Dubrovin valuation ring associated to αi. Let α = min(cid:0)α1, . . . , αr(cid:1). Then α is a v-gauge on A if and value functions αi for i ∈ {1, . . . , r} such that α = min(cid:0)α1, . . . , αr(cid:1). (ii) Let α be a v-gauge on A with ω(α) = r. Then there are uniquely determined vi-Morandi Proof. For each i and j, let Vi be the valuation ring of vi, and let Vij = Vi·Vj, with its associated valuation vij, which is the finest common coarsening of vi and vj. Let eVi = Vi/J(Vij) ⊆ Vij, and define eVj analogously. Note that eVi and eVj are independent in Vij, i.e., eVi·eVj = Vij, since Vi·Vj = Vij. (i) Suppose α is a v-gauge on A. By Th. 2.8, for every pair i, j, the gauges αi and αj have the same vij-coarsening, call it αij. So, for the gauge rings, we have Rαi ⊆ Rαij and Rαj ⊆ Rαij . Since Bi = Rαi and likewise for j, we must have Bij ⊆ Rαij . Hence, Z(Bij) = Bij ∩ K ⊆ Rαij ∩ K = Z(Rαij ) = Vij. Since we always have Vij = Vi·Vj ⊆ Z(Bij), we obtain Vij = Z(Bij). The field Z(Bij) is a finite- degree extension of the field Vij. Since the valuation rings eVi and eVj are independent in Vij ⊆ Z(Bij) and Z(fBi)∩Vij = eVi, it follows that the valuation rings Z(fBi) and Z(fBi) are independent in Z(Bij). Hence, Bi and Bj have the IP. Therefore, B1, . . . , Br have the IP by (3.6). Moreover, since by Th. 2.2 grα(A) ∼=g grα1(A) × . . . × grαr (A) is the simple residue ring of Rαi, we have ω(α) =Pr i=1 ω(αi) =Pr Conversely, suppose B1, . . . , Br have the IP. For any distinct i, j ∈ {1, . . . , r} let ∆ij be the convex subgroup of the divisible hull Γ of Γv associated to Vij and let θij : Γ → Γ/∆ij be the canonical map. Since Bi is the gauge ring Rαi, by Prop. 1.4, Rθij◦αi = Bi·Vij; likewise, Rθij◦αj = Bj·Vij. Since Bi·Vij and Bj·Vij are Dubrovin valuation rings of A integral over Vij, it follows that θij ◦ αi and θij ◦ αj are Morandi value functions. Moreover, since Bi and Bj have the IP, their overrings Bi·Vij and Bj·Vij also have the IP by [MMU, Th. 16.8, p. 92]. But the integrality of Bi·Vij over Vij implies that ξVij ,[A] = 1. By using this or [MMU, Lemma 16.5, p. 90], it follows that Bi·Vij = Bj·Vij. Therefore, θij ◦ αi = θij ◦ αj, because a Morandi value function is completely (3.15) and each Aαi 0 i=1 1 = r. VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 27 determined by its associated Dubrovin valuation ring (see [MMU, Prop. 23.6, p. 135]). Thus, it follows from Th. 2.8 that α is a v-gauge. (ii) By Th. 2.2 there exist vi-gauges αi on A for i ∈ {1, . . . , r} such that α = min(cid:0)α1, . . . , αr(cid:1) and (3.15) holds. Thus ω(α) = ω(α1) + . . . + ω(αr). Since ω(α) = r, we must have each ω(αi) = 1; it follows from [TW1, Prop. 2.5] that each αi is a vi-Morandi value function. The uniqueness of the αi follows from Cor. 2.5. (cid:3) For proofs of some of the following theorems we need the notions of jump rank and jump prime ideals, defined as follows: Let A be a central simple F -algebra and let V be the valuation ring of a nontrivial valuation v on F . For each nonzero prime ideal P of V , let vP be the valuation on F with valuation ring VP , let field Fh,P be a Henselization of F with respect to vP , and let nP be the matrix size of A ⊗F Fh,P . So, 1 ≤ nP ≤ deg A. For prime ideals P ⊆ Q we have nP ≤ nQ since Fh,P embeds in Fh,Q. We say that P is a jump prime ideal of v for A if nP < nQ for every prime ideal Q % P ; we then say that vP is a jump valuation of v for A. The jump rank of v for A is defined to be j(v, A) = the number of jump prime ideals of v for A. Note that if a positive integer m = nP for some P , and P is the union of all the prime ideals Q with nQ = m, then P is a prime ideal of V , since the Q's are linearly ordered, and nP = m since Fh,P is the direct limit of the Fh,Q; so P is the unique jump prime ideal of v with nP = m. Thus, j(v, A) = (cid:12)(cid:12){nP P is a nonzero prime ideal of V }(cid:12)(cid:12), and 1 ≤ j(v, A) ≤ deg A. The jump rank is useful for induction arguments, since it is always finite even when the valuation v has infinite rank. If P1 $ P2 $ . . . $ Pj(v,A) are all the distinct jump prime ideals of v for A, we call Pi the i-th jump prime of v for A. Note that Pj(v,A) = J(V ). If v is the trivial valuation on F , we set j(v, A) = 0. More information on the jump rank can be found in [W1] or [MMU]. integral over VP. Proposition 3.3. Let F be a field, and let v be a nontrivial valuation on F with valuation ring V . Let A be a central simple F -algebra, and let B be a Dubrovin valuation ring of A with center V . (i) There is a unique nonzero prime ideal P of V maximal with the property that B·VP is (ii) Let Q be any prime ideal of V . Then, B·VQ is integral over VQ if and only if Q ⊆ P. (iii) P is a jump prime ideal of v for A. (iv) If j(v, A) = 1, then P = J(V ), so B is integral over V . (v) If P 6= J(V ), then ℓV,VP > 1. Proof. (i) and (v) The existence of P with the maximal property is given in [MMU, Prop. 12.4, p. 72], where it is also proved that if P 6= J(V ) (so B is not integral over V ), then ℓV,VP > 1. The prime ideal P is unique with the maximal property since the prime ideals of the valuation ring V are linearly ordered. (ii) For a prime ideal Q of V , if Q % P, then B·VQ is not integral over VQ by the maximality of P. But if Q ⊆ P, then VQ ⊇ VP, so by (3.11) ξVQ,[A] ≤ ξVP,[A] = 1. Hence, ξVQ,[A] = 1, showing that B·VQ is integral over VQ. (iii) If P = J(V ), then P is a jump prime ideal of v for A. Assume now that P 6= J(V ). Let W = VP and let S = B·W , which is a Dubrovin valuation ring of A with center W . Let 28 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH (3.16) Hence, these residue rings have the same matrix size, i.e., t eB = tB. We have S = MtS (E) for some eB = B/J(S), which is a Dubrovin valuation ring of S = S/J(S). Note that eB = (cid:0)B/J(S)(cid:1)(cid:14) J(B/J(S)) ∼= B/J(B) = B. division ring E. If C is a Dubrovin valuation ring of E with center the valuation ring Z(eB), then MtS (C) and eB are Dubrovin valuation rings of S with the same center. Hence, eB ∼= MtS (C), which implies that eB ∼= MtS (C). So, for the matrix sizes, t eB = tS tC. Thus, we have tS t eB = tB (3.17) (cf. [W1, Th. E(ii)]). Since ξV,[A] = nB/tB, formula (3.10) above can be restated nB = nS(tB/tS)(n eB/t eB) ℓV,W . Since tB/tS ≥ 1 by (3.17) and n eB/t eB ≥ 1, and ℓV,W > 1 from (v), we have nB > nS. Now let P ′ be any prime ideal of V with P $ P ′, and let V ′ = VP ′ and B′ = B·V ′. The prime ideals Q of V ′ coincide with the prime ideals of V lying in P ′, and for any such Q, we have V ′ Q = B·VQ. Hence P is maximal among the prime ideals Q of V ′ with B′·V ′ Q. Therefore, the argument just given showing that nB > nS shows likewise that nB′ > nS. Since this is true for every P ′ % P, this P is a jump prime ideal of v for A. Q = VQ and B′·V ′ Q integral over V ′ (iv) If j(v, A) = 1, then J(V ) is the only jump prime ideal of v, so P = J(V ) and B = B·VP is (cid:3) integral over VP = V . The following general setup occurs repeatedly in the proofs of the next three theorems: Setup 3.4. Let F be a field with a valuation v and let A be a central simple F -algebra with a v-gauge α. So, v is defectless in A by Prop. 1.9. Let w be a coarsening of v. Then w is also defectless in A by Prop. 1.8. Let V be the valuation ring of v, and W the valuation ring of w. Let = W/J(W ), and let u be the valuation associated eV = V /J(W ), which is a valuation ring of F to eV . Let β be the coarsening of α such that βF = w. By Prop. 1.5, α induces a u-gauge α0 0 . Let C1, . . . , Cω(β) be the simple components of Aβ on Aβ for i = 1, . . . , ω(β). As noted after Prop. 1.3 above, each αi 0 . Let αi 0 is a u-gauge on Ci and 0 be the restriction of α0 to Ci w (3.18) for all a1 ∈ C1, . . . , aω(β) ∈ Cω(β). Moreover, 0 ) ∼=g grα1 (3.19) α0(a1, . . . , aω(β)) = min(cid:0)α1 grα0(Aβ 0 (C1) × . . . × gr αω(β) 0 (Cω(β)). 0(a1), . . . , αω(β) 0 (aω(β))(cid:1), By (1.17), the associated graded algebras of α and α0 have the same degree zero part. It then follows from (3.19) that (3.20) ω(α) = ω(α0) = ω(α1 0) + . . . + ω(αω(β) 0 ). We next obtain a description of the simple components Ci. Let (Fh,w, wh) be a Henselization of (F, w). Consider the scalar extension Ah = A ⊗F Fh,w ∼= Mn(Dh), where Dh is a central division Fh,w-algebra. Then, wh is defectless in Dh by definition as w is defectless in A, and the Henselian valuation wh extends to a valuation w′ on Dh. Let Rw′ be the invariant valuation ring of Dh associated to w′ and let Dh = Rw′/J(Rw′), the residue division ring. On the other hand, it follows from [TW1, Cor. 1.26] that βh = β ⊗ wh is a wh-gauge on Ah and grβh(Ah) ∼=g grβ(A). VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 29 Write Ah = EndDh(M ) for some finite-dimensional Dh-vector space M . Since wh is Henselian and is defectless in Dh, [TW1, Th. 3.1] says that βh is an End-gauge as in Ex. 1.1, i.e., there is a w′-norm η on M such that grβh(Ah) = Endgrw′ (Dh)(grη(M )). By [TW1, Prop. 2.1], (3.21) Aβ 0 = (cid:0)Ah(cid:1)βh 0 = (cid:0)Endgrw′ (Dh)(grη(M ))(cid:1)0 ∼= kQi=1 Mri(D0), where D0 is the degree zero part of grw′(Dh), which is Dh; also, k is the number of cosets of Γw′ in Γη. By comparing (3.21) with the decomposition Aβ 0 = C1 × . . . × Cω(β) from (3.19), we conclude that k = ω(β) and that each Ci = Mri(D0), after re-indexing if necessary. Thus, Z(C1) = . . . = Z(Cω(β)) = Z(Dh). Let K denote this common field. (We also have K = Z(S) for any Dubrovin valuation ring S of A with Z(S) = W .) As noted in (3.12), there are ℓV,W extensions of u to K, which we denote u1, . . . , uℓV,W . Let Uj be the valuation ring of uj. By Th. 2.2 applied to the u-gauge αi 0 on Ci, there exist uj-gauges αij 0 on Ci for j ∈ {1, . . . , ℓV,W} such that (3.22) and also We thus have (3.23) Hence, with (3.20), αi 0(a) = min(cid:0)αi1 iℓV,W 0 (a), . . . , α 0 (a)(cid:1) for all a ∈ Ci, grαi 0 (Ci) ∼=g grαi1 0 (Ci) × . . . × gr iℓV,W α 0 (Ci). ω(αi 0) = ω(αi1 iℓV,W 0 ) + . . . + ω(α 0 ). (3.24) ω(α) = ω(α0) = ω(β)Pi=1 ℓV,WPj=1 ω(αij 0 ). Because the Ci are Brauer equivalent central simple K-algebras, we have (3.25) ξUj,[Ci] = ξUj,[C1] for all i ∈ {1, . . . , ω(β)}, j ∈ {1, . . . , ℓV,W}. We claim also that (3.26) ξUj,[Ci] = ξU1,[Ci] for all i ∈ {1, . . . , ω(β)}, j ∈ {1, . . . , ℓV,W}. We check this for Dh, which is Brauer equivalent to Ci. Recall from [JW, Prop. 1.7] that ) acts K = Z(Dh) is a normal field extension of Fh,w w ) transitively on the set {u1, . . . , uℓV,W } of all extensions of u on F with uj = u1◦τ . By [JW, Prop. 1.7], there is d ∈ D× h whose associated automorphism ıd : Dh → Dh (a 7→ dad−1 for a ∈ Rw′) restricts to τ on K. Then, ıd takes Dubrovin valuation rings of Dh with center Uj to those with center U1; so ξUj,[Dh] = ξU1,[Dh]. Thus, from the Brauer equivalence of Ci and Dh, , and hence the Galois group G(K/F to K. Choose τ ∈ G(K/F wh = F w w w as claimed, proving (3.26) ξUj,[Ci] = ξUj,[Dh] = ξU1,[Dh] = ξU1,[Ci], The following diagram illustrates some of the objects considered here. We write v ≥ w to indicate that v is a refinement of w; likewise for α ≥ β. 30 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH A α ≥ β ❏❏❏❏❏❏❏❏ Aβ 0 α0 Ci αi1 0 , . . . iℓV,W , α 0 F v ≥ w Z(Aβ 0 ) ❏❏❏❏❏❏❏❏ K u1, . . . ✻✻✻✻✻ , uℓV,W u W u W Theorem 3.5. Let F be a field with a valuation v and associated valuation ring V , and let A be a central simple F -algebra. Then, for any v-gauge α on A, ω(α) ≥ ξV,[A]. Proof. We write ξV for ξV,[A]. The proof is by induction on ξV . If ξV = 1, then we clearly have ω(α) ≥ ξV . We can thus assume that ξV > 1. By [MMU, Th. 16.14, p. 94], there exist Dubrovin valuation rings R1, . . . , RξV of A having the IP such that each Z(Rt) = V and R1 ∩ . . . ∩ RξV is a Grater ring integral over V . For each t ∈ {1, . . . , ξV }, by Prop. 3.3(i) and (v) there ex- ists a Dubrovin valuation ring St of A containing Rt and minimal with the property that St is integral over Wt = Z(St) and we have ℓV,Wt ≥ 2. Now by [MMU, Cor. 16.6, p. 91], we have S1 = . . . = SξV , hence W1 = . . . = WξV . Let S = St and W = Wt, for all t. The integrality of S over W yields that ξW,[A] = 1. We use the valuation w of W in Setup 3.4; so, β is the w- coarsening of α. Let fRt = Rt/J(S). SinceTξV t=1fRt is integral over eV . The valuation rings Z(fR1), . . . , Z(gRξV ) must include all the extensions of eV to K = Z(S), because Z(fR1) ∩ . . . ∩ Z(gRξV ) is integral over eV . These extensions are thus the valuation rings any j ∈ {1, . . . , ℓV,W}, choose an Rt with Z(fRt) = Uj. It follows from (3.10) with B = Rt that t=1 Ri is integral over V , we haveTξV U1, . . . , UℓV,W of the extensions u1, . . . , uℓV,W of u in Setup 3.4, but with possible repetitions. For ξV = ξW,[A] ξUj,[S] ℓV,W = ξUj,[S] ℓV,W . (3.27) Since ℓV,W ≥ 2, we conclude that each ξUj,[S] < ξV . Since each Ci is Brauer equivalent to S, we have ξUj ,[S] = ξUj,[Ci]. Since each αij 0 constructed in the Setup 3.4 is a uj-gauge on Ci, we have by the induction hypothesis ω(αij 0 ) ≥ ξUj,[Ci]. The equality (3.27) (or (3.26)) shows that ξUj,[Ci] = ξU1,[Ci] = ξU1,[S] for all i, j. Thus, it follows from (3.24) that ω(α) = ω(β)Pi=1 ω(αij 0 ) ≥ ω(β) ℓV,W ξU1,[S] ≥ ℓV,W ξU1,[S] = ξV , ℓV,WPj=1 (cid:3) which completes the proof. Definition 3.6. Let v be a valuation on a field F , and let α be a v-gauge on a central simple F -algebra A. Then, we call α a minimal gauge if ω(α) = ξV,[A]. Remark 3.7. Note that if ω(α) = 1, i.e., Aα 0 is a simple ring, then by [TW1, Prop. 2.5] α is a Morandi value function on A with associated Dubrovin valuation ring Rα. Conversely, if α′ is a Morandi value function on A with α′F = v and v is defectless in A, then by [TW1, Prop. 2.5] α′ is a v-gauge and Rα′ is the Dubrovin valuation ring associated to α′; so, Aα′ is the simple 0 ring Rα′/J(Rα′), whence ω(α′) = 1. VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 31 Theorem 3.8. Let F be a field with a valuation v and associated valuation ring V . Let A be a central simple F -algebra with a v-gauge α. Let β be a coarsening of α, and let w = βF , which is a valuation on F coarser than v. Let W be the valuation ring of w. Then, Consequently, if α is a minimal gauge, then so is β. ω(α)/ω(β) ≥ ξV,[A]/ξW,[A]. Proof. We use the notation of Setup 3.4 with the v, α, w, β given here. By Th. 3.5, we have ω(αij 0 ) ≥ ξUj,[Ci] for i ∈ {1, . . . , ω(β)} and j ∈ {1, . . . , ℓV,W}. Equations (3.25) and (3.26) above It then follows show that ξUj,[Ci] = ξU1,[C1] for each i ∈ {1, . . . , ω(β)} and j ∈ {1, . . . , ℓV,W}. from (3.24) that ω(α) ≥ ω(β) ℓV,W ξU1,[C1]. Choose a Dubrovin valuation ring B of A extending V , and let S = B·W . Then, for the Dubrovin valuation ring eB = B/J(S) the valuation ring Z(eB) is an extension of V /J(W ) to Z(S), so Z(eB) = Uj for some j. Then, (3.10) with this B yields ξV,[A] = ξW,[A]ℓV,W ξUj,[S] = ξW,[A]ℓV,W ξU1,[C1]. Hence, ω(α)(cid:14)ω(β) ≥ ℓV,W ξU1,[C1] = ξW,[A] ℓV,W ξU1,[C1](cid:14)ξW,[A] = ξV,[A](cid:14)ξW,[A]. Suppose α is a minimal gauge. Then, ω(α) = ξV,[A]. The inequality just proved then shows that ω(β) ≤ ξW,[A]. The reverse inequality is given by Th. 3.5. Hence, ω(β) = ξW,[A], showing that β is a minimal gauge. (cid:3) The next theorems give the fundamental connection between minimal gauges and Grater rings. Theorem 3.9. Let F be a field with a valuation v and let A be a central simple F -algebra with a minimal v-gauge α. Then, the gauge ring Rα = {a ∈ A α(a) ≥ 0} is a Grater ring of A with center the valuation ring V of v. Proof. We may assume that the valuation v on F is nontrivial. The proof is by induction on the jump rank j(v, A). If j(v, A) = 1, then any Dubrovin valuation ring of A extending V is integral over V by Prop. 3.3(iv). Thus, 1 = ξV,[A] = ω(α), hence α is a Morandi value function by Remark 3.7. Therefore, Rα is a Dubrovin valuation ring integral over V , so Rα is a Grater ring. We can thus assume j(v, A) > 1. Let Q be the (j(v, A) − 1)-st jump prime ideal of v for A, let W = VQ be the corresponding valuation ring, and let w be the valuation on F associated to W . We write ξV for ξV,[A] and ξW for ξW,[A]. The jump prime ideals of w for A are the same as the jump prime ideals of v for A, except that J(V ) is excluded. Thus, j(w, A) = j(v, A) − 1. Let β be a coarsening of α such that βF = w. We use these v, w, α, β in Setup 3.4. Since α is a minimal gauge, by Th. 3.8 β is also a minimal gauge. Thus, by the induction hypothesis, Rβ is a Grater i=1 Ri, where R1, . . . , RξW are Dubrovin valuation rings having the IP such that each Z(Ri) = W. By [M3, Lemma 3.2], there is an isomorphism ring with center W ; hence, Rβ =TξW (3.28) Aβ 0 ∼−→ Ri/J(Ri) ξWQi=1 given by x + J(Rβ) 7→ (cid:0)x + J(R1), . . . , x + J(RξW )(cid:1). After re-indexing if necessary, we can write Ci = Ri/J(Ri), as in Setup 3.4. By [W1, Cor. E], jump prime ideals of uj for Ci pull back to jump primes ideals of v for A properly containing Q; hence, j(uj , Ci) ≤ j(v, A) − j(w, A) = 1. 32 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Therefore, ξUj,[Ci] = 1 for all i, j by Prop. 3.3(iv). Thus, formula (3.10) reduces to ξV = ξW ℓV,W . Since ω(α) = ξV and ω(β) = ξW , it follows from (3.24) that ω(αij 0 ) = 1 for all i ∈ {1, . . . , ξW} and j ∈ {1, . . . , ℓV,W}. Hence each αij 0 is a Morandi value function by Remark 3.7. Let Sij = {x ∈ Ci αij 0 (x) ≥ 0}, which is a Dubrovin valuation ring of Ci with Z(Sij) = Ui. Let Bij = {a ∈ A a + J(Ri) ∈ Sij}, which is a Dubrovin valuation ring of A with Z(Bij) = V . We prove that the set of Dubrovin valuation rings Bij for i = 1, . . . , ξW and j = 1, . . . , ℓV,W have the IP. By Th. 3.2, Si1, . . . , SiℓV,W have the IP for any i. Thus, Bi1, . . . , BiℓV,W have the IP for each i, by [MMU, Prop. 16.4, p. 90]. Now let i, q ∈ {1, . . . , ξW} with i 6= q. Note that Bij ⊆ Ri and Bqr ⊆ Rq for j, r ∈ {1, . . . , ℓV,W}. Since Ri and Rq are incomparable and have the IP, Bij and Bqr are also incomparable and have the IP by [MMU, Th. 16.8, p. 92]. Therefore, the Dubrovin valuation rings {Bij}i,j are pairwise incomparable and have the IP. We claim that Rα =TξW ℓV,W i=1 j=1 Bij. Since the gauge ring Rα is integral over its center V , it then follows that the intersection of the Bij is a Grater ring, which completes the proof that Rα is a Grater ring. To prove the claim, note first that (3.29) because α(x) ≥ 0 implies β(x) ≥ 0 and β(x) > 0 implies α(x) > 0. Likewise, for each i ∈ {1, . . . , ξW} and j ∈ {1, . . . , ℓV,W} we have J(Rβ) ⊆ J(Rα) ⊆ Rα ⊆ Rβ, Hence, using (3.28) for the first equality as Aβ 0 = Rβ/J(Rβ ) J(Ri) ⊆ J(Bij) ⊆ Bij ⊆ Ri. ξW ,ℓV,WTi,j=1 ξWTi=1 J(Ri) ⊆ Bij ⊆ Ri = Rβ. (3.30) J(Rβ) = ξWTi=1 that a ∈ Rα if and only if a ∈TξW ℓV,W (3.31) Let a ∈ Rβ\J(Rβ). In view of (3.29) and (3.30), the proof of the claim will be completed by showing 0 with R1/J(R1) × . . . × RξW /J(RξW ), 0 = Rβ/J(Rβ) and via the isomorphism (3.28). Thus, by the definition of the gauge α0 on Aβ by (3.18), i=1 j=1 Bij. We identify Aβ α(a) = α0(cid:0)a + J(Rβ)(cid:1) = α0(cid:0)a + J(R1), . . . , a + J(RξW )(cid:1) = min j=1 Sij. Thus, a ∈ Rα if and only if a + J(Ri) ∈ TℓV,W = TℓV,W Hence, α(a) ≥ 0 if and only if αi Rαi a + J(Ri) ∈ Sij if and only if a ∈ Bij. Therefore, a ∈ Rα if and only if a ∈TξW ℓV,W 0(a + J(Ri)) ≥ 0 for i = 1, . . . , ξW . By (3.22), the gauge ring j=1 Sij, for i = 1, . . . , ξW . But i=1 j=1 Bij. This (cid:3) proves the claim, which, as noted above, implies that Rα is a Grater ring. 0(a + J(Ri))(cid:1). 1≤i≤ξW(cid:0)αi 0 Remark 3.10. Note that the Dubrovin valuation rings Bij of Th. 3.9 are uniquely determined by α, because they are the localizations of Rα re its maximal ideals. The following result shows that Th. 3.9 has a converse. Theorem 3.11. Let F be a field with a valuation v and associated valuation ring V . Let A be a central simple F -algebra such that v is defectless in A. Then a subring C of A is a Grater ring of A with center V if and only if C = Rα for some minimal v-gauge α on A. VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 33 The proof requires the existence of minimal gauges on defectless F -algebras, which will be proved in §4. Proof. If α is a minimal v-gauge on A, we have seen in Th. 3.9 that Rα is a Grater ring with center V . For the converse, let C be a Grater ring of A with center V . Because v is defectless in A, Th. 4.3 below shows that there exists a minimal v-gauge β on A. By Th. 3.9, its gauge ring Rβ is a Grater ring of A with center V . By [MMU, Th. 16.15, p. 96], C and Rβ are conjugate in A, i.e., C = qRβq−1 for some q ∈ A×. Composition of β with the inner automorphism ıq : A → A defined by x 7→ q−1xq yields a minimal v-gauge α = β ◦ ıq on A such that Rα = qRβq−1 = C. (cid:3) Note that the gauge α of the theorem is not uniquely determined by C, as the example in §5 below demonstrates. 4. Existence of minimal gauges In this section we prove the existence of minimal gauges on defectless semisimple algebras. First, we extend the concept of minimal gauge to semisimple algebras. Proposition 4.1. Let (F, v) be a valued field and A be a finite-dimensional simple F -algebra. Let v1, . . . , vr be all the extensions of v to Z(A), and let Vi be the valuation ring of vi. Let α be a v-gauge on A. Then, (4.1) ω(α) ≥ ξV1,[A] + . . . + ξVr,[A]. Proof. Let αi be the vi-component of α for i = 1, . . . , r. Then, by the graded algebra isomor- phism (2.1) and the inequality of Th. 3.5 for each i, (4.2) ω(α) = ω(α1) + . . . + ω(αr) ≥ ξV1,[A] + . . . + ξVr,[A]. (cid:3) Definition 4.2. Let v be a valuation on a field F , and let α be a v-gauge on a simple (finite- dimensional) F -algebra A. We say that α is a minimal v-gauge on A if we have equality in (4.1). Note that (4.2) shows that α is a minimal v-gauge if and only if each component αi of α is a minimal vi-gauge. More generally, if A is semisimple, say A = A1 × . . . × Ak with each Ai simple, and β is a v-gauge on A, we say that β is a minimal v-gauge on A if each βi = βAi (as in (1.14)) is a minimal v-gauge on Ai. Theorem 4.3. If v is a valuation on F defectless in a finite-dimensional semisimple F -algebra A, then there exists a minimal v-gauge α on A with Γα ⊆ H(Γv). The general method in proving the theorem is to build up α inductively from minimal gauges for A for valuations on F coarser than v. The proof will begin after Prop. 4.7 and be completed after Lemma 4.9. Lemma 4.4. Let valuation w be a coarsening of v on F . Let A be a semisimple F -algebra with a w-gauge β. Let F ′ be a field containing F with a valuation w′ which is an immediate extension of w, and let v′ be the extension of v to F ′ that refines w′, so v′ is an immediate extension of v. Let A′ = A ⊗F F ′ and let β′ = β ⊗ w′, which is a w′-gauge on A′. Suppose A′ has a v′-gauge α′ whose w′-coarsening is β′. Then α = α′A is a v-gauge on A with w-coarsening β, and grα(A) ∼=g grα′(A′). 34 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Proof. Let u = v/w, which is the valuation on F on F ′w′ immediate extension of w, by [TW1, Cor. 1.26], β′ = β ⊗ w′ is a w′-gauge on A′ with , which coincides with u under the canonical isomorphism F induced by v on F ; likewise let u′ = v′/w′ . Because w′ is a w ∼= F ′w′ w grβ′(A′) ∼=g grβ(A) ⊗grw(F ) grw′(F ′) ∼=g grβ(A). For Γ = H(Γα′) and Λ = H(Γβ′), let ε : Γ → Λ be the map associated to the w-coarsening of v. For each λ ∈ Λ we have the λ-component A′β′ λ defined λ of grβ′(A′), and the u′-value function α′ λ on A′β′ by λ(a + A′β′ α′ if β′(a) = λ, if β′(a) > λ. >λ) = (cid:26) α′(a) λ. Since β = β′A, we can view Aβ ∞ λ is a u′-norm on A′β′ λ λ ⊆ A′β′ λ . So, αλ = α′ ε(γ), we thus have λAβ λ = A′β′ λ, which is a u-norm on Aβ λ by [TW2, Prop. 4.3]. Likewise α = α′A Since α′ is a v′-norm, each α′ induces the u-value function αλ on Aβ λ via the canonical . But since the canonical inclusion grβ(A) ֒→ grβ′(A′) is a graded inclusion; then clearly αλ = α′ isomorphism, we have Aβ λ. Since in addition β is a w-norm on A, by [TW2, Prop. 4.3] α is a v-norm on A. Moreover, α = α′A is surmultiplicative since α′ is surmultiplicative. There is a canonical algebra monomorphism ι : grα(A) ֒→ grα′(A′). γ =(cid:0)Aβ Take any γ ∈ Γ. It follows from the definitions that Aα ε(γ) and αε(γ) = α′ ε(γ)(cid:1)α′ = (cid:0)A′β′ hence, ι is a graded isomorphism. Thus, grα(A) is graded semisimple, since this is true for grα′(A′), as α′ is a v′-gauge. Therefore, α is a v-gauge. The w-coarsening of α = α′A is β′A = β. (cid:3) Lemma 4.5. Let A be a central simple F -algebra with j(v, A) = n > 1. Let P be the (n − 1)-st jump prime ideal of v for A, and let W = VP with its associated valuation w. Let S be a Dubrovin valuation ring of A with Z(S) = W , and let S = S/J(S). Let u = v/w, the residue valuation on F induced by v, and let u1, . . . , ur be the valuations on Z(S) extending u. Then u1, . . . , ur are pairwise independent valuations. γ = (cid:0)Aβ ε(γ)(cid:1)αε(γ) ε(γ)(cid:1)αε(γ) . Since Aβ ε(γ) = A′β′ ε(γ) γ = A′α′ γ ; Aα w γ γ Proof. Let T and B be Dubrovin valuation rings of A with B ⊆ T $ S and Z(B) = V . Let Y = Z(T ) and let y be the valuation of Y . We have V ⊆ Y $ W . Let nB and tB be as given in (3.1) and (3.3). Let eB = B/J(S) which is a Dubrovin valuation ring of S = S/J(S). Recall from (3.17) that t eB = tB. (The proof of (3.17) is valid for any overring S of a Dubrovin valuation ring B.) Using this and ξV,[A] = nB/tB, formula (3.10) yields nB = ξW,[A] n eB ℓV,W . Likewise, by replacing B by T , we have nT = ξW,[A] n eT ℓY,W . Hence, (4.3) nB/nT = (cid:0)n eB/n eT(cid:1)(cid:0)ℓV,W /ℓY,W(cid:1). Because y is a coarsening of v, the Henselization Fh,y embeds in Fh,v. Hence nB/nT is a positive integer, as likewise is n eB/n eT . Also, because the valuation ring eV = V /J(W ) is a refinement of eY = Y /J(W ) in W , this eV has at least as many extensions to Z(S) as fW , i.e., ℓV,W ≥ ℓY,W . But, VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 35 since there are no jump prime ideals between J(Y ) and J(V ), we have nB = nT . Thus, in (4.3) the left side equals 1 and the right side is a product of positive integers. Hence, ℓV,W = ℓY,W . The last equality says that the number of extensions Ui of U = V /J(W ) to Z(S) equals the number of extensions of the coarser valuation Y /J(W ) to Z(S). Hence, any two distinct Ui and Uj have distinct coarsenings to extensions of Y /J(W ). Because this is true for every valuation ring Y with V ⊆ Y $ W , the finest common coarsening of Ui and Uj must be the trivial valuation ring. Hence, Ui and Uj are independent valuation rings in Z(S), so their corresponding valuations ui and uj are independent. (cid:3) y w in Z(D). Recall from [JW, Prop. 1.7] that Z(D) is normal over F The next proposition is the most difficult step in the proof of Th. 4.3. Here is the setup for the proposition: Let A be a central simple F -algebra with v defectless in A and j(v, A) = 2. Let W = VP , where P is the first jump prime ideal of v for A, and let w be the valuation of W . Assume that w is Henselian. Let β be a w-gauge of A with Γβ ⊆ H(Γw). Write A = EndD(M ), where D is the division algebra associated to A, and M is a finite-dimensional right D-vector space. Let y be w the valuation on D extending w on F , and let D = D . Let u be the residue valuation v/w on F induced by v, and let u1, . . . , ur be the extensions of u to Z(D). Let field S be the separable closure of F and that S is abelian Galois over F . Let K be the decomposition field of u1S over u (so K is also the decomposition field of each uiS over u, as G(S/F ) is abelian). For basic properties of decomposition fields, see [Ef, pp. 133-136]. Let L be a subfield of D that is an inertial lift of K over F . That is, L = K and [L:F ] = [K :F ]. Such an L exists (and is unique up to isomorphism) because w is Henselian and K is separable over F , cf. [JW, p. 135]. Let v1, . . . , vr be the extensions of v to L. Let C be the centralizer CD(L). Proposition 4.6. In the situation just described, let α be a v1-gauge on C = CD(L) with Γα ⊆ H(Γv). Then, A has a v-gauge ϕ with w-coarsening β such that Γϕ ⊆ H(Γv) and w w w w w y ω(ϕ) = rω(α)ω(β), where r is the number of extensions of u to Z(D). Moreover v1 on L has extension number 1 in C. Hence, α exists and can be chosen with ω(α) = 1. Proof. Let A′ = EndC (M ) ∼= A⊗F L, which contains A = EndD(M ) canonically, as an F -subalgebra. We will build ϕ as the restriction to A of a suitable End-gauge on A′. w w Since K is Galois over F ), cf. [JW, p. 135]. Let G = G(L/F ). Since G(S/K) is the decomposition group for u1S over u and every valuation on S has a unique extension to the purely inseparable field extension Z(D) of S, we have , its inertial lift L is Galois over F , with G(L/F ) ∼= G(K/F r = (cid:12)(cid:12){extensions of u from F = G(S/F w ): G(S/K) = G = [L:F ]. to Z(D)}(cid:12)(cid:12) = (cid:12)(cid:12){extensions of u from F w to S}(cid:12)(cid:12) w For each i ∈ {1, 2, . . . , r}, since K is the decomposition field of each uiS over u, this uiS is the unique extension of uiK to S, cf. [Ef, Prop. 15.1.2(b), p. 134]. Hence, u1K , . . . , urK are all distinct valuations of K. Let wL = yL, which is the unique extension of the Henselian valuation w to L. The valuations vi of L extending v on F are the composite valuations vi = uiK ∗ wL. Therefore, there are r distinct vi, since the uiK are distinct. The group G acts transitively on the vi since it acts transitively on the uiK , and this action is simply transitive as G = r. Since [L:F ] = r the 36 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Fundamental Inequality shows that each vi is an immediate extension of v. Thus, (4.4) grvi(L) ∼=g grv(F ) for i = 1, . . . , r. Note that since D is Brauer equivalent to A, j(v, D) = j(v, A) = 2 and v has the same jump prime ideals for D as for A; these are j(W ) and j(V ). The valuation ring Y of y on D is a Dubrovin valuation ring of D with Z(Y ) = Y ∩ F = W and Y /J(Y ) = D. By Lemma 4.5 (with D (resp. Y ) for the A (resp. S) of the lemma) the valuation rings u1, . . . , ur of Z(D) are pairwise independent; hence, u1K, . . . , urK are pairwise independent; hence, the finest common coarsening of any distinct vi and vj is wL. Let Γ = H(Γv) and Λ = H(Γw). So each Γvi ⊆ Γ, and for the valuation y on D extending w we have Γy ⊆ Λ. Since w is a coarsening of v on F , there is an epimorphism Γv → Γw. Let ε : Γ → Λ be the unique extension of this map to Γ; then, ε is surjective. Let ΓC,y be the value group of yC. By Skolem-Noether, for each ρ ∈ G there is dρ ∈ D× with dρℓd−1 ρ = ρ(ℓ) for all ℓ ∈ L. Since dρLd−1 ρ = L, we have dρCd−1 ρ = C. Moreover, D = Lρ∈G dρC = Lρ∈G Cdρ. (This is a standard fact about generalized crossed product algebras (see [T, Th. 1.3] or [JW, p. 156])). Since y is a valuation on the division algebra D, there is a canonical epimorphism Γy → G(cid:0)Z(D)/F ζ : Γy → G(K/F w w(cid:1) induced by conjugation by elements of D× (see [JW, Prop. 1.7]). Let ) be the composition of epimorphisms ζ : Γy −→ G(cid:0)Z(D)/F w(cid:1) −→ G(K/F w ). Since K = L, we have ΓC,y ⊆ ker(ζ). Thus, r = [L : F ] = [D : C] ≥ Γy : ΓC,y ≥ Γy : ker(ζ) = G(K/F w ) = [K : F w ] = r. So, equality holds throughout, showing that D is totally ramified over C and ker(ζ) = ΓC,y. Since ζ(y(dρ)) = ρ for all ρ ∈ G, the values y(dρ) are distinct modulo ΓC,y. Thus, there is a disjoint union decomposition (4.5) Γy = Fρ∈G y(dρ) + ΓC,y. Let δρ = y(dρ) ∈ Γy for all ρ ∈ G. The disjoint union decomposition for Γy in (4.5) shows that for any d =Pρ∈G dρcρ ∈ D with all cρ ∈ C, dρcρ(cid:1) = min y(cid:0) Pρ∈G Take any ρ ∈ G, and define vρ : L → Γ ∪ {∞} by ρ∈G(cid:0)y(dρ) + y(cρ)(cid:1) = min ρ∈G(cid:0)y(cρ) + δρ(cid:1). vρ(ℓ) = v1(d−1 ρ ℓdρ) = v1(ρ−1(ℓ)) for all ℓ ∈ L. Then, vρ is a valuation of L extending v on F . Since G acts simply transitively on {v1, . . . , vr} it follows that {v1, . . . , vr} = {vρ ρ ∈ G} and the vρ are distinct for distinct choices of ρ. Because w on F is Henselian, the w-gauge on A = EndD(M ) is an End-gauge by [TW1, Th. 3.1], i.e., β = End(θ) as in Ex. 1.1 for some y-norm θ : M → Λ ∪ {∞}. Let (m1, . . . , mn) be a splitting base for θ of the D-vector space M , and let πj = θ(mj) for j = 1, . . . , n. VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 37 j=1 mjdj) = min 1≤j≤n (πj + y(dj)) for all dj ∈ D. Hence, for any cjρ ∈ C for j = 1, . . . , n and So θ(Pn ρ ∈ G, θ(cid:0) nPj=1 Pρ∈G mjdρcjρ(cid:1) = min = min dρcjρ)(cid:1) 1≤j≤n(cid:0)πj + y(Pρ∈G ρ∈G(cid:0)δρ + y(cjρ)(cid:1)(cid:1) 1≤j≤n(cid:0)πj + min 1≤j≤n, ρ∈G(cid:0)πj + δρ + y(cjρ)(cid:1). min = This shows that (mjdρ)n j=1, ρ∈G is a splitting base for θ as a yC-norm on M . Since we can ad- just the mj by multiplication by any element of D×, we may assume that πi = πj whenever πi + Γy = πj + Γy. For each ρ ∈ G, pick γρ ∈ Γ with For j = 1, . . . , n, pick µj ∈ Γ with ε(γρ) = δρ. ε(µj) = πj. Choose the µj so that µi = µj whenever πi = πj. We now use the v1-gauge α on C to define an "α-v1-norm" η : M → Γ ∪ {∞} as in Lemma 1.2. For all cjρ ∈ C, set η(cid:0) nPj=1 Pρ∈G mjdρcjρ(cid:1) = min 1≤j≤n, ρ∈G(cid:0)µj + γρ + α(cjρ)(cid:1). ε ◦ η(cid:0) nPj=1 Pρ∈G Since α is a v1-gauge on C, its coarsening ε ◦ α is a wL-gauge on C by [TW2, Prop. 4.3]. But since the valuation wL on L extends to a valuation on C, by [TW1, Cor. 3.2] that valuation is the only wL-gauge on C; hence, ε ◦ α = yC. Thus, min i.e., ε ◦ η = θ. Now, let ψ = End(η), the v1-End-gauge on A′ = EndC(M ) determined by η, as in Lemma 1.2. So, for f ∈ A′, ψ(f ) = min mjdρcjρ(cid:1) = m ∈M \{0}(cid:0)η(f (m)) − η(m)(cid:1) = 1≤j≤n, ρ∈G(cid:0)πj + δρ + y(cjρ)(cid:1) = θ(cid:0) nPj=1 Pρ∈G mjdρcjρ(cid:1), 1≤j≤n, ρ∈G(cid:0)η(f (mjdρ)) − µj − γρ(cid:1). min Let ϕ = ψA. We will show that ϕ is the desired v-gauge on A. (4.6) f (mj) = We claim first that ϕ is a v-norm. For this, take any f ∈ A and write nPi=1 Pσ∈G where each dij ∈ D and each cijσ ∈ C. For any σ, ρ ∈ G, we have d−1 in C. That is, dσdρ = dσρtσ,ρ, for some tσ,ρ ∈ C ×. Then, midσcijσ(cid:1)dρ(cid:1) − µj − γρ(cid:17) ϕ(f ) = ψ(f ) = nPi=1 midσcijσ, midij = min min 1≤j≤n, ρ∈G(cid:16)η(cid:0)(cid:0) nPi=1 Pσ∈G 1≤j≤n, ρ∈G(cid:16)η(cid:0) nPi=1 Pσ∈G 1≤j≤n, ρ∈G(cid:16) min 1≤i,j≤n; ρ,σ∈G(cid:0)α(tσ,ρd−1 min min 1≤i≤n, σ∈G(cid:0)α(tσ,ρd−1 midσρ(tσ,ρd−1 ρ cijσdρ)(cid:1) − µj − γρ(cid:17) ρ cijσdρ) + µi + γσρ(cid:1) − µj − γρ(cid:17) ρ cijσdρ) + µi − µj + γσρ − γρ(cid:1). = (4.7) = = σρ dσdρ centralizes L, so lies for all c ∈ C, min 1≤i,j≤n; σ∈G ϕ(eij ◦ λdσ ◦ λcijσ ). 38 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH The choice of D-base (mj)n j=1 of M gives an isomorphism A = EndD(M ) ∼= Mn(D), which we use to interpret formula (4.7). In A we have the "matrix units" eij for i, j ∈ {1, . . . , n}, defined by eij(mj) = mi and eij(mk) = 0 for k 6= j. We also have an embedding λ : D → A given by d 7→ λd, where λd(mj) = mjd for all j. (So, λd(mjb) = mjdb for all b ∈ D, so λd ◦ λb = λdb. Clearly λd ◦ eij = eij ◦ λd for all i, j and all d ∈ D.) This λ corresponds to the diagonal embedding of D in Mn(D). The f in (4.6) above can be described as f =Pn ρ cijσdρ) + µi − µj + γσρ − γρ(cid:1). nPj=1 Pσ∈G j=1Pσ∈G eij ◦ λdσ ◦ λcijσ . Thus, formula (4.7) becomes ϕ(cid:0) nPi=1 This holds for all choices of cijσ in C. In particular, fixing i, j, σ, (4.8) i=1Pn eij ◦ λdσ ◦ λcijσ(cid:1) = ρ∈G(cid:0)α(tσ,ρd−1 eij ◦ λdσ ◦ λcijσ(cid:1) = nPj=1 Pσ∈G ϕ(cid:0) nPi=1 min 1≤i,j≤n; ρ,σ∈G(cid:0)α(tσ,ρd−1 ρ cdρ) + µi − µj + γσρ − γρ(cid:1), ϕ(eij ◦ λdσ ◦ λc) = min so formula (4.8) shows that (4.9) Now, A is a right C-vector space via λ, i.e., f·c = f◦λc for f ∈ A and c ∈ C, and (eij ◦ λdσ )1≤i,j≤n; σ∈G is a C-base of A. Formula (4.9) shows that the direct sum decomposition of A into 1-dimensional right C-vector spaces A =Li,j,σ(eij ◦ λdσ ) · C is a splitting decomposition for the v-value function ϕ on A, i.e., grϕ(A) ∼=g Li,j,σ grϕ(cid:0)(eij ◦ λdσ ) · C(cid:1). Therefore, to show that ϕ is a v-norm on A, it suffices to show that its restriction to each 1- dimensional C-subspace (eij ◦ λdσ ) · C is a v-norm. For this, fix i, j ∈ {1, . . . , n} and σ ∈ G, let h = eij ◦ λdσ , and let H = h·C. For c ∈ C, we have (4.10) (αρ(c)), ϕ(h·c) = ϕ(h ◦ λc) = min ρ∈G where αρ(c) = α(tσ,ρd−1 ρ cdρ) + τρ, with τρ = µi − µj + γσρ − γρ. Since α is a v1-value function on C, we have for any c ∈ C and ℓ ∈ L, αρ(cℓ) = α(tσ,ρd−1 ρ cdρd−1 ρ ℓdρ) = αρ(c) + v1(d−1 ρ cdρ) = αρ(c) + vρ(ℓ). Hence, αρ is a vρ-value function on C. The function g : C → C given by c 7→ tσ,ρd−1 ρ cdρ is an F -vector space isomorphism with αρ(c) = α(g(c)) + τρ for all c ∈ C. So, for any γ ∈ Γ, g maps C αρ >(γ−τρ). Hence, g induces a bijection gr(g) : grαρ(C) → grα(C)(−τρ), in the grade shift notation of (1.1); clearly g is a graded grv(F )- vector space isomorphism. Since grv1(L) = grv(F ) = grvρ(L) by (4.4), and α is a v1-norm on C, we have ≥γ bijectively to C α >γ bijectively to C α ≥(γ−τρ) and C αρ dimgrvρ (L) grαρ(C) = dimgrv(F ) grαρ (C) = dimgrv(F )(cid:0) grα(C)(−τρ)(cid:1) = dimgrv(F ) grα(C) = dimgrv1 (L) grα(C) = dimL C. Thus, αρ is a vρ-norm on C. Hence, ϕρ : H → Γ ∪ {∞} given by ϕρ(h·c) = αρ(c) is a vρ-norm on the 1-dimensional C-vector space H, for every ρ ∈ G. VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 39 We work back from the ϕρ to ϕH . Since ϕH = minρ∈G(ϕρ) by (4.10), there is a graded grv(F )- vector space monomorphism given by Once we verify that Φ is surjective, we will have grϕρ(H) Φ : grϕ(H) ֒→ Lρ∈G >ϕ(b) 7→ (cid:0) . . . , b + H ϕρ >ϕ(b), . . .(cid:1) dimgrv(F ) grϕρ(H) = Pρ∈G dimL H = G dimL H = dimF H. for all b ∈ H \ {0}. dimgrvρ(L) grϕρ(H) eb ϕ = b + H ϕ dimgrv(F ) grϕ(H) = Pρ∈G = Pρ∈G Hence, ϕH is a v-norm on H. For the surjectivity of Φ, consider the wL-coarsening of ϕρ. We have seen that ε ◦ α = yC as wL-gauges on C. Also the equation dσdρ = dσρtσ,ρ yields for the valuation y, y(tσ,ρ) + y(dσρ) − y(dρ) = y(dσ) for all σ, ρ ∈ G. Hence, for c ∈ C, ε ◦ ϕρ(h·c) = ε ◦ α(tσ,ρd−1 ρ cdρ) + ε(τρ) = y(tσ,ρd−1 ρ cdρ) + πi − πj + y(dσρ) − y(dρ) = y(c) + πi − πj + y(dσ). Thus, ε ◦ ϕρ is the same for each ρ ∈ G. Since wL is the finest common coarsening of vρ and vρ′ for all distinct ρ, ρ′ ∈ G, as noted at the beginning of the proof, it follows by an argument just like that for surjectivity of Ψ in the proof of Th. 2.8 that Φ is surjective. Hence ϕH is a v-norm on H, as noted above. Since this is true for each subspace H = (eij ◦ λdσ ) · C in the splitting decomposition of A for ϕ, this ϕ must be a v-norm on A, as claimed. To see that ϕ is not only a v-norm but actually a v-gauge, observe that ϕ = ψA is surmultiplica- tive on A since the v1-gauge ψ is surmultiplicative on all of A′. Moreover, the inclusion A ֒→ A′ yields a canonical graded monomorphism ι : grϕ(A) ֒→ grψ(A′). Because ϕ is a v-norm and ψ is a v1-norm and grv(F ) = grv1(L), we have [grϕ(A) : grv(F )] = [A : F ] = [A′ : L] = [grψ(A′) : grv1(L)] = [grψ(A′) : grv(F )]. Hence, ι is a graded isomorphism. Therefore, grϕ(A) is graded semisimple, since this is true for grψ(A′), as ψ is a gauge. Thus, ϕ is a v-gauge on A. Since ψ and β are End-gauges, we have, for all f ∈ A, ϕ(f ) = ψ(f ) = min and β(f ) = min m ∈M \{0}(cid:0)η(f (m)) − η(m)(cid:1) m ∈M \{0}(cid:0)φ(f (m)) − φ(m)(cid:1). We observed earlier that ε ◦ η = θ. It follows that ε ◦ ϕ = β, i.e., β is the w-coarsening of ϕ. Also, from (4.8), Γϕ ⊆ nSi,j=1 Sσ,ρ ∈G(cid:0)µi − µj + γσρ − γρ + Γα(cid:1) ⊆ Γ = H(Γv). Since grϕ(A) ∼=g grψ(A′), we have ω(ϕ) = ω(ψ). We compute ω(ψ). Partition {1, . . . , n} into a disjoint union Fk ℓ=1 Sℓ according to the coset of Γy containing θ(mj). That is, if j ∈ Sℓ, then Sℓ = {j′ ∈ {1, . . . , n} θ(mj′) − θ(mj) ∈ Γy}. Recall that the πj = θ(mj) have been chosen so that 40 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH πj = πj′ if and only if j and j′ lie in the same Sℓ. Since β = End(θ), by the comments with (1.6) the number of sets Sℓ equals ω(β). We use the Sℓ to decompose grη(M ). Since (mjdρ)n and ]mjdρ = fmjedρ, by Lemma 1.2 (fmjedρ)n and ρ ∈ G, let j=1, ρ∈G is the C-base of M used in building η j=1, ρ∈G is a grα(C)-base of grη(M ). For ℓ = 1, . . . , k Nℓρ = Lj∈Sℓfmjedρ grα(C). kLℓ=1 Lρ∈G So, grη(M ) = Nℓρ, and each Nℓρ is a graded right grα(C)-submodule of grη(M ). We show ε(γ) = πj + δρ + y(c) = πj + y(dρc) ∈ πj + Γy. that their grade sets do not overlap. If γ ∈ ΓNℓρ, then γ = deg(fmjedρec) for some j ∈ Sℓ and c ∈ C × with fmjedρec 6= 0. So, γ = µj + γρ + α(c) ∈ Γ. Then, as ε(γρ) = δρ = y(dρ) and ε ◦ α = yC, Now likewise let γ′ ∈ ΓNℓ′ρ′ , say γ′ = deg(gmj′fdρ′ec′) = µj′ + γρ′ + α(c′). If γ = γ′, then ε(γ) = ε(γ′), ℓ=1Fρ∈G ΓNℓρ, a disjoint Thus, Nℓρ = Nℓ′ρ′ whenever their grade sets intersect. Hence, Γgrη(M ) =Fk Let E = Endgrα(C)(grη(M )) ∼=g grψ(A′) by Lemma 1.2. If f ∈ E0, then f is a degree-preserving so πj + Γy = πj′ + Γy. Hence, Sℓ = Sℓ′, so ℓ = ℓ′ and πj = πj′. Then the equality ε(γ) = ε(γ′) yields y(dρ) + y(c) = y(dρ′) + y(c′). Since the y(dρ) are distinct modulo ΓC,y, it follows that ρ = ρ′. union, so each homogeneous element of grη(M ) lies in some Nℓρ. map, so f must map each Nℓρ to itself, by the disjointness of the grade sets ΓNℓρ. Thus, (4.11) E0 ∼= kQℓ=1 Qρ∈G(cid:0) Endgrα(C)(Nℓρ)(cid:1)0. Since the πj are the same for all j in Sℓ, the µj are likewise the same by hypothesis, hence the base elements fmj edρ of Nℓρ all have the same degree µj + γρ. Therefore, as graded grα(C)-modules, Nℓρ ∼=g grα(C)SℓG(µj + γρ) for any j ∈ Sℓ, in the grade shift notation of (1.1). Hence, Endgrα(C)(cid:0)Nℓρ(cid:1) ∼=g Endgrα(C)(cid:0) grα(C)SℓG(cid:1) ∼=g MSℓG(grα(C)). So, in degree 0,(cid:0) Endgrα(C)(cid:0)Nℓρ(cid:1)(cid:1)0 ∼= MSℓG(grα(C)0), and the number of its simple components coincides with the number of simple components of grα(C), which is ω(α). So, from (4.11), ω(ϕ) = ω(ψ) = number of simple components of E0 = k G ω(α) = ω(β) r ω(α). To complete the proof, we show that α can be chosen with ω(α) = 1. Let T be a Dubrovin valuation ring of D with Z(T ) = V . Then T ⊆ Y since the valuation ring Y of y on D is the unique this valuation is one of the ui; after renumbering if necessary, we may assume that it is u1. Since jump prime ideals of u1 for D pull back to jump prime ideals of v for D strictly containing J(W ) by [W1, Cor. E], and since by hypothesis J(V ) is the only such jump prime ideal for D, we have Dubrovin valuation ring of D with center W . Let eT = T /J(Y ), which is a Dubrovin valuation ring of Y /J(Y ) = D. The valuation of the valuation ring Z(eT ) is an extension to Z(D) of u = v/y. So j(u1, D) = 1. Hence, by Prop. 3.3(iv) eT is integral over its center, which is the valuation ring U1 of u1. Now let B be a Dubrovin valuation ring of C with center Z(B) = V1, the valuation ring of v1 on L = Z(C). Then B ⊆ Y ∩ C, since the valuation ring Y ∩ C is the unique Dubrovin valuation = D, since D is totally ramified ring of C with center WL, the valuation ring of wL. Note that C y VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 41 y y over C re y. Let eB = B/J(Y ∩C), which is a Dubrovin valuation ring of C of the valuation ring Z(eB) restricts on L = K to v1/wL = u1K. Hence, Z(eB) = U1 = Z(eT ), as u1 is the unique extension of u1K to Z(D). Because eB and eT are Dubrovin valuation rings of D with the same center, we have eB ∼= eT . Therefore eB is integral over U1, since this is true for eT . But also U1 is integral over U1 ∩ K, as u1 is the unique extension of u1K to Z(D); hence, eB is integral over U1 ∩ K. Moreover, by [W2, p. 390] the valuation ring Y ∩ C of the division ring C is integral over its center Y ∩ L = WL, It follows by [MMU, Prop. 12.2, p. 70] that B is integral over V1. Let α be the Morandi value function on C with Rα = B. To see that α is a v1-gauge we must check that v1 on L is defectless in C. For this, note that by Th. A.1, there is an Fh,v-isomorphism , so of D. The valuation L ⊗F Fh,v ∼= Lh,v1 × . . . × Lh,vr . Since [L:F ] = r, each factor Lh,vi must be 1-dimensional over Fh,v, i.e., isomorphic to Fh,v. Note that as C = CD(L), the algebras D⊗F L and C are Brauer equivalent. Hence, C ⊗L Lh,v1 is Brauer equivalent to D ⊗F Lh,v1 ∼= D ⊗F Fh,v. Since v on F is defectless in D, v1 on L is defectless in C. It follows by Remark 3.7 that α is a v1-gauge on C with ω(α) = 1. (cid:3) There is a easier version of Prop. 4.6 for jump rank 1 that we will need later: Proposition 4.7. Let A be a central simple F -algebra. Suppose the valuation v on F is defectless in A. Let valuation w on F be a nontrivial coarsening of v, and assume that w is Henselian. Suppose j(v, A) = 1. Let β be a w-gauge on A with Γβ ⊆ H(Γw). Then, there is a v-gauge ϕ on A with w-coarsening β such that ω(ϕ) = ω(β) and Γϕ ⊆ H(Γv). Proof. View A = EndD(M ) where D is a division ring and M is a finite-dimensional right D- vector space. Since w is Henselian, it has a unique extension to a valuation y on D. Moreover, by [TW1, Th. 3.1], β is an End-gauge, as in Ex. 1.1, say β = End(θ) for some y-norm θ on M . Let (m1, . . . , mn) be a splitting base of M for θ, and let πj = θ(mj) for j = 1, . . . , n. Let Λ = H(Γw) and Γ = H(Γv), and ε : Γ → Λ the epimorphism induced by the canonical map Γv → Γw. From (1.8), Γβ = Γgrβ(A) =Sn i,j=1 πi − πj + Γw. By changing θ by replacing each πj by πj − π1 (which does not change End(θ)) we may assume that each πj ∈ Γβ ⊆ H(Γw). Also, by adjusting the mj by multiples in D×, we may assume that πi = πj whenever πi + Γy = πj + Γy. So, the number of distinct πi equals the number of cosets of Γy in Γβ. This number equals ω(β) by (1.6). Because j(v, A) = 1, we have ind(A ⊗F Fh,v) = ind(A ⊗F Fh,w) where Fh,v (resp. Fh,w) is a Henselization of F with respect to v (resp. w). Since Fh,w = F as w is assumed Henselian, it follows that ind(A ⊗F Fh,v) = ind(A). Hence, by Morandi's theorem [M1, Th. 3] the valuation v on F extends to a valuation z on D. Pick any µ1, . . . , µn ∈ Γ with ε(µj) = πj for all j and µj = µi whenever πj = πi. Let η be the z-norm on M given by η(cid:0) nPj=1 mjdj(cid:1) = min 1≤j≤n(cid:0)µj + z(dj)(cid:1) for all d1, . . . , dn ∈ D. Let ϕ = End(η), which is a v-gauge on A since v is defectless in A (see Ex. 1.1). Then ε ◦ ϕ = β since ε ◦ η = θ. Also, if πi + Γy = πj + Γy, then πi = πj, so µi = µj by the choice of the µi, so µi + Γz = µj + Γz. Thus, by (1.6), ω(ϕ) = (cid:12)(cid:12){cosets of Γz in Γϕ}(cid:12)(cid:12) = (cid:12)(cid:12){cosets of Γy in Γθ}(cid:12)(cid:12) = ω(β). 42 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Because each µj ∈ Γ = H(Γv) and Γz ⊆ H(Γv), we have by (1.8) Γϕ = nSi,j=1 (µi − µj) + Γz ⊆ H(Γv). (cid:3) Proof of Th. 4.3 (Central simple case). Suppose A is central simple. We argue by induction on the jump rank j(v, A). If j(v, A) = 0, then v is the trivial valuation on F , and the trivial gauge on A is a minimal v-gauge. If j(v, A) = 1, then ξV,[A] = 1 by Prop. 3.3(iv); so for any Dubrovin valuation ring B of A with Z(B) = V, B is integral over V . Let α be the associated Morandi value function of B. Then Γα = ΓB by (3.13), and ΓB ⊆ H(Γv) by (3.2); hence Γα ⊆ H(Γv). Since v is defectless in A, by Remark 3.7 α is a v-gauge with ω(α) = 1, so α is a minimal v-gauge. Now suppose j(v, A) = n > 1. Let P be the (n−1)-st jump prime ideal of v for A, and let W = VP , with associated valuation w. Then w is defectless in A since v is defectless in A, by Prop. 1.8. Since j(w, A) = n− 1, by induction there is a minimal w-gauge β on A with Γβ ⊆ H(Γw). Let (F ′, wh) be the Henselization of (F, w), let v′ be the valuation of F ′ refining wh and restricting to v on F . Let A′ = A⊗F F ′, which is a central simple F ′-algebra, and let β′ = β ⊗ wh, which is a wh-gauge on A′ with grβ′(A′) ∼=g grβ(A) by [TW1, Cor. 1.26]. Note however that β′ need not be a minimal gauge even though β is minimal. We claim that j(v′, A′) = 2. To see this, let valuation y on F ′ be any coarsening of v′, and let z = yF , which is a coarsening of v. If y = wh or y is coarser than wh, then y is Henselian, h,y = F ′, so of course as wh is Henselian (see [EP, Cor. 4.1.4, p. 90]). Then the Henselization F ′ h,y) = ind(A′). Suppose instead that y is properly finer than wh, so z is a refinement ind(A′ ⊗F ′ F ′ of w. We show that then the Henselizations (F ′ h,y, yh) and (Fh,z, zh) are isomorphic. For this, note first that since the w-coarsening w1 of zh in Fh,z is Henselian and restricts to w in F , there is an F -homomorphism η1 : F ′ → Fh,z with wh = w1 ◦ η1. The Henselian valuation zh on Fh,z pulls back to a valuation z′ on F ′ that refines wh with z′F = z = yF . Hence, z′ = y. Because zh is Henselian and pulls back to y, there is an F -monomorphism η2 : F ′ h,y → Fh,z with yh = zh ◦ η2. But also since yh is Henselian with yhF = yF = z, there is an F -monomorphism η3 : Fh,z → F ′ with zh = yh ◦ η3. Thus, η2 ◦ η3 is an F -homomorphism Fh,z → Fh,z with zh = zh ◦ η2 ◦ η3. By the uniqueness in the universal property for the Henselization (recalled in the Appendix below), we must have η2 ◦ η3 = idFh,z , so η3 is an isomorphism (Fh,z, zh) ∼= (F ′ In particular, (Fh,v, vh) ∼= (F ′ h). So, as there are no jump prime ideals of v for A between P = J(W ) and J(V ), h,y, yh). h,v′, v′ h,y ind(A′ ⊗F ′ F ′ h,y) = ind(A ⊗F Fh,z) = ind(A ⊗F Fh,v) = ind(A′ ⊗F ′ F ′ h,v′); this value is strictly smaller than ind(A′) = ind(A ⊗F F ′) since P is a jump prime ideal. Thus, j(v′, A′) = 2, as claimed. The calculation also shows that wh is the first jump valuation for v′ in A′. Note also that v′ is defectless in A′, since this depends on the defectlessness of the associated h,v′ ∼= A⊗F Fh,v and v is defectless in A. Thus, the division algebra of A′⊗F ′ F ′ hypotheses of Prop. 4.6 are satisfied for the field F ′ with valuations v′ and wh and central simple F ′-algebra A′ with wh-gauge β′. h; but A′⊗F ′ F ′ h,v′ re v′ By Prop. 4.6, A′ has a v′-gauge ϕ whose wh-coarsening is β′ with ω(ϕ) = rω(β′) and Γϕ ⊆ H(Γv′). Let D′ be the associated division algebra of A′, let wD′ be the valuation on D′ extending wh on F ′, and let D′ = D′ wD′ . Let u′ be the residue valuation v′/wh on F ′ wh determined by v′. The integer r in the formula for ω(ϕ) is the number of extensions of u′ to Z(D′). Hence, by (3.12), r = ℓV,W . Let VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 43 ω(α) = ω(ϕ) = rω(β) = ℓV,W ω(β) = ℓV,W ξW,[A]. α = ϕA. By Lemma 4.4, α is a v-gauge on A with w-coarsening β, and grα(A) ∼=g grϕ(A′). Hence, (4.12) The last equality in (4.12) holds since β is a minimal gauge. Since ξV,[A] ≥ ξW,[A]ℓV,W by (3.10), it follows from (4.12) that ω(α) ≤ ξV,[A]. But we always have ω(α) ≥ ξV,[A] by Th. 3.5; so ω(α) = ξV,[A], showing that α is a minimal gauge. Furthermore, Γα = Γϕ ⊆ H(Γv′) = H(Γv). This completes the proof of the central simple case of Th. 4.3. The rest of the proof of the theorem will be given after Lemma 4.9 below. (cid:3) Proposition 4.8. Let A be a central simple F -algebra with v defectless in A. Let w be a valuation on F that is coarser than v, and let β be a minimal w-gauge on A with Γβ ⊆ H(Γw). Then there is a minimal v-gauge α with w-coarsening β and Γα ⊆ H(Γv). Proof. Let V (resp. W ) be the valuation ring of v (resp. w). We argue by induction on the jump rank j(v, A). Clearly, j(w, A) ≤ j(v, A). If j(w, A) = 0, then w is the trivial valuation on F and β is the trivial w-gauge on A. By the case of Th. 4.3 already proved, there is a minimal v-gauge α on A; then the trivial gauge β is a coarsening of α. Thus, we may assume that 1 ≤ j(w, A) ≤ j(v, A). Suppose j(v, A) = 1. Then j(w, A) = 1, so ξW,[A] = 1 by Prop. 3.3(iv). Hence, ω(β) = 1 because β is assumed to be a minimal gauge. Therefore, by Remark 3.7 Rβ is a Dubrovin valuation ring integral over its center, which is W , and β is the Morandi value function of Rβ. Let B be any Dubrovin valuation ring of A with Z(B) = V and B ⊆ Rβ. Such a B is obtainable as the inverse image in Rβ of a Dubrovin valuation ring of Rβ/J(Rβ ) whose center is a valuation ring U of Z(Rβ/J(Rβ )) satisfying U ∩ (W/J(W )) = V /J(W ). Since j(v, A) = 1, by Prop. 3.3(iv) B is integral over V . Let α be the Morandi value function of B. Then Γα = ΓB ⊆ H(Γv) (see (3.2)). Since v is assumed defectless in A, by Remark 3.7 α is a v-gauge with ω(α) = 1; so α is minimal gauge. Because Rα ⊆ Rβ and these Dubrovin valuation rings determine their associated gauges, β is a coarsening of α. Now suppose j(v, A) = n > 1. Let y be the (n − 1)-st jump valuation of v for A. Suppose first that w is coarser than y or w = y. Then, as j(y, A) = n − 1 and y is defectless in A since v is defectless in A, by induction there is a minimal y-gauge η with Γη ⊆ H(Γy) such that β is the w-coarsening of η. The proof of Th. 4.3 (central simple case) shows that there is a minimal v-gauge α with Γα ⊆ H(Γv) and y-coarsening η. The w-coarsening of α is then the w-coarsening of η, which is β. Suppose instead that w is properly finer than y. Let (F ′, wh) be the Henselization of (F, w), and let v′ be the valuation on F ′ refining wh and restricting to v on F . Let β′ = β ⊗ wh, which is a wh-gauge on A′ = A ⊗F F ′, but not necessarily minimal, with Γβ′ = Γβ ⊆ H(Γw) = H(Γwh). Moreover, we claim that j(v′, A′) = 1. To see this, suppose z is any nontrivial valuation of F ′ with z coarser than v′ or z = v′. If z is coarser than wh or z = wh, then z is Henselian, so the h,z) = ind(A′). If z is finer than wh, then as in the proof h,z = F ′ and ind(A′ ⊗F ′ F ′ Henselization F ′ of Th. 4.3 (central simple case) above F ′ h,z = Fh,zF , the Henselization of F re zF . Since there are no jump valuations between w and v, so none between zF and v, ind(A′ ⊗F ′ F ′ h,z) = ind(A ⊗F Fh,zF ) = ind(A ⊗F Fh,v) = ind(A ⊗F F ′) = ind(A′). Since the indices are the same for all z, j(v′, A′) = 1, as claimed. By Prop. 4.7, applied to F ′, v′, w′, β′, there is a v′-gauge α′ of A′ with wh-coarsening β′, such that ω(α′) = ω(β′), and Γα′ ⊆ H(Γv′) = H(Γv). Let α = α′A. By Lemma 4.4, α is a v-gauge on A 44 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH with grα(A) ∼=g grα′(A′), so ω(α) = ω(α′) = ω(β′) = ω(β). Since β is a minimal gauge and ξV,[A] ≥ ξW,[A], α must also be a minimal gauge. Also, Γα = Γα′ ⊆ H(Γv). Finally, since β′ is the wh-coarsening of α′, the w-coarsening of α = α′A is β′A = β. (cid:3) Let F ⊆ L be fields with [L:F ] ≤ ∞, and let v be a nontrivial valuation of F with valua- tion ring V . For each nonzero prime ideal P of V , let s(P ) be the number of valuation rings of L extending the valuation ring VP of F . Clearly, 1 ≤ s(P ) ≤ [L:F ]. Also for prime ideals s(P ) = max j∈J s(Pj). We call P a splitting prime ideal of V in L if s(P ) < s(P ′) for all prime ideals P ′ % P . Define the splitting rank of v in L to be P ⊆ P ′ we have s(P ) ≤ s(P ′). Moreover, if P =Sj∈J Pj for prime ideals P and Pj, j ∈ J, then srk(v, L) = the number of splitting prime ideals of V in L. Note that the maximal ideal J(V ) is always a splitting prime ideal of V in L, so srk(v, L) ≥ 1. If P is a splitting prime ideal, we call the valuation of VP a splitting valuation of V in L. If v is the trivial valuation on F , define srk(v, L) = 0. Lemma 4.9. With the notation above, let v1 and v2 be two different extensions of v to L. Then either v1 and v2 are independent or the finest common coarsening v12 of v1 and v2 restricts to a splitting valuation for v in L. Proof. We argue by induction on n = srk(v, L). If n = 1, let w be the trivial valuation on F . If n > 1, let w be the (n − 1)-st splitting valuation of F for v in L. Let y be a valuation of F coarser than v and strictly finer than w. Because there are no splitting valuations between y and v, y must have the same number of extensions to L as v. Hence, v1 and v2 have distinct y-coarsenings. If n = 1, this shows that v12 must be the trivial valuation, i.e., v1 and v2 are independent valuations. If instead n > 1, this shows that v12F either equals or is coarser than w. Since w is a splitting valuation, it suffices to consider the case when v12F is strictly coarser than w. Then the w-coarsenings w1 of v1 and w2 of v2 are distinct. Hence, the finest common coarsening w12 of w1 and w2 is coarser than or equals to v12. But v12 is coarser than each of w1 and w2, so coarser than or equal to w12. Hence, v12 = w12. Since srk(w, L) = n − 1, the conclusion of the lemma holds for w1 and w2 by induction; hence it also holds for v1 and v2. (cid:3) Proof of Th. 4.3 (completed). It suffices to prove the theorem for the simple components of a semisimple F -algebra. So, assume A is simple. Let L = Z(A), so [L:F ] ≤ ∞. Let v1, . . . , vr be the valuations on L extending v on F and let Vi be the valuation ring of vi. The argument is by induction on n = srk(v, L). If n = 0, then v is the trivial valuation, and the trivial gauge on A is a minimal v-gauge. Assume now that n = 1. Lemma 4.9 then shows that the vi are pairwise independent. By the central simple case of Th. 4.3 proved above, for each i there is a minimal vi-gauge αi of A with Γαi ⊆ H(Γvi) = H(Γv). (Note that each vi is defectless in L, by Prop. 1.6.) Let α = min(α1, . . . , αr). Because the vi are pairwise independent, Cor. 2.10 shows that α is a v-gauge on A. Since each αi is a minimal vi-gauge, α is a minimal v-gauge (see the comments in Def. 4.2). Also, Γα ⊆Sr Now assume that n > 1. Let P1 $ P2 $ . . . $ Pn−1 $ Pn be the splitting prime ideals of V in L, and let W = VPn−1 with associated valuation w. Because v is defectless in A, w is defectless in A, by Prop. 1.8. Since srk(w, L) = n − 1, by induction there is a minimal w-gauge β on A with Γβ ⊆ H(Γw). Let w1, . . . , wℓ be the valuations of L extending w, and for each j let βj be the i=1 Γαi ⊆ H(Γv). VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 45 wj-component of β. The construction of the βj in Th. 2.2 shows that Γβj ⊆ Γβ ⊆ H(Γw) for each j. By Th. 2.8, βj and βk have the same wjk-coarsening for all j, k ∈ {1, . . . , ℓ}. Since β is a minimal w-gauge and β = min(β1, . . . , βℓ), each βj is a minimal wj-gauge. For each i ∈ {1, . . . , r} let j(i) ∈ {1, . . . , ℓ} be the index such that wj(i) is the w-coarsening of vi. For each i, Prop. 4.8 applied to the valuations vi and wj(i) on L shows that there is a minimal vi-gauge αi with wj(i)-coarsening βj(i) and Γαi ⊆ H(Γvi) = H(Γv). Let α = min(α1, . . . , αr). To see that α is a v-gauge, we must check that the αi satisfy the compatibility condition of Th. 2.8. For this, take any distinct i, k ∈ {1, . . . , r}. If j(i) = j(k), then wj(i) = wj(k), which is coarser than both vi and vk, so coarser than (or equal to) vik. Since vikF is a splitting valuation of v in L, the choice of w implies that vik = wj(i) = wj(k). So, the vik-coarsening of αi is βj(i), which is the same as the vik-coarsening βj(k) of αk. Suppose now instead that j(i) 6= j(k), so wj(i) 6= wj(k). The proof of Lemma 4.9 shows that in this case vik = wj(i)j(k). Hence, the vik-coarsening of αi is the wj(i)j(k)-coarsening of the wj(i)-coarsening βj(i) of αi. By Th. 2.8, this coincides with the wj(i)j(k)-coarsening of the wj(k)-coarsening βj(k) of αk. Thus, αi and αk have the same vik-coarsening. Since the compatibility condition thus holds in all cases, Th. 2.8 shows that α is a v-gauge. It is a minimal gauge since each αi is a minimal gauge. Moreover, (cid:3) Γα ⊆Sr i=1 Γαi ⊆ H(Γv). In this section we construct an example of a central simple algebra with multiple non-isomorphic minimal gauges all having the same gauge ring. 5. An example w Example 5.1. Let L be a field with char(L) 6= 2, let x be transcendental over L, and let F = L(x)((y)), the Laurent series field in one variable over L(x). Let w be the complete discrete (so Henselian) y-adic valuation on F , with Γw = Z and F = L(x). Let W be the power series ring L(x)[[x]], which is the valuation ring of w. Let v be the rank 2 valuation on F that is the composite of w with the discrete x-adic valuation on F . Equivalently v is the valuation on F obtained by restriction = L, Γv = Z × Z with from the standard rank 2 Henselian valuation on L((x))((y)). Thus, F right-to-left lexicographic ordering, v(x) = (1, 0), v(y) = (0, 1), and grv(F ) = L[X, X −1, Y, Y −1], a twice iterated Laurent polynomial ring, where X =ex and Y =ey. Let V be the valuation ring of v. Note that w is the rank 1 coarsening of v, and the epimorphism ε : Γv → Γw given by v(c) 7→ w(c) for c ∈ F × is the projection (ℓ, m) 7→ m. Since we will be working primarily with v, we write F for F w . v v Let D = (cid:0)1 + x, y/F ), a quaternion algebra over F with its standard F -base (1, i, j, k), where i2 = 1 + x, j2 = y, and k = ij = −ji. Because w(1 + x) = 0 and 1 + x w is not a square in F , the valuation w has a unique and inertial extension to the field F (i). Therefore, every norm from F (i) to F has w-value in 2Γw. Since w(y) = 1, y is not such a norm. Hence, D is a division algebra. The Henselian valuation w on F therefore has a unique extension to a valuation β on D, with β(1) = β(i) = 0 and β(j) = β(k) = 1 2 . Since β(j) /∈ Γw one can see that D is totally ramified over F (i) for β, while = L(x)(√1 + x), and for all a, b, c, d ∈ F , F (i) is inertial over F ; indeed, Γβ = 1 2(cid:1). (5.1) β(a + bi + cj + dk) = min(cid:0)β(a + bi), β(cj + dk)(cid:1) = min(cid:0)w(a), w(b), w(c) + 1 2 , w(d) + 1 2Z and D = F (i) w β β 46 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH 0 Let K = F (t), where t2 = 1 + x. Thus, K is a quadratic extension field of F , and since 1 + x = 1 in F , v has two extensions to K that are distinguished by whether t = 1 or −1 in K. Let v′ denote the extension of v to K with t = 1. Then, K = F = L and Γv′ = Γv = Z × Z, so gr(K) = gr(F ). Note that as x = (t − 1)(t + 1) and v′(t + 1) = 0, we have v′(t − 1) = v(x) = (1, 0). The rank 1 = L(x)(√1 + x) and coarsening of v′ is the unique, unramified, extension w′ of w to K, with K Γw′ = Z. Also, K is a splitting field of D, as K ∼= F (i), which is a maximal subfield of D. Explicitly, let S = M2(K), and view D as an F -subalgebra of S by identifying w′ 0 −t(cid:1) , i = (cid:0) t Give Q × Q the right-to-left lexicographic ordering. Fix any γ ∈ Q with 0 < γ < 1 δ = (γ, 1 1 0(cid:1), j = (cid:0) 0 y 2 ) ∈ Q × Q. Let α′ be the v-gauge on S given by k = (cid:0) 0 ty −t 0(cid:1). 0 1(cid:1) , 1 = (cid:0) 1 0 α′(cid:0) p q r s(cid:1) = min(cid:0)v′(p), v′(q) − δ, v′(r) + δ, v′(s)(cid:1). 1(cid:1)(cid:9), and identify S = EndK(M ). Then α′ is the v′-gauge End(η) as 0(cid:1),(cid:0) 0 Indeed, let M = K-span(cid:8)(cid:0) 1 q(cid:1) = min(cid:0)v′(p), v′(q) + δ(cid:1). in Ex. 1.1, where η : M → Q× Q∪{∞} is the v′-norm on M given by η(cid:0) p in the notation of (1.7). Let α = α′D, which is a surmultiplicative v-value function on D since the gauge α′ on S is surmultiplicative. While α′ is a v′-gauge, we must still verify that α is a v-gauge. a−bt (cid:1) For this, note that for any z = a + bi + cj + dk ∈ D with a, b, c, d ∈ F , we have z =(cid:0) a+bt (c+dt)y grα′(S) = Endgr(K)(gr(M )) ∼=g M2(gr(K))(0, δ), in S. Thus, as v(y) = (0, 1), 2 , and let Thus, c−dt (5.2) α(z) = min(cid:0)v′(a + bt), v′((c + dt)y) − δ, v′(c − dt) + δ, v′(a − bt)(cid:1) = min(cid:0)v′(a + bt), v′(a − bt), v′(c + dt) + (−γ, 1 2 )(cid:1). 2 ), v′(c − dt) + (γ, 1 So, α(1) = α(i) = 0 and Since v′(1 − t) = (1, 0), we have as γ < 1 2 . So, in grα(D) ⊆ grα′(S), 2 ). 2 )(cid:1) = (−γ, 1 2 )(cid:1) = (γ, 1 2 ), 2 ), (γ, 1 α(j) = α(k) = min(cid:0)(−γ, 1 α(j − k) = min(cid:0)(−γ, 1 ]1 + i =(cid:0) 2 0 ej =ek =(cid:0) 0 ey 0 0(cid:1) ∈ D0, ]1 − i =(cid:0) 0 0 0 0(cid:1) ∈ D(−γ, 1 2 ), 2 ) + (1, 0), (γ, 1 0 2(cid:1) ∈ D0, and ]j − k =(cid:0) 0 0 2 0(cid:1) ∈ D(γ, 1 2 ). Since 1 + i, 1 − i, j, j − k have images in grα(D) which are clearly gr(F )-independent, they com- prise a splitting base of α as a v-value function; this shows that α is a v-norm on D. Moreover, [grα(D): gr(F )] = 4 = [grα′(S): gr(K)] = [grα′(S): gr(F )], since gr(K) = gr(F ). Hence, grα(D) = grα′(S), which is graded simple. Thus, α is a v-gauge on D. Note that (5.3) Also, D0 = S0 = L × L, so ω(α) = 2. From (5.2), we have Γα = Γα′ = Z2 ∪ (δ + Z2) ∪ (−δ + Z2). Rα = {a + bi + cj + dk ∈ D v′(a + bt) ≥ 0, v′(a − bt) ≥ 0, v′(c + dt) ≥ (γ,− 1 2 ), v′(c − dt) ≥ (−γ,− 1 2 )}. VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 47 Let v′(c + dt) = (ℓ, m) ∈ Γv′ = Z × Z. Then, v′(c + dt) ≥ (γ,− 1 2 ) if and only if m ≥ 0, i.e., w′(c + dt) ≥ 0 where w′ is the rank 1 coarsening of v′. Likewise, v′(c − dt) ≥ (−γ,− 1 2 ) if and only if w′(c − dt) ≥ 0. Therefore, each of the infinitely many choices of γ yields the same gauge ring for the associated v-gauge α. But different choices of γ yield non isomorphic gauges since the gauges have different value sets (see (5.3)). Thus, the gauge ring Rα does not determine α. We still must show that α is a minimal gauge. We will show as well that Rα is an intersection of two total valuation rings B1, B2, which are the only Dubrovin valuation rings of D with center V . Since char(F ) 6= 2 and v′(t) = 0, we have min(cid:0)v′(a+ bt), v′(a− bt)(cid:1) = min(cid:0)v(a), v(b)(cid:1). Similarly, min(w′(c + dt), w′(c − dt)) = min(cid:0)w(c), w(d)(cid:1). Thus, the description of Rα simplifies to Rα = {a + bi + cj + dk a, b ∈ V, c, d ∈ W}. The ring Rα lies in the invariant valuation ring Rβ of the valuation β on D extending w on F . From (5.1), we have Rβ = {a + bi + cj + dk a, b, c, d ∈ W} and J(Rβ) = {a + bi + cj + dk a, b ∈ J(W ), c, d ∈ W}. Let π : Rβ → Rβ/J(Rβ) = L(x)(√1 + x) be the canonical projection. Let U1, U2 be the two valuation rings of L(x)(√1 + x) extending the valuation ring U of the x-adic valuation on L(x). The commutative valuation rings Uℓ for ℓ = 1, 2 are Dubrovin valuation rings. Since the invariant valuation ring Rβ is a Dubrovin valuation ring, the pullback rings Bℓ = π−1(Uℓ) for ℓ = 1, 2 are Dubrovin valuation rings of D with center V and fBℓ = Bℓ/J(Rβ ) = Uℓ. Moreover, since Rβ and Uℓ are valuation rings, it is easy to check that Bℓ is a total valuation ring, i.e., for any d ∈ D \ {0}, d ∈ Bℓ or d−1 ∈ Bℓ. Let B = B1. Since B/J(B) ∼= U1/J(U1), which is a field, tB = 1. Also, if (Fh, vh) is a Henselization of (F, v), then 1 + x ∈ F 2 h , since 1 + x = 1 in Fh . Therefore, Fh splits D, which shows that D ⊗F Fh ∼= M2(Fh), hence nB = 2. Thus, vh = F v ξV,[D] = nB/tB = 2 = ω(α), showing that α is a minimal gauge. Another way to calculate the extension number ξV,[D] is by using (3.10) with Rβ for S: Since the valuation rings Rβ and U1 = eB have extension number 1 and the residue valuation ring U = V /J(W ) has two extensions to Rβ/J(Rβ), ξV,[D] = ξRβ,[D] ℓV,W ξU1,[Rβ/J(Rβ )] = 1 · 2 · 1 = 2. w Any inner automorphism ι of D maps the invariant valuation ring Rβ to itself. The automorphism = L(x)-automorphisms of L(x)(√1 + x), so it either induced by ι on Rβ/J(Rβ) is one of the two F preserves or interchanges U1 and U2. Hence, the set of conjugates of B1 in D is {B1, B2}. The Bℓ are therefore the only Dubrovin valuation rings of D with center V . Since the Bℓ are not integral over V (as ξV,[D] 6= 1) the only possible Grater ring of D with center V is B1 ∩ B2. Since Rα is the gauge ring of a minimal v-gauge, it is a Grater ring with center V by Th. 3.9. Hence, Rα = B1∩ B2. (This equality can also be verified directly after first showing that V + V i is the integral closure of V in F (i) and that the Bℓ ∩ F (i) are the valuation rings of F (i) extending V in F .) Note that our example required a valuation of rank at least 2. For if v is a rank 1 valuation on a field F and A is a central simple F -algebra, then for the valuation ring V of v, we have j(v, A) = 1, so ξV,[A] = 1 by Prop. 3.3(iv). Hence, for any minimal v-gauge α on A, we have ω(α) = 1, so by 48 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH Remark 3.7 the gauge ring Rα is a Dubrovin valuation ring integral over its center V , and α is the Morandi value function determined by Rα. Appendix A. Tensor product and Henselization It is well known that if L/F is a finite degree field extension, v is a discrete (rank 1) valuation on F , and v1, . . . , vr are all the extensions of v to L, then L⊗FbF ∼=Qr completion of F (resp. L) with respect to v (resp. vi) (see [B, Ch. VI, §8, no. 6, Prop. 11]). In this appendix we prove an analogous result for valuations of arbitrary rank, replacing the completion by the Henselization. For separable field extensions, this result is implicit in [E, Th. 17.17, p. 135]. We give a full proof here, since this result is essential for many of the arguments in this paper. i=1 bLi, where bF (resp. bLi) is the Let F be a field with a valuation v. A Henselization of (F, v) is a valued field extension (Fh, vh) of (F, v) such that vh is Henselian and for any extension (K, w) of (F, v) with w Henselian there is a unique F -homomorphism η : (Fh, vh) → (K, w) such that vh = w ◦ η. We thus refer to (η(Fh), wη(Fh )) as the Henselization of (F, v) within (K, w). It is clear from the definition that a Henselization of (F, v) is unique up to unique isomorphism. Thus, we sometimes say that (Fh, vh) is "the Henselization" of (F, v). A proof of the existence of a Henselization can be found in [EP, Th. 5.2.2, p. 121]. Theorem A.1. Let F be a field with a valuation v. Let K be a finite-degree field extension of F , and let v1, . . . , vr be all the extensions of v to K. Let (Fh, vh) be a Henselization of (F, v), and let (Kh,i, vi,h) be a Henselization of (Ki, vi) for i = 1, . . . , r. Then, K ⊗F Fh ∼= Kh,1 × . . . × Kh,r. The proof will use the following two lemmas: Lemma A.2. Let F ⊆ N be fields with N Galois over F (possibly of infinite degree), and let G = G(N/F ). Let K and E be subfields of N containing F , with [K :F ] < ∞. Let H = G(N/K) ⊆ G and Z = G(N/E) ⊆ G. Let τ1, . . . , τr be representatives of the distinct Z-H double cosets of G. i=1 ZτiH, a disjoint union.) Then (So, G =Fr K ⊗F E ∼= τ1(K)·E × . . . × τr(K)·E. Proof. Since K is separable over F , we have K = F (a) for some a. Let f be the minimal polynomial of a over F . Then f splits over N , as N is normal over F , say f = (X − a1) . . . (X − an) ∈ N [X] where the ai are distinct and a1 = a. Let A = {a1, . . . , an}. Let f = g1 . . . gr be the irreducible factorization of f in E[X], and fix a root bi of gi for i = 1, . . . , r. The Galois group G acts transitively on A, but A decomposes into r disjoint Z-orbits, A =Fr i=1 Bi, where For each i, choose τ ′ i ∈ G with τ ′ Bi = Z·bi = {roots of gi in N}. i(a) = bi. Then, as H = {σ ∈ G σ(a) = a}, we have Zτ ′ iH = {σ ∈ G σ(a) ∈ Bi}, for i = 1, . . . , r. Thus, Zτ ′ that the double coset representatives τ ′ rH are all the distinct Z-H double cosets in G. Moreover, we may assume i coincide with the τi of the lemma, by replacing bi by τi(a). 1H, . . . , Zτ ′ VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 49 Since gcd(gi, gj ) = 1 for i 6= j, the Chinese Remainder Theorem yields K ⊗F E ∼= F [X]/(f ) ⊗F E ∼= E[X]/f E[X] ∼= E[X](cid:14)(cid:0)(g1) . . . (gr)(cid:1) ∼= E[X]/(g1) × . . . × E[X]/(gr ) ∼= E(b1) × . . . × E(br) ∼= τ1(K)·E × . . . × τr(K)·E. (cid:3) Lemma A.3. Let (Fh, vh) be a Henselization of the valued field (F, v). Let K be any extension field of F lying in the algebraic closure of Fh, and let w be the unique extension of vh to the compositum K·Fh. Then, (K·Fh, w) is a Henselization of (K, wK ). Proof. The valuation w on K·Fh is Henselian since vh is Henselian. Let (Kh, wh) be a Henselization of (K, wK ). The universal property shows that (Kh, wh) embeds in (K·Fh, w). Thus, we may assume that K ⊆ Kh ⊆ K·Fh and wh = wKh. So, whF = wF = v. Since (Fh, vh) is the Henselization of (F, v) within (K·Fh, w), it is also the Henselization of (F, v) within (Kh, wh), by the uniqueness in the universal property for the Henselization; so, Fh ⊆ Kh. Since also K ⊆ Kh, we have K·Fh ⊆ Kh ⊆ K·Fh. Hence, Kh = K·Fh and w = wh, showing that (K·Fh, w) is a Henselization of (K, wK ). Proof of Th. A.1: Assume first that K is separable over F . Let Fsep be a separable closure of F containing Fh, and let vsep be the unique valuation on Fsep extending the Henselian valuation vh. Let G = G(Fsep/F ), H = G(Fsep/K) ⊆ G, and Z = G(Fsep/Fh) ⊆ G. By [EP, Th. 5.2.2, p. 121] and the universal property of the Henselization, Fh is the decomposition field for vsep over v, so Z is the decomposition subgroup of G, i.e., (cid:3) Z = {σ ∈ G vsep ◦ σ = vsep}. Let Ω be the set of all valuations on Fsep extending F . Then, G acts transitively on Ω (see [EP, Th. 3.2.14, p. 68]), while the distinct H-orbits of Ω are Ω1, . . . , Ωr, where Ωi = {w ∈ Ω wK = vi}. For i = 1, . . . , r, choose τi ∈ G with vsep ◦ τiK = vi. Then, {σ ∈ G vsep ◦ σK = vi} = {σ ∈ G vsep ◦ σ ∈ Ωi} = ZτiH. So, G =Fr i=1 ZτiH is the disjoint Z-H double coset decomposition of G. We now apply Lemma A.2 with N = Fsep and E = Fh. (So, the K, H and Z of the lemma are the K, H and Z here.) By the lemma, (A.1) K ⊗F Fh ∼= τi(K)·Fh ∼= where the second isomorphism follows by applying τ −1 isomorphism τ −1 tion of (F, v). The unique extension of the Henselian valuation vh ◦ τi to K · τ −1 (Fh), whose restriction to K is vi by the choice of τi. Therefore, K·τ −1 vsep ◦ τiK·τ −1 by Lemma A.3. The theorem (for K separable over F ) then follows from (A.1). K·τ −1 to the i-th factor. Note that the F - (Fh), vh ◦ τi) is a Henseliza- (Fh) must be (Fh) ∼= Kh,i i maps (Fh, vh) to (τ −1 (Fh), rQi=1 rQi=1 (Fh), vh ◦ τi). Hence (τ −1 i i i i i i i If K is not separable over F , let S be the separable closure of F in K, and let yi = viS for i = 1, . . . , r. Then, y1, . . . , yr are all the extensions of v to S. Since valuations extend uniquely from S to its purely inseparable extension K, we have yi 6= yj for i 6= j. As we just proved, 50 MAURICIO A. FERREIRA AND ADRIAN R. WADSWORTH i=1 Sh,i, where (Sh,i, yi,h) is a Henselization of (S, yi) in Falg , the algebraic closure S ⊗F Fh ∼= Qr of F . Therefore, (A.2) K ⊗F Fh ∼= K ⊗S (S ⊗F Fh) ∼= K ⊗F Sh,i. rQi=1 Because K is purely inseparable over S while Sh,i is separable over S, the fields K and Sh,i are linearly disjoint over S; so, K ⊗S Sh,i is a field, which is isomorphic to the compositum K·Sh,i in Falg . The Henselian valuation yi,h on Sh,i has a unique extension to the field K ⊗S Sh,i whose restriction to K is the unique extension of yi to K, which is vi. By Lemma A.3, K ⊗S Sh,i is a Henselization of K with respect to vi. The theorem thus follows from (A.2). (cid:3) References [B] [D] Bourbaki, N., Commutative Algebra, Chap. 1-7, Springer-Verlag, 2nd printing, 1989. Draxl, P. K., Ostrowski's theorem for Henselian valued skew fields. J. Reine Angew. Math. 354 (1984), 213-218. [Du1] Dubrovin, N. I., Noncommutative valuation rings. (Russian) Trudy Moskov. Mat. Obshch. 45 (1982), 265 -- 280; English transl.: Trans. Moscow Math. Soc. 45 (1984), 273-287. [Du2] Dubrovin, N. I., Noncommutative valuation rings in simple finite-dimensional algebras over a field. (Russian) Mat. Sb. (N.S.) 123(165) (1984), 496 -- 509. English transl.: Math. USSR Sb. 51 (1985) 493-505. [Ef] [E] [EP] [G1] [G2] Efrat, I., Valuations, Orderings, and Milnor K-Theory. Amer. Math. Soc., Providence, RI, 2006. Endler, O., Valuation Theory. Springer-Verlag, New York, 1972. Engler, A. J.; Prestel, A., Valued fields. Springer-Verlag, Berlin, 2005. Grater, J., The "Defektsatz" for central simple algebras. Trans. Amer. Math. Soc. 330 (1992), 823 -- 843. Grater, J., Prime PI-rings in which finitely generated right ideals are principal. Forum Math. 4 (1992), 447 -- 463. [HW] Hwang, Y.-S.; Wadsworth, A. R., Correspondences between valued division algebras and graded division algebras. J. Algebra 220 (1999), 73 -- 114. Jacob, B.; Wadsworth, A. R. Division algebras over Henselian fields. J. Algebra 128 (1990), 126 -- 179. [JW] [MMU] Marubayashi, H.; Miyamoto, H.; Ueda, A., Non-commutative valuation rings and semi-hereditary orders. Kluwer, Dordrecht, 1997. [M1] Morandi, P. J., The Henselization of a valued division algebra. J. Algebra 122 (1989), 232-243. [M2] Morandi, P. J., Value functions on central simple algebras. Trans. Amer. Math. Soc. 315 (1989), 605 -- 622. [M3] Morandi, P. J., An approximation theorem for Dubrovin valuation rings. Math. Z. 207 (1991), 71 -- 81. [M4] Morandi, P. J., On defective division algebras, in K-theory and algebraic geometry: connections with quadratic forms and division algebras (Santa Barbara, CA, 1992), eds. B. Jacob and A. Rosenberg, pp. 359 -- 367, Proc. Sympos. Pure Math., 58, Part 2, Amer. Math. Soc., Providence, RI, 1995. [RTW] Renard, J.-F.; Tignol, J.-P.; Wadsworth, A. R., Graded Hermitian forms and Springer's theorem. Indag. Math. (N.S.) 18 (2007), 97 -- 134. [R] [T] Ribenboim, P., Le th´eor`eme d'approximation pour les valuations de Krull. Math. Z. 68 (1957), 1 -- 18. J.-P.; Generalized Tignol, No. 106, Universit´e Catholique de Louvain-La-Neuve, Belgium, 1987, unpublished; http://perso.uclouvain.be/jean-pierre.tignol/Rapport106.pdf. in Seminaires Math´ematiques products, crossed (nouvelle s´erie), available at [TW1] Tignol, J.-P.; Wadsworth A. R., Value functions and associated graded rings for semisimple algebras, Trans. Amer. Math. Soc. 362 (2010), 687 -- 726. [TW2] Tignol, J.-P; Wadsworth, A. R. Valuations on algebras with involution. Math. Ann. 351 (2011) 109 -- 148. [TW3] Tignol, J.-P; Wadsworth, A. R., Value functions on simple algebras, and associated graded algebras, book in preparation. [W1] Wadsworth, A. R., Dubrovin valuation rings and Henselization. Math. Ann. 283 (1989), 301 -- 328. VALUE FUNCTIONS AND DUBROVIN VALUATION RINGS ON SIMPLE ALGEBRAS 51 [W2] Wadsworth, A. R., Valuation theory on finite dimensional division algebras. Valuation theory and its appli- cations, Vol. I (Saskatoon, SK, 1999), eds. F.-V. Kuhlmann et al., pp. 385 -- 449, Fields Inst. Commun., 32, Amer. Math. Soc., Providence, RI, 2002. [Z] Zhao, Y., On the intersection property of Dubrovin valuation rings. Proc. Amer. Math. Soc. 125 (1997), 2825 -- 2830. Departamento de Ciencias Exatas, Universidade Estadual de Feira de Santana, Avenida Transnordes- tina, S/N, Novo Horizonte, Feira de Santana, Bahia 44036-900 Brazil E-mail address: [email protected] Department of Mathematics 0112, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0112 USA E-mail address: [email protected]
1112.1532	1	1112	2011-12-07T11:47:41	On the centralizer of an $I$-matrix in $M_2(R/I)$, $I$ a principal ideal and $R$ a UFD	[ "math.RA" ]	The concept of an $I$-matrix in the full $2\times 2$ matrix ring $M_2(R/I)$, where $R$ is an arbitrary UFD and $I$ is a nonzero ideal in $R$, was introduced in \cite{mar}. Moreover a concrete description of the centralizer of an $I$-matrix $\hat B$ in $M_2(R/I)$ as the sum of two subrings $\mathcal S_1$ and $\mathcal S_2$ of $M_2(R/I)$ was also given, where $\mathcal S_1$ is the image (under the natural epimorphism from $M_2(R)$ to $M_2(R/I)$) of the centralizer in $M_2(R)$ of a pre-image of $\hat B$, and where the entries in $\mathcal S_2$ are intersections of certain annihilators of elements arising from the entries of $\hat B$. In the present paper, we obtain results for the case when $I$ is a principal ideal $<k>$, $k\in R$ a nonzero nonunit. Mainly we solve two problems. Firstly we find necessary and sufficient conditions for when $\mathcal S_1\subseteq\mathcal S_2$, for when $\mathcal S_2\subseteq \mathcal S_1$ and for when $\mathcal S_1=\mathcal S_2$. Secondly we provide a formula for the number of elements in the centralizer of $\hat B$ for the case when $R/<k>$ is finite.	math.RA	math	On the centralizer of an I-matrix in M2(R/I), I a principal ideal and R a UFD Magdaleen S. Marais Abstract. The concept of an I-matrix in the full 2 × 2 matrix ring M2(R/I), where R is an arbitrary UFD and I is a nonzero ideal in R, was introduced in [10]. Moreover a concrete description of the centralizer of an I-matrix bB in M2(R/I) as the sum of two subrings S1 and S2 of M2(R/I) was also given, where S1 is the image (under the natural epimorphism from M2(R) to M2(R/I)) of the centralizer in M2(R) of a pre-image of bB, and where the entries in S2 are intersections of certain annihilators of elements arising from the entries of bB. In the present paper, we obtain results for the case when I is a principal ideal hki, k ∈ R a nonzero nonunit. Mainly we solve two problems. Firstly we find necessary and sufficient conditions for when S1 ⊆ S2, for when S2 ⊆ S1 and for when S1 = S2. Secondly we provide a formula for the number of elements in the centralizer of bB for the case when R/hki is finite. 1. Introduction We denote the centralizer of an element s in an arbitrary ring S by CenS(s). Know- ing that Mn(R), the full n × n matrix ring over a commutative ring R, is a prime example of a non-commutative ring, it is surprising that a concrete description of CenMn(R)(B) for an arbitrary B ∈ Mn(R) has not yet been found. If R[x] is the polynomial ring in the variable x over R, then (1) {f (B) f (x) ∈ R[x]} ⊆ CenMn(R)(B). 2000 Mathematics Subject Classification. 16S50, 15A33, 16D20. Key words and phrases. Centralizer, I-matrix, matrix ring, unique factorization domain, principal ideal domain, enumerate. This research is part of the author's research for her doctoral dissertation which was con- ducted at Stellenbosch University under the direction of L. van Wyk. The financial assistance of the National Research Foundation (NRF) towards this research is hereby acknowledged. Opin- ions expressed and conclusions arrived at are those of the author and are not necessarily to be attributed to the National Research Foundation. 1 2 MAGDALEEN S. MARAIS In fact, it is known that (see [7]) {f (B) f (x) ∈ R[x]} = CenMn(R)(CenMn(R)(B)). The most progress, finding a concrete description of CenMn(R)(B), has been made for the case when the underlying ring R is a field (see [6], [8], [9], [11] and [13]). The following well-known result in this case provides a necessary and sufficient condition for equality in (1). Theorem 1.1. If B is an n × n matrix over a field F , then CenMn(F )(B) = {f (B) f (x) ∈ F [x]} if and only if the minimum polynomial of B coincides with the characteristic poly- nomial of B. The concept of an I-matrix in the full 2 × 2 matrix ring M2(R/I), where R is a UFD and I an ideal in R was introduced in [10]. In this paper, unless stated otherwise, we assume that R is a UFD, I is a nonzero ideal in R and k := gcd(I) 6= 0. Let θI : R → R/I and ΘI : M2(R) → M2(R/I) be the natural epimorphism and induced epimorphism respectively. We denote the image θI (b) of b ∈ R by bI and the image ΘI (B) of B ∈ M2(R) by bBI . However, if there is no ambiguity, then we simply write θ, Θ, b and bB respectively. Definition 1.1. We call a matrix " eI hI # ∈ M2(R/I) an I-matrix fI gI if heI − hI , fIi = htI i or heI − hI , gI i = htI i or h fI , gI i = htI i, where tk. If R is a PID, then every matrix in M2(R/I) is an I-matrix. A concrete description of the centralizer of an I-matrix, as the sum of two subrings of M2(R/I), was given by the following result in [10]: fI gI Theorem 1.2. Let R be a UFD, I a nonzero ideal in R, and let bBI = " eI hI # ∈ M2(R/I) be an I-matrix, then Cen(bB) = S1 + S2, where # . S2 =" ann( f ) ∩ ann(e − h) ann(g) ∩ ann(e − h) S1 = Θ(Cen(B)) and ann( f ) ∩ ann(g) ann( f ) ∩ ann(g) Unfortunately the concrete description in Theorem 1.2 could not be generalized to n × n-matrices, for n ≥ 3, in the sense of Proposition 1.2. In [10], for R a UFD, a matrix was given for every factor ring R/I with zero divisors and every n ≥ 3 for which equality in (2) does not hold. Proposition 1.2. Let R be a commutative ring and let B = [bij ] ∈ Mn(R). Then (2) Θ(Cen(B)) + [Aij ] ⊆ Cen(bB), ON THE CENTRALIZER OF AN I-MATRIX IN M2(R/I), I A PRINCIPAL IDEAL AND R A UFD 3 where Aij = \k, k6=j ann(bjk)\ \k, k6=i ann(bki)\ ann(bii − bjj ). Regarding Theorem 1.2, an example was also provided in [10], where S1 6⊆ S2 and S2 6⊆ S1, from which the following questions arise: When is S1 6⊆ S2, when is S2 6⊆ S1 and when is S1 = S2? In Section 2 this questions will be answered for the case when I ⊂ R is a principal ideal hki generated by a nonzero nonunit k ∈ R. The problem of enumerating the number of matrices with given characteristics over a finite ring has been treated extensively in the literature. Formulas have been found, for example, for the number of matrices with a given characteristic polynomial [12]; the number of matrices over a finite field that are cyclic [1] or symmetric [4]; and the number of matrices over the ring of integers Z modulo m, Zm, that are nilpotent [2]. By using the results in [3], some of the above mentioned results, where the matrices over a finite field that satisfy some property are enumerated by rank, can be extended to matrices over certain finite rings that satisfy the property under consideration. Naturally the question whether it is possible to enumerate the number of ma- trices in CenMn(R)(B), denoted by CenMn(R)(B), when R is a finite commutative ring and B ∈ Mn(R), arises. Using the fact that the dimension of CenMn(F )(B) is known by the following result, due to Frobenius, the answer is straightforward in the case when R is a finite field F . Theorem 1.3. Let B ∈ Mn(F ), and suppose that f1, . . . , fl ∈ F [x] are the invariant factors of B, where fi divides fi−1, for i = 2, . . . , l. Then the dimension of CenMn(F )(B) is given by (2i − 1)(deg fi). lXi=1 For example, if n = 2, then the number of elements in CenMn(F )(B) is F 2, if B is a nonscalar matrix, and it is F 4 if B is a scalar matrix. Unfortunately the answer is not that easy in the case when R is a finite ring that is not a field. In Section 3 we define an equivalence relation on M2(R/hki) and we use this relation, together with Theorem 1.2, and the results in Section 2, to obtain a formula for the number of matrices in CenM2(R/hki)(bB) when R is a UFD and R/hki is finite, k is a nonzero nonunit element in R and bB ∈ M2(R/hki). 2. Containment considerations regarding the centralizer of a hki-matrix In this section we answer the following questions: Regarding Theorem 1.2, when is S1 6⊆ S2, when is S2 6⊆ S1 and when is S1 = S2? 4 MAGDALEEN S. MARAIS We need the following preliminary definitions and results in Theorem 2.7, the main result of this section. Since the minimum polynomial and characteristic polynomial of any 2 × 2 non- scalar matrix over a field coincide, Theorem 1.1 can be written in the following form for the 2 × 2 case. 0 0 a b f (i) M2(F ), if e = h, f = 0 and g = 0 (i.e. B is a scalar matrix)  a, b ∈ F) , if e 6= h, f = 0 and g = 0 Corollary 2.1. Let B =" e g h # ∈ M2(F ), F a field. Then b #(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (ii)(" a 0 b a − g−1b(e − h) #(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (iii)(" a f −1gb a − f −1b(e − h) #(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (iv)(" g h # ∈ M2(R), R a UFD. Then CenM2(R)(B) 0 # + vE(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) a, b ∈ F) , if f = 0 and g 6= 0 a, b ∈ F) , if f 6= 0. (ii)(m−1w" e − h f g Lemma 2.2. Let B =" e f (i) M2(R), if e = h, f = 0 and g = 0 (i.e. B is a scalar matrix) The following result, giving a concrete description of the centralizer of a matrix v, w ∈ R) , if at least one of e − h, f, g is nonzero, where m−1 is the inverse of m := gcd(e − h, f, g) in the quotient field of R. The following four results can be easily proved. Lemma 2.3. Let S be a subring of a ring T and let s ∈ S. Then CenS(s) = S ∩ CenT (s). Lemma 2.4. Let B ∈ Mn(R), where R is a commutative ring. Then CenM2(R)(BT ) = (CenM2(R)(B))T . Lemma 2.5. Let R be a UFD. Suppose b, k ∈ R, k a nonzero nonunit, and δ = gcd(b, k). Then hti = θ−1(ann(bhki)), where t = δ−1k ∈ R, with δ−1 the inverse of δ in the quotient field of R. CenM2(F )(B) = in M2(R), was proved in [10]: =  ON THE CENTRALIZER OF AN I-MATRIX IN M2(R/I), I A PRINCIPAL IDEAL AND R A UFD 5 Lemma 2.6. Let R be a UFD and let k, x, y ∈ R, then ann( d) = ann(x) ∩ ann(y) in R/hki, with gcd(x, y) = d. We are now in a position to prove Theorem 2.6. B =" e (a) Theorem 2.7. Let R be a UFD, k = pn1 1 pn2 2 · · · pnm m and let f g h # ∈ M2(R) be such that bB is a hki-matrix. Then CenM2(R/hki)(bB) = Θhki(CenM2(R)(B)) if and only if B is a scalar matrix or satisfies the following conditions for every i, i = 1, 2, . . . , m: (i) pi is not a divisor of at least one of the elements e − h, f and g; pick such an element a, and call the remaining two elements b and c, say. (ii) gcd(b, c, k) = 1 or ahgcd(b,c,k)i is invertible in R/hgcd(b, c, k)i; (3) (4) (5) (b) (c) Cen(bB) =" Θ(Cen(B)) =" if and only if f = 0 and g = 0; ann( f ) ∩ ann(g) ann(g) ∩ ann(e − h) ann( f ) ∩ ann(e − h) ann( f ) ∩ ann(g) # ann( f ) ∩ ann(g) ann(g) ∩ ann(e − h) ann( f ) ∩ ann(e − h) ann( f ) ∩ ann(g) # if and only if f = 0, g = 0 and (e − h is invertible or e − h = 0). Proof. (a) Since (3) follows trivially if B is a scalar matrix, we assume that B is a non-scalar matrix. Suppose that conditions (i) and (ii) are satisfied. If (6) annM2(R/hki)(θhki(gcd(f, g))) = 0hki, annM2(R/hki)(θhki(gcd(f, e − h))) = 0hki (7) and annM2(R/hki)(θhki(gcd(g, e − h))) = 0hki, then the result follows from Theorem 1.2 and Lemma 2.6. Thus suppose that at least one of the annihilators in (6) and (7) is nonzero. We now show that " 0hki 0hki ann(θhki(gcd(g, e − h))) 0hki 0hki ann(θhki(gcd(f, e − h))) # ," 0hki 0hki # , (8) " 0hki 0hki ann(θhki(gcd(f, g))) # ∈ Θhki(CenM2(R)(B)). 0hki 6 MAGDALEEN S. MARAIS Since then, because Θhki(CenM2(R)(B)) is a ring, (3) follows from Theorem 1.2 and Lemma 2.6. If annM2(R/hki)(θhki(gcd(g, e − h))) 6= 0hki, then, by Lemma 2.5, 1 6= gcd(g, e − h, k) := δ and annM2(R/hki)(θhki(gcd(e − h, g))) = h([kδ−1)hkii. To accomplish our objective, we show that for each dhki ∈ annM2(R/hki)(θhki(gcd(g, e − d′ hki = dhki, h))) there is a d′ since then hki ∈ annM2(R/hki)(θhki(gcd(g, e − h))) such that fhki Θhki " 0 gd′ f d′ (e − h)d′ #! =" 0hki 0hki dhki 0hki # , so that we therefore can conclude from Lemma 2.2(ii) that " 0hki 0hki annM2(R/hki)(θhki(gcd(g, e − h))) 0hki # ∈ Θhki(CenM2(R)(B)). Thus, let dhki be an arbitrary element in annM2(R/hki)(θhki(gcd(g, e − h))), i.e. sup- pose dhki := shki([kδ−1)hki for some shki ∈ R/hki. Since fhδi is invertible in R/hδi, fhδi = 1hδi which implies that tf = 1+vδ by (ii), there is a thδi ∈ R/hδi such that thδi for some v ∈ R. Hence f td = (1 + vδ)(skδ−1 + wk) = skδ−1 + (w + vs + vδw)k. In other words, if we set d′ d′ hki := (btd)hki then fhki hki = fhki(btd)hki = (\skδ−1)hki = dhki. It can similarly be shown that each of the other two sets in (8) is contained in Θhki(CenM2(R)(B)). Conversely, suppose B does not satisfy both of the conditions (i) and (ii). We distinguish between the following cases: (a′) B does not satisfy (i), i.e. gcd(e − h, f, g, k) 6= 1; (b′) B satisfies (i), but not (ii). (a′) Suppose there is a prime pi in the prime factorization of k such that pie−h, f, g. We distinguish between the following two cases: (i′) f = 0 or g = 0; (ii′) f, g 6= 0. (i′) Since pie − h, f, g, direct verification shows that Because θhki(pn1 pni+1 i+1 · · · pnm m ) 6= 0hki, it follows that the entries in po- sition (1, 2) and position (2, 1) of bAhki only have nonzero pre-images in R. Since B is a non-scalar matrix, it follows from Lemma 2.2(ii) that every matrix in CenM2(R)(B) bAhki 0 θ(pn1 1 · · · pni−1 := " ∈ CenM2(R/hki)(bBhki). 1 · · · pni−1 i−1 pni−1 i i−1 pni−1 i pni+1 i+1 · · · pnm m ) θ(pn1 1 · · · pni−1 i−1 pni−1 i 0 pni+1 i+1 · · · pnm m ) # ON THE CENTRALIZER OF AN I-MATRIX IN M2(R/I), I A PRINCIPAL IDEAL AND R A UFD 7 has 0 in position (1, 2) if f = 0 and 0 in position (2, 1) if g = 0. Therefore bAhki 6∈ Θhki(CenM2(R)(B)) if f = 0 or g = 0. (ii′) Since f, g 6= 0 and pif, g it follows that (9) f = cpr i and g = dps i i i 0hki 0hki for some s, r ≥ 1 and c, d ∈ R such that pi ∤ c, d. Now, r ≤ s or s ≤ r. Let us first assume that r ≤ s. Because pie − h, f, g direct verification shows that θhki(pn1 i−1 pni−1 1 · · · pni−1 pni+1 i+1 · · · pnm m ) 0hki # ∈ CenM2(R/hki)(bBhki). bAhki :=" We now show that bAhki 6∈ Θhki(CenM2(R)(B)). Firstly note that the set of all the pre-images of bAhki is ker θhki # . Thus, if bAhki ∈ Θhki(CenM2(R)(B)), then, according to Corollary 2.1(iv) and Lem- ma 2.3 there is a pre-image" m + κ3 κ4 # ∈ M2(R) of bAhki, where κ1, κ2, κ3, κ4 ∈ ker θhki, such that " gf −1b a − (e − h)f −1b # m + κ3 κ4 # =" κ1 pni+1 i−1 pni−1 i+1 · · · pnm κ1 pni+1 i−1 pni−1 i+1 · · · pnm ker θhki pni+1 i+1 · · · pnm 1 · · · pni−1 pn1 1 · · · pni−1 pn1 1 · · · pni−1 pn1 m + ker θhki i−1 pni−1 ker θhki " κ2 a κ2 b i i i i−1 pni−1 1 · · · pni−1 pni+1 i+1 · · · pnm in M2(R) for some a, b ∈ R. In other words, there are a, b ∈ R such that κ1 = a, κ2 = b and pn1 m + κ3 = gf −1b. But then, consider- ing (9) and keeping in mind that r ≤ s, gf −1b ∈ R and pi ∤ c, d, we have that gf −1b = dps in R. Be- pni+1 cause pni i+1 i · · · pnm i i is the ideal generated by pni 1 · · · pni−1 i )−1κ2 ∈ hpni i−1 pni−1 m +κ3, it follows that pn1 i (cpr 1 · · · pni−1 i i, where hpni pni+1 i+1 · · · pnm i i, which implies that pni+1 i+1 · · · pnm 1 · · · pni−1 pn1 m + κ3 6= gf −1b. m + κ3 6∈ hpni i−1 pni−1 i−1 pni−1 ∤ pn1 i i i i If s ≤ r one can similarly show that Thus we have a contradiction. Therefore bAhki 6∈ Θhki(CenM2(R)(B)). # ∈ CenM2(R/hki)(bBhki), bAhki :=" 0hki and that bAhki 6∈ Θhki(CenM2(R)(B)), by using Lemma 2.4 and Corollary 2.1(iv) instead of Corollary 2.1(iv). pni+1 i+1 · · · pnm m ) 1 · · · pni−1 i−1 pni−1 θhki(pn1 i 0hki 0hki (b′) Suppose B satisfies (i), but not (ii). Then, for each i ∈ {1, . . . , m}, at least one of the following cases is true: 8 MAGDALEEN S. MARAIS (i′) gcd(e − h, f, g, k) = 1 , 1 6= gcd(e − h, g, k) := δ and fhδi is not invertible in R/hδi; (ii′) gcd(e − h, f, g, k) = 1 , 1 6= gcd(e − h, f, k) := δ and ghδi is not invertible in R/hδi; (iii′) gcd(e − h, f, g, k) = 1 , 1 6= gcd(f, g, k) := δ and ehδi − hhδi is not invertible in R/hδi; We now show that (3) does not follow in each of the above cases. (i′) In this case Lemma 2.5 implies that annM2(R/hki)(θhki(gcd(g, e − h))) = h([kδ−1)hkii. Note that since δ is not a unit, hkδ−1i 6= hki. By Theorem 1.2 it follows that ([kδ−1)hki 0hki # ∈ CenM2(R/hki)(bB). 0hki bAhki :=" 0hki " ker θhki ker θhki If we can show that bAhki 6∈ Θhki(CenM2(R)(B)), then we are finished. Now, k−1δ + ker θhki # ker θhki is the set of all the pre-images of bAhki in R. Therefore, taking into account that gcd(e − h, f, g, k) = 1, if bAhki ∈ Θhki(CenM2(R)(B)) it follows from Corol- # ∈ M2(R) of bAhki, where lary 2.2(ii) that there is a pre-image" κ1 # =" a gb a − (e − h)b # κ1, κ2, κ3, κ4 ∈ ker θhki, such that " κ1 kδ−1 + κ2 kδ−1 + κ2 κ3 κ4 κ3 κ4 f b for some a, b ∈ R. Hence, gb = κ3 and (e − h)b = κ1 − κ4, which implies that b = skδ−1 for some s ∈ R. But then, since f b = kδ−1 + κ2, we have that f b = f skδ−1 = kδ−1 + κ2 ⇔ f s = 1 + tδ for some t ∈ R ⇔ fhδishδi = 1hδi. Since fhδi is not invertible in R/hδi, according to assumption, we have a contradic- tion. Therefore bAhki 6∈ Θhki(CenM2(R)(B)) and so we conclude that CenM2(R/hki)(bBhki) 6⊆ Θhki(CenM2(R)(B)). (ii′ and iii′) It follows similarly to case (i′) that CenM2(R/hki)(bB) 6⊆ Θ(CenM2(R)(B)). ON THE CENTRALIZER OF AN I-MATRIX IN M2(R/I), I A PRINCIPAL IDEAL AND R A UFD 9 (b) Suppose f , g = 0, then f, g ∈ hki, and so by Corollary 2.2(ii) Θ(Cen(B)) = Θ (" a = Θ (" a ⊆ " = " R/hki ann(e − h) f b gb a − (e − h)b #(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) a, b ∈ R)! 0 a − (e − h)b #(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) a, b ∈ R)! # ann(e − h) R/hki 0 ann( f ) ∩ ann(g) ann( f ) ∩ ann(e − h) ann(g) ∩ ann(e − h) ann( f ) ∩ ann(g) # . Conversely, suppose Θ(Cen(B)) ⊆" Since" a 0 0 a # ∈ Θ(Cen(B)) for every a ∈ R/hki it follows that ann( f ) ∩ ann(e − h) ann( f ) ∩ ann(g) ann(g) ∩ ann(e − h) ann( f ) ∩ ann(g) #. ann( f ) ∩ ann(g) = R/hki which implies that ann( f ) = R/hki and ann(g) = R/hki and so f , g = 0. (c) Using (b) and (a), it follows that Θ(Cen(B)) =" ⇔ Θ(Cen(B)) ⊆" ann( f ) ∩ ann(g) ann(g) ∩ ann(e − h) ann( f ) ∩ ann(e − h) ann( f ) ∩ ann(g) ann( f ) ∩ ann(g) ann(g) ∩ ann(e − h) ann( f ) ∩ ann(e − h) ann( f ) ∩ ann(g) # # and " ann( f ) ∩ ann(g) ann(g) ∩ ann(e − h) ann( f ) ∩ ann(e − h) ann( f ) ∩ ann(g) # ⊆ Θ(Cen(B)) ⇔ f , g = 0 and " ann( f ) ∩ ann(g) ann(g) ∩ ann(e − h) ann( f ) ∩ ann(g) ⇔ f , g = 0 and (e − h is invertible in R/hki or e − h = 0). ann( f ) ∩ ann(e − h) # ⊆ Θ(Cen(B)) (cid:3) 0 0 0 x2 0 # and B′′ = " 1 + xyz #, Example 2.8. Let R = F [x, y, z], k = x3y2z and let B =" x2y2 x + 1 0 #. Note that bB, bB′ and bB′′ are B′ = " x2y2 a + dx2y2b #(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) a, b ∈ F [x, y, z]/hx3y2zi) . hx3y2zi-matrices. Since gcd(x2y2, x2) = x2 and ( [x + 1)hx2i is invertible in R/hx2i, it follows from Lemma 2.2(ii) and Theorem 2.7(a) that Cen(bBhki) = Θ(Cen(B)) = (" a cx2b ( [x + 1)b 0 0 10 MAGDALEEN S. MARAIS Furthermore, it follows from Theorem 2.7(b) that Cen(bB′ hki) =" R/hx3y2zi hcxzi R/hx3y2zi # hcxzi and, since θhx3y2zi(1 + xyz) is invertible in R/hx3y2zi, from Theorem 2.7(c) that hki)) = " ann( f ) ∩ ann(g) ann(g) ∩ ann(e − h) ann( f ) ∩ ann(e − h) ann( f ) ∩ ann(g) # Cen(bB′′) = Θ(Cen(B′′ = " R/hx3y2zi 0 0 R/hx3y2zi # . The following result is well-known. Lemma 2.9. Let R be a PID. Then an element b ∈ R/hki is invertible if and only if gcd(b, k) = 1. Using Lemma 2.9 and the fact that every matrix in M2(R) is a hki-matrix if R is a PID, we simplify Theorem 2.7(a) for the case when R is a PID. Corollary 2.10. Let R be a PID and let B =" e Cen(bB) = Θ(Cen(B)) if and only if B is a scalar matrix or gcd(e − h, f, g, k) = 1. f g h # ∈ M2(R). Then Note that although Corollary 2.11 is not a characterization of the hki-matrices for which (3) is true, it is easier to verify if Corollary 2.11 applies to a specific matrix in M2(R) than to verify if Theorem 2.7(a) applies to a specific matrix in M2(R). Corollary 2.11. Let R be a UFD, k ∈ R and B =" e g h # ∈ M2(R). If at f least one of the three elements e − h, f and g is invertible in R/hki, then Cen(bB) = Θ(Cen(B)). Proof. It follows trivially that bB is a hki-matrix. Without loss of generality, let us suppose that f is invertible in R/hki. Then, by Lemma 2.9 gcd(f, k) = 1. Hence condition (i) in Theorem 2.7(a) is satisfied. Now, suppose that gcd(e − h, g, k) = δ. If δ is a unit, then condition (ii) is also satisfied. Thus suppose that δ is not a unit. Then, since fhki is invertible in R/hki and δk, it follows that there is a t ∈ R such that tf = 1 + sk = 1 + svδ for some s, v ∈ R which implies that thδi fhδi = 1hδi. Therefore condition (ii) in Theorem 2.7(a) is also satisfied. (cid:3) ON THE CENTRALIZER OF AN I-MATRIX IN M2(R/I), I A PRINCIPAL IDEAL AND R A UFD11 3. The number of matrices in the centralizer of a matrix in M2(R/hki), R a UFD and R/hki finite In this section k ∈ R will always be a nonzero nonunit such that R/hki is finite and we will always denote the number of elements in a ring S by S. The purpose of this section is to determine the number of matrices in CenM2(R/hki)(B), where R is a UFD, R/hki is finite and B ∈ M2(R/hki). To reach our goal, we first need some preliminary results. g h # ∈ M2(R) and let d := gcd(e − Definition 3.1. Let k ∈ R, let B =" e h, f, g, k). We define the relation ∼ on CenM2(R/hki)(bBhki) as follows: for bAhki,bChki ∈ CenM2(R/hki)(bBhki), iff f bAhki ∼ bChki bAhki − bChki ∈ M2(h \(kd−1)hkii). It follows immediately that ∼ is an equivalence relation. hki and the set We denote the equivalence class of bAhki by bA∗ Since (10) hki bA∗ ann( fhki) ∩ ann(ghki) ann( fhki) ∩ ann(ghki) ann(ehki − hhki) ∩ ann(ghki) ann(ehki − hhki) ∩ ann( fhki) M2(h \(kd−1)hkii) ⊆" # , it follows from Theorem 1.2 that M2(h \(kd−1)hkii) ⊆ CenM2(R/hki)(bBhki). Therefore each equivalence class in (CenM2(R/hki)(bBhki))∗ has h \(kd−1)hkii4 elements. We define addition ⊞ and multiplication ⊡ on (CenM2(R/hki)(bBhki))∗ by ⊡ bC∗ hki = (bAhki ⊡ bChki)∗. ⊞ bC∗ hki = (bAhki + bChki)∗ triple h(CenM2(R/hki)(bBhki))∗, ⊞, ⊡i is a ring, which we sometimes, if the context is clear, denote by (CenM2(R/hki)(bBhki))∗. It is easy to show that the binary operations ⊞ and ⊡ are well-defined and that the Using the following well-known result, Corollary 3.3 can easily be proved. and hki bA∗ Theorem 3.2. If A1, . . . , Am are ideals in a ring S (not necessarily commuta- tive or with a unit), then there is a monomorphism of rings φ : S/(A1 ∩· · ·∩Am) → S/A1 ⊕ · · · ⊕ S/Am defined by φ(s + (A1 ∩ · · · ∩ Am)) = (s + A1, . . . , s + Am). If S2 + Ai = S for all i and Ai + Aj = S for all i 6= j, then φ is an isomorphism of rings. of all equivalence classes by {bA∗ hki bAhki ∈ (CenM2(R/hki)(bBhki))} (CenM2(R/hki)(bBhki))∗. 12 MAGDALEEN S. MARAIS Corollary 3.3. Let R/hki be finite, and let k = pn1 1 pn2 2 · · · pnm m , with p1, . . . , pm different primes and n1, . . . , nm ≥ 1. Then (i) φ : R/hki → R/hpn1 1 i ⊕ R/hpn2 2 i ⊕ · · · ⊕ R/hpnm m i defined by φ(r) = (θhp n1 1 i(r), θhp n2 2 i(r), · · · , θhpnm m i(r)) is an isomorphism. (ii) Φ : M2(R/hki) → M2(R/hpn1 1 i) ⊕ · · · ⊕ M2(R/hpnm m i) defined by Φ([bij]) = (Θhp n1 1 i([bij ]), . . . , Θhpnm m i([bij ])) is an isomorphism. We need the following trivial results in the next Lemma 3.6. Lemma 3.4. Let S, S1, . . . , Sm be rings, s ∈ S and let Γ : S → S1 ⊕ · · · ⊕ Sm defined by Γ(s) = (s1, . . . , sm) be an isomorphism. Then t ∈ CenS(s) if and only if ti ∈ CenSi(si), for all i. Lemma 3.5. Let R/hki be finite. An element b ∈ R/hki is invertible if and only if gcd(b, k) = 1. Lemma 3.6. Let B =" e = 1, then f g h # ∈ M2(R) and let k ∈ R. If gcd(e − h, f, g, k) Cen(bBhki) = R/hki2. m , where p1, . . . , pm are different primes and 2 · · · pnm Proof. Suppose k = pn1 1 pn2 ni ≥ 1 for all i. It follows from Lemma 3.3(ii) and Lemma 3.4 that Therefore, CenM2(R/hp ni i i). CenM2(R/hp ni i i). ni i i)(bBhp i i)(bBhp ni mMi=1 CenM2(R/hki)(bBhki) ∼= mYi=1 CenM2(R/hki)(bBhki) = i i)(bBhp CenM2(R/hki)(bBhki) = mYi=1 If we can show that CenM2(R/hp from Lemma 3.3(ii) and Lemma 3.4, that ni i i) = R/hpni ni i i2, for all i, it follows, again R/hpni i i2 = R/hki2. Let pi be an arbitrary prime in the prime factorization of k. Since gcd(e − h, f, g, k) = 1, it follows that pi ∤ f or pi ∤ g or pi ∤ e − h. Thus, by Lemma 3.5, at least one of fhp i i is invertible in R/hpni i i − hhp i i or ehp i i, ghp i i. ni ni ni ni ON THE CENTRALIZER OF AN I-MATRIX IN M2(R/I), I A PRINCIPAL IDEAL AND R A UFD13 If f is invertible in R/hpni = 1, it follows from Corollary 2.11 and Lemma 2.2(ii) that i i with inverse t, say, then given that gcd(e − h, f, g, pni i ) It can be similarly shown that if g is invertible in R/hpni i i with inverse t, say, then CenM2(R/hp (11) (12) CenM2(R/hp (13) CenM2(R/hp ni 0 1 # + b" 0 0 1 # + b" 0 ni g i i)(bB) = Cen " e = (a" 1 0 i i)(bB) = (a" 1 0 i i)(bB) = (a" 1 0 ni f 1 1 tg t f th #! h #! = Cen " te tg −t(e − h) #(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 1 −t(e − h) #(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 1 #(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) g h # ∈ M2(R) and let −t f −tg f 0 1 # + b" 0 i i) . i i) ; a, b ∈ R/hpni a, b ∈ R/hpni a, b ∈ R/hpni i i) . It is easy to see that the number of elements in the sets in (11), (12) and (13) are R/hpni i i2. and if e − h is invertible in R/hpni i i with inverse t, say, then Lemma 3.7. Let k ∈ R, let B =" e B′ =" d−1(e − h) d−1f d−1g 0 # , where d := gcd(e − h, f, g, k), then Proof. Since ghki fhki hkd−1i). (CenM2(R/hki)(bBhki))∗ ∼= CenM2(R/hkd−1i)(cB′ hki ∈ (CenM2(R/hki)(bBhki))∗ ⇔ bAhki ∈ CenM2(R/hki)(bBhki) bA∗ ⇔ bAhki ∈ CenM2(R/hki) " ehki − hhki 0 # −" e − h f ⇔ A" e − h f ⇔ AB′ − B′A ∈ M2(hkd−1i) ⇔ bAhkd−1i ∈ CenM2(R/hkd−1i)(cB′ hki = bC∗ bA∗ hki ⇔ bAhki − bChki ∈ M2(h[kd−1 0hki #! 0 # A ∈ M2(hki) ⇔ A − C ∈ M2(hkd−1i) ⇔ bAhkd−1i = bChkd−1i. hkii) g g and (cid:3) hkd−1i) 14 MAGDALEEN S. MARAIS it follows that Γ : (CenM2(R/hki)(bBhki))∗ → CenM2(R/hkd−1i)(cB′ hkd−1i), defined by is a well-defined function which is 1 − 1 and onto. It can be easily shown that Γ is a homomorphism. (cid:3) Γ(bA∗) = bAhkd−1i, We are finally able to determine the number of elements in the centralizer of a matrix in M2(R/hki), if R is a UFD and R/hki is finite. Theorem 3.8. Suppose R is a UFD, k ∈ R is a nonzero nonunit such that R/hki is finite, and B =" e g h # ∈ M2(R), f then where d :=gcd(e − h, f, g, k). CenM2(R/hki)(bBhki) = R/hkd−1i2 · h([kd−1)hkii4, Proof. With B′ as in Lemma 3.7, it follows from Lemma 3.6 that CenM2(R/hkd−1i)(cB′ hkd−1i) = R/hkd−1i2. Since each equivalence class in (CenM2(R/hki)(bBhki))∗ has cardinality h([kd−1)hkii4, it follows that and so Lemma 3.7 implies that CenM2(R/hki)(bBhki) = (CenM2(R/hki)(bBhki))∗h([kd−1)hkii4, CenM2(R/hki)(bBhki) = CenM2(R/hkd−1i)(cB′ = R/hkd−1i2h([kd−1)hkii4. hkd−1i)h([kd−1)hkii4 Example 3.9. Let R = Z[i], k = 12 so that R/hki = Z12[i] (see [5], p. 604, (cid:3) Using the fact that every matrix is a hki-matrix if R is a PID, note that, according to Lemma 2.2(ii) and Theorem 1.2 Theorem 1) and let b9i CenM2(Z12[i])(bBh12i) = Θh12i (" a = (" a + b4c c3ib +c4n # . bB =" b4i 3 +b6i bi a − 3ib #(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (1 +b2i)b + c4m a −c3ib (1 + 2i)b 3ib h4i h4i 0 # a, b ∈ Z[i])! +" h4i #(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) a, b, c, m, n ∈ Z12[i]) . ON THE CENTRALIZER OF AN I-MATRIX IN M2(R/I), I A PRINCIPAL IDEAL AND R A UFD15 Now, since gcd(3i, 3 + 6i, 9i, 12) = 3, let d = 3 so that kd−1 = 12 · 3−1 = 4. Since Z[i]/h4i = {a + ib a, b ∈ Z4} = 16 and h4h12ii = 9 it follows from Theorem 3.8 that CenM2(Z12[i])(bBh12i) = 162 · 94 = 1679616. For 2 × 2 matrices over a factor ring of Z we have the following result. Corollary 3.10. Let bB =" e g d = gcd(e − h, f, g, k). f h # ∈ M2(Zk), then Cen(bB) = (kd)2, where Proof. According to Theorem 3.8 CenM2(Zk)(bBhki) = Zkd−1 2h([kd−1)hkii4 = (kd−1)2d4 = (kd)2. Example 3.11. Let bBh12i = " 2h12i 8h12i # . Since gcd(6, 2, 4, 12) = 2, it 2h12i 4h12i (cid:3) follows that CenM2(Z12)(bBh12i) = (12 · 2)2 = 242 = 576. Remark 3.12. A natural example to include in this section, if such an example exists, would be one of a UFD R, which is not a PID, and a nonzero nonunit k ∈ R, such that R/hki is finite. Unfortunately we could not find such an example. Neither have we been able to prove that if R is UFD and k ∈ R is a nonzero nonunit such that R/hki is finite, then R is a PID. References [1] E.R. Berlekamp, Distribution of cyclic matrices in a finite field, Duke Math. J. 33 (1966), 45 -- 48. [2] D. Bollman and H. Ram´ιrez, On the number of nilpotent matrices over Zm, Reine Angew. Math. 238 (1969), 85 -- 88. [3] D. Bollman and H. Ram´ιrez, On the enumeration of matrices over finite commutative rings, Amer. Math. Monthly 76 (1969), 1019 -- 1023. [4] L. Carlitz, Representations by quadratic forms in a finite field, Duke Math. J. 21 (1954), 123 -- 137. [5] G. Dresden and W.M. Dym`acek, Finding factors of factor rings over the Gaussian integers, Amer. Math. Monthly 112(7) (2005), 602 -- 611. [6] P.R. Halmos, Finite-dimensional vector spaces, 2nd edition, Van Nostrand Company, Prince- ton, N.J., 1958. [7] X. Hou, On the centralizer of the centralizer of a matrix, Linear Algebra Appl. 256 (1997), 251 -- 261. [8] T.W. Hungerford, Algebra, Holt, Rinehart and Winston, New York, London, 1974. 16 MAGDALEEN S. MARAIS [9] N. Jacobson, Lectures in abstract algebra, Vol. II, Van Nostrand Company, Princeton, N.J., 1953. [10] M.S. Marais, The Centralizer of an I-matrix in M2(R/I), R a UFD, Algebra Colloquium, to appear. [11] G.C. Nelson and T. Ton-That, Multiplicatively closed bases for C(A), Note Mat. 26(2) (2006), 81 -- 104. [12] I. Reiner, On the number of matrices with given characteristic polynomial, Illinois J. Math. 5 (1961), 324 -- 329. [13] D.A. Suprunenko and R.I. Tyshkevich, Commutative matrices, Academic Press, New York, 1968. E-mail address: [email protected] African Institute for Mathematical Sciences, 6 Melrose Rd, Muizenberg, 7945, Cape Town, South Africa
1101.1563	1	1101	2011-01-08T01:20:51	Gr\"obner-Shirshov bases for categories	[ "math.RA" ]	In this paper we establish Composition-Diamond lemma for small categories. We give Gr\"obner-Shirshov bases for simplicial category and cyclic category.	math.RA	math	Grobner-Shirshov bases for categories∗ L. A. Bokut† School of Mathematical Sciences, South China Normal University Guangzhou 510631, P. R. China Sobolev Institute of Mathematics, Russian Academy of Sciences Siberian Branch, Novosibirsk 630090, Russia [email protected] Yuqun Chen‡ and Yu Li School of Mathematical Sciences, South China Normal University Guangzhou 510631, P. R. China [email protected] [email protected] Abstract: In this paper we establish Composition-Diamond lemma for small categories. We give Grobner-Shirshov bases for simplicial category and cyclic category. Key words: Grobner-Shirshov basis, simplicial category, cyclic category. AMS 2000 Subject Classification: 16S15, 13P10, 18G30, 16E40 1 Introduction This paper devotes to Grobner-Shirshov bases for small categories (all categories below are supposed to be small) presented by a graph (=quiver) and defining relations (see, Maclane [57]). As important examples, we use the simpicial and the cyclic categories (see, for example, Maclane [58], Gelfand, Manin [43]). In an above presentation, a category is viewed as a "monoid with several objects". A free category C(X), generated by a graph X, is just "free partial monoid of partial words" u = xi1 . . . xin, n ≥ 0, xi ∈ X and all product defined in C(X). A relation is an expression u = v, u, v ∈ C(X), where sources and targets of u, v are coincident respectively. The same as for semigroups, we may use two equivalent languages: Grobner-Shirshov bases language and rewriting systems language. Since we are using the former, we need a Composition-Diamond lemma (CD-lemma for short) for a free associative partial algebra kC(X) over a field k, where kC(X) is just a linear combination of uniform (with the same sources and targets) partial ∗Supported by the NNSF of China (Nos.10771077, 10911120389). †Supported by RFBR 09-01-00157, LSS -- 3669.2010.1 and SB RAS Integration grant No. 2009.97 (Russia). ‡Corresponding author. 1 words. Then it is a routing matter to establish CD-lemma for kC(X). It is a free category ("semigroup") partial algebra of a free category. Remark that in the literature one is usually used a language of rewriting system, see, for example, Malbos [53]. Let us stress that the partial associative algebras presented by graphs and defining relations are closely related to the well known quotients of the path algebras from representation theory of finitely dimensional algebras, see, for example, Assem, Simson, Skowro´nski [1]. In this respect Grobner-Shirshov bases for categories are closely related to non-commutative Grobner bases for quotients of path algebras, see Farkas, Feustel, Green [42]. Rewriting system language for non-commutative Grobner bases of quotients of path algebras was used by Kobayashi [48]. Main new results of this paper are Grobner-Shirshov bases for the simplicial and cyclic categories. All algebras are assumed to be over a field. 2 A short survey on Grobner-Shirshov bases What is now called Grobner and Grobner-Shirshov bases theory was initiated by A. I. Shirshov (1921-1981) [66, 67], 1962 for non-associative and Lie algebras, by H. Hiron- aka [45, 46], 1964 for quotients of commutative infinite series algebras (both formal and convergent), and by B. Buchberger [32, 33], 1965, 1970 for commutative algebras. English translation of selected works of A. I. Shirshov, including [66, 67], is recently published [68]. Remark that Shirshov's approach was a most universal as we understand now since Lie algebra case becomes a model for many classes of non-commutative and non-associative algebras (with multiple operations), starting with associative algebras (see below). Hi- ronaka's papers on resolution of singularities of algebraic varieties become famous very soon and Hironaka got Fields Medal due to them few years latter. B. Buchberger's thesis influenced very much many specialists in computer sciences, as well as in commutative algebras and algebraic geometry, for huge important applications of his bases, named him under his supervisor W. Grobner (1898-1980). Original Shirshov's approach for Lie algebras [67], 1962, based on a notion of com- position [f, g]w of two monic Lie polynomials f, g relative to associative word w, i.e., f, g are elements of a free Lie algebra Lie(X) regarded as the subspace of Lie poly- nomials of the free associative algebra khXi, and w ∈ X ∗, the free monoid generated by X. The definition of Lie composition relies on a definition of associative composi- tion (f, g)w as (monic) associative polynomials (after worked out into f, g all Lie brackets [x, y] = xy−yx) relative to degree-lexicographical order on X ∗. Namely, (f, g)w = f b−ag, where w = acb, ¯f = ac, ¯g = cb, a, b, c ∈ X ∗, c 6= 1. Here ¯f means the leading (maximal) associative word of f . Then (f, g)w belongs to associative ideal Id(f, g) of khXi generated by f, g, and the leading word of (f, g)w is less than w. Now we need to put some Lie brackets [f b] − [ag] on f b − ag in such a way that the result would belong to Lie ideal generated by f, g (so we can not trouble bracketing into f, g) and the leading associa- tive monomial of [f b] − [ag] must be less than w. To overcome these obstacles Shirshov used his previous paper [64], 1958 with a new linear basis of free Lie algebra Lie(X). As it happened the same linear basis of Lie(X) was discovered in the paper Chen, Fox, Lyndon [34], 1958. Now this basis is called Lyndon-Shirshov basis, or, by a mistake, 2 Lyndon basis. It consists of non-associative Lyndon-Shirshov words (NLSW) [u] in X, that are in one-one correspondence with associative Lyndon-Shirshov words (ALSW) u in X. The latter is defined as by a property u = vw > wv for any v, w 6= 1. Shirshov [64], 1958 introduced and used the following properties of both associative and non-associative Lyndon-Shirshov words: 1) For any ALSW u there is a unique bracketing [u] such that [u] is a NLSW. There are two algorithms for bracketing an ALSW. He mostly used "down-to-up al- gorithm" to rewrite an ALSW u on a new alphabet Xu = {xi(xβ)j, i > β, j ≥ 0, xβ is the minimal letter in u}; the result u′ is again ALSW on Xu with the lex-order xi > xixβ > ((xixβ)xβ) > . . . . It is Shirshov's rewriting or elimination algorithm from his famous paper Shirshov [63], 1953, on what is now called Shirshov-Witt theorem (any subalgebra of a free Lie algebra is free). This rewriting was rediscovered by Lazard [49], 1960 and now called as Lazard elimination (it is better to call Lazard-Shirshov elimination). There is "up-to-down algorithm" (see Shirshov [65], 1958, Chen, Fox, Lyndon [34], 1958): [u] = [[v][w]], where w is the longest proper end of u that is ALSW, in this case v is also an ALSW. 2) Leading associative word of NLSW [u] is just u (with the coefficient 1). 3) Leading associative word of any Lie polynomial is an associative Lyndon-Shirshov word. 4) A non-commutative polynomial f is a Lie polynomial if and only if f = f0 → · · · → fi → fi+1 → · · · → fn = 0, where fi → fi+1 = fi − αi[ui], ¯fi = ui is an ALSW, αi is the leading coefficient of fi, i = 0, 1, . . . . 5) Any associative word c 6= 1 is the unique product of (not strictly) increasing sequence of associative Lyndon-Shirshov words: c = c1c2 · · · cn, c1 ≤ · · · ≤ cn, ci are ALSW's. 6) If u = avb, where u, v are ALSW, a, b ∈ X ∗, then there is a relative bracketing [u]v = [a[v]b] of u relative to v, such that the leading associative word of [u]v is just u. Namely, [u] = [a[vc]d], cd = b, [u]v = [a[[v[c1]] . . . [cn]]d], c = c1c2 · · · cn as above. 7) If ac and cb are ASLW's and c 6= 1 then acb is an ALSW as well. If a, b are ALSW's and a > b, then ab is an ALSW as well. Property 5) was known to Chen, Fox, Lyndon [34], 1958 as well. Lyndon [52], 1954, was actually the first for definition of associative "Lyndon-Shirshov" words. To the best of our knowledge, for many years, until PhD thesis by Viennot [69], 1978, no one mentioned the Lyndon's discovery in 1954. On the other hand, there were dozens of papers and some books on Lie algebras that mentioned both associative and non-associative "Lyndon- Shirshov" words as Shirshov's regular words, see, for example, P. M. Cohn [38], 1965, Bahturin [2], 1978. Now a Lie composition [f, g]w of monic Lie polynomials f, g relative to a word w = ¯f b = a¯g = acb, c 6= 1 is defined by Shirshov [67], 1962, as follows [f, g]w = [f b] ¯f − [ag]¯g, where [f b] ¯f means the result of substitution f for [ ¯f ] into the relative bracketing [w]ac of w with respect to ¯f = ac, the same for [ag]¯g. 3 According to the definition and properties above, any Lie composition [f, g]w is an ele- ment of the Lie ideal generated by f, g, and the leading associative word of the composition is less than w. The composition above is now called composition of intersection. Shirshov avoided what is now called composition of inclusion [f, g]w = f − [agb]¯g, w = ¯f = a¯gb, assuming that any system S of Lie polynomials is reduced (irreducible) in a sense that leading associative word of any polynomial from S does not contain leading associative words of another polynomials from S. This assumption relies on his algorithm of elimi- nation of leading words for Lie polynomials below. For associative polynomials the elimination algorithm is just non-commutative version of the Euclidean elimination algorithm. For Lie polynomial case Shirshov [67], 1962, defined the elimination of a leading word as follows: If w = avb, where w, v are ALSW's, and v is the leading word of some monic Lie polynomial f , then the transformation [w] 7→ [w] − [af b]v is called an elimination of leading word of f into [w]. The result of Lie elimination is a Lie polynomial with a leading associative word less than w. Then Shirshov [67], 1962, formulated an algorithm to add to an initial reduced system of Lie polynomials S a "non-trivial" composition [f, g]w, where f, g belong to S. Non- triviality of a Lie polynomial h relative to S means that h is not going to zero using "elimination of leading words of S". Actually, he defines to add to S not just a composition but rather the result of elimination of leading words of S into the composition in order to have a reduced system as well. Then Shirshov proved the following Composition Lemma. Let S be a reduced subset of Lie(X). If f belongs to the Lie ideal generated by S, then the leading associative word ¯f contains, as a subword, some leading associative word of a reduced multi-composition of elements of S. He constantly used the following clear Corollary. The set of all irreducible NLSW's [u] such that u does not contain any leading associative word of a reduced multi-composition of elements of S is a linear basis of the quotient algebra Lie(X)/Id(S). Some later (see Bokut [8], 1972) the Shirshov Composition lemma was reformulated in the following form: Let S be a closed under composition set of monic Lie polynomials (it means that any composition [f, g]w of intersection and inclusion of elements of S is [aisibi] = aisibi < w, ai, bi ∈ X ∗, si ∈ S, αi ∈ k). If trivial, i.e., [f, g]w = P αi[aisibi], f ∈ Id(S), then ¯f = a¯sb for some s ∈ S. And S-irreducible NLSW's is a linear basis of the quotient algebra Lie(X)/Id(S). The modern form of Shirshov's lemma is the following (see, for example, Bokut, Chen [12]). Shirshov's Composition-Diamond Lemma for Lie algebras. Let Lie(X) be a free Lie algebra over a field, S monic subset of Lie(X) relative to some monomial order on X ∗. Then the following conditions are equivalent: 1) S is a Grobner-Shirshov basis (i.e., any composition of intersection and inclusion of elements of S is trivial). 4 2) If f ∈ Id(S), then ¯f = a¯sb for some s ∈ S, a, b ∈ X ∗. 3) Irr(S) = {[u][u] is an NLSW and u does not contain any ¯s, s ∈ S} is a linear basis of the Lie algebra Lie(XS) with defining relations S. The proof of the Shirshov's Composition-Diamond lemma for Lie algebras becomes a model for proofs of number of Composition-Diamond lemmas for many classes of algebras. An idea of his proof is to rewrite any element of Lie ideal, generated by S in a form X αi[aisibi], where each si ∈ S, ai, bi ∈ X ∗, αi ∈ k such that i) leading words of each [aisibi] is equal to aisibi (in this case an expression [asb] is called normal Lie S-word in X) and ii) a1s1b1 > a2s2b2 > . . . . Now let F (X) be a free algebra of a variety (or category) of algebras. Following the idea of Shirshov's proof, one needs 1) to define appropriate linear basis (normal words) of F(X), 2) to define monomial order of normal words, 3) to define compositions of element of S (they may be compositions of intersection, inclusion and left (right) multiplication, or may be else), 4) prove two key lemmas: Key Lemma 1. Let S be a Grobner-Shirshov basis (any composition of polynomials from S is trivial). Then any S-word is a linear combination of normal S-words. Key Lemma 2. Let S be a Grobner-Shirshov basis, [a1s1b1] and [a2s2b2] normal S-words, s1, s2 ∈ S. If a1s1b1 = a2s2b2, then [a1s1b1] − [a2s2b2] is going to zero by elimination of leading words of S (elimination means composition of inclusion). There are number of CD-lemmas that realized Shirshov's approach to them. Shirshov [67], 1962, assumed implicitly that his approach, based on the definition of composition of any (not necessary Lie) polynomials, is equally valid for associative al- gebras as well (the first author is a witness that Shirshov understood it very clearly and explicitly; only lack of non-trivial applications prevents him from publication this approach for associative algebras). Explicitly it was done by Bokut [9] and Bergman [5]. CD-lemma for associative algebras is formulated and proved in the same way as for Lie algebras. Composition-Diamond Lemma for associative algebras. Let khXi be a free associative algebra over a field k and a set X. Let us fix some monomial order on X ∗. Then the following conditions are equivalent for any monic subset S of khXi: 1) S is a Grobner-Shirshov basis (that is any composition of intersection and inclusion is trivial). 2) If f ∈ Id(S), then ¯f = a¯sb for some s ∈ S, a, b ∈ X ∗. 3) Irr(S) = {u ∈ X ∗u 6= a¯sb, s ∈ S, a, b ∈ X ∗} is a linear basis of the factor algebra khXSi = khXi/Id(X). There are a lot of applications of Shirshov's CD-lemmas for Lie and associative algebras. Let us mention some connected to the Malcev embedding problem for semigroup algebras 5 (Bokut [6, 7], 1969, there is a semigroup S such that the multiplication semigroup of the semigroup algebra k(S), where k is a field, is embeddable into a group, but k(S) is not embeddable into any division algebra), the unsolvability of the word problem for Lie algebras (Bokut [8]), Grobner-Shirshov bases for semisimple Lie algebras (Bokut, Klein [25, 26, 27, 28]), Kac-Moody algebras (Poroshenko [59, 60, 61]), finite Coxeter groups (Bokut, Shiao [30]), braid groups in different set of generators (Bokut, Chainikov, Shum [23], Bokut [10], Bokut [11]), quantum algebra of type An (Bokut, Malcolmson [29]), Chinese monoids (Chen, Qiu [35]). There are applications of Shirshov's CD-lemma [66], 1962 for free anti-commutative non-associative algebras: there are two anti-commutative Grobner-Shirshov bases of a free Lie algebra, one gives the Hall basis (Bokut, Chen, Li [17]), another the Lyndon- Shirshov basis (Bokut, Chen, Li [18]). Bokut, Chen, Mo [20] proved and reproved some embedding theorems for associative algebras, Lie algebras, groups, semigroups, differential algebras, using Shirshov's CD- lemmas for associative and Lie algebras. Bahturin, Olshanskii [3] found embeddings without distortion of associative algebras and Lie algebras into 2-generated simple algebras. They also used Shirshov's CD-lemmas for associative and Lie algebras. Mikhalev [54] used Shirshov's approach and CD-lemma for associative algebras to prove CD-lemma for colored Lie super-algebras. Mikhalev, Zolotykh [56] proved CD-lemma for free associative algebra over a commuta- tive algebra.A Free object in this category is k[Y ] ⊗ khXi, tensor product of a polynomial algebra and a free associative algebra. Here one needs to use several compositions of intersection and inclusion. Bokut, Fong, Ke [24] proved CD-lemma for free associative conformal (in a sense of V. Kac [47] algebra C(X, (n), n = 0, 1, . . . , D, N(a, b), a, b ∈ X) of a fixed locality N(a, b). A linear basis of free associative conformal algebra was constructed by M. Roitman [62]. Any normal conformal word has a form [u] = a1(n1)[a2(n2)[. . . [ak(nk)Diak+1] . . . ]], where aj ∈ X, nj < N(aj, aj+1), i ≥ 0. The same word without brackets is called the leading associative word of [u]. One needs to use external multi-operator semigroup as a set of leading associative words of conformal polynomials (it is the same as for Lie algebras), several compositions of inclusion and intersection, and new compositions of left (right) multiplication (last compositions are absent into classical cases). Also in the CD-lemma for conformal algebras we have 1) ⇒ 2) ⇔ 3), but in general from 2) does not follow 1). Here conditions 2) and 3) are formulated in terms of associative leading words, the same as for Lie algebras. We see that CD-lemma for associative conformal algebras has a lot of in common with CD-lemma for Lie algebras, but there are also some differences. Though PBW-theorem is not valid for Lie conformal algebras (M. Roitman [62]), some generic intersection compositions for universal enveloping algebra U(L) of any Lie conformal algebra L are trivial (it is called "1/2 PBW theorem"). Bokut, Chen, Zhang [22] proved CD-lemma for associative n-conformal algebras, where instead of one derivation D and polynomial algebra k[D] one has n derivations D1, . . . , Dn and polynomial algebra k[D1, . . . , Dn]. This case is treated in the same way as for n = 1. A more general case, the associative H-conformal algebra (or H-pseudo-algebra in a sense of Bakalov, D'Andrea, Kac [4]), where H is any Hopf algebra, is still open. Mikhalev, Vasilieva [55] proved CD-lemma for free supercommutative polynomial alge- 6 bras. Here they use compositions of multiplication as well. Bokut, Chen, Li [16] proved CD-lemma for free pre-Lie algebras (also known as Vinberg- Koszul-Gerstenhaber right-symmetric algebras). Bokut, Chen, Liu [19] proved CD-lemma for free dialgebras in a sense of Loday [50]. Here conditions 1) and 2) are not equivalent but from 1) follows 2). The cases of associative conformal algebras and dialgebras show that definition of Grobner-Shirshov bases by condition 1) is in general preferable than the one using 2). Bokut, Shum [31] proved CD-lemma for free Γ-associative algebras, where Γ is a group. It has applications to the Malcev problem above and to Bruhat normal forms for algebraic groups. Eisenbud, Peeva, Sturmfels [41] found non-commutative Grobner basis of any commu- tative algebra (extending any commutative Grobner basis to a non-commutative one). Bokut, Chen, Chen [14] proved CD-lemma for Lie algebras over commutative algebras. Here one needs to establish Key Lemma 1 in a more strong form -- any Lie S-word is a linear combination of S-words of the form [asb]¯s in the sense of Shirshov's special Lie bracketing. As an application they proved Cohn's conjecture [37] for the case of characteristics 2, 3 and 5 (that some Cohn's examples of Lie algebras over commutative algebras are not embeddable into associative algebras over the same commutative algebras). Bokut, Chen, Deng [15] proved CD-lemma for free associative Rota-Baxter algebras. As an application, Chen and Mo [36] proved that any dendriform algebra is embeddable into universal enveloping Rota-Baxter algebra. It was Li Guo's conjecture, [44]. Bokut, Chen, Chen [13] proved CD-lemma for tensor product of two free associa- tive algebras. As an application they extended any Mikhalev-Zolotyh commutative-non- commutative Grobner-Shirshov basis laying into tensor product k[Y ] ⊗ khXi to non- commutative-non-commutative Grobner-Shirshov basis laying into khY i ⊗ khXi (a la Eisenbud-Peeva-Sturfels above). They also gave another proof of the Eisenbud-Peeva- Sturmfekls theorem above. As we mentioned in introduction, Farkas, Feustel, Green [42] proved CD-lemma for path algebras. Drensky, Holtkamp [40] proved CD-lemma for nonassociative algebras with multiple linear operators. Bokut, Chen, Qiu [21] proved CD-lemma for associative algebras with multiple linear operators. Dotsenko, Khoroshkin [39] proved CD-lemma for operads. 3 Free categories and free category partial algebras Let X = (V (X), E(X)) be an oriented (multi) graph. Then the free category on X is C(X) = (Ob(X), Arr(X)), where Ob(X) = V (X), and Arr(X) is the set of all paths ("words") of X including the empty paths 1v, v ∈ V (X). It is easy to check C(X) has the following universal property. Let C be a category and ΓC the graph relative to C i.e., V (ΓC ) = Ob(C ) and E(ΓC ) = mor(C ). Let e : X → C(X) be a mono graph morphism of the graph X to the graph ΓC(X), where e = (e1, e2), and e1 is a mapping on V (X), e2 on E(X), both e1 and e2 are mono. For any graph morphism 7 b from X to ΓC , where b = (b1, b2), and b1 is mono, there exists a unique category morphism (a functor) f : C(X) → C , such that the corresponding diagram is commutative i.e., f e = b. Therefore each category C is a homomorphic image of a free category C(X) for some graph X and thus C is isomorphic to C(X)/ρ(S) for some set S, where S = {(u, v)u, v have the same sources and the same targets} ⊆ Arr(X) × Arr(X) and ρ(S) the congruence of C(X) generated by S. If this is the case, X is called the generating set of C and S the relation set of C and we denote C = C(XS). Let C be a category and k a field. Let kC = {f = n Xi=1 αiµiαi ∈ k, µi ∈ mor(C ), n ≥ 0, µi (0 ≤ i ≤ n) have the same domains and the same codomains}. Note that in kC , for f, g ∈ kC , f + g is defined only if f, g have the same domain and the same codomain. A multiplication • in kC is defined by linearly extending the usual compositions of morphisms of the category C . Then (kC , •) is called the category partial algebra over k relative to C and kC(X) the free category partial algebra generated by the graph X. 4 Composition-Diamond lemma for categories Let X be a oriented (multi) graph, C(X) the free category generated by X and kC(X) the free category partial algebra. Since we only consider the morphisms of the free category C(X), we write C(X) just for Arr(X). Note that for f, g ∈ kC(X) if we write gf , it means gf is defined. A well ordering > on C(X) is called monomial if it satisfies the following conditions: In fact, there are many u > v ⇒ uw > vw and wu > wv, for any u, v, w ∈ C(X). monomial orders on C(X). For example, let E(X) be a well ordered set. Then the deg-lex order > on C(X) is defined by the following way: for any words u = x1 · · · xm, v = y1 · · · yn ∈ C(X), m = u, n = v, u > v ⇐⇒ u > v or (u = v and x1 = y1, x2 = y2, . . . , xt = yt, xt+1 > yt+1 for some 0 ≤ t < n). It is easy to check that > is a monomial order on C(X). In the following sections, we will see other monomial orders. Now, we suppose that > is a fixed monomial order on C(X). Given a nonzero polynomial f ∈ kC(X), it has a word ¯f ∈ C(X) such that f = αf + P αiui, where f > ui, 0 6= α, αi ∈ k, ui ∈ C(X). We call f the leading term of f and f is monic if α = 1. Let S ⊂ kC(X) be a set of monic polynomials, s ∈ S and u ∈ C(X). We define S-word us by induction: (i) us = s is an S-word of s-length 1. 8 (ii) Suppose that us is an S-word of s-length m and v is a word of length n, i.e., the number of edges in v is n. Then usv and vus are S-words of s length m + n. Note that for any S-word us = asb, where a, b ∈ C(X), we have asb = a¯sb. Let f, g be monic polynomials in kC(X). Suppose that there exist w, a, b ∈ C(X) such that w = ¯f = a¯gb. Then we define the composition of inclusion (f, g)w = f − agb. For the case that w = ¯f b = a¯g, w, a, b ∈ C(X), the composition of intersection is defined as follows: (f, g)w = f b − ag. It is clear that (f, g)w ∈ Id(f, g) and (f, g)w < w, where Id(f, g) is the ideal of kC(X) generated by f, g. The composition (f, g)w is trivial modulo (S, w), if (f, g)w = Xi αiaisibi where each αi ∈ k, ai, bi ∈ C(X), si ∈ S, aisibi an S-word and ai ¯sibi < ¯f . If this is the case, then we write (f, g)w ≡ 0 mod(S, w). In general, for p, q ∈ kC(X), we write p ≡ q mod(S, w) which means that p − q = P αiaisibi, where each αi ∈ k, ai, bi ∈ C(X), si ∈ S, aisibi an S-word and ai ¯sibi < w. Definition 4.1 Let S ⊂ kC(X) be a nonempty set of monic polynomials. Then S is called a Grobner-Shirshov basis in kC(X) if any composition (f, g)w with f, g ∈ S is trivial modulo (S, w), i.e., (f, g)w ≡ 0 mod(S, w). Lemma 4.2 Let a1s1b1, a2s2b2 be monic S-words. If S is a Grobner-Shirshov basis in kC(X) and w = a1s1b1 = a2s2b2, then a1s1b1 ≡ a2s2b2 mod(S, w). Proof. There are three cases to consider. Case 1. Suppose that subwords ¯s1 and ¯s2 of w are disjoint, say, a2 ≥ a1 + ¯s1. Then, we can assume that a2 = a1¯s1c and b1 = c¯s2b2 for some c ∈ C(X), and so, w = a1¯s1c¯s2b2. Now, a1s1b1 − a2s2b2 = a1s1c¯s2b2 − a1¯s1cs2b2 = a1s1c(¯s2 − s2)b2 + a1(s1 − ¯s1)cs2b2. Since s2 − s2 < ¯s2 and s1 − s1 < ¯s1, we conclude that a1s1b1 − a2s2b2 = Xi αiuis1vi +Xj βjujs2vj 9 for some αi, βj ∈ k, S-words uis1vi and ujs2vj such that ui¯s1vi, uj ¯s2vj < w. This shows that a1s1b1 ≡ a2s2b2 mod(S, w). Case 2. Suppose that the subword ¯s1 of w contains ¯s2 as a subword. We may assume that ¯s1 = a¯s2b, a2 = a1a and b2 = bb1, that is, w = a1a¯s2bb1 for some S-word as2b. We have a1s1b1 − a2s2b2 = a1s1b1 − a1as2bb1 = a1s1 − as2bb1 = a1(s1, s2)s1b1 ≡ 0 mod(S, w) since S is a Grobner-Shirshov basis. Case 3. ¯s1 and ¯s2 have a nonempty intersection as a subword of w. We may assume that a2 = a1a, b1 = bb2, w1 = ¯s1b = a¯s2. Then, we have a1s1b1 − a2s2b2 = a1s1bb2 − a1as2b2 = a1(s1b − as2)b2 = a1(s1, s2)w1b2 ≡ 0 mod(S, w) This completes the proof. (cid:3) Lemma 4.3 Let S ⊂ kC(X) be a subset of monic polynomials and Irr(S) = {u ∈ C(X)u 6= a¯sb, a, b ∈ C(X), s ∈ S}. Then for any f ∈ kC(X), f = Xui≤ ¯f αiui + Xaj sj bj ≤ ¯f βjajsjbj where each αi, βj ∈ k, ui ∈ Irr(S) and ajsjbj an S-word. αiui ∈ kC(X), where 0 6= αi ∈ k and u1 > u2 > · · · . If u1 ∈ Irr(S), Proof. Let f = Pi then let f1 = f − α1u1. If u1 6∈ Irr(S), then there exist some s ∈ S and a1, b1 ∈ C(X), such that ¯f = u1 = a1 ¯s1b1. Let f1 = f − α1a1s1b1. In both cases, we have ¯f1 < ¯f . Then the result follows from the induction on ¯f . (cid:3) Theorem 4.4 (Composition-Diamond lemma for categories) Let S ⊂ kC(X) be a nonempty set of monic polynomials and < a monomial order on C(X). Let Id(S) be the ideal of kC(X) generated by S. Then the following statements are equivalent: (i) S is a Grobner-Shirshov basis in kC(X). (ii) f ∈ Id(S) ⇒ ¯f = a¯sb for some s ∈ S and a, b ∈ C(X). (ii)′ f ∈ Id(S) ⇒ f = α1a1s1b1 + α2a2s2b2 + · · · , where each αi ∈ k, aisibi is an S-word and a1¯s1b1 > a2¯s2b2 > · · · . 10 (iii) Irr(S) = {u ∈ C(X)u 6= a¯sb a, b ∈ C(X), s ∈ S} is a linear basis of the partial algebra kC(X)/Id(S) = kC(XS). Proof. (i) ⇒ (ii). Let S be a Grobner-Shirshov basis and 0 6= f ∈ Id(S). Then, we have f = n Xi=1 αiaisibi, where each αi ∈ k, ai, bi ∈ C(X), si ∈ S and aisibi an S-word. Let wi = aisibi, w1 = w2 = · · · = wl > wl+1 ≥ · · · , l ≥ 1. We will use the induction on l and w1 to prove that f = asb for some s ∈ S and a, b ∈ C(X). If l = 1, then f = a1s1b1 = a1s1b1 and hence the result holds. Assume that l ≥ 2. Then, by Lemma 4.2, we have a1s1b1 ≡ a2s2b2 mod(S, w1). Thus, if α1 + α2 6= 0 or l > 2, then the result holds by induction on l. For the case α1 + α2 = 0 and l = 2, we use the induction on w1. Now, the result follows. (ii) ⇒ (ii)′. Assume (ii) and 0 6= f ∈ Id(S). Let f = α1f + · · · . Then, by (ii), f = a1s1b1. Therefore, f1 = f − α1a1s1b1, f1 < f , f1 ∈ Id(S). Now, by using induction on f , we have (ii)′. (ii)′ ⇒ (ii). This part is clear. αiui = 0 in kC(XS), where αi ∈ k, ui ∈ Irr(S). It αiui ∈ Id(S) in kC(X). Then all αi must be equal to zero. Otherwise, (ii) ⇒ (iii). Suppose that Pi means that Pi Pi Now, by Lemma 4.3, (iii) follows. αiui = uj ∈ Irr(S) for some j which contradicts (ii). (iii) ⇒ (i). For any f, g ∈ S , by Lemma 4.3 and (iii), we have (f, g)w ≡ 0 mod(S, w). Therefore, S is a Grobner-Shirshov basis. (cid:3) Remark. If the category in Theorem 4.4 has only one object, then Theorem 4.4 is exact Composition-Diamond lemma for free associative algebras. 5 Grobner-Shirshov bases for the simplicial category and the cyclic category In this section, we give Grobner-Shirshov bases for the simplicial category and the cyclic category respectively. 11 For each non-negative integer p, let [p] denote the set {0, 1, 2, . . . , p} of integers in their usual order. A (weakly) monotonic map µ : [q] → [p] is a function on [q] to [p] such that i ≤ j implies µ(i) ≤ µ(j). The objects [p] with morphisms all weakly monotonic maps µ constitute a category L called simplicial category. It is convenient to use two special families of monotonic maps εi q : [q − 1] → [q], ηi q : [q + 1] → [q] defined for i = 0, 1, ...q (and for q > 0 in the case of εi) by εi q(j) = (cid:26) j, j + 1, ηi q(j) = (cid:26) j, j − 1, if i > j, if i ≤ j, if i ≥ j, if j > i. Let X = (V (X), E(X)) be an oriented (multi) graph, where V (X) = {[p] p ∈ Z +∪{0}} q : [q + 1] → [q] p > 0, 0 ≤ i ≤ p, 0 ≤ j ≤ q}. Let and E(X) = {εi S ⊆ C(X) × C(X) be the relation set consisting of the following: p : [p − 1] → [p], ηj fq+1,q : gq,q+1 : hq−1,q : q = εj q+1εj−1 εi qηi ηj q+1 = ηi ηj q−1εi q =   j > i, j ≥ i, q+1εi q, qηj+1 q+1, q−1ηj−1 εi q−2, 1q−1, q−2, q−1ηj εi−1 j > i, i = j, i > j + 1. i = j + 1, Then the simplicial category L is just the category C(XS) generated by X with defining relation S, see Maclane [58], Theorem VIII. 5.2. We will give another proof in what follows. p > ηj q iff p > q or (p = q and i < j). We order C(X) by the following way. q ∈ {ηi Firstly, for any ηi Secondly, for each u = ηi1 pp ≥ 0, 0 ≤ i ≤ p}, ηi p1ηi2 pn ∈ {ηi p2 · · · ηin p, ηj pp ≥ 0, 0 ≤ i ≤ p}* (all possible words pp ≥ 0, 0 ≤ i ≤ p}, including the empty word 1v, v ∈ Ob(X)), let wt(u) = pn, ηin−1 pp ≥ 0, 0 ≤ i ≤ p}, u > v iff wt(u) > wt(v) p1). Then for any u, v ∈ {ηi pn−1, · · · , ηi1 on {ηi (n, ηin lexicographically. q ∈ {εi Thirdly, for any εi Finally, for each u = v0εi1 p, εj let wt(u) = (n, v0, v1, · · · , vn, εi1 p, p ∈ Z +, 0 ≤ i ≤ p}, εi p > εj p1v1εi2 p2 · · · εin p1, · · · , εin pnvn ∈ C(X), n ≥ 0, vj ∈ {ηi pn). Then for any u, v ∈ C(X), q iff p > q or (p = q and i < j). pp ≥ 0, 0 ≤ i ≤ p}, u ≻1 v ⇔ wt(u) > wt(v) lexicographically. It is easy to check that the ≻1 is a monomial order on C(X). Then we have the following theorem. Theorem 5.1 Let X, S be defined as the above, the generating set and the relation set of the quotient category C(XS) respectively. Then with the order ≻1 on C(X), S is a Grobner-Shirshov basis for the category partial algebra kC(XS). 12 Proof. According to the order ≻1, ¯fq+1,q = εi q+1εj−1 q , ¯gq,q+1 = ηj qηi q+1 and ¯hq−1,q = ηj q−1εi q. So, all the possible compositions of S are the following: (a) (fq+2,q+1, fq+1,q ) εk q+2 εi q+1 j−1 q ε , k ≤ i ≤ j − 1; (b) (gq−1,q, gq,q+1) ηk q−1 j q ηi η q+1 (c) (hq,q+1, fq+1,q ) q εi ηk q+1 j−1 q ε , , i ≤ j ≤ k; i ≤ j − 1; (d) (gq−2,q−1, hq−1,q ) , j ≤ k. ηk q−2 η j q−1 εi q We will prove that all possible compositions are trivial. Here, we only give the proof of the (b). For others cases, the proofs are similar. Let us consider the following subcases of the case (b): (I) i < j < k; (II) i < j, j = k, or j = k + 1; (III) i < k, k + 1 < j; (IV) j > k + 1, i = k, k + 1; (V) j > i > k + 1. For subcase (I), (hq,q+1, fq+1,q) ηk q εi q+1 j−1 q ε For subcase (II), (hq,q+1, fq+1,q) ηk q εi q+1 j−1 q ε For subcase (III), (hq,q+1, fq+1,q ) q εi ηk q+1 j−1 q ε For subcase (IV), (hq,q+1, fq+1,q) ηk q εi q+1 j−1 q ε For subcase (V), (hq,q+1, fq+1,q ) ηk q εi q+1 j−1 q ε = εi qηk−1 q−1 εj−1 q − ηk q εj q+1εi q qεj−1 q−1ηk−2 ≡ εi ≡ 0 mod(S, ηk q−2 − εj q εi qηk−1 q−1 εi q+1εj−1 q q = εi qηk−1 q−1 εj−1 q − ηk q εj q+1εi q ≡ 0 mod(S, ηk q εi q+1εj−1 q ). ). = εi qηk−1 q−1 εj−1 q − ηk q εj q+1εi q qεj−2 q−1ηk−1 ≡ εi ≡ 0 mod(S, ηk q−2 − εj−1 q ηk q+1εj−1 q−1εi q ). q εi q = εj−1 q − ηk q εj q+1εi q q − εj−1 ≡ εj−1 q ηk ≡ 0 mod(S, ηk q−1εi q εi q q+1εj−1 q ). = εi−1 q ηk q−1εj−1 q − ηk q εj q+1εi q q−1ηk q εj−2 ≡ εi−1 ≡ 0 mod(S, ηk q εi q−2 − εj−1 q ηk q+1εj−1 q−1εi ). q q 13 Therefore S is a Grobner-Shirshov basis of the category partial algebra kC(XS). (cid:3) By Theorem 4.4, Irr(S) = {εi1 q−1p ≥ i1 > ... > im ≥ 0, 0 ≤ j1 < ... < jn < q, and q − n + m = p} is a linear basis of the category partial algebra kC(XS). Therefore, we have the following corollaries. p ...εim p−m+1ηj1 q−n...ηjn Corollary 5.2 (Maclane [58], Lemma VIII. 5.1) In the category C(XS), each morphism µ : [q] → [p] can be uniquely represented as p ...εim εi1 p−m+1ηj1 q−n...ηjn q−1, where p ≥ i1 > ... > im ≥ 0, 0 ≤ j1 < ... < jn < q, and q − n + m = p. Corollary 5.3 (Maclane [58], Theorem VIII. 5.2) L = C(XS). The cyclic category is defined by generators and defining relations as follows, see [43]. Let Y = (V (Y ), E(Y )) be an oriented (multi) graph, where V (Y ) = {[p] p ∈ Z + ∪ {0}}, and E(Y ) = {εi q : [q + 1] → [q], tq : [q] → [q] p > 0, 0 ≤ i ≤ p, 0 ≤ j ≤ q}. Let S ⊆ C(Y ) × C(Y ) be the set consisting of the following relations: p : [p − 1] → [p], ηj fq+1,q : gq,q+1 : hq−1,q : ρ1 : ρ2 : ρ3 : q = εj q+1εj−1 εi ηj qηi q+1 = ηi ηj q−1εi q =   q = εi−1 tqεi q = ηi−1 tqηi q tq+1 q = 1q. q tq−1, tq+1, j > i, j ≥ i, q+1εi q, qηj+1 q+1, q−1ηj−1 εi q−2, 1q−1, q−2, i = 1, ..., q, q−1ηj εi−1 ı = 1, ..., q, j > i, i = j, i > j + 1, i = j + 1, The category C(Y S) is called cyclic category, denoted by Λ. An order on C(Y ) is defined by the following way. Firstly, for any ti p, tj Secondly, for any ηi q ∈ {ηi Thirdly, for each u = w0ηi1 q ∈ {tqq ≥ 0}∗, (tp)i > (tq)j iff i > j or (i = j and p > q). p, ηj pp ≥ 0, 0 ≤ i ≤ p}, ηi p2 · · · wn−1ηin pnwn ∈ {tq, ηi p1w1ηi2 p > ηj wi ∈ {tqq ≥ 0}∗, let wt(u) = (n, w0, w1, · · · , wn, ηin {tq, ηi pq, p ≥ 0, 0 ≤ i ≤ p}, u > v iff wt(u) > wt(v) lexicographically. pn−1, · · · , ηi1 pn, ηin−1 q iff p > q or (p = q and i < j). pq, p ≥ 0, 0 ≤ i ≤ p},where p1). Then for any u, v ∈ Fourthly, for any εi p, εj q ∈ {εi p, p ∈ Z +, 0 ≤ i ≤ p}, εi p > εj q iff p > q or (p = q and i < j). Finally, for each u = v0εi1 p1v1εi2 p2 · · · εin pnvn ∈ C(Y ), n ≥ 0, vj ∈ {tq, ηi pq, p ≥ 0, 0 ≤ i ≤ p}*, let wt(u) = (n, v0, v1, · · · , vn, εi1 p1, · · · , εin pn). Then for any u, v ∈ C(Y ), u ≻2 v ⇔ wt(u) > wt(v) lexicographically. It is also easy to check the order ≻2 is a monomial order on C(Y ), which is an extension of ≻1. Then we have the following theorem. 14 Theorem 5.4 Let Y , S be defined as the above, the generating set and the relation set of cyclic category C(Y S) respectively. Let SC = S ∪ {ρ4, ρ5}, where ρ4 : ρ5 : Then tqε0 tqη0 q = εq q, q t2 q = ηq q+1. (1) With the order ≻2 on C(Y ), SC is a Grobner-Shirshov basis for the cyclic category partial algebra kC(Y S). (2) For each morphism µ : [q] → [p] in the cyclic category Λ = C(Y S), µ can be uniquely represented as p ...εim εi1 p−m+1ηj1 q−n...ηjn q−1tk q , where p ≥ i1 > ... > im ≥ 0, 0 ≤ j1 < ... < jn < q, 0 ≤ k ≤ q and q − n + m = p. Proof. ¯ρ1 = tqεi It is easy to check that ¯fq+1,q = εi q, ¯ρ2 = tqηi q, ¯ρ3 = tq+1 q , ¯ρ4 = tqε0 q+1εj−1 , ¯gq,q+1 = ηj q, and ¯ρ5 = tqη0 q . q qηi q+1, ¯hq−1,q = ηj q−1εi q, First of all, we prove Id(S) = Id(SC). q q qεq−1 =tq (ρ3, ρ1)tq+1 εq q q tq q ≡ tqη0 q+1 − ηq ηq polynomial tqη0 and thus ρ5 ∈ Id(S). q tq qtq q ≡ tqε0 q−1 −εq tq−1 −εq q tq q , ρ4 and tqη0 q+1 − ηq q+1. Therefore (tqη0 q is tqη0 q ≡ tqε0 q tq q+1 − ηq It suffices to show ρ4, ρ5 ∈ Id(S). Since tq+1 − q ∈ Id(S). Clearly, the leading term of the q = ρ4 and (ρ3, ρ2)tq+1 qηq−1 q −εq =tq ηq q q q q tq q+1 − ηq q , ρ3)tq η0 = −tqη0 q + ηq q t2 q+1 q tq+2 q+1 Secondly, we prove that all possible compositions of SC are trivial which are the fol- lowing: (a) (fq+2,q+1, fq+1,q ) εk q+2 εi q+1 j−1 q ε , k ≤ i ≤ j − 1; (b) (gq−1,q, gq,q+1) ηk q−1 j q ηi η q+1 (c) (hq,q+1, fq+1,q ) ηk q εi q+1 j−1 q ε , , i ≤ j ≤ k; i ≤ j − 1; (d) (gq−2,q−1, hq−1,q ) , j ≤ k; ηk q−2 η j q−1 εi q (e) (ρ1, fq+1,q ) tq+1εi q+1 j−1 q ε , j > i and i = 1, 2, . . . , q; (f) (ρ3, ρ1)tq+1 q , εi q i = 1, 2, . . . , q; (g) (ρ2, gq,q+1)tq ηj q ηi q+1 (h) (ρ2, hq,q+1)tq ηj q εi q+1 , , j ≥ i and j = 1, 2, . . . , q; j ≥ i and j = 1, 2, . . . , q; i = 1, 2, . . . , q; (i) (ρ3, ρ2)tq+1 q ηi q (j) (ρ3, ρ4)tq+1 q ε0 q (k) (ρ3, ρ5)tq+1 q η0 q , ; ; 15 (l) (ρ4, fq+1,q ) tq+1ε0 q+1 j−1 q ε , j > 0; (m) (ρ5, gq,q+1) tq η0 q η0 q+1 (n) (ρ5, hq,q+1) tq η0 q εi q+1 ; , i ≥ 0. Here, we only give the proof of the case (n) (ρ6, hq,q+1) . The others can be tq η0 q εi q+1 similarly proved. Let us consider the following subcases of the case (n): (I) i = 0; (II) i = 1; (III) i > 1. For subcase (I), (ρ6, hq,q+1) For subcase (II), (ρ6, hq,q+1) tq η0 q ε0 q+1 tq η0 q ε1 q+1 For subcase (III), (ρ6, hq,q+1) tq η0 q εi q+1 = ηq q+1 − tq q+1ε0 q t2 q tq+1εq+1 q εq q+1 − tq ≡ ηq ≡ ηq ≡ 0 mod(S, tqη0 q+1tq − tq q ε0 q+1). = ηq q+1 − tq q t2 q+1ε1 q tq+1ε0 q εq+1 q+1tq − tq ≡ ηq ≡ ηq ≡ 0 mod(S, tqη0 q+1tq − tq q ε1 q+1). = ηq q t2 q εi−2 q+1εi q+1t2 q+1 − tqεi−1 q − εi−2 ≡ ηq ≡ 0 mod(S, tqη0 q η0 tq−1η0 q εi q+1). q q−1 q−1 Thus SC is a Grobner-Shirshov basis of the category partial algebra kC(Y S). Now, by Theorem 4.4, for each morphism µ : [q] → [p] in Λ = C(Y S) can be uniquely represented as p ...εim εi1 p−m+1ηj1 q−n...ηjn q−1tk q , where p ≥ i1 > ... > im ≥ 0, 0 ≤ j1 < ... < jn < q, 0 ≤ k ≤ q and q − n + m = p. (cid:3) Remark. According to Loday [51], the uniqueness property in Theorem 5.4 (2) was known. References [1] I. Assem, D. Simson, A. Skowro´nski, Elements of the representation theory of asso- ciative algebras, London Mathematical Society Student Texts 65, 2006. 16 [2] Yu. A. Bahturin, Lectures on Lie Algebras, Akademie-Verlag, Berlin, 1978. [3] Yu. Bahturin, A. Olshanskii, Filtrations and distortion in infinite-dimensional alge- bras, arXiv:1002.0015. [4] B. Bakalov, A. D'Andrea, V.G. Kac, Theory of finite pseudoalgebras, Adv. Math., 162(1)(2001), 1-140. [5] G. M. Bergman, The diamond lemma for ring theory, Adv. Math., 29(1978), 178-218. [6] L. A. Bokut, Groups of fractions of multiplication semigroups of certain rings I, II, III, Sibir. Math. J., 10(1969), 246-286, 744-799, 800-819. [7] L. A. Bokut, On the Malcev problem, Sibir. Math. J., 10(5)(1969), 965-1005. [8] L. A. Bokut, Unsolvability of the word problem, and subalgebras of finitely presented Lie algebras, Izv. Akad. Nauk. SSSR Ser. Mat., 36(1972), 1173-1219. [9] L. A. Bokut, Imbeddings into simple associative algebras, Algebra i Logika, 15(1976), 117-142. [10] L. A. Bokut, Grobner-Shirshov bases for braid groups in Artin-Garside generators, J. Symbolic Computation, 43(2008), 397-405. [11] L. A. Bokut, Grobner-Shirshov bases for the braid group in the Birman-Ko-Lee generators, J. Algebra, 321(2009), 361-379. [12] L. A. Bokut, Yuqun Chen, Grobner-Shirshov bases for Lie algebras: after A.I. Shir- shov, Southeast Asian Bull. Math., 31(2007), 1057-1076. [13] L. A. Bokut, Yuqun Chen, Yongshan Chen, Composition-Diamond lemma for tensor product of free algebras, J. Algebra, 323(2010), 2520-2537. [14] L. A. Bokut, Yuqun Chen, Yongshan Chen, Grobner-Shirshov bases for Lie algebras over a commutative algebra, arXiv:1006.3217 [15] L. A. Bokut, Yuqun Chen, Xueming Deng, Grobner-Shirshov bases for Rota-Baxter algebras, Siberian Mathematical Journal, 51(6)(2010), 978C988. [16] L. A. Bokut, Yuqun Chen, Yu Li, Grobner-Shirshov bases for Vinberg-Koszul- Gerstenhaber right-symmetric algebras, Fundamental and Applied Mathematics, 14(8)(2008), 55-67. (in Russian) [17] L. A. Bokut, Yuqun Chen, Yu Li, Anti-commutative Grobner-Shirshov basis of a free Lie algebra, Sci. China, 52(2009), 244-253. [18] L. A. Bokut, Yuqun Chen, Yu Li, Anti-commutative Grobner-Shirshov basis of a free Lie algebra relative to Lyndon-Shirshov words, preprint. [19] L. A. Bokut, Yuqun Chen, Cihua Liu, Grobner-Shirshov bases for dialgebras, Inter- national Journal of Algebra and Computation, 20(3)(2010), 391-415. 17 [20] L. A. Bokut, Yuqun Chen, Qiuhui Mo, Grobner-Shirshov bases and embeddings of algebras, International Journal of Algebra and Computation, 20(7)(2010), 875-900. [21] L. A. Bokut, Yuqun Chen, Jianjun Qiu, Grobner-Shirshov bases for associative al- gebras with multiple operators and free Rota-Baxter algebras, Journal of Pure and Applied Algebra, 214(2010), 89-100. [22] L. A. Bokut, Yuqun Chen, Guangliang Zhang, Composition-Diamond lemma for associative n-conformal algebras, arXiv:0903.0892 [23] L. A. Bokut, V. V. Chainikov, K. P. Shum, Markov and Artin normal form theorem for braid groups, Comm. Algebra, 35(2007), 2105-2115. [24] L. A. Bokut, Y. Fong, W. F. Ke, Composition Diamond Lemma for associative conformal algebras, J. Algebra, 272(2004), 739-774. [25] L. A. Bokut, A. A. Klein, Serre relations and Grobner-Shirshov bases for simple Lie algebras I, Internat. J. Algebra Comput., 6(4)(1996), 389-400. [26] L. A. Bokut, A. A. Klein, Grobner-Shirshov bases for the exceptional Lie algebras E6, E7, E8, Algebras and Combinatorics, An International Congress, ICAC'97, Hong Kong, Eds. Kar-Ping Shum, Earl J Taft, Zhe-Xian Wan, Springer, 1997, 37-46. [27] L. A. Bokut, A. A. Klein, Grobner-Shirshov bases for exceptional Lie algebras I, J. Pure and Applied Algebra, 133(1998), 51-57. [28] L. A. Bokut, A. A. Klein, Serre relations and Grobner-Shirshov bases for simple Lie algebras II, Internat. J. Algebra Comput., 6(4)(1996), 401-412. [29] L. A. Bokut, P. Malcolmson, Grobner-Shirshov bases for relations of a Lie algebra and its enveloping algebra, Algebra and Combinatorics (Hong Kong), 47-54, Springer, Singapore, 1999. [30] L. A. Bokut, L.-S. Shiao, Grobner-Shirshov bases for Coxeter groups, Comm. Alge- bra, 29(2001), 4305-4319. [31] L. A. Bokut, K. P. Shum, Relative Grobner-Shirshov bases for algebras and groups, St Petersburg Math. J., 19(2008), N6, 867-881. [32] B. Buchberger, An algorithm for finding a basis for the residue class ring of a zero- dimensional polynomial ideal [in German], Ph.D. thesis, University of Innsbruck, Aus- tria, (1965). [33] B. Buchberger, An algorithmical criteria for the solvability of algebraic systems of equations [in German], Aequationes Math., 4(1970), 374-383. [34] K.-T. Chen, R. Fox, R. Lyndon, Free differential calculus IV: The quotient group of the lower central series, Ann. Math., 68(1958) 81-95. [35] Chen Yuqun, Jianjun Qiu, Grobner-Shirshov basis for the Chinese monoid, Journal of Algebra and its Applications,7(5)(2008), 623-628. 18 [36] Yuqun Chen, Qiuhui Mo, Embedding dendriform algebra into its universal enveloping Rota-Baxter algebra, Proc. Amer Math. Soc., to appear. arxiv.org/abs/1005.2717 [37] P. M. Cohn, A remark on the Birkhoff-Witt theorem, Journal London Math. Soc., 38(1963), 197-203 [38] P. M. Cohn, Universal Algebra, Harper and Row, 1965. [39] V. Dotsenko, A. Khoroshkin, Grobner bases for operads, Duke Mathematical Journal, 153(2)(2010), 363-396. [40] V. Drensky, R. Holtkamp, Planar trees, free nonassociative algebras, invariants, and elliptic integrals, Algebra Discrete Math., 2(2008), 1-41. [41] D. Eisenbud, I. Peeva, B. Sturmfels, Non-commutative Grobner bases for commuta- tive algebras, Proc. Amer. Math. Soc., 126 (3) (1998) 687-691. [42] D. R. Farkas, C. D. Feustel, E. L. Green, Synergy in the theories of Grobner bases and path algebras, Can. J. Math., 45(1993), 727-739. [43] S. I. Gelfand, Y. I. Manin, Homological Algebra, Springer-Verlag, 1999. [44] Li Guo, private communication, 2009. [45] H. Hironaka, Resolution of singulatities of an algebraic variety over a field if charac- teristic zero, I, Ann. Math., 79(1964), 109-203. [46] H. Hironaka, Resolution of singulatities of an algebraic variety over a field if charac- teristic zero, II, Ann. Math., 79(1964), 205-326. [47] V. Kac, Vertex algebras for beginners, University Lecture Series, Vol. 10, AMS, Providence, RI, 1996. [48] Y. Kobayashi, Grobner bases on path algebras and the Hochschild cohomology alge- bras, Sci. Math. Japonicae 64(2006), 411-437. [49] M. Lazard, Groupes, anneaux de Lie et probl`eme de Burnside. Istituto Matematico dell' Universit`a di Roma, 1960. [50] J.-L. Loday, Dialgebras, in Dialgebras and Related Operads, Lecture Notes in Math- ematics, Vol. 1763 (Springer Verlag, Berlin, 2001), 7-66. [51] J.-L. Loday, private communication, 2010. [52] R. C. Lyndon, On Burnside's problem I, Trans. Amer. Math. Soc., 77(1954), 202-215. [53] P. Malbos, Rewriting systems and Hochschild-Mitchell homology, Electr. Notes Theor. Comput. Sci., 81(2003), 59-72. [54] A. A. Mikhalev, A composition lemma and the word problem for color Lie superal- gebras, Moscow Univ. Math. Bull., 44(1989), no.5, 87-90. 19 [55] A. A. Mikhalev, E. A. Vasilieva, Standard bases of ideals of free supercommutative polynomail algebra (ε-Grobner bases), Proc. Second International Taiwan-Moscow Algebra Workshop, Springer-Verlag, 2003. [56] A. A. Mikhalev, A. A. Zolotykh, Standard Grobner-Shirshov bases of free algebras over rings, I. Free associative algebras, International Journal of Algebra and Compu- tation, 8(6)(1998), 689-726. [57] Saunders Maclane, Categories for the working mathematician, Second Edition. [58] Saunders Maclane, Homology, Springer-Verlag, 1963. [59] E. Poroshenko, Grobner-Shirshov bases for Kac-Moody algebras A(1) n , For- mal Power Series and Algebraic Combinatorics, 12th International Conference, FP- SAC'00, Moscow, Russia, June 2000, Proceedings, D. Krob, A. A. Mikhalev, A. V. Mikhalev (Eds), Springer, 2000, 552-563. n and B(1) [60] E. Poroshenko, Grobner-Shirshov bases for Kac-Moody algebras of the type A(1) n , Comm. Algebra, 30(6)(2002), 2617-2637. [61] E. Poroshenko, Grobner-Shirshov bases for Kac-Moody algebras C (1) n and D(1) nik Novosibirsk State University (Math., Mekh., Inform.), 2(2002), N1, 58-70. n , Vest- [62] M. Roitman, On free conformal and vertex algebras, J. Algebra, 217(2)(1999), 496- 527. [63] A. I. Shirshov, Subalgebras of free Lie algebras, Mat. Sbornik N. S., 33(1953), 441- 452. [64] A. I. Shirshov, On free Lie rings, Mat. Sb., 45 (1958), 2, 113-122 (in Russian). [65] A. I. Shirshov, Some Problems in the theory of rings that are nearly associative, Uspekhi Mat. Nauk 13(1958), no. 6 (84), 3-20. [66] A. I. Shirshov, Some algorithmic problem for ε-algebras, Sibirsk. Mat. Z., 3(1962), 132-137. [67] A. I. Shirshov, Some algorithmic problem for Lie algebras, Sibirsk. Mat. Z., 3(2)(1962), 292-296 (in Russian); English translation in SIGSAM Bull., 33(2)(1999), 3-6. [68] Selected works of A.I. Shirshov, Eds L.A. Bokut, V. Latyshev, I. Shestakov, E. Zel- manov, Trs M. Bremner, M. Kochetov, Birkhauser, Basel, Boston, Berlin, 2009. [69] G. Viennot, Algebras de Lie libres et monoid libres. Bases des Lie algebres et facror- izations des monoides libres. Lecture Notes in Mathematics, 691. Berlin-Geldelberg- New York, Springer-Verlag, 124 p. (1978). 20
1309.7511	1	1309	2013-09-28T23:32:01	Representing finite distributive lattices as congruence lattices of lattices	[ "math.RA" ]	Dilworth's theorem. Every finite distributive lattice $D$ can be represented as the congruence lattice of a finite lattice $L$. We want: Every finite distributive lattice $D$ can be represented as the congruence lattice of a nice finite lattice $L$. nice = sectionally complemented, uniform, semimodular, given automorphism group, regular, uniform, isoform	math.RA	math	Representing finite distributive lattices as congruence lattices of lattices Tulane University G. Gratzer ι β γ α β K ω Con K γ α J(Con K) α β γ 1 ι α ω ι α K Con K J(Con K) ι α ω ι α K Con K J(Con K) 2 Dilworth's theorem. Every finite distributive lattice D can be represented as the congruence lattice of a finite lattice L. We want: Every finite distributive lattice D can be rep- resented as the congruence lattice of a nice finite lattice L. G. Gratzer and E. T. Schmidt, 1962: nice = sectionally complemented 3 K Con K nice = semimodular 4 K Con K nice = uniform 5 K Con K nice = minimal 6 A finite distributive lattice D is determined by the poset J(D) of join-irreducible elements. So a representation of a finite distributive lat- tice D as the congruence lattice of a lattice L is really a representation of a finite poset P (= J(D)) as the poset of join-irreducible con- gruences of a finite lattice L. We want: Every finite poset P can be represented as the poset of join-irreducible congruences of a nice finite lattice L. 7 There are two types of representation theo- rems: (a) The straight representation theorems. (b) The congruence-preserving extension re- sults. 8 Let K be a finite lattice. A finite lattice L is a congruence-preserving extension of K, if L is an extension and every congruence Θ of K has exactly one extension Θ to L -- that is, ΘK = Θ. Of course, then the congruence lattice of K is isomorphic to the congruence lattice of L. K L 9 We could say that the congruence lattice of L is "naturally isomorphic" to the congruence lattice of K or that the algebraic reasons deter- mining the congruence lattice of K are carried over to L. Examples: K L K L 10 Not examples: K L K L 11 nice = sectionally complemented 1. The results Theorem 1 (G. Gratzer and E. T. Schmidt, 1962). Every finite distributive lattice D can be represented as the congruence lattice of a finite sectionally complemented lattice L. Theorem 2 (G. Gratzer and E. T. Schmidt, 1999). Every finite lattice K has a finite, sec- tionally complemented, congruence-preserving extension L. 12 nice = sectionally complemented 2. Basic technique: Chopped lattices Let M be a poset satisfying the following two conditions: (i) inf{a, b} exists in M , for any a, b ∈ M ; (ii) sup{a, b} exists, for any a, b ∈ M having a common upper bound in M . We define in M : and a ∧ b = inf{a, b}; a ∨ b = sup{a, b}, whenever sup{a, b} exists in M . This makes M into a partial lattice, called a chopped lattice. 13 Examples: The second chopped lattice has a 32 element ideal lattice! 14 Lemma 1 (G. Gratzer and H. Lakser). Let M be a finite chopped lattice. Then, for every congruence relation Θ, there exists exactly one congruence relation Θ of Id M such that, for a, b ∈ M , (a] ≡ (b] (Θ) iff a ≡ b (Θ). 15 Problem. Let M be a sectionally complemented chopped lattice. Under what conditions is Id M sectionally complemented? 16 nice = sectionally complemented 3. The representation theorem We use chopped lattices to prove Theorem 1. Let D be a finite distributive lattice, and form the finite poset P = J(D). Lemma 2. Let D be a finite distributive lattice. Then there exists a finite chopped lattice M such that Con M is isomorphic to D. Let us illustrate this with D, the four-element chain, 0 ≺ c ≺ b ≺ a; so J(D) is the three-element chain c ≺ b ≺ a. 17 Take the finite set M0 = {a, b, c}∪{0} and make it a meet-semilattice by defining inf{x, y} = 0, if x 6= y. Note that the congruence relations of M0 are in one-to-one correspondence with subsets of {a, b, c}. a, c M0 b, 0 We must force that a ≡ 0 implies that b ≡ 0 implies that c ≡ 0. 18 To accomplish this, we use the lattice N6 = N (p, q). p(q) N6 = N (p, q) p1 q2 0 q q1 In N (p, q), p1 ≡ 0 "implies" that q1 ≡ 0, but q1 ≡ 0 "does not imply" that p1 ≡ 0. We construct the finite chopped lattice M by "inserting" N (p, q) in M0, for a, b and for b, c, by appropriately doubling b and c. 19 p1 a, c M0 b, 0 p(q) q2 0 q q1 a(b) b(c) J(D) a = a1 = a2 b2 a b c b b1 0 c c2 c1 M 20 nice = sectionally complemented 4. Basic technique: Cubic extensions For a finite lattice K, let { Ki i ∈ I } be the subdirect factors of K. For each Ki, we select S(Ki), a finite simple extension of Ki. Now we form: R(K) = Y( S(Ki) I ∈ I ), and call it a cubic extension of K. Then there is a one-to-one correspondence be- tween subsets of I and congruences Θ of R(K); hence, the congruence lattice of R(K) is a fi- nite Boolean lattice. Every congruence of K extends to R(K). Each S(Ki) can be chosen to be sectionally comple- mented. Then the cubic extension is section- ally complemented. 21 nice = minimal The lattice L constructed by R. P. Dilworth to represent D is very large, it has O(22n) ele- ments. Theorem 3 (G. Gratzer, H. Lakser, and E. T. Schmidt). Let D be a finite distributive lattice with n join-irreducible elements. Then there exists a planar lattice L of O(n2) elements with Con L ∼= D. We illustrate the proof of this result. 22 This diagram shows a distributive lattice D, P = J(D), and the chain C we form from P . The chain is of length 2P = 6, and the prime intervals are "marked" with elements of P as shown. We call this "marking" coloring. p2 D p1 p3 p2 P p1 p3 p3 p3 p2 p2 p1 p1 C 23 To construct L, we take C2. If both lower edges of a covering square in C2 have the same color, we add an element to make it a covering M3. If in C2 we have a covering C2 × C3, where the lower C2 is colored by p, the lower C3 is colored by q twice, where p > q, then we add an element to make it an N5,5. p p p p M3 p > q q q q p q p N5 5 , 24 Here is what we obtain: p2 D p1 p3 q p > q q p N5,5 p3 p3 p2 p1 p3 P p2 C p2 p1 p2 p2 C p1 p1 p1 p3 p3 p p M3 25 Theorem 4 (G. Gratzer, I. Rival, and N. Za- guia, 1995). Let α be a real number satisfying the following condition: Every distributive lat- tice D with n join-irreducible elements can be represented as the congruence lattice of a lat- tice L with O(nα) elements. Then α ≥ 2. 26 nice = semimodular 1. Representation theorem Theorem 5 (G. Gratzer, H. Lakser, and E. T. Schmidt, 1998). Every finite distributive lattice D can be represented as the congruence lattice of a finite semimodular lattice S. In fact, S can be constructed as a planar lattice of size O(n3), where n is the number of join-irreducible ele- ments of D. 27 The proof of this result is very similar to the proof of Theorem 3. The basic building block is S8; we show two views of this lattice: S8 p p p < q q p The first view shows S8 with its only nontrivial congruence indicated with a dashed line; the second views shows S8 as it is used in the con- struction. 28 To illustrate the construction, take the follow- ing distributive lattice D; the poset J(D) is also shown. D p2 p3 p1 q J(D) Now we form two chains and their direct prod- ucts. We augment a covering C2 × C2 colored by the same element on both sides by an el- ement to form M3. For p1 ≺ p2, we replace the covering C3 × C3 colored by p2, p1 and p1, p1 by S8. We make a similar replacement for p1 ≺ p3, to obtain S: 29 D S p1 p1 p1 p4 p1 p3 p2 p3 p1 p1 J(D) p4 p1 p4 p3 p2 p2 p1 p p < q q p p S8 30 nice = semimodular 2. Constructing a congruence-preserving semimodular extension Theorem 6 (G. Gratzer and E. T. Schmidt, 2001). Every finite lattice K has a congruence- preserving embedding into a finite semimodu- lar lattice L. The proof starts out with the cubic extension R(K) of K, where we choose each S(Ki) semi- modular. So the cubic extension is semimodu- lar. The congruences then are represented in a dual ideal F of R(K) that is Boolean. By glu- ing a suitable modular lattice M to R(K). The congruences are then represented on a dual ideal E′ of M that is a chain, so the proof is completed by gluing the lattice S to the con- struct: 31 S C E ′ M F bR(K) 32 nice = given automorphism group Theorem 7 (The Independence Theorem, V. A. Baranskiı and A. Urquhart, 1979). Let D be a finite distributive lattice with more than one el- ement, and let G be a finite group. Then there exists a finite lattice L such that the congru- ence lattice of L is isomorphic to D and the automorphism group of L is isomorphic to G. This is a representation theorem. There is also a congruence-preserving extension variant for this result. 33 Theorem 8 (The Strong Independence The- orem, G. Gratzer and E. T. Schmidt, 1995). Let K be a finite lattice with more than one element and let G be a finite group. Then K has a congruence-preserving extension L whose automorphism group is isomorphic to G. 34 nice = regular 1. The result Let L be a lattice. We call a congruence re- lation Θ of L regular, if any congruence class of Θ determines the congruence. Let us call the lattice L regular, if all congruences of L are regular. Sectionally complemented lattices are regular, so we already have a representation theorem. We also have a congruence-preserving exten- sion version: Theorem 9 (G. Gratzer and E. T. Schmidt, 2001). Every finite lattice L has a congruence- preserving embedding into a finite regular lat- tice L. 35 nice = regular 2. Basic technique: Boolean triples For a bounded lattice K, let us call the triple hx, y, zi ∈ L3 boolean iff the following equations hold: x = (x ∨ y) ∧ (x ∨ z), y = (y ∨ x) ∧ (y ∨ z), z = (z ∨ x) ∧ (z ∨ y). We denote by M3hKi ⊆ K3 the poset of all boolean triples of K. M3hKi is a bounded lat- tice. We identify the lattice K with the in- terval [h0, 0, 0i, h1, 0, 0i] under the isomorphism x 7→ hx, 0, 0i. 36 G. Gratzer and F. Wehrung, 1997: Theorem 10. M3hKi is a congruence-preserv- ing extension of K. 37 nice = uniform Let L be a lattice. We call a congruence re- lation Θ of L uniform, if any two congruence classes of Θ are of the same size (cardinality). Let us call the lattice L uniform, if all congru- ences of L are uniform. Theorem 11 (G. Gratzer, E. T. Schmidt, and K. Thomsen, 2002). Every finite distributive lattice D can be represented as the congruence lattice of a finite uniform lattice L. A uniform lattice is always regular, so the lat- tice L of this theorem is also regular. 38 same size not isomorphic The uniform construction for the four-element chain. 39 Why not draw less trivial examples? If we start with the lattice: then the uniform lattice we conctruct has 153,125 elements (and order dimension 7). 40 nice = isoform Let L be a lattice. We call a congruence re- lation Θ of L isoform, if any two congruence classes of Θ are isomorphic (as lattices). Let us call the lattice L isoform, if all congruences of L are isoform. Theorem 12 (G. Gratzer and E. T. Schmidt, 2002). Every finite distributive lattice D can be represented as the congruence lattice of a finite, isoform lattice L. Since isomorphic lattices are of the same size, this theorem is a stronger version of the theo- rem for uniform lattices. Theorem 13 (G. Gratzer, E. T. Schmidt, and R. W. Quackenbush, 2004). Every finite lattice K has a congruence-preserving extension to a finite isoform lattice L. 41 Simultaneous representations of two distributive lattices Let L be a lattice and let K be a sublattice of L. Then the restriction map rs : Con L → Con K is a {0, 1, ∧}-homomorphism. If K is an ideal, then rs is a {0, 1}-homomorphism. G. Gratzer and H. Lakser, 1986: Theorem 14. Let D and E be finite distribu- tive lattices; let D have more than one ele- ment. Let ϕ be a {0, 1}-homomorphism of D into E. Then there exists a (sectionally com- plemented) finite lattice L and an ideal K of L such that D ∼= Con L, E ∼= Con K, and ϕ is represented by rs, the restriction map. 42 Example: D b i c a i E f d e o ϕ : D → E o J(D) b c J(E) i a d e P ϕ : J(D) → J(E) 43 Building block: aR(aL) aL aL(aM ) aM (aR) aM aR aL2 aL1 aM2 aM1 aR1 aR2 We take six: for o, b, c, d, e, i. 44 We code J(D) and J(E) with the middle ele- ments (the M -s), for example, a ≺ b is coded by aM (bM ) aM b M aM2 aM1 b M2 b M1 N (aM , bM ) P ϕ is an equivalence: one direction we code with the L-s, the other, with the R-s. 45 New problems: Make K and L "nice". 46 Sectionally complemented lattices: G. Gratzer and H. Lakser, Notes on section- ally complemented lattices. I. Characterizing the 1960 sectional complement. Acta Math. Hungar. 108 (2005), 115 -- 125. G. Gratzer and H. Lakser, Notes on section- ally complemented lattices. II. Generalizing the 1960 sectional complement with an application to congruence restrictions. Acta Math. Hungar. 108 (2005), 251 -- 258. G. Gratzer and E. T. Schmidt, Finite lattices with isoform congruences. Tatra Mt. Math. Publ. 27 (2003), 111 -- 124. Isoform lattices: G. Gratzer, E. T. Schmidt, and R. W. Quackenbush, Acta Sci. Math. (Szeged) 70 (2004), 473 -- 494. 47 p q1 1 0 q q2 p2 p1 p3 p2 p3 p3 p2 p2 p1 p1 p1 p3 D P C Φ2 Φ1 Φ2 Φ1 p q1 1 0 q q2 h1, 1, 1i h1, 1, 1i h1, 0, 0i h0, 1, 0i h0, 0, 1i D h1, b, bi x h1, a, ai h1, 0, 0i M3hKi Da,b h0, 0, 0i h0, 1, 0i hx, 0, 0i D h0, b, 0i Ia,b h0, a, 0i h0, 0, 0i h0, 0, 1i M3[D] M3hKi M3hKi p 1 p 2 p 3 p 1 p 2 V p 3 Ka,b Da,b Ia,b ϕ1 ϕ2 Dc,d Ic,d Kc, d q1 q2 q3 4 M3h[a, b]i The four building blocks. Ka,b I ψ Da,b D Dc,d Ic,d Kc, d Ia,b q1 q2 q3 M3h[a, b]i 2 Two gluings. U 2 2 4 e2 A L B pA = pB Ka,b M3hKi A p 1 p 2 p 3 Da,b M Dc,d Ia,b q1 Ic,d h q2 q3 Kc, d M3h[a, b]i e1 0A = 0B The final gluing. Φ2 h Φ1
1201.0591	1	1201	2012-01-03T08:43:33	Uniformly Flat Semimodules	[ "math.RA", "math.CT" ]	We revisit the notion of flatness for semimodules over semirings. In particular, we introduce and study a new notion of uniformly flat semimodules based on the exactness of the tensor functor. We also investigate the relations between this notion and other notions of flatness for semimodules in the literature.	math.RA	math	Uniformly Flat Semimodules∗ Jawad Y. Abuhlail† Department of Mathematics and Statistics King Fahd University of Petroleum & Minerals [email protected] 2 1 0 2 n a J 3 ] . A R h t a m [ 1 v 1 9 5 0 . 1 0 2 1 : v i X r a Abstract We revisit the notion of flatness for semimodules over semirings. In particular, we introduce and study a new notion of uniformly flat semimodules based on the exactness of the tensor functor. We also investigate the relations between this notion and other notions of flatness for semimodules in the literature. Introduction The homological classification of monoids, suggested by L. A. Skornjakov [Sko1969a, Sko1969b], is still an ongoing project attracting the attention of many experts in Semigroup Theory and Universal Algebra. Many papers were devoted to study the category ActS of right S-acts over a monoid S (a right S-act is a set A with a map µ : A × S −→ A such that a(st) = (as)t and a1S = a for all a ∈ A and s, t ∈ S); for more information see the encyclopedic manuscript of Kilp et al. [KKM2000]. The philosophy in several of these papers is to model the theory of modules over rings (e.g. [AF1974], [Wis1991]) by studying the interplay between the (categorical) properties of ActS and the (algebraic) properties of S. Another approach to study Abelian monoids is to consider them as semimodules over the semiring N0 of nonnegative integers [Gol1999a]. This provides us with a richer struc- ture, motivates a non-additive version of the theory of modules over rings and opens the door for developing non-Abelian homological algebra [Ina1997]. It is worth mentioning that this approach is supported by the important role that semirings and semimodules play in emerging areas of research like idempotent analysis, tropical geometry and several aspects of theoretical physics [LM2005], [KM1997] in addition to many applications in several branches of mathematics and computer science (e.g. [GM2008], [Gol1999a], [HW1998]). Although some notions of flatness which are different for S-acts (e.g. [KKM2000, Chapter III], [B-F2009] and the papers cited there) coincide for semimodules as shown by Katsov [Kat2004a], several notions of flatness which turn out to be the same for modules are in fact different for semimodules (e.g. flatness and mono-flatness [KN2011]). This results in a rich theory of flatness for semimodules. In this manuscript, we revisit some of these notions and introduce a new notion of uniformly flat semimodules based on the exactness of the tensor product functor simulating the classical notion of flat modules over rings. ∗MSC2010: 16Y60 Keywords: Semirings, Semimodules, Flat Semimodules, Tensor Products, Exact Sequences †The author would like to acknowledge the support provided by the Deanship of Scientific Research (DSR) at King Fahd University of Petroleum & Minerals (KFUPM) for funding this work through project No. FT100004. 1 The motivation for introducing a new notion of flatness for semimodules can be understood in light of the following observations: the notions of flat and k-flat semimodules introduced in [Alt2004] use Takahashi's tensor products of semimodules [Tak1982a] which are not the natural tensor products. Among the mains disadvantages of such tensor products is that the category of semimodules over a commutative semiring is not monoidal and that the tensor functor is not left adjoint to the hom functor as one would expect. In fact, Takahashi's tensor products solve the universal problem related to such structures in the subcategory of cancellative semimodules, but they fail to provide a universal solution in the whole category of semimodules (see Section 2 for more details). Moreover, several results use Takahashi's notion of exact sequences of semimodules [Tak1981] (see also [Gol1999a]), which we believe is not natural as well; for more details see the recent manuscript [Abu]. On the other hand, while the notion of flat semimodules introduced in [Kat2004a] is quite natural, it does not provide a notion of relative flatness w.r.t. a given family of semimodules which showed to be important in studying several notions related to pure exact sequences of modules over rings (e.g. [Wis1991]). This motivated us to introduce a new notion of flatness, namely that of uniformly flat semimodules, using the natural tensor products of semimodules [Kat1997] and what we believe is a more appropriate notion of exact sequences of semimodules introduced recently in [Abu]. This paper is organized as follows. In Section one, we recall some preliminaries about semirings, semimodules and exact sequences of semimodules. In Section two, we recall the construction of the natural tensor products of semimodules over semirings, clarify their con- nection with Takahashi's tensor products and study some of their properties. In Section 3, we introduce the notion of uniformly flat semimodules and investigate its connection with other notions of flatness in the literature. We also generalize several results known for modules over rings to semimodules over semirings. 1 Preliminaries As pointed out in [KN2011]: "when investigating semirings and their representations, one should undoubtedly use methods and techniques of both ring and lattice theory as well as diverse techniques and methods of categorical and universal algebra." For the convenience of the readers who might have different backgrounds, and to make this manuscript as much self-contained as possible, we collect in this section some definitions, remarks and results that will be used in the sequel. For unexplained terminology, our main references are [Mac1998] for Category Theory, [Gra2008] for Universal Algebra and [Wis1991] for Ring and Module Theory. Semirings and Semimodules Semirings (semimodules) are roughly speaking, rings (modules) without subtraction. Recall that a semigroup (S, ∗) is said to be cancellative iff for any s1, s2, s ∈ S we have [s1 ∗ s = s2 ∗ s =⇒ s1 = s2] and [s ∗ s1 = s ∗ s2 =⇒ s1 = s2]. Definition 1.1. A semiring is an algebraic structure (S, +, ·, 0, 1) consisting of a non-empty set S with two binary operations "+" (addition) and "·" (multiplication) satisfying the fol- lowing conditions: 1. (S, +, 0) is an Abelian monoid with neutral element 0S; 2 2. (S, ·, 1) is a monoid with neutral element 1; 3. x · (y + z) = x · y + x · z and (y + z) · x = y · x + z · x for all x, y, z ∈ S; 4. 0 · s = 0 = s · 0 for every s ∈ S (i.e. 0 is absorbing). 1.2. Let S, S′ be semirings. A map f : S → S′ is said to be a morphism of semirings iff for all s1, s2 ∈ S : f (s1 + s2) = f (s1) + f (s2), f (s1s2) = f (s1)f (s2), f (0S) = 0S′ and f (1S) = 1S′. 1.3. Let (S, +, ·) be a semiring. We say that S is cancellative iff the additive semigroup (S, +) is cancellative; commutative iff the multiplicative semigroup (S, ·) is commutative; semifield iff (S\{0}, ·, 1) is a commutative group. Examples 1.4. Rings are indeed semirings. The first natural example of a commutative semir- ing is (N0, +, ·), the set of nonnegative integers. The semirings (R+ 0 , +, ·) are indeed semifields. Moreover, for any ring R we have a semiring structure (, +, ·) on the set Ideal(R) of ideals of R with the usual addition and multiplication of ideals of R. For more examples, the reader may refer to [Gol1999a]. 0 , +, ·) and (Q+ Definition 1.5. Let S be a semiring. A right S-semimodule is an algebraic structure (M, +, 0M ) consisting of a non-empty set M, a binary operation "+" along with a right S-action M × S −→ M, (m, s) 7→ ms, such that: 1. (M, +, 0M ) is an Abelian monoid with neutral element 0M ; 2. (ms)s′ = m(ss′), (m + m′)s = ms + m′s and m(s + s′) = ms + ms′ for all s, s′ ∈ S and m, m′ ∈ M ; 3. m1S = m and m0S = 0M = 0M s for all m ∈ M and s ∈ S. 1.6. 1. Let M and M ′ be right S-semimodules. A map f : M → M ′ is said to be a morphism of S-semimodules (or S-linear ) iff for all m1, m2 ∈ M and s ∈ S : f (m1 + m2) = f (m1) + f (m2) and f (ms) = f (m)s. The set HomS(M, M ′) of S-linear maps from M to M ′ is clearly an Abelian monoid under addition. The category of right S-semimodules is denoted by SS. Analogously, one can define the category SS of left S-semimodules. A right S-semimodule MS is said to be cancellative iff the semigroup (M, +) is cancellative. With CSS ⊆ SS (resp. S CS ⊆ S S) we denote the full subcategory of cancellative right (left) S-semimodules. 1.7. Let S be a semiring and M a right S-semimodule. A non-empty subset L ⊆ M is said to be an S-subsemimodule, and we write L ≤S M, iff L is closed under "+M " and ls ∈ L for all l ∈ L and s ∈ S. Example 1.8. Every Abelian monoid (M, +, 0M ) is an N0-semimodule in the obvious way. Moreover, the categories AbMon of Abelian monoids and the category SN0 of N0-semimodules are isomorphic. 3 1.9. Let S, T be semirings, M a left S-semimodule and a right T -semimodule. We say that M is an (S, T )-bisemimodule iff (sm)t = s(mt) for all s ∈ S, m ∈ M and t ∈ T. For (S, T )- bisemimodules M, M ′, we call an S-linear T -linear map f : M → N ′ a morphism of (S, T )- bisemimodules (or (S, T )-bilinear ). The set Hom(S,T )(M, M ′) of (S, T )-bilinear maps from M to M ′ is clearly an Abelian monoid under addition. The category of (S, T )-bisemimodules will be denoted by S ST . Throughout, and unless otherwise explicitly specified, S is an associative semiring with 1S 6= 0S. We mean by an S-semimodule a right S-semimodule unless something different is mentioned explicitly. 1.10. Let M be an S-semimodule. An equivalent relation ≡ on M is said to be an S- congruence iff for any m, m′, m1, m′ 2 ∈ M and s ∈ S we have 1, m2, m′ [(m1 ≡ m′ 1 ∧ m2 ≡ m′ 2) ⇒ m1 + m2 ≡ m′ 1 + m′ 2] and [m ≡ m′ ⇒ ms ≡ m′s]. Every S-subsemimodule L ≤S M induces two S-congruences on M given by m1 ≡L m2 ⇔ m1 + l1 = m2 + l2 for some l1, l2 ∈ L; m2 ≡[L] m2 ⇐⇒ m1 + l1 + m′′ = m2 + l2 + m′′ for some l1, l2 ∈ L and m′′ ∈ M. We call the S-semimodule M/L := M/≡L the quotient (factor ) semimodule of M by L. If M is cancellative, then L and M/L are cancellative. On the other hand, M/≡[L] is obviously cancellative. 1.11. Let M be an S-semimodule and recall the S-congruence relation ≡[0] on M defined by m ≡[0] m′ ⇐⇒ m + m′′ = m′ + m′′ for some m′′ ∈ M. The quotient S-semimodule M/ ∼ is indeed cancellative and we have a canonical surjection cM : M −→ c(M ) with Ker(cM ) = {m ∈ M m + m′′ = m′′ for some m′′ ∈ M }. The class of cancellative right S-semimodules is a reflective subcategory of SS in the sense that the functor c : SS −→ CSS, M 7→ M/ ≡[0] is left adjoint to the embedding functor CSS ֒→ SS, i.e. for any S-semimodule M and any can- cellative S-semimodule N we have a natural isomorphism of Abelian monoids HomS(c(M ), N ) ≃ HomS(M, N ) [Tak1981, Page 517]. Proposition 1.12. The category SS and its full subcategory CSS have kernels and cokernels, where for any morphism of S-semimodules f : M → N we have Ker(f ) = {m ∈ M f (m) = 0} and Coker(f ) = N/f (M ). 1.13. We call a subset Y ⊆ N subtractive iff Y = Y , the subtractive closure of Y, where Y = {n ∈ N n + y1 = y2 for some y1, y2 ∈ Y }. An S-semimodule M is said to be completely subtractive iff every S-subsemimodule of M is subtractive. 4 We call a morphism of S-semimodules γ : M −→ N : k-uniform iff for any m1, m2 ∈ M : γ(m1) = γ(m2) =⇒ ∃ k1, k2 ∈ Ker(γ) s.t. m1 + k1 = m2 + k2; (1) i-uniform iff γ(M ) = γ(M ); uniform iff γ is k-uniform and i-uniform. Remark 1.14. The uniform (k-uniform, i-uniform) morphisms of semimodules were called regular (k-regular, i-regular ) by Takahashi [Tak1982c]. We think that our terminology avoids confusion sine a regular monomorphism (regular epimorphism) has a different well-established meaning in the language of Category Theory. 1.15. Let M be an S-semimodule, L ≤S M an S-subsemimodule and consider the factor semimodule M/L. Then we have a surjective uniform morphism of S-semimodules πL := M → M/L, m 7→ [m] with Ker(πL) = {m ∈ M m + l1 = l2 for some l1, l2 ∈ L} = L; in particular, L = Ker(πL) if and only if L ⊆ M is subtractive. In [Abu] we introduced a new notion of exact sequences of semimodules. Takahashi's exact sequences [Tak1981] shall be called semi-exact in the sequel: Definition 1.16. We call a sequence of S-semimodules f −→ M g −→ N L exact iff f (L) = Ker(g) and g is k-uniform. An exact sequence 0 −→ L f −→ M g −→ N −→ 0 is called a short exact sequence. 1.17. ([Abu]) We call a sequence of S-semimodules L f → M g → N : proper-exact iff f (L) = Ker(g); semi-exact iff f (L) = Ker(g); quasi-exact iff f (L) = Ker(g) and g is k-uniform. 1.18. We call a (possibly infinite) sequence of S-semimodules (2) (3) · · · → Mi−1 fi−1→ Mi fi→ Mi+1 fi+1→ Mi+2 → · · · (4) chain complex iff fj+1 ◦ fj = 0 for every j; exact (resp. proper-exact, semi-exact, quasi-exact) iff each partial sequence with three terms Mj fj→ Mj+1 fj+1→ Mj+2 is exact (resp. proper-exact, semi-exact, quasi-exact); 5 1.19. An S-semimodule N is said to be a retract of an S-semimodule M iff there exist a (surjective) S-linear map θ : M −→ N and an (injective) S-linear map ψ : N −→ M such that θ ◦ ψ = idN (equivalently, N ≃ α(M ) for some idempotent endomorphism α ∈ End(MS)). On the other hand, N is a direct summand of M (i.e. M = N ⊕ N ′ for some S-subsemimodule N ′ of M ) if and only if there exists α ∈ Comp(End(MS)) s.t. α(M ) = N where for any semiring T we set Comp(T ) = {t ∈ T ∃ et ∈ T with t + et = idT and tet = 0T = ett}. Indeed, every direct summand of M is a retract of M ; the converse is not true in general (cf. [Gol1999a, Proposition 16.6]). Lemma 1.20. ([Abu, Proposition 3.10, Corollary 3.11]) Let A, B and C be S-semimodules. 1. 0 −→ A f −→ B is exact if and only if f is injective. 2. B g −→ C −→ 0 is exact if and only if g is surjective. 3. 0 −→ A f −→ B g −→ C is semi-exact and f is uniform if and only if A = Ker(g). 4. A f −→ B g −→ C −→ 0 is semi-exact and g is uniform if and only if C = Coker(f ). 5. 0 −→ A f −→ B g −→ C −→ 0 is exact if and only if A = Ker(g) and C = Coker(f ). The following technical result follows immediately from the definitions and [Tak1983, Lemmas 1.11, 1.15]. Lemma 1.21. 1. Consider a commutative diagram of S-semimodules with π ◦ι = idN and π′ ◦ ι′ = idN ′ : N ι M π N γ γ γ N ′ ι′ M ′ π′ / N ′ i-uniform). If γ is uniform (resp. k-uniform, i-uniform), then eγ is uniform (resp. k-uniform, 2. Consider a commutative diagram of S-semimodules with π ◦ i = idN , π′ ◦ i′ = idN ′ and π′′ ◦ i′′ = idN ′′ : N ι M π N f f f N ′ ι′ M ′ π′ / N ′ g g g / N ′′ ι′′ / M ′′ π′′ / N ′′ f −→ M ′ g −→ M ′′ is exact (resp. proper-exact, semi-exact, quasi-exact), then eg −→ N ′′ is exact (resp. proper-exact, semi-exact, quasi-exact). 6 If M ef N −→ N ′ / / / / / / / / / / / / / Some redundant assumptions in [Abu, Lemma 4.5] do not hold in some situations which we will handle in this paper. A slight adjustment of the proof of the above mentioned result yields Lemma 1.22. Consider the following commutative diagram of S-semimodules L1 α1 L2 f1 f2 M1 α2 / M2 g1 g2 N1 α3 / N2 1. Let the second sequence be quasi-exact (i.e. f2(L2) = Ker(g2) and g2 is k-uniform) and g1, α1 be surjective. (a) Let g1 ◦ f1 = 0. If α2 is injective, then α3 is injective. (b) If α3 is surjective (and α2 is i-uniform), then α2 is a semi-epimorphism (surjec- tive). 2. Let the first row be semi-exact (i.e. f1(L1) = Ker(g1)) and f2 be injective. (a) Let f1, α2 be cancellative and g1 be k-uniform. If α1, α3 are injective, then α2 is injective. (b) Let g2◦f2 = 0. If Ker(α3) = 0, and α2 is surjective, then α1 is a semi-epimorphism. If moreover, α1 or f1 is i-uniform, then α1 is surjective. 2 Tensor products of semimodules Tensor products of semimodules were introduced and investigated by Takahashi [Tak1981]. However, they did not provide a solution to the universal problem related to such structures in the whole category of semimodules. On the other hand, Katsov [Kat1997] considered a different tensor product in the category of semimodules (over a commutative semiring) which solved several of the problems that Takahashi's tensor products had. It is worth mentioning, as Katsov pointed out, that his construction of the tensor product and the elementary results related to it seem to be folklore (e.g. Grillet [Gril1969] gave an explicit construction of a non-associative tensor product in the variety Sgr of semigroups and suggested that the same construction works for all algebraic varieties of Universal Algebra). Varieties in which the tensor products behave nicely were considered by F. Linton [Lin1966] (see also [Bor1994b, Theorem 3.10.3]). Construction of tensor products As before, S denotes an associative semiring with 1S 6= 0S. With S S and SS we denoted the categories of left and right S-semimodule, respectively. For the convention of the reader, we recall the construction of tensor products of semimodules and some of its properties (e.g. [Kat1997], [Kat2004a], [KN2011]): 2.1. Let MS be a right S-semimodule and SN a left S-semimodule. An S-balanced map g : M ×N → G, where G is an Abelian monoid, is a bilinear map such that g(ms, n) = g(m, sn) for all m ∈ M, s ∈ S and n ∈ N. Let F be the free Abelian monoid with basis M × N. 7 / / / / / / Every element of F can be written uniquely as a linear combination of elements of the set {δ(m,n) (m, n) ∈ M × N } where δ(m,n) is the Kronecker delta function. Let σ ⊆ F × F be the congruence relation generated by the set of all ordered pairs {(δ(m1+m2,n), δ(m1,n) + δ(m2,n)), (δ(m,n1+n2), δ(m,n1) + δ(m,n2)), (δ(ms,n), δ(m,sn))}, where m1, m2, m ∈ M, n1, n2, n ∈ N, s ∈ S and consider canonical maps πσ : F −→ F/σ and τ := πσ ◦ ι : M × N → F/σ. Let G be an Abelian monoid and β : M × N → G an S-balanced map. Since F is free over M × N, the map β induces a unique map β′ : F → G. Since Ker(πσ) = σ ⊆ Ker(g), there exists a unique map γ : F/σ → G such that γ ◦ πσ = β′ (given by γ(f ) = β′(f ), for every f ∈ F ) and so γ ◦ τ = γ ◦ πσ ◦ ι = β′ ◦ ι = β : M × N ι ✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺✺ #●●●●●●●●●● β′ β / G τ z✉✉✉✉✉✉✉✉✉✉ πσ F γ F/σ M × N τ ⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧⑧ γ M ⊗S N ✼✼✼✼✼✼✼✼✼✼✼✼✼✼ β / G (5) So, M ⊗S N := (F/σ, τ ) is solution for the following universal problem: For every Abelian monoid G with an S-balanced map β : M × N → G, there exists a unique morphism of monoids γ : M ⊗S N → G that completes the right triangle in (5) commutatively. 2.2. Let MS a right S-semimodule, SN a left S-semimodule and F the free Abelian monoid with basis M × N. Let N (M ) ⊆ F × F be the symmetric S-subsemimodule generated by the set of elements of the form (δ(m1+m2,n), δ(m1,n) + δ(m2,n)), (δ(m,n1+n2), δ(m,n1) + δ(m,n2)), (δ(m1,n) + δ(m2,n), δ(m1+m2,n)), (δ(m,n1) + δ(m,n2), δ(m,n1+n2)), (δ(ms,n), δ(m,sn)), (δ(m,sn), δ(ms,n)), and consider the S-congruence relation on F defined by f ρg ⇐⇒ f + h = g + h′ for some (h, h′) ∈ N (M ). Takahashi's tensor product of M and N is defined as M ⊠S N = F/ρ. Notice that there is an S-balanced map eτ : M × N −→ M ⊠S N, (m, n) 7→ m ⊠S n := (m, n)/ρ with the following universal property [Tak1982a]: for every cancellative Abelian monoid eG and every S-balanced map eβ : M × N −→ eG there exists a unique morphism of monoids eγ : M ⊠S N −→ eG such that eγ ◦ eτ = eG. The above mentioned property means that − ⊠S − plays the role of a tensor product w.r.t. cancellative semimodules. On the other hand, notice that for every Abelian monoid G, we 8 _ # z / / have a commutative diagram M ⊗S N cM ⊗S N w♣♣♣♣♣♣♣♣♣♣♣ c(M ⊗S N ) M × N τ xqqqqqqqqqq β #●●●●●●●●● / G γ c(γ) (6) cG !❈❈❈❈❈❈❈❈ / c(G) which suggests that c(− ⊗S −) plays the same role. The above observations motivates the following connection between the bifunctors − ⊗S − and − ⊠S −, where CAbMon is the category of cancellative Abelian monoids: Theorem 2.3. We have an equivalence of functors − ⊠S − ≈ c(−) ◦ (− ⊗S −) : SS × S S −→ CAbMon. In particular, for every right S-semimodule MS and every left S-semimodule SN, we have a natural isomorphism of Abelian monoids M ⊠S N ≃ c(M ⊗S N ). Proof. Let MS be a right S-semimodule, SN a left S-semimodule and consider the Abelian monoids (M ⊗S N ; τ ), (M ⊠S N ;eτ ) along with the canonical morphisms of monoids cM ⊗S N : M ⊗S N −→ c(M ⊗S N ) and c(γ) : c(M ⊗S N ) −→ M ⊠S N. Since eG = c(M ⊗S N ) is cancellative and eβ := cM ⊗S N ◦ τ : M × N −→ c(M ⊗ N ) is S- balanced, there exists a unique morphism of monoids eγ : M ⊠S N −→ c(M ⊗S N ) such that eγ ◦eτ = eβ = cM ⊗S N ◦τ . On other hand, for G = M ⊠S N, the map β := eτ : M ×N −→ M ⊠S N such that γ ◦ τ = β = eτ . Consider the morphisms of monoids is S-balanced and so there exists a unique morphism of monoids γ : M ⊗S N −→ M ⊠S N ϕ : M ⊠S N θ : M ⊗S N Notice that eγ −→ c(M ⊗S N ) cM ⊗S N −→ c(M ⊗S N ) c(γ) −→ M ⊠S N ; c(γ) −→ M ⊠S N eγ −→ c(M ⊗S N ). ϕ ◦ eτ = c(γ) ◦ eγ ◦ eτ = c(γ) ◦ eβ = c(γ) ◦ cM ⊗S N ◦ τ = γ ◦ τ = eτ . Since idM ⊠S N : M ⊠S N −→ M ⊠S N is the unique morphism of monoids satisfying this property, we conclude that c(γ) ◦ eγ = idM ⊠S N . On the other hand, we have θ ◦ τ = eγ ◦ c(γ) ◦ cM ⊗S N ◦ τ = eγ ◦ γ ◦ τ = eγ ◦ eτ = cM ⊗S N ◦ τ . Although τ is not an epimorphism (in general), τ (M × N ) is a generating set for M ⊗S N, whence θ = cM ⊗S N and so eγ ◦ c(γ) ◦ cM ⊗S N = idc(M ⊗S N ) ◦ cM ⊗S N . Since cM ⊗S N is an epimorphism, we conclude that eγ ◦ c(γ) = idc(M ⊗S N ). One can easily check that this isomorphism is natural in MS and SN.(cid:4) 9 x # w / ! / Remarks 2.4. Let S and T be semirings. 1. For every right S-semimodule M and left S-semimodule SN we have canonical isomor- phisms of Abelian monoids M ϑr M≃ M ⊗S S and N ϑl M≃ S ⊗S N, whence M ⊠S S ≃ c(M ⊗S S) ≃ c(M ) and S ⊠S N ≃ c(S ⊗S N ) ≃ c(N ). 2. If M is a right S-semimodule, N is an (S, T )-bisemimodule and X is a left T -semimodule, then we have a canonical isomorphism of Abelian monoids (M ⊗S N ) ⊗T X ≃ M ⊗S (N ⊗T X). Proposition 2.5. (cf. [KN2011]) Let M be a right S-semimodule and N a left S-semimodule. 1. If M is a (T, S)-bisemimodule, then M ⊗S − : S S −→ T S is left adjoint to HomT (M, −) : T S −→ SS, i.e. for every left S-semimodule X and every left T -semimodule Y, we have a canonical isomorphism of Abelian monoids that is natural in SX and T Y : HomT (M ⊗S X, Y ) ς l ≃ HomS(X, HomT (M, Y )). 2. If N is an (S, T )- bisemimodule, then −⊗S N : SS −→ ST is left adjoint to HomT (N, −) : for every right S-semimodule X and every right T -semimodule Y we ST −→ SS, i.e. have a canonical isomorphism of Abelian monoids that is natural in XS and YT : HomT (X ⊗S N, Y ) ς r ≃ HomS(X, HomT (N, Y )). As a consequence of Lemma 2.3 and Proposition 2.5 we recover [Tak1982c, Corollary 4.5]: Corollary 2.6. Let M be a right S-semimodule and N a left S-semimodule. 1. If M is a (T, S)-bisemimodule, SX a left S-semimodule and Y ∈ T CS a canecllative left T -semimodule, then we have a canonical isomorphism HomT (M ⊠S X, Y ) ≃ HomS(X, HomT (M, Y )). 2. If N is an (S, T )- bisemimodule, X is a right S-semimodule and Y ∈ CST a cancellative right T -semimodule, then we have a canonical isomorphism HomT (X ⊠S N, Y ) ς r ≃ HomS(X, HomT (N, Y )). Proof. We prove "1". The proof of "2" is similar. The required isomorphism is given by HomT (M ⊠S X, Y ) = HomT (c(M ⊗S X), Y ) ≃ HomT (M ⊗S X, Y ) ≃ HomS(X, HomT (M, Y )).(cid:4) Definition 2.7. A category C is said to be (finitely) complete iff every functor F : D −→ C, with D a small (finite) category, has a limit. Dually, C is said to be (finitely) cocomplete iff every functor F : D −→ C with D a small (finite) category has a colimit. 10 Taking into account the fact that SS is a variety (in the sense of Universal Algebra) we have (e.g. [Sch1972, Theorem 21.6.4]): Proposition 2.8. The category SS of right S-semimodules is complete (has equalizers and products) and cocomplete (has coequalizers and coproducts). 2.9. Let J be a directed set. The directed limit (inductive limit, filtered colimit) of a directed system of S-semimodules (Mj, {fjj′ : Mj −→ Mj′ j ≤ j′})J can be constructed as follows: consider the disjoint union ` Mj}J and the congruence relation on ` (Mj × {j}), the embeddings {ιj : Mj −→ ` Mj = S j∈J Mj : j∈J j∈J j∈J (x, j) ∼ (x′, j′) ⇐⇒ ∃ j′′ ≥ j, j′ s.t. x ∈ Mj, x′ ∈ Mj′ and fjj′′(x) = fj′j′′(x′). (7) We define lim −→ Mj := a j∈J Mj/ ∼ and γj : Mj −→ lim −→ Mj, m 7→ [(m, j)]. Notice that lim −→ and Mj is an S-semimodule with [(m, j)]s = [(ms, j)] for all s ∈ S and m ∈ Mj [(m, j)] + [(m′, j′)] = [(fjj′′(m) + fj′j′′(m′), j′′)], where j′′ ≥ j, j′. 2.10. Let J be a directed set. The inverse limit (projective limit) of an inverse system of S-semimodules (Mj, {fjj′ : Mj′ −→ Mj j ≤ j′})J is given by: Mj = {(mj)j∈J mj = fjj′(mj′) whenever j ≤ j′}. lim ←− The proof of the following important observation is straightforward: Proposition 2.11. Every S-semimodule M is a direct limit of its finitely generated S- subsemimodules. Lemma 2.12. ([Abu-b]) Let J be a directed set and (Xj, {fjj′ : Xj −→ Mj′})J , (Yj, {gjj′ : Yj −→ Yj′})J be directed systems of S-semimodules. Let {hj : Xj −→ Yj}J be a class of S-linear morphisms satisfying hj′ ◦ fjj′ = gjj′ ◦ hj for all j, j′ ∈ J with j ≤ j′. 1. There exists a unique morphism h : (lim −→ h ◦ fj. Xj, fj) −→ (lim −→ Yj, gj ) which satisfies gj ◦ hj = 2. If hj is injective (surjective) for every j ∈ J, then h is injective (surjective). 3. If hj is uniform (resp. k-uniform, i-uniform) for every j ∈ J, then h is uniform (resp. k-uniform, i-uniform). Proposition 2.13. Let (Lj, {fjj′})J , (Mj, {gjj′})J and (Nj, {hjj′})J be directed systems of S-semimodules. 1. If {Lj αj−→ Mj βj−→ Nj}J is a class of exact (resp. semi-exact, proper-exact, quasi-exact) sequences of S-semimodules, with αj′ ◦ fjj′ = gjj′ ◦ αj and βj′ ◦ gjj′ = hjj′ ◦ βj for all Nj is j ∈ J, then the induced sequence of S-semimodules lim −→ α−→ lim −→ −→ lim −→ Mj Lj β exact (resp. semi-exact, proper-exact, quasi-exact). 11 2. If {0 −→ Lj Lj α−→ lim −→ Mj αj−→ Mj βj−→ Nj −→ 0}J is a class of short exact (resp. semi-exact, proper-exact, quasi-exact) sequences of S-semimodules, with αj′ ◦ fjj′ = gjj′ ◦ αj and βj′ ◦ gjj′ = hjj′ ◦ βj for all j ∈ J, then the induced sequence of S-semimodules 0 −→ lim Nj −→ 0 is exact (resp. semi-exact, proper-exact, quasi- −→ exact). In particular, Ker(β) ≃ lim −→ Ker(βj) and Coker(α) ≃ lim −→ β −→ lim −→ Coker(αj). Lemma 2.14. Let (Mj, {fjj′})J be a directed system of left S-semimodules with associated directed system of S-linear maps fj : Mj −→ lim −→ Mj and let X be a left S-semimodule. 1. (HomS(X, Mj ), (X, fjj′))J is a directed system of Abelian monoids. Moreover, (X, fj) : Mj) is a directed system of morphisms of Abelian HomS(X, Mj ) −→ HomS(X, lim −→ monoids and induces a morphism of Abelian monoids ψX = lim −→ (X, fj) : lim −→ HomS(X, Mj) −→ HomS(X, lim −→ Mj), [(αj, j)] 7→ [(fj ◦ αj, j)]]. (8) 2. If SX is finitely generated, then ψX is injective. Proof. The first statement is obvious. Assume that SX is finitely generated. Suppose that ψX([(αj, j)]) = ψX ([(α′ j′ for some αj ∈ HomS(X, Mj), j ∈ HomS(X, Mj′) and j, j′ ∈ J. Since SX is finitely generated, there exists j′′ ≥ j, j′ such α′ that fjj′′ ◦ αj = fj′j′′ ◦ αj′′, i.e. (X, fj′j′′)(αj′) = (X, fjj′′)(αj), whence [(αj, j)] = [(αj′, j′)].(cid:4) j′, j′)]), i.e. fj ◦ αj = fj′ ◦ α′ Proposition 2.15. (cf. [Bor1994a, Proposition 3.2.2]) Let C, D be arbitrary categories and C F −→ D G −→ C be functors such that (F, G) is an adjoint pair. 1. F preserves all colimits which turn out to exist in C. 2. G preserves all limits which turn out to exist in D. The following results can be obtained as a direct consequence of Propositions 2.15 and 2.5. Corollary 2.16. Let S, T be semirings and T FS a (T, S)-bisemimodule. 1. F ⊗S − : S S −→ T S preserves all colimits. (a) For every family of left S-semimodules {Xλ}Λ, we have a canonical isomorphism of left T -semimodules F ⊗S M Xλ ≃ M λ∈Λ λ∈Λ (F ⊗S Xλ). (b) For any directed system of left S-semimodules (Xj, {fjj′})J , we have an isomor- phism of left T -semimodules F ⊗S lim −→ Xj ≃ lim −→ (F ⊗S Xj). (c) F ⊗S − preserves coequalizers. 12 (d) F ⊗S − preserves cokernels (uniform quotients). 2. HomT (F, −) : T S −→ S S preserves all limits. (a) For every family of left T -semimodules {Yλ}Λ, we have a canonical isomorphism of left S-semimodules HomT (F, Y Yλ) ≃ Y λ∈Λ λ∈Λ HomT (F, Yλ). (b) For any inverse system of left T -semimodules (Xj, {fjj′})J , we have an isomor- phism of left S-semimodules HomT (F, lim ←− Xj) ≃ lim ←− HomT (F, Xj). (c) HomT (F, −) preserves equalizers; (d) HomT (F, −) preserves kernels (uniform subsemimodules). 3. HomT (−, F ) : T S −→ SS preserves all limits. (a) For every family of left T -semimodules {Yλ}Λ, we have a canonical isomorphism of right S-semimodules HomT (M Yλ, F, ) ≃ Y λ∈Λ λ∈Λ HomT (Yλ, F ). (b) For any directed system of left T -semimodules (Xj, {fjj′})J , we have an isomor- phism of right S-semimodules HomT (lim −→ Xj, F ) ≃ lim ←− HomT (Xj, F ). (c) HomT (−, F ) converts coequalizers into equalizers; (d) HomT (F, −) converts cokernels into kernels (uniform quotients into uniform sub- semimodules). Corollary 2.16 allows us to improve [Tak1982a, Theorem 2.6]. Proposition 2.17. Let T GS an S-bisemimodule and consider the functor HomT (G, −) : T S −→ S S. Let 0 −→ L f → M g → N (9) be a sequence of left T -semimodules and consider the following sequence of left S-semimodules 0 −→ HomT (G, L) (G,f ) → HomT (G, M ) (G,g) −→ HomT (G, N ). (10) 1. If 0 −→ L f → M is exact and f is uniform, then 0 −→ HomT (G, L) (G,f ) → HomT (G, M ) is exact and (G, f ) is uniform. 2. If (9) is semi-exact and f is uniform, then (10) is semi-exact (proper exact) and (G, f ) is uniform. 13 3. If (9) is exact and HomT (G, −) preserves k-uniform morphisms, then (10) is exact. Proof. 1. The following implications are obvious: 0 −→ L f → M is exact =⇒ f is injective =⇒ (G, f ) is injective =⇒ 0 −→ HomT (G, L) f is uniform and consider the exact sequence of S-semimodules (G,f ) → HomT (G, M ) is exact. Assume that 0 −→ L f −→ M πL−→ M/L −→ 0. Notice that L = Ker(πL) by Lemma 1.20 (5). By Corollary 2.16, HomT (G, −) preserves kernels and so (G, f ) = ker(G, πL) whence uniform. 2. Apply Lemma 1.20 (3): The semi-exactness of (9) and the uniformity of f are equivalent to L ≃ Ker(g). Since HomT (G, −) preserves kernels, we deduce that HomT (G, L) = Ker((G, g)) which is equivalent to the semi-exactness of (10) and the uniformity of (G, f ). Notice that (G, f )(HomT (G, L)) = (G, f )(HomT (G, L)) = Ker(G, g), i.e. (10) is proper exact. 3. The statement follows directly from "2" and the assumption on HomT (G, −).(cid:4) Proposition 2.18. Let T GS be a (T, S)-bisemimodule and consider the functor HomT (−, G) : T S −→ SS. Let f → M g → N −→ 0 L be a sequence of left T -semimodules and consider the sequence of right S-semimodules 0 −→ HomT (N, G) (g,G) → HomT (M, G) (f,G) −→ HomT (L, G). (11) (12) 1. If M g → N −→ 0 is exact and g is uniform, then 0 −→ HomT (N, G) (g,G) → HomT (M, G) is exact and (g, G) is uniform. 2. If (11) is semi-exact and g is uniform, then (12) is semi-exact (proper-exact) and (g, G) is uniform. 3. If (11) is exact and HomT (−, G) converts i-uniform morphisms into k-uniform ones, then (12) is exact. Proof. 1. The following implications are clear: M g → N −→ 0 is exact =⇒ g is surjective =⇒ (g, G) is injective =⇒ 0 −→ HomT (N, G) g is uniform and consider the exact sequence of S-semimodules (g,G) → HomT (M, G) is exact. Assume that 0 −→ Ker(g) ι −→ M g −→ N −→ 0. Notice that N ≃ Coker(ι). By Corollary 2.16, HomT (−, G) converts cokernels into kernels, we conclude that (g, G) = ker((f, G)) whence uniform. 2. Apply Lemma 1.20: L f → M g → N −→ 0 is semi-exact and f is uniform ⇐⇒ M ≃ Coker(f ). Since the contravariant functor HomT (−, G) converts cokernels into kernels, it follows that HomT (N, G) = Ker((f, G)) which is in turn equivalent to (12) being semi- exact and (g, G) being uniform. Notice that (g, G)(HomS(N, G)) = (g, G)(HomS(N, G)) = Ker((f, G)), i.e. (12) is proper-exact. 14 3. This follows immediately from "2" and the assumption on HomT (−, G).(cid:4) Proposition 2.19. Let T GS be a (T, S)-bisemimodule and consider the functor G ⊗S − : S S −→ T S. Let f → M g → N → 0 L be a sequence of left S-semimodules and consider the sequence of left T -semimodules G ⊗S L idG⊗S f → G ⊗S M idG⊗S g −→ G ⊗S N → 0 (13) (14) 1. If M g → N → 0 is exact and g is uniform, then G ⊗S M idG⊗S g −→ G ⊗S N → 0 is exact and idG ⊗S g is uniform. Proposition 2.20. If (13) is semi-exact and g is uniform, then (13) is semi-exact and idG ⊗S g is uniform. Proposition 2.21. If (13) is exact and G ⊗S − preserves i-uniform morphisms, then (13) is exact. Proof. The following implications are obvious: M idG ⊗S g is surjective =⇒ G ⊗S M and consider the exact sequence of S-semimodules idG⊗S g g → N → 0 is exact =⇒ g is surjective =⇒ −→ G ⊗S N → 0 is exact. Assume that g is uniform 0 −→ Ker(g) ι−→ M g −→ N −→ 0. Then N ≃ Coker(ι). By Corollary 2.16, G ⊗S − preserves cokernels and so idG ⊗S g = coker(idG ⊗S ι) whence uniform. Proof. Apply Lemma 1.20: The assumptions on (13) are equivalent to N = Coker(f ) by Lemma 1.20. Since G ⊗S − preserves cokernels, we conclude that G ⊗S N = Coker(idG ⊗S f ), i.e. (14) is semi-exact and idG ⊗S g is uniform. Proof. This follows directly form "2" and the assumption on G ⊗S −.(cid:4) We say that an S-semimodule P is projective iff for every surjective morphism of S- (P,g) semimodules M −→ HomS(P, N ) −→ 0 is surjective. It is well-known that SP is projective if and only if P is a retract of a free S-semimodule (e.g. [Tak1983, Theorem 1.9], [Gol1999a, Proposition 17.16]). −→ N −→ 0, the induced morphism of Abelian monoids HomS(P, M ) g The proof of the following lemma is straightforward: Lemma 2.22. 1. Let {fλ : Lλ −→ Mλ}Λ be a family of left S-semimodule morphisms and Mλ. Then f is uniform (resp. consider the induced S-linear map f : L k-uniform, i-uniform) if and only if fλ is uniform (resp. k-uniform, i-uniform) for S L every λ ∈ Λ. In particular, L S Mλ for every λ ∈ Λ. Mλ if and only if Lλ ≤u Lλ −→ L Lλ ≤u λ∈Λ λ∈Λ λ∈Λ λ∈Λ 2. A morphism ϕ : L −→ M of left S-semimodules is uniform (resp. k-uniform, i-uniform) if and only if idF ⊗S ϕ : F ⊗S L −→ F ⊗S M is uniform (resp. k-uniform, i-uniform) for every non-zero free right S-semimodule F 6= 0. 3. If PS is projective and ϕ : L −→ M is a uniform (resp. k-uniform, i-uniform) morphism of left S-semimodules, then idF ⊗S ϕ : P ⊗S L −→ P ⊗S M is uniform (resp. k-uniform, i-uniform). 15 It is well-known, that for every (finitely generated) S-semimodule X, there is a free S- semimodule S(J), for some (finite) index set J, and a surjective S-linear map S(J) −→ X −→ 0. Definition 2.23. We call a left S-semimodule X : X uniformly finitely generated iff there exists a uniform surjective S-linear map Sn −→ g −→ 0 for some n ∈ N; uniformly finitely presented iff SX is uniformly finitely generated and for any exact se- quence of S-semimodules 0 −→ K f −→ Sn g −→ X −→ 0, the S-semimodule K is finitely generated. Remark 2.24. Takahashi [Tak1983] defined an S-semimodule X to be normal iff there exists a projective S-semimodule P and a uniform surjective S-linear map P −→ X −→ 0 (called a projective presentation of X). Indeed, every uniformly finitely generated S-semimodule is normal. ε Proposition 2.25. If SX is uniformly finitely presented, then there exist m, n ∈ N and an exact sequence of S-semimodules Sm f −→ Sn g −→ X −→ 0. Proof. Since SX is uniformly finitely generated, there exists a uniform surjective S-linear map eg : Sn −→ X. Let K = Ker(g) and consider the exact sequence of left S-semimodules 0 −→ K ker(g) −→ Sn eg −→ X −→ 0. By assumption, SK is finitely generated and so there exists a surjective S-linear map π : Sm −→ K for some m ∈ N. Notice that ef := ker(g) ◦ π is i-uniform by [Abu, Lemma 3.8 "1-c"] and eg is uniform by assumption. Indeed, ef (Sm) = Ker(eg) and so the following sequence is exact Sm f −→ Sn g −→ X −→ 0.(cid:4) f −→ M with M ∈ M, the induced morphism of Abelian monoids HomS(M, Q) Definition 2.26. ([Abu-b]) We say that a right S-semimodule Q is (uniformly) M-injective, where M is a class of right S-semimodules, iff for every (uniform) injective morphism 0 −→ (f,Q) −→ L If SQ is (uniformly) M -injective for every M ∈ HomS(L, Q) is surjective (and uniform). S S, then we say that SQ is (uniformly) injective. In fact, SQ is uniformly injective if and only if HomS(−, Q) preserves exact sequences. 3 Flat Semimodules As before, S is a semiring with 1S 6= 0S. If M is a left S-semimodule, then we write S M to indicate that U is a uniform (subtractive) S-subsemimodule of M (i.e. the U ≤u embedding map U ι ֒→ M is uniform). The following definition applies to any variety in the sense of Universal Algebra (e.g. [BR2004]): Definition 3.1. We say that a right S-semimodule F is flat iff F = lim −→ (filtered colimit) of finitely presented projective right S-semimodules. Fi, a directed limit 16 Lemma 3.2. (cf. [Kat2004a]) The following are equivalent for a right S-semimodule FS : 1. F ⊗S − is left exact (i.e. preserves finite limits); 2. F ⊗S − preserves pullbacks and equalizers; 3. FS is pullback-flat, i.e. F ⊗S − preserves pullbacks; 4. FS is L-flat, i.e. F ≃ lim −→ S-semimodules; 5. FS is flat. Fλ, a filtered (directed) colimit of finitely generated free Although the above definition is quite natural, a notion of flatness w.r.t. to a family of semimodules is important. This motivates introducing the following notion. Definition 3.3. Let F be a right S-semimodule and M a class of left S-semimodules. We say that F is uniformly flat w.r.t. M (or uniformly M-flat) iff for every exact sequence of left S-semimodules 0 −→ L −→ M −→ N −→ 0, f g with M ∈ M, the following sequence of Abelian monoids is exact 0 −→ F ⊗S L idF ⊗S f −→ F ⊗S M idF ⊗S g −→ F ⊗S N −→ 0. (15) If FS is uniformly M -flat for every left S-semimodule SM, then we say that F is uniformly flat. Theorem 3.4. Let F be a right S-semimodule. 1. Let SM be a left S-semimodule. Then FS is uniformly M -flat if and only if for every U ≤u S M we have F ⊗S U ≤u S F ⊗S M. 2. FS is uniformly flat if and only if FS ⊗ − preserves uniform subsemimodule. Proof. We need only to prove "1". (=⇒) Assume that FS is uniformly M -flat. Let U ≤u S M and consider the exact sequence of S-semimodules 0 −→ U ι−→ M π−→ M/U −→ 0, where ι is the canonical embedding and idF ⊗S ι π is the canonical uniform surjection. By assumption, the sequence 0 −→ F ⊗S U −→ idF ⊗S π −→ F ⊗S M/U −→ 0 is exact; in particular, F ⊗S U ≤u F ⊗S M is a uniform F ⊗S M submonoid. −→ M (⇐=) Let 0 −→ L i.e. L ≃ Ker(g) and N ≃ Coker(f ). By Proposition 2.19 "2", the sequence F ⊗S L F ⊗S M idF ⊗S f is injective and uniform, whence (15) is exact.(cid:4) −→ F ⊗S N −→ 0 is proper exact and idF ⊗S g is uniform. By assumption, −→ N −→ 0 be an exact sequence of left S-semimodules, idF ⊗S f −→ idF ⊗S g f g Corollary 3.5. 1. Let M be a left S-semimodule. Any retract of a uniformly M -flat right S-semimodule is uniformly M -flat. 2. Any retract of a uniformly flat right S-semimodule is uniformly flat. 17 Proof. We need only to prove "1". Let M be a left S-semimodule and U ≤u S M. Let FS be a uniformly M -flat right S-semimodule and eF a retract of F. Then there exist S-linear maps eF −→ eF such that θ ◦ ψ = id eF . Consider the commutative diagram −→ F ψ θ F ⊗S U id F ⊗S ιU F ⊗S M ψ⊗SidU ψ⊗SidM F ⊗S U idF ⊗S ιU F ⊗S M θ⊗SidU θ⊗S idM F ⊗S U id F ⊗S ιU / F ⊗S M Indeed, (θ ⊗S idU ) ◦ (ψ ⊗S idU ) = id eF ⊗S U and (θ ⊗S idM ) ◦ (ψ ⊗S idM ) = id eF ⊗S M , i.e. eF ⊗S U is a retract of F ⊗S U and eF ⊗S M is a retract of F ⊗S M. Since FS is flat, It follows that id eF ⊗S ιU is idF ⊗S ιU : F ⊗S U −→ F ⊗S M is injective and uniform. S eF ⊗S M. Consequently, eF injective and indeed uniform by Lemma 1.21 "1", i.e. eF ⊗S U ≤u is uniformly M -flat.(cid:4) Proposition 3.6. Let {Fλ}Λ be a family of right S-semimodules. 1. Let M be a left S-semimodule. Then L λ∈Λ uniformly M -flat for every λ ∈ Λ. Fλ is uniformly M -flat if and only if Fλ is 2. L λ∈Λ Fλ is uniformly flat if and only if Fλ is uniformly flat for every λ ∈ Λ. Fλ and consider the projections πλ : λ∈Λ Proof. We need only to prove "1". Let F := L F −→ Fλ, (fλ)Λ 7→ fλ for λ ∈ Λ. Let U ≤u that Fλ is M -flat for every λ ∈ Λ. Then Fλ ⊗S U ≤u L L L Every Fλ, λ ∈ Λ, is a retract of L (Fλ ⊗S U ) ≃ L S L (Fλ ⊗S U ) ≤u S L (Fλ ⊗S M ) ≃ L Fλ⊗ is uniformly M -flat. On the other hand, assume that L (Fλ ⊗S M ) by Lemma 2.22. Since L Fλ ⊗S M, we conclude that L Fλ ⊗S U ≤u λ∈Λ λ∈Λ λ∈Λ λ∈Λ λ∈Λ λ∈Λ λ∈Λ λ∈Λ λ∈Λ λ∈Λ S M be a uniform S-subsemimodule. Assume S Fλ ⊗S M for every λ ∈ Λ, whence Fλ ⊗S U and Fλ ⊗S M. It follows that Fλ is uniformly M -flat. Fλ whence uniformly M -flat by Corollary 3.5.(cid:4) λ∈Λ Lemma 3.7. Let SX be a left S-semimodule, SYT an (S, T )-bisemimodule, T Z a uniformly flat left T -module and consider the following map of Abelian monoids νX,Y,Z : HomS(X, Y ) ⊗T Z −→ HomS(X, Y ⊗T Z), f ⊗T z 7→ f (−) ⊗S z]. 1. If SX is uniformly finitely generated, then ν X,Y,Z is injective and uniform. 2. If SX is uniformly finitely presented, then ν X,Y,Z is an isomorphism. Proof. 1. Since SX is uniformly finitely generated, there exists a uniform surjective S- linear map Sn eg −→ X −→ 0. 18 / / / / / By Proposition 2.18, HomS(X, Y ) ≤u ֒→ HomS(Sn, Y ) ⊗T Z since T Z is uniformly flat and we have a commutative diagram S HomS(Sn, Y ), whence HomS(X, Y )⊗T Z (eg,Y )⊗T idZ 0 0 / HomS(X, Y ) ⊗T Z (g,Y )⊗T idZ HomS(Sn, Y ) ⊗T Z νX,Y,Z νSn,Y,Z / HomS(X, Y ⊗T Z) (g,Y ⊗T Z) / HomS(Sn, Y ⊗T Z) Notice that ν Sn,Y,Z is an isomorphism, whence νX,Y,Z is injective. Moreover, it follows by [Abu, Lemma 3.8 (1)] that ν X,Y,Z is uniform. 2. Since SX is finitely presented, there exists by Proposition 2.25 an exact sequence of S- f g semimodules Sm −→ X −→ 0 for some m, n ∈ N. By Proposition 2.18 and the uniform flatness of T Z we obtain the following commutative diagram with proper-exact rows −→ Sn 0 0 / HomS(X, Y ) ⊗T Z (g,Y )⊗T idZ HomS(Sn, Y ) ⊗T Z ( f ,Y )⊗T idZ HomS(Sm, Y ) ⊗T Z νX,Y,Z νSn,Y,Z νSm,Y,Z / HomS(X, Y ⊗T Z) (g,Y ⊗T Z) / HomS(Sn, Y ⊗T Z) ( f ,Y ⊗T Z) / HomS(Sm, Y ⊗T Z) Notice that ν Sm,Y,Z and ν Sn,Y,Z are isomorphisms and so it follows by Lemma 1.22 "3" that ν X,Y,Z is surjective. Notice that ν X,Y,Z is injective by "1" whence an isomorphism.(cid:4) Applying Lemma 3.7 to S = T and Y = S, considered as a bisemimodule over itself in the canonical way, we obtain with X ∗ = HomS(X, S) : Proposition 3.8. Let SX be a uniformly finitely presented S-semimodule, SZ a uniformly flat left S-semimodule and consider the following morphism of Abelian monoids νX,Z : X ∗ ⊗S Z −→ HomS(X, Z), f ⊗S z 7→ f (−)z]. If SX is uniformly finitely generated (uniformly finitely presented), then ν X,Z is injective and uniform (an isomorphism). Definition 3.9. Let M be a left S-semimodule. We say that a right S-semimodule FS is M -mono-flat [Kat2004a] (or M -k-flat [Alt2004]) iff F ⊗S L ≤S F ⊗S M for every S- subsemimodule L ≤S M. If FS is M -mono-flat for every left S-semimodule M, then we call FS mono-flat (or k-flat). Notation. For every left S-semimodule M, we set IS(M ) := {G ∈ SS G ⊗S U idG⊗S ι −→ G ⊗S M is i-uniform ∀ U ≤u S M }; Remark 3.10. Let M be a left S-semimodule. If FS ∈ IS(M ) and M -mono-flat, then F is uniformly M -flat. The following result is straightforward (cf. [Alt2004, Proposition 4.1]): Proposition 3.11. Let M be a left S-semimodule and FS ∈ IS(M ). Then F is uniformly M -flat if and only if F ⊗S L ≤S F ⊗S M for every finitely generated S-subsemimodule L ≤S M. 19 / / / / / / / / / / / / / Proposition 3.12. Let 0 −→ M1 semimodules and assume that FS is uniformly M -flat. −→ M γ δ −→ M2 −→ 0 be an exact sequence of left S- 1. FS is uniformly M1-flat. 2. If FS ∈ IS(M2), then F is uniformly M2-flat. Proof. Assume that FS is uniformly M -flat. 1. Let U ≤u so F ⊗S U ≤u M1-flat. S M1. Since M1 ≤u S F ⊗S M and S F ⊗S M1 (e.g. [Abu, Lemma 3.8 (1-b)]). Consequently, FS is uniformly S M, whence F ⊗S U ≤u S M, we have U ≤u 2. Let U ≤u S M2 and consider eU := {m ∈ M δ(m) ∈ U }. Then eU ≤u commutative diagram of left S-semimodules with exact rows and columns S M and we have a 0 / U ι δ 0 U ι / M δ / M2 / 0 / 0 0 0 γ γ / M1 / M1 Tensoring with FS, we obtain a commutative diagram of Abelian monoids 0 0 F ⊗S M1 idF ⊗S γ / F ⊗S U idF ⊗Sδ F ⊗S U 0 / F ⊗S M1 idF ⊗S γ / F ⊗S M idF ⊗S δ / F ⊗S M2 idF ⊗Sι idF ⊗S ι / 0 / 0 Since FS is uniformly flat, the second row is exact. By Proposition 2.19, the first row is semi-exact and idF ⊗S eδ is uniform. It follows by Lemma 1.22 "1-a" that idF ⊗S ι is injective. Since F ∈ IS(M ), we have F ⊗S U ≤u S F ⊗S M2. Consequently, FS is uniformly M2-flat.(cid:4) Let f −→ B g −→ C A (16) be a sequence of left S-semimodules. The proof of the following result is straightforward: Proposition 3.13. 1. If FS is a free right S-semimodule and (16) is exact (resp. semi- exact, quasi-exact, proper-exact), then the sequence F ⊗S A idF ⊗S f −→ F ⊗S B idF ⊗S g −→ F ⊗S C (17) of Abelian monoids is exact (resp. semi-exact, proper-exact, quasi-exact). 2. Every free S-semimodule is uniformly flat. 20 / / / / / / / / / / / / / / / / / Corollary 3.14. 1. If PS is a projective right S-semimodule and (16) is exact (resp. semi- exact, quasi-exact, proper-exact), then the sequence P ⊗S A id⊗S f −→ P ⊗S B id⊗S g −→ P ⊗S C (18) of Abelian monoids is exact (resp. semi-exact, proper-exact, quasi-exact). 2. Every projective S-semimodule is uniformly flat. Proof. 1. This follows directly from Proposition 3.13 and Lemma 1.21 "2". 2. This follows directly from the definition and "1".(cid:4) Definition 3.15. Let Q be a right S-semimodule. We say that QS (uniformly) cogenerates a class M of right semimodules iff the following holds: for every morphisms ι : U −→ M with M ∈ M, if (ι, Q) : HomS(M, Q) −→ HomS(U, Q) is surjective (and uniform), then 0 −→ U ι−→ M is injective (and uniform). If Q (uniformly) cogenerates all left S-semimodules, then we say that Q is a (uniform) cogenerator in SS. Example 3.16. The assumption that S has an injective cogenerator might be empty. For example the semiring N0 has no injective cogenerators. Proposition 3.17. Let T FS be a (T, S)-bisemimodule, M a left S-semimodule and X a left T -semimodule. 1. Let T X be uniformly F ⊗S M -injective. If FS is uniformly M -flat, then HomT (F, X) is uniformly M -injective. 2. Let M be uniformly X-cogenerated. If HomT (F, X) is uniformly M -injective, then FS is uniformly M -flat. Proof. Let M be a left S-semimodule, U ≤u diagram S M and consider the following commutative HomS(M, HomT (F, X)) (−,HomT (F,X)) HomS(U, HomT (F, X)) ≃ ≃ HomT (F ⊗S M, X) (idF ⊗S ιU ,X) / HomT (F ⊗S U, X) 1. Let T X be uniformly F ⊗S M -injective. If FS is uniformly M -flat, then F ⊗S U ≤u T F ⊗S M, whence (idF ⊗S ιU , X) is surjective and uniform. Consequently, (−, HomT (F, X)) is surjective and uniform. This means that HomT (F, X) is uniformly M -injective. 2. Let M be uniformly X-cogenerated. If HomT (F, X) is uniformly M -injective, then (−, HomT (F, X)) is surjective and uniform, whence (idF ⊗S ιU , X) is surjective and uniform. So, idF ⊗S ιU is injective and uniform. This means that FS is uniformly M -flat.(cid:4) Theorem 3.18. Let T FS be a (T, S)-bisemimodule and assume that T S has a uniformly injective-cogenerator Q. Then FS is uniformly flat if and only if HomT (F, Q) is uniformly injective. 21 / / / The analogous of Baer's criterion for injective modules over rings "M is R-injective =⇒ M is injective" might fail for semimodules over semirings. Example 3.19. ([Ili2008]) The semifield (Q+, +, ·) has only two ideals {0} and Q+ whence every semimodule is Q+-injective. However, {0} is the only injective Q+-semimodule (e.g. by [Ili2008]). The above example motivates the following definitions: Definition 3.20. We say that the semiring S is a left (uniformly) Baer's semiring iff every (uniformly) injective left S-semimodule is (uniformly) injective. The right (uniformly) Baer- injective semirings can be defined analogously. Proposition 3.21. Let T FS be a (T, S)-bisemimodule and assume that T S has a uniformly injective cogenerator Q. If S is a left uniformly Baer semiring, then the following are equiv- alent: 1. FS is uniformly flat; 2. For every uniform left ideal SI ≤u S, we have F ⊗S I ≤u S F ⊗S S. Proof. We need only to prove "2" =⇒ "1". Let SI ≤u assumption, 0 −→ F ⊗S I S S be a left uniform ideal. By idF ⊗S ι −→ F ⊗S S is exact and idF ⊗S ι is a uniform morphism (idF ⊗S ι,Q) of left T -semimodule, whence HomT (F ⊗S I, Q) −→ HomS(F ⊗S S, Q) −→ 0 is ex- act and (idF ⊗S ι, Q) is uniform. Notice that HomT (F ⊗S I, Q) ≃ HomS(I, HomT (F, Q)) (ι,HomT (F,Q)) and HomS(F ⊗S S, Q) ≃ HomS(S, HomT (F, Q)), whence HomS(I, HomT (F, Q)) HomS(S, HomT (F, Q)) −→ 0 is exact and (ι, HomT (F, Q)) is uniform, i.e. HomT (F, Q) is uni- formly S-injective. Since S is a left uniformly Baer semiring, we conclude that HomT (F, Q) is uniformly injective as a left S-semimodule, whence FS is uniformly flat by Theorem 3.18.(cid:4) −→ Theorem 3.22. Let (Fj , {fjj′})J be a directed system of right S-semimodules. 1. If each Fj is uniformly M -flat, for some left S-semimodule M, then lim −→ M -flat. Fj is uniformly 2. If each Fj is uniformly flat, then lim −→ Fj is uniformly flat. S M. Then Fj ⊗S U ≤u Proof. We need only to prove "1". Assume that Fj is uniformly M -flat for every j ∈ J. Let U ≤u S Fj ⊗S M for each j ∈ J. It follows by Corollary 2.16 that lim (Fj ⊗S U ) −→ and lim −→ (Fj ⊗S M ) and so we are done (note that lim −→ Fj ⊗S M ≃ lim −→ Fj ⊗S U ≃ lim −→ (Fj ⊗S U ) ≤u S lim −→ (Fj ⊗S M )).(cid:4) Corollary 3.23. If every finitely generated subsemimodule of an S-semimodule F is uniformly flat, then F is uniformly flat. Proof. This follows directly from Theorem 3.22 and the fact that every semimodule is the direct limit of its finitely generated subsemimodules (cf. Proposition 2.11).(cid:4) As a direct consequence of Theorem 3.22 we obtain: Corollary 3.24. Every flat S-semimodule is uniformly flat. 22 We finish this manuscript with the following open question: Question: When is every uniformly flat S-semimodule flat? Acknowledgments. The author thanks H. Al-Thani, Y. Katsov and A. Patchkoria for providing him with several related papers and preprints. References [Abu] J. Abuhlail, Exact sequences in non-exact categories (An application to semimod- ules), submitted, arXiv:1111.0330. [Abu-b] J. Abuhlail, The category of semimodules over semirings (revisited), under prepa- ration. [AF1974] F. W. Anderson and K. R. Fuller, Rings and Categories of Modules, Springer (1974). [Alt2004] H. Al-Thani, Flat Semimodules, Int. J. Math. Math. Sci., 2004 (17) (2004), 873- 880. [Alt2003] H. Al-Thani, Injective semimodules, J. Inst. Math. Comput. Sci. Math. Ser. 16 (3) (2003), 143 -- 152. [AHS2004] J. Ad´amek, H. Herrlich and G. E. Strecker, Abstract and Concrete Cate- gories; The Joy of Cats 2004. Dover Publications Edition (2009) (available at: http://katmat.math.uni-bremen.de/acc). [BB2004] F. Borceux and D. Bourn, Mal'cev, protomodular, homological and semi-Abelian categories, Mathematics and its Application 566, Kluwer Academic Publishing, (2004). [B-F2009] S. Bulman-Fleming and A. Gilmour, Flatness properties of diagonal acts over monoids, Semigroup Forum 79 (2) (2009), 298 -- 314, [Bor1994a] F. Borceux, Handbook of Categorical Algebra. I, Basic Category Theory, Cam- bridge Univ. Press (1994). [Bor1994b] F. Borceux, Handbook of Categorical Algebra. II, Basic Category Theory, Cam- bridge Univ. Press (1994). [BR2004] F. Borceux and J. Rosick´y, On Von Neumann varieties, Theory and Applications of Categories 13 (1) (2004), 5 -- 20. [GM2008] M. Gondran and M. Minoux, Graphs, dioids and semirings. New models and algorithms, Operations Research/Computer Science Interfaces Series 41, Springer, New York (2008). [Gol1999a] J. Golan, Semirings and Their Applications, Kluwer Academic Publishers, Dor- drecht (1999). 23 [Gra2008] G. Gratzer, Universal algebra. With appendices by Gratzer, Bjarni J´onsson, Wal- ter Taylor, Robert W. Quackenbush, Gunter H. Wenzel, and Gratzer and W. A. Lampe. Revised reprint of the 1979 second edition. Springer, New York (2008). [Gril1969] P. A. Grillet, The tensor product of semigroups, Trans. Amer. Math. Soc. 138 (1969) 267 -- 280. [HW1998] U. Hebisch and H. J. Weinert, Semirings: algebraic theory and applications in computer science, World Scientific Publishing Co., Inc., River Edge, NJ (1998). [Ili2008] [Ili2010] S. N. Il'in, On the applicability of two theorems from the theory of rings and modules to semirings, Math. Notes 83 (3 -- 4) (2008), 492 -- 499. S. N. Il'in, Direct sums of injective semimodules and direct products of projective semimodules over semirings; Russian Math. 54 (10) (2010), 27 -- 37. [Ina1997] H. Inassaridze, Non-abelian homological algebra and its applications, Mathematics and its Applications 421, Kluwer Academic Publishers, Dordrecht (1997). [Kat1997] Y. Katsov, Tensor products and injective envelopes of semimodules over additively regular semirings, Algebra Colloq. 4 (2) 121 -- 131 (1997). [Kat2004a] Y. Katsov, On flat semimodules over semirings, Algebra Universalis 51 (2-3), 287-299 (2004). [KN2011] Y. Katsov, T. G. Nam, Morita equivalence and homological characterization of rings, J. Alg. Appl. 10 (3), 445 -- 473 (2011). [KKM2000] M. Kilp, U. Knauer and A. V. Mikhalev, Monoids, Acts and Categories, De Gruyter Expositions in Mathematics 29, Walter de Gruyter: Berlin, 2000. [KM1997] V. N. Kolokoltsov, V. P. Maslov, Idempotent analysis and its applications (with an appendix by Pierre Del Moral), translated from Russian, Mathematics and its Applications 401, Kluwer Academic Publishers Group, Dordrecht (1997). [Lin1966] F. Linton, Autonomous equational categories, J. Math. Mech. 15 (1966), 637 -- 642. [LM2005] G. L. Litvinov and V. P. Maslov (editors), Idempotent mathematics and mathe- matical physics, Papers from the International Workshop held in Vienna, February 3 -- 10, 2003. Contemporary Mathematics, 377. American Mathematical Society, Providence, RI (2005). [Mac1998] S. Mac Lane, Categories for the working mathematician. Second edition. Graduate Texts in Mathematics 5, Springer-Verlag (1998). [Pat2003] A. Patchkoria, Extensions of semimodules and the Takahashi functor ExtΛ(C, A), Homology Homotopy Appl. 5 (1), 387 -- 406 (2003). [Sch1972] H. Schubert, Categories, Springer Verlag (1972). [Sko1969a] L. A. Skornjakov, On homological classification of monoids (in Russian), Sib. Math. J. 10 (1969), 1139 -- 1143, correction: ibid. 12 (1971), 689. [Sko1969b] L. A. Skornjakov, Characterization of a category of acts, Math. Sb. 80 (1969), 492 -- 502. 24 [Ste1971] B. Stenstrom, Flatness and localization over monoids, Math. Nachr. 48 (1971), 315 -- 334. [Tak1981] M. Takahashi, On the bordism categories. II. Elementary properties of semimod- ules. Math. Sem. Notes Kobe Univ. 9 (2) (1981), 495-530. [Tak1982a] M. Takahashi, On the bordism categories. III. Functors Hom and for semimodules. Math. Sem. Notes Kobe Univ. 10 (2) (1982), pp. 551-562. [Tak1982b] M. Takahashi, Extensions of semimodules. I. Math. Sem. Notes Kobe Univ. 10 (2) (1982), pp. 563-592. [Tak1982c] M. Takahashi, Completeness and c-cocompleteness of the category of semimodules, Math. Sem. Notes Kobe Univ. 10 (2) (1982), 551-562. [Tak1983] M. Takahashi, Extensions of semimodules. II. Math. Sem. Notes Kobe Univ. 11 (1) (1983), pp. 83-118. [Wis1991] R. Wisbauer, Foundations of Module and Ring Theory, a Handbook for Study and Research, Gordon and Breach Science Publishers (1991). 25
1207.6367	1	1207	2012-07-26T18:51:51	Limit T-subspaces and the central polynomials in n variables of the Grassmann algebra	[ "math.RA" ]	Let F<X> be the free unitary associative algebra over a field F on the set X = {x_1, x_2, ...}. A vector subspace V of F<X> is called a T-subspace (or a T-space) if V is closed under all endomorphisms of F<X>. A T-subspace V in F<X> is limit if every larger T-subspace W \gneqq V is finitely generated (as a T-subspace) but V itself is not. Recently Brand\~ao Jr., Koshlukov, Krasilnikov and Silva have proved that over an infinite field F of characteristic p>2 the T-subspace C(G) of the central polynomials of the infinite dimensional Grassmann algebra G is a limit T-subspace. They conjectured that this limit T-subspace in F<X> is unique, that is, there are no limit T-subspaces in F<X> other than C(G). In the present article we prove that this is not the case. We construct infinitely many limit T-subspaces R_k (k \ge 1) in the algebra F<X> over an infinite field F of characteristic p>2. For each k \ge 1, the limit T-subspace R_k arises from the central polynomials in 2k variables of the Grassmann algebra G.	math.RA	math	Limit T -subspaces and the central polynomials in n variables of the Grassmann algebra Dimas Jos´e Gon¸calves∗, Alexei Krasilnikov†, Irina Sviridova‡ Departamento de Matem´atica, Universidade de Bras´ılia, 70910-900 Bras´ılia, DF, Brazil Abstract Let F hXi be the free unitary associative algebra over a field F on the set X = {x1, x2, . . .}. A vector subspace V of F hXi is called a T -subspace (or a T -space) if V is closed under all endomorphisms of F hXi. A T - subspace V in F hXi is limit if every larger T -subspace W (cid:9) V is finitely generated (as a T -subspace) but V itself is not. Recently Brandao Jr., Koshlukov, Krasilnikov and Silva have proved that over an infinite field F of characteristic p > 2 the T -subspace C(G) of the central polynomials of the infinite dimensional Grassmann algebra G is a limit T -subspace. They conjectured that this limit T -subspace in F hXi is unique, that is, there are no limit T -subspaces in F hXi other than C(G). In the present article we prove that this is not the case. We construct infinitely many limit T -subspaces Rk (k ≥ 1) in the algebra F hXi over an infinite field F of characteristic p > 2. For each k ≥ 1, the limit T -subspace Rk arises from the central polynomials in 2k variables of the Grassmann algebra G. 2000 AMS MSC Classification: 16R10, 16R40, 16R50 Keywords: polynomial identities, central polynomials, Grassmann algebra, T - subspace 1 Introduction Let F be a field, X a non-empty set and let F hXi be the free unitary associative algebra over F on the set X. Recall that a T-ideal of F hXi is an ideal closed ∗e-mail [email protected] †e-mail [email protected] ‡e-mail [email protected] 1 under all endomorphisms of F hXi. Similarly, a T -subspace (or a T -space) is a vector subspace in F hXi closed under all endomorphisms of F hXi. Let I be a T -ideal in F hXi. A subset S ⊂ I generates I as a T -ideal if I is the minimal T -ideal in F hXi containing S. A T -subspace of F hXi generated by S (as a T -subspace) is defined in a similar way. It is clear that the T -ideal (T -subspace) generated by S is the ideal (vector subspace) generated by all the polynomials f (g1, . . . , gm), where f = f (x1, . . . , xm) ∈ S and gi ∈ F hXi for all i. Note that if I is a T -ideal in F hXi then T -ideals and T -subspaces can be defined in the quotient algebra F hXi/I in a natural way. We refer to [9, 10, 12, 18, 20, 25] for the terminology and basic results concerning T -ideals and algebras with polynomial identities and to [4, 8, 16, 17, 18] for an account of the results concerning T -subspaces. From now on we write X for {x1, x2, . . .} and Xn for {x1, . . . , xn}, Xn ⊂ X. If F is a field of characteristic 0 then every T -ideal in F hXi is finitely generated (as a T -ideal); this is a celebrated result of Kemer [19, 20] that solves the Specht problem. Moreover, over such a field F each T -subspace in F hXi is finitely generated; this has been proved more recently by Shchigolev [28]. Very recently Belov [7] has proved that, for each Noetherian commutative and associative unitary ring K and each n ∈ N, each T -ideal in KhXni is finitely generated. On the other hand, over a field F of characteristic p > 0 there are T -ideals in F hXi that are not finitely generated. This has been proved by Belov [5], Grishin [13] and Shchigolev [26] (see also [6, 14, 18]). The construction of such T -ideals uses the non-finitely generated T -subspaces in F hXi constructed by Grishin [13] for p = 2 and by Shchigolev [27] for p > 2 (see also [14]). Shchigolev [27] also constructed non-finitely generated T -subspaces in F hXni, where n > 1 and F is a field of characteristic p > 2. A T -subspace V ∗ in F hXi is called limit if every larger T -subspace W (cid:9) V ∗ is finitely generated as a T -subspace but V ∗ itself is not. A limit T -ideal is defined in a similar way. It follows easily from Zorn's lemma that if a T -subspace V is not finitely generated then it is contained in some limit T -subspace V ∗. Similarly, each non-finitely generated T -ideal is contained in a limit T -ideal. In this sense limit T -subspaces (T -ideals) form a "border" between those T - subspaces (T -ideals) which are finitely generated and those which are not. By [5, 13, 26], over a field F of characteristic p > 0 the algebra F hXi contains non-finitely generated T -ideals; therefore, it contains at least one limit T -ideal. No example of a limit T -ideal is known so far. Even the cardinality of the set of limit T -ideals in F hXi is unknown; it is possible that, for a given field F of characteristic p > 0, there is only one limit T -ideal. The non-finitely generated T -ideals constructed in [1] come closer to being limit than any other known non-finitely generated T -ideal. However, it is unlikely that these T -ideals are limit. About limit T -subspaces in F hXi we know more than about limit T -ideals. Recently Brandao Jr., Koshlukov, Krasilnikov and Silva [8] have found the first example of a limit T -subspace in F hXi over an infinite field F of characteristic p > 2. To state their result precisely we need some definitions. 2 For an associative algebra A, let Z(A) denote the centre of A, Z(A) = {z ∈ A za = az for all a ∈ A}. A polynomial f (x1, . . . , xn) is a central polynomial for A if f (a1, . . . , an) ∈ Z(A) for all a1, . . . , an ∈ A. For a given algebra A, its central polynomials form a T -subspace C(A) in F hXi. However, not every T -subspace can be obtained as the T -subspace of the central polynomials of some algebra. Let V be the vector space over a field F of characteristic 6= 2, with a count- able infinite basis e1, e2, . . . and let Vs denote the subspace of V spanned by e1, . . . , es (s = 2, 3, . . .). Let G and Gs denote the unitary Grassmann algebras of V and Vs, respectively. Then as a vector space G has a basis that consists of 1 and of all monomials ei1ei2 . . . eik , i1 < i2 < · · · < ik, k ≥ 1. The mul- tiplication in G is induced by eiej = −ejei for all i and j. The algebra Gs is the subalgebra of G generated by e1, . . . , es, and dim Gs = 2s. We refer to G and Gs (s = 2, 3, . . .) as to the infinite dimensional Grassmann algebra and the finite dimensional Grassmann algebras, respectively. The result of [8] concerning a limit T -subspace is as follows: Theorem 1 ([8]). Let F be an infinite field of characteristic p > 2 and let G be the infinite dimensional Grassmann algebra over F . Then the vector space C(G) of the central polynomials of the algebra G is a limit T-space in F hXi. It was conjectured in [8] that a limit T -subspace in F hXi is unique, that is, C(G) is the only limit T -subspace in F hXi. In the present article we show that this is not the case. Our first main result is as follows. Theorem 2. Over an infinite field F of characteristic p > 2 the algebra F hXi contains infinitely many limit T -subspaces. Let F be an infinite field of characteristic p > 0. In order to prove Theorem 2 and to find infinitely many limit T -subspaces in F hXi we first find limit T - subspaces in F hXni for n = 2k, k ≥ 1. Let Cn = C(G) ∩ F hXni be the set of the central polynomials in at most n variables of the unitary Grassmann algebra G. Our second main result is as follows. Theorem 3. Let F be an infinite field of characteristic p > 2. If n = 2k, k ≥ 1, then Cn is a limit T -subspace in F hXni. If n = 2k + 1, k > 1, then Cn is finitely generated as a T -subspace in F hXni. Remark. We do not know whether the T -subspace C3 is finitely generated. Define [a, b] = ab − ba, [a, b, c] = [[a, b], c]. For k ≥ 1, let T (3,k) denote the T -ideal in F hXi generated by [x1, x2, x3] and [x1, x2][x3, x4] . . . [x2k−1, x2k] and let Rk denote the T -subspace in F hXi generated by C2k and T (3,k+1). Theorem 2 follows immediately from our third main result that is as follows. Theorem 4. Let F be an infinite field of characteristic p > 2. For each k ≥ 1, Rk is a limit T -subspace in F hXi. If k 6= l then Rk 6= Rl. 3 Now we modify the conjecture made in [8]. Problem 1. Let F be an infinite field of characteristic p > 2. Is each limit T -subspace in F hXi equal to either C(G) or Rk for some k? In other words, are C(G) and Rk (k ≥ 1) the only limit T -subspaces in F hXi? In the proof of Theorems 3 and 4 we will use the following theorem that has been proved independently by Bekh-Ochir and Rankin [4], by Brandao Jr., Koshlukov, Krasilnikov and Silva [8] and by Grishin [15]. Let q(x1, x2) = xp−1 1 [x1, x2]xp−1 2 , qk(x1, . . . , x2k) = q(x1, x2) · · · q(x2k−1, x2k). Theorem 5 ([4], [8], [15]). Over an infinite field F of a characteristic p > 2 the vector space C(G) of the central polynomials of G is generated (as a T-space in F hXi) by the polynomial x1[x2, x3, x4] and the polynomials xp 1 , xp 1 q1(x2, x3) , xp 1 q2(x2, x3, x4, x5) , . . . , xp 1 qn(x2, . . . , x2n+1) , . . . . In order to prove Theorems 3 and 4 we need some auxiliary results. Define, for each l ≥ 0, q(l)(x1, x2) = xpl−1 1 [x1, x2]xpl−1 2 , q(l) k (x1, . . . , x2k) = q(l)(x1, x2) · · · q(l)(x2k−1, x2k). Recall that Cn = C(G) ∩ F hXni. To prove Theorem 3 we need the following assertions that are also of independent interest. Proposition 6. If n = 2k, k > 1, then Cn is generated as a T -subspace in F hXni by the polynomials x1[x2, x3, x4], xp 1, xp 1q1(x2, x3), . . . , xp 1qk−1(x2, . . . , x2k−1) together with the polynomials {q(l) k (x1, . . . , x2k) l = 1, 2, . . .}. If n = 2k + 1, k > 1, then Cn is generated as a T -subspace in F hXni by the polynomials x1[x2, x3, x4], xp 1, xp 1q1(x2, x3), . . . , xp 1qk(x2, . . . , x2k+1). Let T (3) denote the T -ideal in F hXi generated by [x1, x2, x3]. Define T (3) n = T (3) ∩ F hXni. We deduce Proposition 6 from the following. 4 Proposition 7. If n = 2k, k ≥ 1, then Cn/T (3) in F hXni/T (3) by the polynomials n n is generated as a T -subspace xp 1 + T (3) n , xp 1q1(x2, x3) + T (3) n , . . . , xp 1qk−1(x2, . . . , x2k−1) + T (3) n (1) together with the polynomials {q(l) k (x1, . . . , x2k) + T (3) n l = 1, 2, . . .}. (2) If n = 2k + 1, k ≥ 1, then the T -subspace Cn/T (3) by the polynomials n in F hXni/T (3) n is generated xp 1 + T (3) n , xp 1q1(x2, x3) + T (3) n , . . . , xp 1qk(x2, . . . , x2k+1) + T (3) n . (3) Remarks. 1. For each k ≥ 1, the limit T -subspace Rk does not coincide with the T -subspace C(A) of all central polynomials of any algebra A. Indeed, suppose that Rk = C(A) for some A. Let T (A) be the T -ideal of all polynomial identities of A. Then, for each f ∈ C(A) and each g ∈ F hXi, we have [f, g] ∈ T (A). Since [x1, x2] ∈ Rk = C(A), we have [x1, x2, x3] ∈ T (A). It follows that T (3) ⊆ T (A). It is well-known that if a T -ideal T in the free unitary algebra F hXi over an infinite field F contains T (3) then either T = T (3) or T = T (3,n) for some n (see, for instance, [11, Proof of Corollary 7]). Hence, either T (A) = T (3) or T (A) = T (3,n) for some n. Note that T (3) = T (G) and T (3,n) = T (G2n−1) (see, for example, [11]) so we have either T (A) = T (G) or T (A) = T (G2n−1) for some n. For an associative algebra B, we have f (x1, . . . , xr) ∈ C(B) if and only if [f (x1, . . . , xr), xr+1] ∈ T (B). It follows that if B1, B2 are algebras such that T (B1) = T (B2) then C(B1) = C(B2). In particular, if T (A) = T (G) then C(A) = C(G), and if T (A) = T (G2n−1) then C(A) = C(G2n−1). However, x1[x2, x3] . . . [x2k+2, x2k+3] ∈ Rk \ C(G) so Rk 6= C(G). Furthermore, the T -subspaces C(Gs) of the central polynomials of the finite dimensional Grassmann algebras Gs (s = 2, 3, . . .) have been de- scribed recently by Bekh-Ochir and Rankin [3] and by Koshlukov, Krasilnikov and Silva [21]; these T -subspaces are finitely generated and do not coincide with Rk. This contradiction proves that Rk 6= C(A) for any algebra A, as claimed. 2. For an associative unitary algebra A, let Cn(A) and Tn(A) denote the set of the central polynomials and the set of the polynomial identities in n variables x1, . . . , xn of A, respectively; that is, Cn(A) = C(A) ∩ F hXni and Tn(A) = T (A) ∩ F hXni. Then Cn(A) is a T -subspace and Tn(A) is a T -ideal in F hXni. Note that, by Belov's result [7], the T -ideal Tn(A) is finitely generated for each algebra A over a Noetherian ring and each positive integer n. On the other hand, there exist unitary algebras A over an infinite field F of characteristic p > 2 such that, for some n > 1, the T -subspace Cn(A) of the central polynomials 5 of A in n variables is not finitely generated. Moreover, such an algebra A can be finite dimensional. Indeed, take A = Gs, where s ≥ n. It can be checked that C(Gs) ∩ F hXni = Cn if s ≥ n. By Proposition 9, the T -subspace C2k(Gs) in F hX2ki is not finitely generated provided that s ≥ 2k. However, the following problem remains open. Problem 2. Does there exist a finite dimensional algebra A over an infinite field F of characteristic p > 0 such that the T -subspace C(A) of all central polynomials of A in F hXi is not finitely generated? Note that a similar problem for the T -ideal T (A) of all polynomial identities of a finite dimensional algebra A over an infinite field F of characteristic p > 0 remains open as well; it is one of the most interesting and long-standing open problems in the area. 2 Preliminaries Let hSiT S denote the T -subspace generated by a set S ⊆ F hXi. Then hSiT S is the span of all polynomials f (g1, . . . , gn), where f ∈ S and gi ∈ F hXi for all i. It is clear that for any polynomials f1, . . . , fs ∈ F hXi we have hf1, . . . , fsiT S = hf1iT S + . . . + hfsiT S. Recall that a polynomial f (x1, . . . , xn) ∈ F hXi is called a polynomial identity in an algebra A over F if f (a1, . . . , an) = 0 for all a1, . . . , an ∈ A. For a given algebra A, its polynomial identities form a T -ideal T (A) in F hXi and for every T -ideal I in F hXi there is an algebra A such that I = T (A), that is, I is the ideal of all polynomial identities satisfied in A. Note that a polynomial f = f (x1, . . . , xn) is central for an algebra A if and only if [f, xn+1] is a polynomial identity of A. Let f = f (x1, . . . , xn) ∈ F hXi. Then f = P0≤i1,...,in fi1...in , where each polynomial fi1...in is multihomogeneous of degree is in xs (s = 1, . . . , n). We refer to the polynomials fi1...in as to the multihomogeneous components of the polynomial f. Note that if F is an infinite field, V is a T -ideal in F hXi and f ∈ V then fi1...in ∈ V for all i1, . . . , in (see, for instance, [2, 9, 12, 25]). Similarly, if V is a T -subspace in F hXi and f ∈ V then all the multihomogeneous components fi1...in of f belong to V . Over an infinite field F the T -ideal T (G) of the polynomial identities of the infinite dimensional unitary Grassmann algebra G coincides with T (3). This was proved by Krakowski and Regev [22] if F is of characteristic 0 (see also [23]) and by several authors in the general case, see for example [11]. It is well known (see, for example, [22, 23]) that over any field F we have [g1, g2][g1, g3] + T (3) = T (3); [g1, g2][g3, g4] + T (3) = −[g3, g2][g1, g4] + T (3); [gm 1 , g2] + T (3) = mgm−1 [g1, g2] + T (3) 1 (4) for all g1, g2, g3, g4 ∈ F hXi. Also it is well known (see, for instance, [8, 17]) that a basis of the vector space F hXi/T (3) over F is formed by the elements of the 6 form xm1 i1 · · · xmd id [xj1 , xj2 ] · · · [xj2s−1 , xj2s ] + T (3), where d, s ≥ 0, i1 < . . . < id, j1 < . . . < j2s. Define T (3) n = T (3) ∩ F hXni. We claim that if n < 2i then T (3,i) ∩ F hXni = T (3) n . (5) (6) Indeed, a basis of the vector space (F hXni+T (3))/T (3) is formed by the elements of the form (5) such that 1 ≤ i1 < . . . < id ≤ n, 1 ≤ j1 < . . . < j2s ≤ n. In particular, we have 2s ≤ n. On the other hand, it can be easily checked that T (3,i)/T (3) is contained in the linear span of the elements of the form (5) such that s ≥ i. Since n < 2i, we have ((F hXni + T (3))/T (3)) ∩ (T (3,i)/T (3)) = {0}, that is, T (3,i)∩F hXni ⊆ T (3). It follows immediately that T (3,i)∩F hXni ⊆ T (3) n . Since T (3) if n < 2i, as claimed. n ⊆ T (3,i) ∩ F hXni for all i, we have T (3,i) ∩ F hXni = T (3) n Let F be a field of characteristic p > 2. It is well known (see, for example, [24, 4, 8, 16]) that, for each g, g1, . . . , gn ∈ F hXi, we have gp + T (3) is central in F hXi/T (3); (g1 · · · gn)p + T (3) = gp 1 · · · gp (g1 + . . . + gn)p + T (3) = gp n + T (3); 1 + . . . + gp n + T (3). (7) Let F be an infinite field of characteristic p > 2. Let Q(k,l) be the T -subspace k (x1, . . . , x2k)iT S. Note that the in F hXi generated by q(l) (l ≥ 0), Q(k,l) = hq(l) multihomogeneous component of the polynomial k q(l) k (1 + x1, . . . , 1 + x2k) = (1 + x1)pl−1[x1, x2](1 + x2)pl−1 . . . (1 + x2k−1)pl−1[x2k−1, x2k](1 + x2k)pl−1 of degree pl−1 in all the variables x1, . . . , x2k is equal to γ q(l−1) k (x1, . . . , x2k) = γ xpl−1−1 1 [x1, x2]xpl−1−1 2 . . . xpl−1−1 2k−1 [x2k−1, x2k]xpl−1−1 2k , where γ = (cid:0) pl−1 pl−1−1(cid:1) so Q(k,l−1) ⊆ Q(k,l). Hence, for each l > 0 we have ≡ 1 (mod p). It follows that q(l−1) k 2k l X i=0 Q(k,i) = Q(k,l). ∈ Q(k,l) for all l > 0 (8) The following lemma is a reformulation of a result of Grishin and Tsybulya [16, Theorem 1.3, item 1)]. 7 Lemma 8. Let F be an infinite field of characteristic p > 2. Let k ≥ 1, ai ≥ 1 for all i = 1, 2 . . . , 2k and let m = xa1−1 1 xa2−1 2 . . . xa2k −1 2k [x1, x2] . . . [x2k−1, x2k] ∈ F hXi. Suppose that, for some i0, 1 ≤ i0 ≤ 2k, we have ai0 = plb, where l ≥ 0 and b is coprime to p. Suppose also that, for each i, 1 ≤ i ≤ 2k, we have ai ≡ 0 (mod pl). Then hmiT S + T (3) = Q(k,l) + T (3). 3 Proof of Propositions 6 and 7 In the rest of the paper, F will denote an infinite field of characteristic p > 2. Proof of Proposition 7 Let U be the T -subspace of F hXni defined as follows: i) T (3) n ⊂ U ; ii) the T -subspace U/T (3) n of F hXni/T (3) n is generated by the polynomials (1) and (2) if n = 2k and by the polynomials (3) if n = 2k + 1. To prove the proposition we have to show that Cn/T (3) Cn = U ). It can be easily seen that U/T (3) prove that Cn/T (3) (equivalently, Cn ⊆ U ). n ⊆ U/T (3) n ⊆ Cn/T (3) n n = U/T (3) (equivalently, n . Thus, it remains to n Let h be an arbitrary element of Cn. We are going to check that h + T (3) U/T (3) n . Since h ∈ C(G), it follows from Theorem 5 that n ∈ h = X j αj vp j + X i,j αij wp ij qi(f (ij) 1 , . . . , f (ij) 2i ) + h′, where vj, wij , f (ij) may assume that vj , wij, f (ij) s s ∈ F hXi, αj , αij ∈ F , h′ ∈ T (3). Note that h ∈ F hXni so we , h′ ∈ F hXni for all i, j, s. It follows that h + T (3) n = X j αj vp j + X i,j αij wp ij qi(f (ij) 1 , . . . , f (ij) 2i ) + T (3) n . Recall that T (3,i) is the T -ideal in F hXi generated by the polynomials for [x1, x2, x3] and [x1, x2] . . . [x2i−1, x2i]. By (6), we have T (3,i) ∩ F hXni = T (3) each i such that 2i > n. Since, for each i, j, n ij qi(f (ij) wp 1 , . . . , f (ij) 2i ) ∈ T (3,i), 8 we have X i> n 2 X j It follows that αij wp ij qi(f (ij) 1 , . . . , f (ij) 2i ) ∈ T (3,i) ∩ F hXni = T (3) n . h + T (3) n = X j αj vp j + X i≤ n 2 αij wp ij qi(f (ij) 1 , . . . , f (ij) 2i ) + T (3) n . X j If n = 2k + 1 (k ≥ 1) then we have h + T (3) n = X j αj vp j + k X i=1 X j αij wp ij qi(f (ij) 1 , . . . , f (ij) 2i ) + T (3) n so h + T (3) n ∈ U/T (3) n , as required. If n = 2k (k ≥ 1) then we have h + T (3) n = h1 + h2 + T (3) n , where and h1 = X j αjvp j + k−1 X i=1 X j αij wp ij qi(f (ij) 1 , . . . , f (ij) 2i ) h2 = X j αkj wp kj qk(f (kj) 1 , . . . , f (kj) 2k ). It is clear that h1 +T (3) (1); hence, h1 + T (3) h2 + T (3) n n belongs to the T -subspace generated by the polynomials n . On the other hand, it can be easily seen that n ∈ U/T (3) is a linear combination of polynomials of the form m + T (3) n , where m = xb1 1 · · · xb2k 2k [x1, x2] · · · [x2k−1, x2k]. 2k belongs to We claim that, for each m of this form, the polynomial m + T (3) U/T (3) 2k . Indeed, by Lemma 8, we have hmiT S + T (3) = hq(l) 2k belongs to the T -subspace of F hX2ki/T (3) Since both m and q(l) m+T (3) for some l ≥ 0. If l ≥ 1 then m+ T (3) polynomial of the form (2). If l = 0 then m + T (3) F hX2ki/T (3) 2k generated by q(1) the T -subspace generated by q(0) the T -subspace generated by q(1) component of q(1) This proves our claim. k iT S + T (3) for some l ≥ 0. k are polynomials in x1, . . . , x2k, this equality implies that k +T (3) 2k 2k is a 2k belongs to the T -subspace of 2k . Indeed, in this case m + T (3) 2k belongs to 2k and the latter T -subspace is contained in 2k because q(0) is equal to the multilinear 2k ∈ U/T (3) 2k . k (1 + x1, . . . , 1 + x2k). It follows that, again, m+ T (3) k + T (3) k +T (3) k + T (3) 2k ∈ U/T (3) 2k because, for l ≥ 1, q(l) 2k that is generated by q(l) k + T (3) k It follows that h2 + T (3) n ∈ U/T (3) n and, therefore, h + T (3) n ∈ U/T (3) n , as required. Thus, Cn ⊆ U for each n. This completes the proof of Proposition 7. 9 Proof of Proposition 6 It is clear that the polynomial x1[x2, x3, x4]x5 generates T (3) as a T -subspace in F hXi. Since g1[g2, g3, g4]g5 = g1[g2, g3, g4, g5]+ g1g5[g2, g3, g4] for all gi ∈ F hXi, the polynomial x1[x2, x3, x4] generates T (3) as a T -subspace in F hXi as well. It follows that x1[x2, x3, x4] generates T (3) n as a T -subspace in F hXni for each n ≥ 4. Proposition 6 follows immediately from Proposition 7 and the observation above. 4 Proof of Theorem 3 If n = 2k + 1, k > 1, then Theorem 3 follows immediately from Proposition 6. Suppose that n = 2k, k ≥ 1. Then Theorem 3 is an immediate consequence of the following two propositions. Proposition 9. For all k ≥ 1, C2k is not finitely generated as a T -subspace in F hX2ki. Proposition 10. Let k ≥ 1 and let W be a T -subspace of F hX2ki such that C2k $ W . Then W is a finitely generated T -subspace in F hX2ki. Proof of Proposition 9 The proof is based on a result of Grishin and Tsybulya [16, Theorem 3.1]. By Proposition 7, C2k is generated as a T -subspace in F hX2ki by T (3) 2k to- gether with the polynomials xp 1, xp 1q1(x2, x3), . . . , xp 1qk−1(x2, . . . , x2k−1) (9) and {q(l) k (x1, . . . , x2k) l = 1, 2, . . .}. Let Vl be the T -subspace of F hX2ki generated by T (3) nomials (9) and the polynomials {q(i) 2k together with the poly- k (x1, . . . , x2k) i ≤ l}. Then we have C2k = [ l≥1 Vl. (10) Also, it is clear that V1 ⊆ V2 ⊆ . . . . Let U (k−1) be the T -subspace in F hXi generated by the polynomials (9). The following proposition is a particular case of [16, Theorem 3.1]. Proposition 11 ([16]). For each l ≥ 1, (Q(k,l+1) + T (3))/T (3) 6⊆ (U (k−1) + Q(k,l) + T (3,k+1))/T (3). 10 Remark. T (3,k+1)/T (3) are denoted in [16] by Pi<k CD(i) The T -subspaces (U (k−1) + T (3))/T (3), (Q(k,l) + T (3))/T (3) and pl and C(k+1), respectively. p , C(k) Since the T -subspace Q(k,l+1) is generated by the polynomial q(l+1) and k T (3) ⊂ T (3,k+1), Proposition 11 immediately implies that q(l+1) k /∈ U (k−1) + Q(k,l) + T (3,k+1). Further, since T (3) 2k ⊂ T (3) ⊂ T (3,k+1), we have Vl ⊂ U (k−1) + X i≤l Q(k,i) + T (3,k+1) = U (k−1) + Q(k,l) + T (3,k+1) (recall that, by (8), Pi≤l Q(k,i) = Q(k,l)). l ≥ 1; on the other hand, q(l+1) k ∈ Vl+1 by the definition of Vl+1. Hence, It follows that q(l+1) k /∈ Vl for all V1 $ V2 $ . . . . (11) It follows immediately from (10) and (11) that C2k is not finitely generated as a T -subspace in F hX2ki. The proof of Proposition 9 is completed. Proof of the Proposition 10 For all integers i1, . . . , it such that 1 ≤ i1 < . . . < it ≤ n and all integers a1, . . . , an ≥ 0 such that ai1, . . . , ait ≥ 1, define x to be the monomial a2 a1 2 ...xan 1 x n xi1 xi2 ...xit 1 xa2 xa1 2 . . . xan n xi1 xi2 . . . xit = xb1 1 xb2 2 . . . xbn n ∈ F hXi, where bj = aj − 1 if j ∈ {i1, i2, . . . , it} and bj = aj otherwise. Lemma 12. Let f (x1, . . . , xn) ∈ F hXi be a multihomogeneous polynomial of the form f = α xa1 1 . . . xan n + X 1≤i1<...<i2t≤n α(i1,...,i2t) xa1 1 . . . xan n xi1 . . . xi2t [xi1 , xi2 ] . . . [xi2t−1 , xi2t ] (12) where α, α(i1,...,i2t) ∈ F . Let L = hf iT S + h[x1, x2]iT S + T (3). Suppose that ai = 1 for some i, 1 ≤ i ≤ n. Then either L = F hXi or L = h[x1, x2]iT S + T (3) or L = hx1[x2, x3] . . . [x2θ, x2θ+1]iT S + h[x1, x2]iT S + T (3) for some θ ≤ n−1 2 . Proof. Note that each multihomogeneous polynomial f (x1, . . . , xn) ∈ F hXi can be written, modulo T (3), in the form (12). Hence, we can assume without loss of generality (permuting the free generators x1, . . . , xn if necessary) that a1 = 1. Note that if α 6= 0, then f (x1, 1, . . . , 1) = αx1 ∈ L so L = hx1iT S = F hXi. Suppose that α = 0. 11 We claim that we may assume without loss of generality that f is of the form f (x1, . . . , xn) = x1g(x2, . . . , xn), where g = X 2≤i1<...<i2t≤n t≥1 α(i1,...,i2t) xa2 2 . . . xan n xi1 . . . xi2t [xi1 , xi2 ] . . . [xi2t−1 , xi2t ]. (13) [xi1 , xi2 ] . . . [xi2t−1 , xi2t ] in (12). If i1 > 1 Indeed, consider a term m = x then a1 1 ...xan n xi1 ...xi2t xa2 2 . . . xan n xi1 . . . xi2t m = x1 [xi1 , xi2 ] . . . [xi2t−1 , xi2t ]. (14) Suppose that i1 = 1; then m = m′[x1, xi2 ] . . . [xi2t−1 , xi2t ], where m′ = x We have a2 2 ...xan n xi2 ...xi2t . m + T (3) = m′[x1, xi2 ] . . . [xi2t−1 , xi2t ] + T (3) = [m′x1, xi2 ] . . . [xi2t−1 , xi2t ] − x1[m′, xi2 ] . . . [xi2t−1 , xi2t ] + T (3) = [m′x1[xi3 , xi4 ] . . . [xi2t−1 , xi2t ], xi2 ] − x1[m′, xi2 ] . . . [xi2t−1 , xi2t ] + T (3). Hence, m = −x1[m′, xi2 ] . . . [xi2t−1 , xi2t ] + h, (15) where h ∈ h[x1, x2]iT S + T (3). It follows easily from (14) and (15) that there exists a multihomogeneous polynomial g1 = g1(x2, . . . , xn) ∈ F hXi such that f = x1g1 + h1, where h1 ∈ h[x1, x2]iT S + T (3). Further, there is a multihomogeneous polynomial g of the (mod T (3)); then f = x1g + h2, where h2 ∈ form (13) such that g ≡ g1 h[x1, x2]iT S + T (3). It follows that L = hx1g(x2, . . . , xn)iT S + h[x1, x2]iT S + T (3). Thus, we can assume without loss of generality that f = x1g(x2, . . . , xn), where g is of the form (13), as claimed. If f = 0 then L = h[x1, x2]iT S + T (3). Suppose that f 6= 0. Let θ = min {t 2 . We can assume that α(i1,...,i2t) 6= 0}. It is clear that 2θ + 1 ≤ n so θ ≤ n−1 α(2,...,2θ+1) 6= 0; then f = x1(cid:16)α(2,...,2θ+1) n [x2, x3] . . . [x2θ, x2θ+1] xa2 2 . . . xan x2 . . . x2θ+1 xa2 2 . . . xan n xi1 . . . xi2t α(i1,...,i2t) + X 2≤i1<...<i2t ≤n t≥θ, i2t >2θ+1 [xi1 , xi2 ] . . . [xi2t−1 , xi2t ](cid:17). (16) Let f1(x1, . . . , x2θ+1) = f (x1, x2, . . . , x2θ+1, 1, . . . , 1) ∈ L; then f1 = α(2,...,2θ+1) x1 xa2 2 . . . xan x2 . . . x2θ+1 n [x2, x3] . . . [x2θ, x2θ+1]. It can be easily seen that the multihomogeneous component of degree 1 in the variables x1, x2, . . . , x2θ+1 of the polynomial f1(x1, x2 +1, . . . , x2θ+1 +1) is equal to α(2,...,2θ+1) x1[x2, x3] . . . [x2θ, x2θ+1]. 12 It follows that x1[x2, x3] . . . [x2θ, x2θ+1] ∈ L; hence, hx1[x2, x3] . . . [x2θ, x2θ+1]iT S + h[x1, x2]iT S + T (3) ⊆ L. On the other hand, it is clear that the polynomial f of the form (16) belongs to the T -subspace of F hXi generated by x1[x2, x3] . . . [x2θ, x2θ+1]; it follows that hf iT S ⊆ hx1[x2, x3] . . . [x2θ, x2θ+1]iT S and, therefore, L ⊆ hx1[x2, x3] . . . [x2θ, x2θ+1]iT S + h[x1, x2]iT S + T (3). Thus, L = hx1[x2, x3] . . . [x2θ, x2θ+1]iT S + h[x1, x2]iT S + T (3). The proof of Lemma 12 is completed. Proposition 13. Let W be a T -subspace of F hX2ki such that C2k $ W . Then W = F hX2ki or W is generated as a T -subspace by the polynomials xp 1, xp 1q1(x2, x3), . . . , xp 1qλ−1(x2, . . . , x2λ−1), x1[x2, x3, x4], x1[x2, x3] . . . [x2λ, x2λ+1], for some positive integer λ ≤ k − 1. Proof. It is well-known that over a field F of characteristic 0 each T -ideal in F hXi can be generated by its multilinear polynomials. It is easy to check that the same is true for each T -subspace in F hXi. Over an infinite field F of charac- teristic p > 0 each T -ideal in F hXi can be generated by all its multihomogeneous polynomials f (x1, . . . , xn) such that, for each i, 1 ≤ i ≤ n, degxi f = psi for some integer si (see, for instance, [2]). Again, the same is true for each T -subspace in F hXi. Let f (x1, . . . , x2k) ∈ W \ C2k be an arbitrary multihomogeneous polynomial such that, for each i (1 ≤ i ≤ 2k), we have either degxif = psi or degxif = 0. We may assume that degxif = psi for i = 1, . . . , l and degxif = 0 for i = l+1, . . . , 2k (that is, f = f (x1, . . . , xl)). Then we have f +T (3) 2k = α m+ X 1≤i1<...<i2t≤l α(i1,...,i2t) m xi1 . . . xi2t [xi1 , xi2 ] . . . [xi2t−1 , xi2t ]+T (3) 2k , where α, α(i1,...,i2t) ∈ F , m = xps1 1 . . . xpsl l . If si > 0 for all i = 1, . . . , l then it can be easily seen that f ∈ C(G) so f ∈ C2k, a contradiction with the choice of f . Thus, si = 0 for some i, 1 ≤ i ≤ l. Let Lf be the T -subspace of F hXi generated by f , [x1, x2] and T (3). By Lemma 12, we have either Lf = F hXi or Lf = hx1[x2, x3] . . . [x2θ, x2θ+1]iT S + h[x1, x2]iT S + T (3) for some θ < k (since f /∈ C2k, we have Lf 6= h[x1, x2]iT S + T (3)). Note that if k = 1 (that is, f = f (x1, x2)) then the only possible case is Lf = F hXi. It is clear that if Lf = F hXi for some f ∈ W \ C2k then x1 ∈ W so W = F hX2ki. Suppose that W 6= F hX2ki; then k > 1 and Lf 6= F hXi for all 13 f ∈ W \ C2k. For each f ∈ W \ C2k satisfying the conditions of Lemma 12, the T -subspace Lf in F hXi can be generated, by Lemma 12, by the polynomials [x1, x2], x1[x2, x3x4] and x1[x2, x3] . . . [x2θ, x2θ+1] (17) for some θ = θf < k. Since the polynomials (17) belong to F hX2ki (recall that k > 1), the T -subspace in F hX2ki generated by f , [x1, x2] and T (3) is also generated (as a T -subspace in F hX2ki) by the polynomials (17). Note that [x1, x2] and x1[x2, x3, x4] belong to C2k so the T -subspace Vf in F hX2ki generated by f and C2k can be generated by C2k and x1[x2, x3] . . . [x2θ, x2θ+1] for some θ = θf < k. Let λ = min {θ x1[x2, x3] . . . [x2θ, x2θ+1] ∈ W }. Since W is the sum of the T -subspaces Vf for all suitable multihomogeneous polynomials f ∈ W \ C2k and each Vf is generated by C2k and x1[x2, x3] . . . [x2θ, x2θ+1] for some θ = θf < k, W can be generated as a T -subspace in F hX2ki by C2k and x1[x2, x3] . . . [x2λ, x2λ+1]. Now it follows easily from Proposition 6 that W can be generated by the polynomials xp 1, xp 1q1(x2, x3), . . . , xp 1qλ−1(x2, . . . , x2λ−1) together with the polynomials x1[x2, x3, x4] and x1[x2, x3] . . . [x2λ, x2λ+1], where we note that λ < k. This completes the proof of Proposition 13. Proposition 10 follows immediately from Proposition 13. The proof of The- orem 3 is completed. 5 Proof of Theorem 4 Proposition 14. For each k ≥ 1, Rk is not finitely generated as a T -subspace in F hXi. Proof. Recall that Rk is the T -subspace in F hXi generated by C2k and T (3,k+1). By Proposition 7, C2k is generated as a T -subspace in F hX2ki by T (3) 2k together with the polynomials (9) and the polynomials {q(l) k (x1, . . . , x2k) l = 1, 2, . . .}. Since T (3) 2k ⊂ T (3) ⊂ T (3,k+1), we have Rk = U (k−1) + X l≥1 Q(k,l) + T (3,k+1), where U (k−1) and Q(k,l) are the T -subspaces in F hXi generated by the polyno- mials (9) and by the polynomial q(l) k (x1, . . . , x2k), respectively. 14 Let Vl = U (k−1) + Pi≤l Q(k,i) + T (3,k+1). Then Rk = [ l≥1 Vl (18) and V1 ⊆ V2 ⊆ . . . . Recall that, by (8), Pi≤l Q(k,i) = Q(k,l) so Vl = U (k−1) + Q(k,l) + T (3,k+1). By Proposition 11, Q(k,l+1) 6⊆ Vl for all l ≥ 1 so V1 $ V2 $ . . . . (19) The result follows immediately from (18) and (19). Lemma 15. Let f = f (x1, . . . , xn) ∈ F hXi be a multihomogeneous polynomial of the form f = α xps1 1 . . . xpsn n + X i1<...<i2t α(i1,...,i2t) xps1 . . . xpsn 1 n xi1 . . . xi2t [xi1 , xi2 ] . . . [xi2t−1 , xi2t ], (20) where α, α(i1,...,i2t) ∈ F , si ≥ 0 for all i. Let L = hf iT S + Rk, k ≥ 1. Then one of the following holds: 1. L = F hXi; 2. L = Rk; 3. L = hx1[x2, x3] . . . [x2θ, x2θ+1]iT S + Rk for some θ, 1 ≤ θ ≤ k; 4. L = hxps 1 q(s) k (x2, . . . , x2k+1)iT S + Rk for some s ≥ 1. Proof. Note that each multihomogeneous polynomial f (x1, . . . , xn) ∈ F hXi of degree psi in xi (1 ≤ i ≤ n) can be written, modulo T (3), in the form (20). Hence, we can assume without loss of generality (permuting the free generators x1, . . . , xn if necessary) that s1 ≤ si for all i. Write s = s1. Suppose that s = 0. Then, by Lemma 12, we have either hf iT S + h[x1, x2]iT S + T (3) = F hXi hf iT S + h[x1, x2]iT S + T (3) = h[x1, x2]iT S + T (3) or or hf iT S + h[x1, x2]iT S + T (3) = hx1[x2, x3] . . . [x2θ, x2θ+1]iT S + h[x1, x2]iT S + T (3) for some θ. Since h[x1, x2]iT S + T (3) ⊂ Rk and x1[x2, x3] . . . [x2θ, x2θ+1] ∈ Rk if θ > k, we have either L = F hXi or L = Rk or L = hx1[x2, x3] . . . [x2θ, x2θ+1]iT S + Rk for some θ ≤ k. 15 Now suppose that s > 0; then si > 0 for all i, 1 ≤ i ≤ n. It can be easily seen that, by (7), xps1 1 . . . xpsn n ∈ (cid:0)hxp 1iT S + T (3)(cid:1) ⊂ Rk and, for all t < k, [xi1 , xi2 ] . . . [xi2t−1 , xi2t ] ∈ (cid:16)hxp xps1 . . . xpsn 1 n xi1 . . . xi2t Also we have xps1 [xi1 , xi2 ] . . . [xi2t−1 , xi2t ] ∈ T (3,k+1) ⊂ Rk for each t > k. It follows that we can assume without loss of generality that the polynomial f is of the form 1qt(x2, . . . , x2t+1)iT S + T (3)(cid:17) ⊂ Rk. ...xpsn n 1 xi1 ...xi2t f = X 1≤i1<...<i2k ≤n α(i1,...,i2k) xps1 . . . xpsn 1 n xi1 . . . xi2k [xi1 , xi2 ] . . . [xi2k−1 , xi2k ]. (21) Note that if n < 2k then f = 0 and if n = 2k then f = α(1,2,...,2k) xps1 . . . xps2k 1 2k x1x2 . . . x2k [x1, x2] . . . [x2k−1, x2k] so, by Lemma 8, we have f ∈ Q(k,s) + T (3), where s = s1 > 0. In both cases we have f ∈ Rk and L = Rk. Suppose that n > 2k. We claim that we may assume that f is of the form f (x1, . . . , xn) = xps 1 g(x2, . . . , xn), (22) where g = X 2≤i1<...<i2k ≤n α(i1,...,i2k) xps2 . . . xpsn 2 n xi1 . . . xi2k [xi1 , xi2 ] . . . [xi2k−1 , xi2k ]. Indeed, consider a term m = xps1 1 ...xpsn n xi1 ...xi2k [xi1 , xi2 ] . . . [xi2k−1 , xi2k ] in (21). If i1 > 1 then m = xps 1 xps2 . . . xpsn 2 n xi1 . . . xi2k [xi1 , xi2 ] . . . [xi2k−1 , xi2k ]. (23) Suppose that i1 = 1. Let ai = psi for all i. Then m + T (3,k+1) = xps −1 1 xps2 . . . xpsn 2 n xi2 . . . xi2k [x1, xi2 ] . . . [xi2k−1 , xi2k ] + T (3,k+1) = x aj1 j1 · · · x ajl jl xa1−1 1 · · · x ai2k i2k −1 [x1, xi2 ] · · · [xi2k−1 , xi2k ] + T (3,k+1) = xa1−1 1 x aj1 j1 . . . x ajl jl [x1, xi2 ]x ai2 −1 i2 m′ + T (3,k+1), where m′ = x ai3 −1 i3 [xi3 , xi4 ]x ai4 −1 i4 . . . x ai2k−1 i2k−1 −1 [xi2k−1 , xi2k ]x ai2k i2k −1 , 16 {j1, . . . , jl} = {1, . . . , n} \ {1, i2, . . . , i2k}, l = n − 2k > 0. Suppose that a1 = aj1 = aj2 = . . . = ajz and ajz+1, ajz+2, . . . , ajl > a1. Let u = x1xj1 · · · xjz x a′ jz+1 jz+1 · · · x a′ jl jl , where a′ i = ai/ps for all i. Let h = h(x1, . . . , x2k) = xa1−1 1 ai2 −1 [x1, x2]x 2 . . . x ai2k−1 2k−1 −1 [x2k−1, x2k]x ai2k −1 2k . By (4), h ∈ C(G); hence, h ∈ C2k ⊂ Rk. It follows that h(u, xi2 , . . . , xi2k ) ∈ Rk, that is, ups−1[u, xi2 ]x ai2 −1 i2 m′ ∈ Rk. (24) Since, by (7), [vp 1, v2] ∈ T (3) ⊂ T (3,k+1) for all v1, v2 ∈ F hXi, we have ai2 −1 i2 m′ + T (3,k+1) · · · x ups−1[u, xi2 ]x = (x1xj1 · · · xjz )ps−1 x = (x1xj1 · · · xjz )ps−1 x + (x1xj1 · · · xjz )ps−1 x = m + xps ajz+1 jz+1 ajz+1 jz+1 ajz+1 jz+1 · · · xps−1 · · · x ajz+1 jz+1 1 xps −1 j1 · · · x x jz ajl jl ajl jl ajl jl [x1xj1 . . . xjz , xi2 ]x [x1, xi2 ]xj1 . . . xjz x x1[xj1 . . . xjz , xi2 ]x m′ + T (3,k+1) ai2 −1 i2 ai2 −1 i2 m′ ai2 −1 i2 m′ + T (3,k+1) · · · x ajl jl [xj1 . . . xjz , xi2 ]x ai2 −1 i2 m′ + T (3,k+1) where the second summand is not present if z = 0 (that is, if aji > a1 for all i), in which case m ∈ Rk. Since 1 xps −1 xps j1 · · · xps−1 jz ajz+1 jz+1 x · · · x = xps 1 X 2≤i1<...<i2k β(i1,...,i2k) ajl [xj1 . . . xjz , xi2 ]x jl xps2 . . . xpsn 2 n xi1 . . . xi2k ai2 −1 i2 m′ + T (3,k+1) [xi1 , xi2 ] . . . [xi2k−1 , xi2k ] + T (3,k+1) for some β(i1,...,i2k) ∈ F , we have m + xps 1 X 2≤i1<...<i2k β(i1,...,i2k) xps2 . . . xpsn 2 n xi1 . . . xi2k [xi1 , xi2 ] . . . [xi2k−1 , xi2k ] ∈ Rk. (25) It is clear that, using (23) and (25), we can write f = f1 + f2, where f1 = xps 1   X 2≤i1<...<i2k γ(i1,...,i2k) xps2 . . . xpsn 2 n xi1 . . . xi2k [xi1 , xi2 ] . . . [xi2k−1 , xi2k ]  is of the form (22) and f2 ∈ Rk. Then we have hf iT S + Rk = hf1iT S + Rk. Thus, we can assume (replacing f with f1) that the polynomial f is of the form (22), as claimed. 17 If f = 0 then L = Rk. Suppose that f 6= 0. Then we can assume without loss of generality that α(2,3,...,2k+1) 6= 0. It follows that the T -subspace hf iT S contains the polynomial h(x1, . . . , x2k+1) = α−1 = xps . . . xps2k+1 −1 1 xps2 −1 2k+1 2 (2,3,...,2k+1)f (x1, . . . , x2k+1, 1, 1, . . . , 1) [x2, x3] . . . [x2k, x2k+1]. Then hf iT S + Rk also contains the homogeneous component of the polynomial h(x1 + 1, . . . , x2k+1 + 1) of degree ps in each variable xi (i = 1, 2, . . . , 2k + 1), that is equal, modulo T (3), to 1 xps −1 2k+1[x2, x3] . . . [x2k, x2k+1], 2 where γ = Q2k+1 . . . xps −1 γ xps i=2 (cid:0)psi −1 ps−1(cid:1) ≡ 1 (mod p). It follows that xps 1 q(s) k (x2, . . . , x2k+1) ∈ hf iT S + Rk. On the other hand, for all i1, . . . , i2k such that 2 ≤ i1 < . . . < i2k ≤ n, we have xps 1 xps2 . . . xpsn 2 n xi1 . . . xi2k [xi1 , xi2 ] . . . [xi2k−1 , xi2k ] ∈ hxps 1 q(s) k (x2, . . . , x2k+1)iT S + T (3,k+1) (recall that si ≥ s for all i) so f ∈ hxps 1 q(s) k (x2, . . . , x2k+1)iT S + Rk. Thus, hf iT S + Rk = hxps 1 q(s) k (x2, . . . , x2k+1)iT S + Rk, where s ≥ 1. The proof of Lemma 15 is completed. Proposition 16. Let W be a T -subspace of F hXi such that Rk $ W. Then one of the following holds: 1. W = F hXi. 2. W is generated as a T -subspace by the polynomials xp 1, xp 1q1(x2, x3), . . . , xp 1qλ−1(x2, . . . , x2λ−1), x1[x2, x3, x4], x1[x2, x3] . . . [x2λ, x2λ+1] for some λ ≤ k. 3. W is generated as a T -subspace by the polynomials xp 1, xp 1q1(x2, x3), . . . , xp 1qk−1(x2, . . . , x2k−1), {q(l) k (x1, . . . , x2k) 1 ≤ l ≤ µ − 1}, xpµ 1 q(µ) k (x2, . . . , x2k+1), x1[x2, x3, x4], x1[x2, x3] . . . [x2k+2, x2k+3] for some µ ≥ 1. 18 Proof. Let f = f (x1, . . . , xn) be an arbitrary polynomial in W \ Rk satisfying the conditions of Lemma 15, that is, an arbitrary multihomogeneous polynomial such that degxif = psi for some si ≥ 0 (1 ≤ i ≤ n). Let Lf = hf iT S + Rk. By Lemma 15, we have either Lf = F hXi or Lf = hx1[x2, x3] . . . [x2θ, x2θ+1]iT S + Rk for some θ ≤ k or for some s ≥ 1. Lf = hxps 1 q(s) k (x2, . . . , x2k+1)iT S + Rk Note that W is generated as a T -subspace in F hXi by Rk together with the polynomials f ∈ W \ Rk satisfying the conditions of Lemma 15. It follows that W = P Lf , where the sum is taken over all the polynomials f ∈ W \ Rk satisfying these conditions. It is clear that if Lf = F hXi for some f ∈ W \ Rk then W = F hXi. Suppose that Lf 6= F hXi for all f ∈ W \ Rk. Let, for some f ∈ W \ Rk, we have Lf = hx1[x2, x3] . . . [x2θ, x2θ+1]iT S + Rk, θ ≤ k. Define λ = min {θ x1[x2, x3] . . . [x2θ, x2θ+1] ∈ W }; note that λ ≤ k. We have x1[x2, x3] . . . [x2θ, x2θ+1] ∈ hx1[x2, x3] . . . [x2λ, x2λ+1]iT S for all θ ≥ λ and xps 1 q(s) k (x2, . . . , x2k+1) ∈ hx1[x2, x3] . . . [x2λ, x2λ+1]iT S + T (3) for all s. Hence, W = hx1[x2, x3] . . . [x2λ, x2λ+1]iT S + Rk, where λ ≤ k. follows that W is generated as a T -subspace by the polynomials It xp 1, xp 1q1(x2, x3), . . . , xp 1qλ−1(x2, . . . , x2λ−1), x1[x2, x3, x4], x1[x2, x3] . . . [x2λ, x2λ+1], λ ≤ k. Now suppose that, for all f ∈ W \ Rk satisfying the conditions of Lemma 15, we have Lf = hxps 1 q(s) k (x2, . . . , x2k+1)iT S + Rk for some s = sf ≥ 1. Note that if s ≤ r then xpr 1 q(r) k (x2, . . . , x2k+1) ∈ hxps 1 q(s) k (x2, . . . , x2k+1)iT S + T (3). Take µ = min {s xps hxpµ generated as a T -subspace in F hXi by the polynomials k (x2, . . . , x2k+1) ∈ W }. Then we have W = Rk + k (x2, . . . , x2k+1)iT S and it is straightforward to check that W can be 1 q(µ) 1 q(s) xp 1, xp 1q1(x2, x3), . . . , xp 1qk−1(x2, . . . , x2k−1) 19 and the polynomials {q(l) together with the polynomials k (x1, . . . , x2k) 1 ≤ l ≤ µ − 1}, xpµ 1 q(µ) k (x2, . . . , x2k+1) x1[x2, x3, x4] and x1[x2, x3] . . . [x2k+2, x2k+3]. This completes the proof of Proposition 16. Proposition 16 immediately implies the following corollary. Corollary 17. Let W be a T -subspace of F hXi such that Rk $ W (k ≥ 1). Then W is a finitely generated T -subspace in F hXi. Proposition 18. If k 6= l then Rk 6= Rl. Proof. Suppose, in order to get a contradiction, that Rk = Rl for some k, l, k < l. Then we have C(G) ⊆ Rl. Indeed, by Theorem 5, the T -subspace C(G) is generated by the polynomial 1qn(x2, . . . , x2n+1), . . . . 1q1(x2, x3), . . . , xp 1, xp x1[x2, x3, x4] and the polynomials xp Clearly, x1[x2, x3, x4] ∈ T (3) ⊂ Rl. Further, xp 1, xp 1q1(x2, x3), . . . , xp 1ql−1(x2, . . . , x2l−1) ∈ Rl by the definition of Rl and xp 1qk+1(x2, . . . , x2k+3), xp 1qk+2(x2, . . . , x2k+5), . . . ∈ T (3,k+1) ⊆ Rk = Rl by the definition of T (3,k+1). Since k < l, we have xp 1, xp 1q1(x2, x3), . . . , xp 1qk(x2, . . . , x2k+1), xp 1qk+1(x2, . . . , x2k+3), . . . ∈ Rl. Hence, all the generators of the T -subspace C(G) belong to Rl so C(G) ⊆ Rl, as claimed. Note that T (3,k+1) ⊆ Rl and T (3,k+1) 6⊆ C(G) so C(G) $ Rl. By Theorem 1, C(G) is a limit T -subspace so each T -subspace W such that C(G) $ W is finitely generated. In particular, Rl is a finitely generated T -subspace. On the other hand, by Proposition 14, the T -subspace Rl is not finitely generated. This contradiction proves that Rk 6= Rl if k 6= l, as required. Theorem 4 follows immediately from Proposition 14, Corollary 17 and Propo- sition 18. 6 Acknowledgement This work was partially supported by DPP/UnB and by CNPq-FAPDF PRONEX grant 2009/00091-0 (193.000.580/2009). The work of the second and the third authors was partially supported by CNPq; the work of the third author was also partially supported by FEMAT. Thanks are due to the referee whose remarks and suggestions improved the paper. 20 References [1] E.V. Aladova, A.N. Krasilnikov, Polynomial Trans. Amer. Math. Soc. 361 (2009) 5629–5646. identities in nil-algebras, [2] Yu.A. Bahturin, Identical relations in Lie algebras. Transl. from Russian. VNU Science Press, b.v., Utrecht, 1987. [3] C. Bekh-Ochir, S.A. Rankin, The central polynomials of the finite di- mensional unitary and nonunitary Grassmann algebras, Asian-European J. Math. 3 (2010) 235–249. [4] C. Bekh-Ochir, S.A. Rankin, The central polynomials of the infinite di- mensional unitary and nonunitary Grassmann algebras, J. Algebra Appl. 9 (2010) 687–704. [5] A.Ya. Belov, On non-Specht varieties, Fundam. Prikl. Mat. 5 (1999) 47–66. [6] A.Ya. Belov, Counterexamples to the Specht problem, Sb. Math. 191 (2000) 329–340. [7] A.Ya. Belov, The local finite basis property and the local representability of varieties of associative rings, Izv. Math. 74 (2010) 1–126. [8] A. Brandao Jr., P. Koshlukov, A. Krasilnikov, E.A. Silva, The central poly- nomials for the Grassmann algebra, Israel J. Math. 179 (2010) 127-144. [9] V. Drensky, Free algebras and PI-algebras, Graduate course in algebra, Springer, Singapore, 1999. [10] V. Drensky, E. Formanek, Polynomial identity rings, Advanced Courses in Mathematics. CRM Barcelona. Birkhauser Verlag, Basel, 2004. [11] A. Giambruno, P. Koshlukov, On the identities of the Grassmann algebras in characteristic p > 0, Israel J. Math. 122 (2001) 305–316. [12] A. Giambruno, M. Zaicev, Polynomial identities and asymptotic methods, Mathematical Surveys and Monographs, 122, American Mathematical So- ciety, Providence, RI, 2005. [13] A.V. Grishin, Examples of T -spaces and T -ideals of characteristic 2 without the finite basis property, Fundam. Prikl. Mat. 5 (1999) 101–118. [14] A.V. Grishin, On non-Spechtianness of the variety of associative rings that satisfy the identity x32 = 0, Electron. Res. Announc. Amer. Math. Soc. 6 (2000) 50–51 (electronic). [15] A.V. Grishin, On the structure of the centre of a relatively free Grassmann algebra, Russ. Math. Surv. 65 (2010) 781–782. [16] A.V. Grishin, L.M. Tsybulya, On the multiplicative and T-space structure of the relatively free Grassmann algebra, Sb. Math. 200 (2009) 1299–1338. 21 [17] A.V. Grishin, V.V. Shchigolev, T -spaces and their applications, J. Math. Sci. (N. Y.) 134 (2006) 1799–1878. [18] A. Kanel-Belov, L.H. Rowen, Computational aspects of polynomial identi- ties, A K Peters, Ltd., Wellesley, MA, 2005. [19] A.R. Kemer, Finite basability of identities of associative algebras, Algebra and Logic 26 (1987) 362–397. [20] A.R. Kemer, Ideal of identities of associative algebras, Translations of Mathematical Monographs, 87, American Mathematical Society, Provi- dence, RI, 1991. [21] P. Koshlukov, A. Krasilnikov, E.A. Silva, The central polynomials for the finite dimensional Grassmann algebras, Algebra and Discrete Mathematics 2009 no.3 69–76. [22] D. Krakowski, A. Regev, The polynomial identities of the Grassmann al- gebra, Trans. Amer. Math. Soc. 181 (1973) 429–438. [23] V.N. Latyshev, On the choice of basis in a T-ideal, Sibirsk. Mat. Zh. 4 (1963) 1122–1126. [24] A. Regev, Grassmann algebras over finite fields, Commun. Algebra 19 (1991) 1829–1849. [25] L.H. Rowen, Polynomial identities in ring theory, Academic Press, 1980. [26] V.V. Shchigolev, Examples of infinitely based T -ideals, Fundam. Prikl. Mat. 5 (1999) 307–312. [27] V.V. Shchigolev, Examples of infinitely basable T -spaces, Sb. Math. 191 (2000) 459–476. [28] V.V. Shchigolev, Finite basis property of T -spaces over fields of character- istic zero, Izv. Math. 65 (2001) 1041–1071. 22
1105.0033	1	1105	2011-04-30T00:15:00	Hopf algebras of GK-dimension two with vanishing Ext-group	[ "math.RA" ]	We construct and study a family of finitely generated Hopf algebra domains $H$ of Gelfand-Kirillov dimension two such that $\Ext^1_H(k,k)=0$. Consequently, we answer a question of Goodearl and the second-named author.	math.RA	math	HOPF ALGEBRAS OF GK-DIMENSION TWO WITH VANISHING EXT-GROUP D.-G. WANG, J.J. ZHANG AND G. ZHUANG Abstract. We construct and study a family of finitely generated Hopf alge- bra domains H of Gelfand-Kirillov dimension two such that Ext1 H (k, k) = 0. Consequently, we answer a question of Goodearl and the second-named author. 0. Introduction Analysis of Hopf algebras of low Gelfand-Kirillov dimension (or GK-dimension for short) is an important step in understanding basic properties and algebraic structures of general Hopf algebras. Noetherian prime regular Hopf algebras of GK-dimension one were studied in [BZ, Li, LWZ]. The study of noetherian Hopf algebra domains (or Hopf domains, for short) of GK-dimension two was started by Goodearl and the second-named author [GZ] in which a classification was obtained under the extra hypothesis (♮) [Theorem 1.4]. Let k be an algebraically closed field of characteristic zero. Recall from [GZ] that (♮) means the following non-vanishing condition of the first Ext-group (♮): Ext1 H (H k,H k) 6= 0, where H k denotes the trivial left H-module. A well-known example of Hopf domains of GK-dimension two is the quantized enveloping algebra of the positive Borel subalgebra of sl2(k), which is isomorphic to A(1, q) := khx±1, yi/(xy−qyx) where q is a nonzero scalar and ∆(x) = x⊗x, ∆(y) = y⊗ 1 + x⊗ y (defined in Example 1.1). One can check the condition (♮) by verifying Ext1 It is natural to ask if every Hopf domain . A(1,q)(k, k) = (k if k ⊕ k if q 6= 1 q = 1 of GK-dimension two satisfies (♮), see [GZ, Question 0.3]. Our first goal is to construct finitely generated noetherian Hopf domains H of GK-dimension two such that Ext1 H (k, k) = 0, or equivalently, that the condition (♮) fails. As a consequence, [GZ, Question 0.3] is answered negatively. We investigate a family of Hopf algebras of GK-dimension two with vanish- ing Ext1 H (k, k), denoted by K({pi},{qi},{αi}, M ) (see Section 2). A subfamily of which, denoted by B(n,{pi}s 1), is a modification of B(n, p0, p1,··· , ps, q) introduced in [GZ]. We conjecture that these B(n,{pi}s 1) are the only pointed Hopf domains of GK-dimension two that are missing from the list given in [GZ, Theorem 0.1]. The second goal of the paper is to prove the following the- orem which provides an evidence to the conjecture. We say that H satisfies the hypothesis Ω′ if 1, q,{αi}s 1, q,{αi}s Ω′ : H does not contain A(1, q) as Hopf subalgebra for any q being either a primitive 5th or a primitive 7th root of unity. 2000 Mathematics Subject Classification. Primary 16P90, 16W30; Secondary 16A24, 16A55. Key words and phrases. Hopf algebra, Gelfand-Kirillov dimension, pointed, noetherian. 1 2 D.-G. WANG, J.J. ZHANG AND G. ZHUANG Theorem 0.1. Let H be a Hopf domain of GK-dimension two such that it is finitely generated by grouplike and skew primitive elements as an algebra and that Ext1 1, q,{αi}s 1) with αi 6= αj for some distinct integers i and j. H (k, k) = 0. If H satisfies Ω′, then H is isomorphic to B(n,{pi}s There are seven families of noetherian Hopf domains of GK-dimension two which satisfy Ext1 H (k, k) 6= 0 [Theorem 1.4]. As an immediate consequence, we have Corollary 0.2. Let H be a Hopf domain of GK-dimension two such that it is finitely generated by grouplike and skew primitive elements. If H satisfies Ω′, then H is isomorphic to either the algebra in Theorem 0.1 or one of the algebras in the seven families listed in Theorem 1.4. 1, q,{αi}s It is unknown whether all finitely generated Hopf domains of GK-dimension two are generated by grouplike and skew primitive elements. If it is affirmative, Corollary 0.2 provides a classification of finitely generated Hopf domains of GK- dimension two. Some basic properties of B(n,{pi}s 1) are listed in the next theorem. Theorem 0.3. Let H be the algebra B(n,{pi}s (a) H is finitely generated over its affine center. (b) injdim H = 2. (c) gldim H < ∞ if and only if gldim H = 2 if and only if s = 2 and α1 6= α2. Although many statements hold over arbitrary base field k, we assume that k is algebraically closed of characteristic zero for simplicity. All vector spaces, algebras, tensor products and linear maps are taken over k. Usually H denotes a Hopf algebra over k. Our basic reference for Hopf algebras is the book [Mo] and we denote counit, coproduct, and antipode by the symbols ǫ, ∆, and S, respectively. 1, q,{αi}s 1). Then Acknowledgments. The authors thank Nicol´as Andruskiewitsch for Remark 4.3 and thank Ken Goodearl, Martin Lorenz and Don Passman for Proposition 1.5 and their proofs. J.J. Zhang thanks Ken Brown and Ken Goodearl for many valuable conversations on the subject during the last few years. A part of research was done when J.J. Zhang was visiting Fudan University in Fall quarter of 2009, Spring quarter of 2010 and Fall quarter of 2010. J.J. Zhang and G. Zhuang were supported by the US National Science Foundation. 1. Review and some classification results This section is divided into two parts. The first part is a review of the work on a classification of Hopf domains of GK-dimension two under the condition (♮). The second part concerns a classification of pointed Hopf domains of GK-dimension two with GKdim C0 6= 1 where {Ci} denotes the coradical filtration of H. 1.1. Goodearl-Zhang's work. We collect some examples of Hopf algebras of GK- dimension two and state the main result of [GZ]. Everything in this subsection is from [GZ]. A nonzero element y ∈ H is skew primitive, or more precisely, (1, g)- primitive, if ∆(y) = y ⊗ 1 + g ⊗ y where g is a grouplike element in H. Such a g (uniquely determined by y) is called the weight of y and denoted by µ(y). Note that (g − 1) is always a skew primitive element of weight g. A skew primitive is called trivial if it is of the form c(g − 1) HOPF ALGEBRAS OF GK-DIMENSION TWO 3 for c ∈ k× := k \{0} and for a grouplike element g. In most cases, a skew primitive element is meant to be nontrivial. Example 1.1. Let n ∈ Z and q ∈ k×, and set A = khx±1, y xy = qyxi. There is a unique Hopf algebra structure on A under which x is grouplike and y is skew primitive, with ∆(y) = y ⊗ 1 + xn ⊗ y. This Hopf algebra is denoted by A(n, q). By [GZ, Construction 1.1], if m ∈ Z and r ∈ k×, then A(m, r) ∼= A(n, q) if and only if either (m, r) = (n, q) or (m, r) = (−n, q−1). Example 1.2. Let n, p0, p1, . . . , ps be positive integers and q ∈ k× with the fol- lowing properties: (a) s ≥ 2 and 1 < p1 < p2 < ··· < ps; (b) p0 n and p0, p1, . . . , ps are pairwise relatively prime; (c) q is a primitive ℓ-th root of unity, where ℓ = (n/p0)p1p2 ··· ps. Set m = p1p2 ··· ps and mi = m/pi for i = 1, . . . , s. Choose an indeterminate y, and consider the subalgebra A = k[y1, . . . , ys] ⊂ k[y] where yi = ymi for i = 1, . . . , s. The k-algebra automorphism of k[y] sending y 7→ qy restricts to an automorphism σ of A. There is a unique Hopf algebra structure on the skew Laurent polynomial ring B = A[x±1; σ] such that x is grouplike and the yi are skew primitive, with ∆(yi) = yi ⊗ 1 + xmin ⊗ yi for i = 1, . . . , s. The Hopf algebra B has GK-dimension two. We shall denote it B(n, p0, . . . , ps, q) [GZ, Construction 1.2]. Example 1.3. Let n be a positive integer and set C = k[y±1](cid:2)x; (yn − y) d dy(cid:3). There is a unique Hopf algebra structure on C such that ∆(y) = y ⊗ y and ∆(x) = x⊗ yn−1 + 1 ⊗ x. This Hopf algebra is denoted by C(n). For m, n ∈ Z>0, the Hopf algebras C(m) and C(n) are isomorphic if and only if m = n [GZ, Construction 1.4]. Here is the main result of [GZ]. An algebra is called affine if it is finitely generated as an algebra over k. The condition (♮) is defined in the introduction. Theorem 1.4. [GZ, Theorem 0.1] Let H be a Hopf domain of GK-dimension two satisfying (♮). Then H is noetherian if and only if H is affine, if and only if H is isomorphic to one of the following: (I) The group algebra kΓ, where Γ is either (Ia) the free abelian group Z2, or (Ib) the nontrivial semidirect product Z ⋊ Z. (II) The enveloping algebra U (g), where g is either (IIa) the 2-dimensional abelian Lie algebra over k, or (IIb) the Lie algebra over k with basis {x, y} and [x, y] = y. (III) The Hopf algebras A(n, q) from Example 1.1, for n ≥ 0. (IV) The Hopf algebras B(n, p0, . . . , ps, q) from Example 1.2. (V) The Hopf algebras C(n) from Construction from Example 1.3, for n ≥ 2. Aside from the cases A(0, q) ∼= A(0, q−1), the Hopf algebras listed above are pairwise non-isomorphic. It would be convenient if every noetherian Hopf algebra domain of GK-dimension two satisfied (♮), but the algebras defined in Section 2 are counterexamples. 1.2. Partial results on pointed Hopf domains of GK-dimension two. In this subsection we start a classification of pointed Hopf domains H of GK-dimension 4 D.-G. WANG, J.J. ZHANG AND G. ZHUANG strictly less than three. Note that we do not assume that H satisfies the condition (♮) in this subsection. Since H is pointed, the coradical C0 of H is a group algebra kG where G consists of all grouplike elements in H. Since kG is a subalgebra of H, GKdim kG < 3. Then GKdim kG is either 0, or 1, or 2. We consider these three subcases. 1.2.1. GKdim C0 = 2. The following proposition was proposed by Goodearl and the second-named author and proved by Goodearl, Passman and Lorenz. We thank them for sharing their proofs with us. Recall that the centeralizer of an element g in a group Γ is defined to be CΓ(g) = {h ∈ Γ hg = gh}. The centralizer of a subset in Γ is defined similarly. The finite conjugate center of a group Γ is defined to be [Pa, p. 115] ∆(Γ) = {x ∈ Γ [Γ : CΓ(x)] < ∞}. Please do not confuse ∆(Γ) with the coproduct ∆ and with conventions ∆(B(V )) and ∆+(B(V )) introduced and used locally in Section 4. Proposition 1.5 (Goodearl-Zhang). If the group algebra kΓ is an affine domain of GK-dimension two, then Γ is either Z2 or the nontrivial semidirect product Z ⋊ Z = hx, y xy = y−1xi. Namely, kΓ is in Theorem 1.4(I). The following nice proof is due to Lorenz. Proof of Proposition 1.5 (Lorenz). First of all, since kΓ is an affine domain, Γ is finitely generated and torsionfree. Since GKdim(kΓ) = 2, Γ is abelian-by-finite with Hirsch number 2. Let A = ∆(Γ) be the finite conjugate center of Γ. By [Pa, Lemma 1.6, p. 117] A is abelian, and hence A ∼= Z2. In this case A is also the largest abelian subgroup of Γ. Since A is abelian and Γ is abelian-by-finite, we have A = CΓ(A) and G := Γ/A is finite. If G = {1} then Γ = Z2. So it remains to consider that case that G 6= {1}. Let f be the homomorphism defined by G ֒→ GL(A) −→ {±1} g 7−→ gA 7−→ det gA where gA ∈ GL(A) is given by the conjugation of g on A. We claim that f is an isomorphism, or equivalently, det gA = −1 for all 1 6= g ∈ G. Write g = γA with γ 6∈ A. Since G is finite, there is an n such that gn = 1, or γn ∈ A. But γn 6= 1 since Γ is torsionfree. So gA has a nontrivial fixed point γn ∈ A. This implies that 0 det gA(cid:19), and hence we must have det gA = −1. gA has Jordon canonical form (cid:18)1 This proves the claim. Pick any x ∈ Γ whose image is g that generates G. Note that x2 ∈ Z(Γ), the center of Γ. Since xA(= gA) has eigenvalues 1 and −1, Z(Γ) has rank 1, or equivalently, is infinite cyclic. Hence the subgroup hx, Z(Γ)i of Γ is infinite cyclic too. Without loss of generality we assume that hx, Z(Γ)i is generated by x. Consequently, Z(Γ) is generated by x2. Furthermore, A/Z(Γ) is infinite cyclic as well, with a generator y. Since xA has the eigenvalue −1 in y, we have 0 xyx−1 = y−1x2r HOPF ALGEBRAS OF GK-DIMENSION TWO 5 for some r. Replacing y by yx2s for a suitable s, we may assume that r = 0 or r = −1. If r = −1, then (xy)2 = 1, which contradicts to the fact Γ is torsionfree. Thus we must have r = 0 and xyx−1 = y−1. Since Γ is generated by x and y, Γ = hx, y xy = y−1x} = Z ⋊ Z. (cid:3) Lemma 1.6. Let H be a pointed Hopf domain. If GKdim H < GKdim C0 + 1, then H = C0. Proof. Suppose on contrary that H 6= C0. Then there is a nonzero skew primitive element y ∈ C1 \ C0 such that ∆(y) = y ⊗ 1 + g ⊗ y and g−1yg = λy + τ (g − 1) for some λ, τ ∈ k [WZZ1, Lemma 2.5]. By [WZZ1, Theorem 0.2], the hypothesis that GKdim H < GKdim C0 + 1 implies that λ is a pth primitive root of unity for some p ≥ 2. Since λ 6= 1, we may assume g−1yg − λy = 0 by [WZZ1, Lemma 2.5]. Since the Hopf subalgebra K generated by g±1 and y is a noncommutative domain, GKdim K ≥ 2 by [GZ, Lemma 4.5], whence GKdim K = 2 and K is isomorphic to A(1, λ) defined in Example 1.1. As a consequence, yp is a nontrivial skew primitive element. Since g−pypgp = yp, [WZZ1, Theorem 0.2] implies that GKdim H ≥ GKdim C0 + 1, a contradiction. (cid:3) Theorem 1.7. Let H be an affine pointed Hopf domain of GK-dimension strictly less than three. If the coradical C0 is of GK-dimension two, then H is isomorphic to kΓ where Γ is either Z2 or Z ⋊ Z as given in Theorem 1.4(I). Proof. The assertion follows from Lemma 1.6 and Proposition 1.5. (cid:3) 1.2.2. GKdim C0 = 0. In this case C0 = k and H is a connected Hopf algebra. Part (a) of the following lemma is due to Le Bruyn (unpublished). Lemma 1.8. Let H be a connected Hopf algebra and K be the associated graded Hopf algebra grC H with respect to the coradical of H. GKdim H. (a) H is a domain. (b) K is a connected graded Hopf algebra that is a domain with GKdim K ≤ (c) Let m be the graded maximal ideal of K. Then grm K is a universal en- veloping algebra U (g) where g is a graded Lie algebra generated in degree one with dimension no more than GKdim K. Proof. (c) This is a consequence of [GZ, Proposition 3.4(a)]. (b) Since U (g) is a domain (for any g), so is grm K. By definition, K is a connected N-graded Hopf algebra. Then Ti mK = 0, and so the filtration {mi K}i≥0 is exhaustive and separated. Therefore K is a domain. By [KL, Lemma 6.5], GKdim K ≤ GKdim H. (a) This is a consequence of (b). (cid:3) Theorem 1.9. Let H be a connected Hopf domain of GK-dimension strictly less than three. Then H is isomorphic to U (h) for a Lie algebra h of dimension no more than 2. If GKdim H ≥ 2, then H is isomorphic to Hopf algebras in Theorem 1.4(II). Proof. To avoid triviality we assume that GKdim H ≥ 2. By [Zh, Theorem 1.1], H contains a Hopf subalgebra U (h) for some 2-dimensional Lie algebra h. Retain the notation from Lemma 1.8, we have grm K = U (g) and dim g ≥ dim g1 ≥ dim K1 = dim C1/C0 ≥ dim h = 2. 6 D.-G. WANG, J.J. ZHANG AND G. ZHUANG By Lemma 1.8(b,c), dim g < 3, whence dim g = dim g1 = dim K1 = 2. Since g is graded and generated in degree 1, it must be a 2-dimensional abelian Lie algebra. Therefore grm(K) = U (g) = U (g1) is commutative and generated by g1. Since K is connected graded and since dim K1 = 2, K1 = g1 = m/m2. This implies that K ∼= grm(K) = U (g). As a consequence, H is generated by primitive elements. The assertion follows from [Mo, Theorem 5.6.5]. (cid:3) 1.2.3. GKdim C0 = 1. The most difficult case is when GKdim C0 = 1. Since we assume H is a domain, so is C0. Thus C0 is commutative [GZ, Lemma 4.5]. By [GZ, Proposition 2.1], C0 = kG where G is abelian of rank one. Lemma 1.10. Let H be a pointed Hopf algebra that is finitely generated as algebra over k. Then C0 is finitely generated. Proof. Suppose H is generated by a finite dimensional subcoalgebra V . Then V is a pointed coalgebra. Let F be the free (pointed) Hopf algebra generated by V , which is defined in [Ta2]. By the universal property of F , there is a Hopf algebra sujective map F → H. It follows from [Ta2, Theorem 35] that the coradical F0 of F is generated by the coradical of V . Hence F0 is finitely generated. By [Mo, Corollary 5.3.5], G(H) is a quotient of G(F ). Therefore G(H) is finitely generated and the assertion follows. (cid:3) See Theorem 6.2 for a result in this subcase. To conclude this section we state a result of [WZZ2]. An algebra A is called PI if it satisfies a polynomial identity and A is called locally PI if every affine subalgebra of A is PI. Theorem 1.11. [WZZ2, Theorem 7.2] Let H be an affine pointed Hopf domain such that GKdim H < 3 and that C0 = kZ. If H is not PI, then H is isomorphic to one of following (a) The Hopf algebra A(n, q) of Example 1.1 where n > 0 and q is not a root of unity. (b) The Hopf algebra C(n) of Example 1.3 for n ≥ 2. Note that the proof of [WZZ2, Theorem 7.2] does not use anything in this paper. Combining [WZZ2, Theorem 7.2] with [GZ, Lemma 4.5] and Theorems 1.7 and 1.9 we have the following Corollary. Corollary 1.12. [WZZ2, Theorem 0.1] Let H be an affine pointed Hopf domain of GKdimension strictly less than three. If H is not PI, then H is isomorphic to one of following (a) The enveloping algebra U (g) of 2-dimensional non-abelian solvable Lie al- gebra g as in Theorem 1.4(IIb). (b) The Hopf algebra A(n, q) of Example 1.1 where n > 0 and q is not a root of unity. (c) The Hopf algebra C(n) of Example 1.3 for n ≥ 2. As a consequence of the above Corollary, there is no affine pointed Hopf domain of GK-dimension strictly between 2 and 3 [WZZ2, Corollary 7.3]. HOPF ALGEBRAS OF GK-DIMENSION TWO 7 2. Definition and elementary properties By the last section only un-classified (and more interesting) affine pointed Hopf H (k, k) = 0, domains of GKdimension two are PI and satisfy GKdim C0 = 1 and Ext1 which will occupy our attention for the rest of the paper. for all i < j; j = q−nj i i are primitive pi-th roots of unity; In this section we construct and study our main object -- a class of Hopf domains H (k, k) = 0. We first introduce a more general class, denoted by K, with Ext1 dependent on a set of parameters with various conditions listed below. Suppose (I2.0.1) s ≥ 2 and M ≥ 2 are two integers; (I2.0.2) n1,··· , ns, p1,··· , ps are positive integers such that M = nipi for any i; (I2.0.3) q1,··· , qs are nonzero scalars in k; (I2.0.4) for each i, both qi and qni (I2.0.5) qni (I2.0.6) α1,··· , αs are scalars in k. There are two more conditions to consider. We will see soon in Lemma 2.3(b,c) that the Hopf algebra K is a domain if and only if (I2.0.7) gcd(pi, pj) = 1 for all i 6= j, and that K satisfies the vanishing condition Ext1 (I2.0.8) αi 6= αj for some i 6= j. In the rest of this section we fix a parameter set i=1,{qi}s {s, M,{ni}s i=1,{pi}s K(k, k) = 0 if and only if i=1,{αi}s i=1} satisfying (I2.0.1-I2.0.6). Let K be the algebra generated by x±1, y1,··· , ys subject to the following relations (I2.0.9) xx−1 = x−1x = 1, (I2.0.10) yix = qixyi for all i, (I2.0.11) yjyi = qni (I2.0.12) ypj j = ypi It is easy to see that the parameters {αi}s i=1 can be replaced by {0, α2−α1,··· , αs− α1} without changing the algebra. In other words, we may assume that α1 = 0. If pj = 1 for some j, then relation (I2.0.12) says that yj is generated by yi and x, so we can remove yj from the generating set without changing the algebra K. By choosing a minimal generating set we may assume that j yiyj for all i < j, i + (αj − αi)(xM − 1) for all i < j. (I2.0.13) pi ≥ 2 for all i. Lemma 2.1. The algebra K has a k-linear basis of monomials {xw0 yw1 1 yw2 2 ··· yws s } where w0 ∈ Z, w1 ∈ N and 0 ≤ wi ≤ pi − 1 for all 2 ≤ i ≤ s. Proof. We use Bergman's Diamond Lemma [Be, Theorem 1.2]. Define a linear order on the set of generators as follows x−1 < x < y1 < ··· < ys. By using relations in (I2.0.9)-(I2.0.12) it is easy to see that the algebra K is gener- ated by x±1, y1,··· , ys subject to the following relations, with leading monomials in the left-hand side of the equations, (I2.1.1) xx−1 = 1, (I2.1.2) x−1x = 1, 8 D.-G. WANG, J.J. ZHANG AND G. ZHUANG j = yp1 j = q−nj i x−1yi for all i, 1 + αj(xM − 1) for all j > 1 (where we assume α1 = 0). (I2.1.3) yix = qixyi for all i, (I2.1.4) yix−1 = q−1 (I2.1.5) yjyi = qij yiyj for all i < j where qij = qni (I2.1.6) ypj Using these relations, every element in K can be written as a linear combina- tion of monomials listed the assertion. Therefore the given set of monomials {xw0yw1 To prove these monomials form a basis, it suffices to show that all ambiguities generated by relations (I2.1.1-I2.1.6) can be resolved (see Diamond Lemma [Be, Theorem 1.2]). The rest of the proof amounts to verifying the required statement. The first ambiguity is created between (I2.1.1) and (I2.1.2), which can be resolved s } span the algebra K. ··· yws 1 yw2 2 , i as follows. (xx−1)x = 1x = x, and x(x−1x) = x1 = x. To save space we only resolve two more ambiguities. As noted before we may assume that α1 = 0. The ambiguity between (I2.1.3) and (I2.1.6) is obtained from the monomial ypj j x. It is easy to see that ypj−1 j (yjx) = qpj j x(ypj j ) = x(yp1 1 + αj(xM − 1)) and that (ypj j )x = (yp1 1 + αj (xM − 1))x = x(yp1 1 + αj(xM − 1)). So the ambiguity is resolved. The ambiguity between (I2.1.5) and (I2.1.6) can be resolved as below. For any ypj−1 j i < j, and 1 + αj(xM − 1)) (yjyi) = qpj = (q−pi = (yp1 j ) = yi(yp1 1 + αj(xM − 1))yi ij yi(ypj 1i yp1 1 + αj(xM − 1))yi 1 + αj(xM − 1))yi. So the ambiguity is resolved. (ypj j )yi = (yp1 It is routine to check that all other ambiguities can be resolved and therefore the (cid:3) assertion follows. The coalgebra structure of K is defined by the following rules (I2.1.7) ∆(x) = x ⊗ x, ǫ(x) = 1, (I2.1.8) ∆(x−1) = x−1 ⊗ x−1, ǫ(x−1) = 1, (I2.1.9) ∆(yi) = yi ⊗ 1 + xni ⊗ yi, ǫ(yi) = 0. Lemma 2.2. The algebra K is a Hopf algebra using the rules defined by (I2.1.7)- (I2.1.9) and the antipode is determined by the following rules S(x) = x−1, S(x−1) = x, S(yi) = −x−niyi = −qni i yix−ni for all i. Proof. It is easy to verify that rules (I2.1.7)-(I2.1.9) define algebra homomorphisms ∆ : K → K ⊗ K and ǫ : K → k since both of them maps relations of K to zero. Coassociativity and counit axioms hold since these axioms hold for the generators. This proves that K is a bialgebra. HOPF ALGEBRAS OF GK-DIMENSION TWO 9 Note that S extends to an algebra anti-automorphism of K. To check K is a Hopf algebra we only need to apply the antipode axiom to the generators, which can be verified directly. (cid:3) The Hopf algebra K is denoted by K({pi},{qi},{αi}, M ) if we need to indicate the parameters. Note that ni = M/pi for all i = 1,··· , s. Lemma 2.3. Let K be K({pi},{qi},{αi}, M ). (a) gcd(pi, ni) = 1 for all i. (b) If K is a domain (or, more generally, K has a quotient Hopf algebra domain i = 1 and gcd(pi, pj) = 1 for all j = qnj K′ of GK-dimension two), then qni i 6= j. As a consequence, (I2.0.7) holds. (c) K satisfies (♮) if and only if αi = αj for all i 6= j. j = ypi Proof. (a) Since qi and qni is a primitive pi-th root of unity, we have ypj j yiyj implies that yj(yi + γxni ) = qni i are both primitive pi-th root of unity, gcd(pi, ni) = 1. (b) First we assume K is a domain. Fix any i 6= j. Since k is algebraically closed, there is a γ such that αj − αi = γpi. We re-write the relation ypj j = i + (αj − αi)(xM − 1) as ypj ypi i xni yi and qni j = (yi + γxni)pi − γpi. The relation i yjyi = qni j (yi + γxni)yj. Thus the subalgebra Y generated by a := yi + γxni and b := yj has GK-dimension at most one. To see this, note that kha, bi/(ba − qni j ab) is a domain of GK-dimension two and that Y is a proper quotient of kha, bi/(ba − qni j ab). Since K is a domain, so is Y . By [GZ, Lemma 4.5], Y is commutative, and whence, qni = 1. Consequently, pj divides ni. By part (a), gcd(pi, pj) = gcd(pi, ni) = 1. i + γpi(xpini − 1). Since yixni = qni j = q−nj i If K has a quotient Hopf algebra K′ which is a domain of GK-dimension two, the proof can be modified so that ab = ba in K′ where a is the image of yi + γxni and b is the image of yj in K′. Further that yiyj = qni j ba in K′. If qni 6= 1, then either a or b is 0 in K′. In either cases, equation (I2.1.6) j implies that the image of yp1 is in coradical of K′, which is C0(K′) = k[x, x−1]. 1 Consequently, the image of ypj is also in C0(K′) for all j. Therefore GKdim K′ = 1, a contradiction. Thus qni j yiyj in K implies that ab = qni j = 1, which leads to the conclusion. j (c) If αi = αj for all i, j, then K/(y1,··· , ys) is isomorphic to k[x, x−1], which is an infinite dimensional commutative Hopf algebra. Hence (♮) holds following [GZ, Theorem 3.8(c)]. Suppose αi 6= αi for some i 6= j. As noted before we may assume pi ≥ 2 for all j to avoid triviality. Then qi is not 1 for each i. Relation (I2.0.10) implies that yi ∈ [K, K] for all i. Thus K/[K, K] is isomorphic to k[x, x−1]/(xM − 1) by relation (I2.0.12), which is finite dimensional. By [GZ, Theorem 3.8(c)], (♮) fails. (cid:3) Proposition 2.4. The following are equivalent. (a) K({pi},{qi},{αi}, M ) is a domain. (b) gcd(pi, pj) = 1 for all i 6= j. (Consequently, pj ni for all i 6= j). (c) There exists a nonzero scalar q such that qi = qmi for each i where mi = (p1 ··· ps)/pi, and in this case K({pi},{qi},{0}, M ) is isomorphic to a do- main B(n, n, p1,··· , ps, q) for n := M/(p1 ··· ps). (d) K({pi},{qi},{0}, M ) is a domain. Proof. (a) ⇒ (b) This is Lemma 2.3(b). 10 D.-G. WANG, J.J. ZHANG AND G. ZHUANG (b) ⇒ (c) We consider the algebra K := K({pi},{qi},{0}, M ) under the hypoth- esis that gcd(pi, pj) = 1 for all i 6= j. Since M = pini = pjnj and gcd(pj, pi) = 1, pj ni for all i 6= j. Then qni j = 1 and consequently, yi commutes with yj. Let A be the subalgebra of K generated by y1,··· , ys. Then we have relations yiyj = yjyi, i = ypj ypi j for all i 6= j. By Lemma 2.1, there is no other relations in Y . By the proof of [GZ, Construction 1.2], Y is isomorphic to a subalgebra of k[y, y−1] by identifying yi with ymi. Further Y S−1 is isomorphic to k[y, y−1] where S is the set of all monomials in y1,··· , ys. Using the relations (I2.0.9)-(I2.0.12) it is easy to see that S is an Ore set of K and the localization KS−1 equals k[y, y−1][x, x−1; σ] where σ is a graded algebra automorphism of k[y, y−1]. Let q be the scalar such that yx = qxy. Then yix = ymix = qmixymi = qmixyi for all i. By comparing the above equation with (I2.0.10), we obtain that qi = qmi. Recall that we assume αi = 0 for all i. In this case the algebra K({pi},{qi},{0}, M ) is exactly the algebra B(n, n, p1,··· , ps, q) in Example 1.2 (with p0 = n). The assertion follows and K is a domain. (c) ⇒ (d) Clear. (d) ⇒ (a) Define an N-filtration on K({pi},{qi},{αi}, M ) by setting deg(x) = 0, deg yi = mi. Then the associated graded ring is isomorphic to K({pi},{qi},{0}, M ). Since K({pi},{qi},{0}, M ) is a domain, so is K({pi},{qi},{αi}, M ). (cid:3) By Proposition 2.4, the Hopf algebra K({10, 15},{q3, q−2},{0, 1}, 30) is not a domain where q is a primitive 30th root of unity. The choice of {qs} is not unique even if all other parameters are fixed. For example, K({10, 15},{q3, q8},{0, 1}, 30) is also a Hopf algebra of the same kind. Convention 2.5. Suppose (I2.0.1)-(I2.0.7) hold for the parameter set used for the algebra K. Re-arranging {pi}s i=1 we may assume that 1 < p1 < p2 < ··· < ps. Let ℓ = p1 ··· ps, mi = ℓ/pi and n = M/ℓ. By Proposition 2.4 and Example 1.2, there is an ℓ-th root of unity q such that qi = qmi for every i. As a consequence, since M/(pipj) is an integer for i < j, qni j = (qmj )ni = q ℓM pj pi = (qℓ) M pj pi = 1 for all i < j. denoted by B(n,{pi}s is generated by x±1, y1,··· , ys and subject to the relations In this case, the algebra K({pi},{qi},{αi}, M ) is a domain and 1, q,{αi}s 1) 1). In other words, the algebra B(n,{pi}s 1, q,{αi}s xx−1 = x−1x = 1, yix = qmixyi for all i, yjyi = yiyj for all i < j, ypj j = ypi i + (αj − αi)(xM − 1) for all i < j, with comultiplication and counit determined by (I2.1.7)-(I2.1.9) and antipode de- termined by rules in Lemma 2.2. If αi = αj for all i, j, then B(n,{pi}s 1, q,{αi}s 1) is just the algebra B(n, n, p1,··· , ps, q). Note that n = M/(p1 ··· ps) and that we have removed p0 from the above B convention since p0 = n. We continue to work on the algebra K({pi},{qi},{αi}, M ) without assuming (I2.0.7) although our main interest is about B(n,{pi}s 1, q,{αi}s 1). HOPF ALGEBRAS OF GK-DIMENSION TWO 11 Lemma 2.6. The coalgebra structure of K({ps},{qs},{αs}, M ) is independent of {αi}. In particular, K({ps},{qs},{αs}, M ) is isomorphic to K({ps},{qs},{0}, M ) as coalgebras. Proof. We use the k-linear basis given in Lemma 2.1. Then the coproduct of K({ps},{qs},{αs}, M ) and the coproduct of K({ps},{qs},{0}, M ) coincide. Hence the assertion follows. (cid:3) By Lemma 2.3(c), K({ps},{qs},{αs}, M ) (when αi 6= αj) is not isomorphic to K({ps},{qs},{0}, M ) as algebras. Theorem 2.7. Let K := K({pi},{qi},{αi}, M ) be defined as above. (a) The algebra K is affine and noetherian. (b) K is pointed and the coradical of K is C0 = k[x, x−1] ∼= kZ. (c) K is finitely generated over its affine center. (d) GKdim K = 2. (e) injdim K = 2. (f) gldim K is finite if and only if gldim K = 2 if and only if s = 2 and α1 6= α2. Proof. (a) By definition, K is affine. It is noetherian since K is a factor ring of an iterated Ore extension k[x±1][y1, σ1]··· [ys, σs] where automorphisms σi can be read off from relations (I2.0.9)-(I2.0.11). (b) Using Lemma 2.1 it can be checked directly that C0(K) = k[x, x−1], so it is pointed. (c) Let Z be the center of K and let T be the subalgebra of Z generated by central elements yp1 1 , xM and x−M . By Lemma 2.1 K = X0≤w0<M,0≤wi<pi,∀i T xw0yw1 1 ··· yws s . Hence K is finitely generated over its center Z and Z is affine. (d) Since T is isomorphic to k[s, s−1][t] which has GK-dimension two and since K is finitely generated over T , GKdim K = 2. (e) This is a consequence of (a,d) and [WZ, Theorem 0.1]. (f) Suppose αi = αj for some i 6= j. Without loss of generality, we may assume that α1 = α2. Let K0 be the Hopf subalgebra generated by x±1, y1 and y2. Then it follows from Lemma 2.1 that K is a left and a right free K0-module. By [MR, Proposition 2.2(i)] or equivalently, by [LL], projdim kK0 ≤ projdim kK gldim K0 ≤ gldim K. Let Y be the subalgebra generated by y1 and y2. Then K0 = Y [x, x−1]. Since Y is a connected graded domain of GK-dimension one and it is not isomorphic to the polynomial ring k[y], gldim Y = ∞. By [MR, Theorem 7.5.3(ii)], gldim K0 = ∞. Consequently, gldim K = ∞. Next we consider the case when s ≥ 3 and αi 6= αj for all i 6= j. We will deal with the case s = 2 at the end. Let K1 be the Hopf subalgebra generated by x±1, y1, y2 and y3. It follows from Lemma 2.1 that K is a left and a right free K1-module. The argument in the previous paragraph shows that it suffices to show gldim K1 = ∞. In other words, we may assume s = 3. Let A be the subalgebra generated by y1, y2, y3 i − ypj and x±M . Note that relation (I2.0.12) implies that xM = 1 + (αi− αj)−1(ypi j ). 12 D.-G. WANG, J.J. ZHANG AND G. ZHUANG From this it is easy to check that A is isomorphic to BS−1 where B is a connected graded algebra generated by y1, y2, y3 subject to the relations (i) yiyj = qni (ii) (α1 − α2)−1(yp1 1 − yp2 j yjyi for all i, j, 2 ) = (α2 − α3)−1(yp2 1 − yp3 3 ). And S consists of all powers of the elements {1 + (αi − αj)−1(ypi j ) i 6= j}. Since B is connected graded, PI of GK-dimension two and since B is not isomorphic to a skew polynomial ring, gldim B = ∞ [StZ, Theorem 3.5] and flatdim kB = projdim kB = ∞ where kB is the module B/B≥0. Since kBS−1 = kA where kA is a 1-dimensional right A-module. By [MR, Proposition 7.4.2(iii)], flatdim kA = ∞, so 3 ) = (α1 − α3)−1(yp1 i − ypj 2 − yp3 gldim A ≥ projdim kA = flatdim kA = ∞. Finally note that K is a free A-module K = ⊕nm−1 projdim kK ≥ projdim kA = ∞. assume that α1 = 0 and α2 = 1. So we have a relation The remaining case is when s = 2 and α1 6= α2. After a scalar change we may i=0 xiA by Lemma 2.1. Thus 1 − yp2 yp1 2 = 1 − xM . 1 + yp2 [y1, y2]S−1 where S consists of all powers of the element Let A be the algebra kq {1 − yp1 2 }. Then A has global dimension 2. The algebra K is a direct sum ⊕M−1 i=0 xiA which is in fact a crossed product A ⋆ (Z/(M )). By [MR, Theorem 7.5.6(iii)], gldim K = gldim A = 2. (cid:3) n1 2 Theorem 0.3 is a consequence of Theorem 2.7. Part (f) of Theorem 2.7 suggests the following questions. Question 2.8. Suppose H is a noetherian affine Hopf algebra of GK-dimension n. (a) Is there a function f of n such that the global dimension of H is either infinite or bounded by f (n)? (b) Assume H is a domain (or a prime algebra) with finite global dimension. Is there a function f of n such that the minimal number of generators of H is bounded by f (n)? Lemma 2.9. Let K be as in Theorem 2.7. (a) If y is a (1, g)-primitive elements not in C0, then either g = xni or g = xM and y is a linear combination of {y1,··· , ys, yM 1 } modulo C0. Consequently, the set {n1,··· , ns, M} is an invariant of K. (b) Suppose that n1,··· , ns are distinct. Then every Hopf automorphism f of K is of the form φ : x → x, yi → ciyi, for all i, where ci ∈ k× satisfies i = cpj cpi (c) Suppose K is a domain. If K′ = K({p′i},{q′i},{α′i}, M′) is another algebra and f : K → K′ is a Hopf surjective map. Then f is an isomorphism. Up to a permutation of {1, 2,··· , s}, there is a scalar c ∈ k× such that p′i = pi, q′i = qi, α′i = cαi for all i. i = 1 for all i when αk 6= αl for some k, l. for all i, j and cpi j Proof. (a) We use induction on s. When s = 0, the statement is trivial. Suppose now s ≥ 1 and assume that the assertion holds for Ks−1 where Ks−1 is the Hopf subalgebra generated by x±1, y1,··· , ys−1. Let F be a skew (1, g)-primitive element in K but not in Ks−1 where g is a grouplike element. Write F =Pw s where i=0 fiyi HOPF ALGEBRAS OF GK-DIMENSION TWO 13 fi ∈ Ks−1 and w < ps and fw 6= 0. If w > 1, then ∆(F ) = ∆(fwyw w−1 fiyi s) s + Xi=0 s ⊗ 1 + = ∆(fw)(yw and since F is (1, g)-primitive, w−1 Xj=1 (cid:18)w j(cid:19)qs (xnsjyj s ⊗ yw−j s ) + xnsw ⊗ yw s ) + ∆( fiyi s), w−1 Xi=0 w w fiyi Xi=0 ∆(F ) = F ⊗ 1 + g ⊗ F = ( Xi=0 s) ⊗ 1 + g ⊗ ( s , we have(cid:0)w Since K is free over Ks−1 with basis 1, ys,··· , yps ∆(fw)(xns⊗1) = 0 1(cid:1)qs . Since(cid:0)w by comparing the coefficient of the term ys⊗yw−1 (xns⊗1) is invertible, 1(cid:1)qs ∆(fw) = 0. Consequently, fw = 0, yielding a contradiction. Therefore w = 1 and F = f0 + f1ys. Then ∆(F ) = F ⊗ 1 + g ⊗ F implies that fiyi s). s ∆(f1) = f1 ⊗ 1, ∆(f1)(xns ⊗ 1) = g ⊗ f1, ∆(f0) = f0 ⊗ 1 + g ⊗ f0. to cyp1 These equations imply that f1 ∈ k, g = xns and f0 is a (1, g)-primitive. The assertion follows by induction. (b) Let φ be any Hopf automorphism of K. Since {ni} are distinct, it follows from part (a) that every (1, xni)-primitive element is of the form cyi + d(xni − 1) for some c, d ∈ k. Hence φ sends yi to ciyi + di(xni − 1) for i = 1,··· , s and sends yp1 1 1 + d(xM − 1). The equation cyp1 1 + d(xM − 1) = φ(yp1 implies that d = d1 = 0 and c = cp1 i = c for all i. Applying φ to ∆(yi) we see that φ(xni ) = xni , which implies that φ is an identity on C0. If αk 6= αl for some k < l, then cp i = c = 1 by the equation (I2.1.6). The assertion follows. 1 ) = φ(y1)p1 = (c1y1 + d1(xn1 − 1))p1 1 . Similar, di = 0 and cpi (c) When K is a domain, condition (I2.0.7) implies that {ni} are distinct. If f is surjective, then f is also injective since K is a domain and GKdim K = GKdim K′ = 2. Hence f is an isomorphism. Similar to the proof of (b), one sees that f sends x to x, yi to ciyi up to a permutation. Thus {pi} = {p′i}, {qi} = {q′i}, M = M′, and cpi for all i, j. We may assume that α1 = α′1 = 0. Then α′j = cαj where c = cpj for all i, j. (cid:3) i = cpj j = cpi j i 3. Preliminary analysis on skew primitive elements It remains to prove Theorem 0.1 and Corollary 0.2 in the rest of the paper. A basic idea is to analyze skew primitive elements in more details. The analysis sometimes is tedious but necessary, and will be useful for the study of pointed Hopf algebras of GK-dimension three or higher. Let H be a pointed Hopf algebra and let C0 denote the coradical of H. We will need to use some concepts introduced in [WZZ1]. Suppose y is a skew primitive element with ∆(y) = y ⊗ 1 + g ⊗ y where the grouplike element g(= µ(y)) is the weight of y. Let Tg−1 denote the inverse conjugation by g sending a → g−1ag for 14 D.-G. WANG, J.J. ZHANG AND G. ZHUANG all a ∈ H. A nonzero scalar λ is called the commutator of y of level n (or more generally of finite level) if (Tg−1 − λIdH )n(y) ∈ C0, and (Tg−1 − λIdH )n−1(y) 6∈ C0. If the commutator of y of finite level exists, it is denoted by γ(y). When n = 1, γ(y) is called the commutator of y. The weight commutator of y is the pair (µ(y), γ(y)) which is denoted by ω(y). Given a grouplike element g, let Pg,∗,∗ denote the span of all (1, g)-primitive elements in H. It is clear that Pg,∗,∗∩ C0 = k(g− 1). Let P ′g,∗,∗ denote the quotient space Pg,∗,∗/k(g − 1). Given a nonzero scalar λ and an integer n, let Pg,λ,n denote the span of all (1, g)-primitive elements having commutator λ of level at most n. Let Pg,λ,∗ be the union of Pg,λ,n for all n. Similarly let P ′g,λ,n (respectively, P ′g,λ,∗ ) denote Pg,λ,n/k(g − 1) (respectively, Pg,λ,∗/k(g − 1)). Note that Pg,λ,0 = k(g − 1) and whence P ′g,λ,0 = 0. The total space of nontrivial skew primitive elements is defined to be (I3.0.1) P ′T :=Mg P ′g,∗,∗ =Mg Mλ P ′g,λ,∗ (see Lemma 3.2(b)). Let Rn be the set of primitive nth roots of unity in k and let √ denote Sn≥2 Rn -- the set of all roots of unity in k which is not 1. Let P ′M :=Mg Mλ6∈√ P ′g,λ,∗. In this case the major weight g is unique and the major commutator λ is also unique. For simplicity, we say H has a unique major skew primitive element in this case. By definition, when H has a unique major skew primitive element, two major skew primitive elements are linearly dependent in the quotient space P ′M . The major weights play a special role connecting non-major weights. Lemma 3.2. Suppose GKdim H < ∞. is a direct sum of P ′g,λ,∗ (a) Pg,∗,∗ is a sum of Pg,λ,∗ for all λ ∈ k×. (b) P ′g,∗,∗ (c) y ∈ Pg,λ,1 means that y is (1, g)-primitive and g−1yg = λy + τ (g − 1) for (d) If Pg,λ,n = Pg,λ,n+1 for some n, then Pg,λ,n = Pg,λ,∗. A similar statement some τ ∈ k. holds when P is replaced by P ′. for all λ ∈ k×. Proof. (a) This is [WZZ1, Lemma 3.7(a)], see also its proof. (b) If y ∈ Pg,λ,∗ ∩Pλ′6=λ Pg,λ′ ,∗, then y ∈ k(g − 1). Hence Pλ P ′g,λ,∗ sum. The assertion follows from part (a). (c,d) Easy. is a direct (cid:3) Definition 3.1. (a) If y ∈ Pg,λ,∗ \ C0 for some λ 6∈ √ , then y is called a major skew primitive element, g is called a major weight of H and λ a major commutator of H. (b) If P ′M is 1-dimensional, then there is only one grouplike element g and only one scalar γ 6∈ √ such that P ′M = P ′g,λ,∗ = P ′g,λ,1. HOPF ALGEBRAS OF GK-DIMENSION TWO 15 Lemma 3.3. In parts (b) and (c) suppose that the coradical C0 of H is commutative and that GKdim H < GKdim C0 + 2 < ∞. (a) Suppose y is a nontrivial skew primitive element in Hopf domain H such that γ(y) is a primitive pth root of unity for some p > 1. Then y0 := yp is a major skew primitive element (after choosing y properly). As a consequence, µ(y)p is a major weight and 1 is a major commutator and µ(y0)−1y0µ(y0) = y0. = dim P ′g,λ,1 = 1. (b) The dimension of P ′M is at most one. As a consequence, there is at most one pair (g, λ) with λ 6∈ √ such that dim P ′g,λ,∗ 6= 0; and in this case, dim P ′g,λ,∗ (c) If H is a pointed Hopf domain, then the dimension of P ′M is 1 and the major weight and the major commutator are unique. Consequently, there is exactly one pair (g, λ) with λ 6∈ √ such that dim P ′g,λ,∗ 6= 0; and further, dim P ′g,λ,∗ = dim P ′g,λ,1 = 1. Proof. (a) By Lemma 3.2(d), Pg,λ,1 6= Pg,λ,0 where (g, λ) denotes (µ(y), γ(y)). By choosing a different y if necessary, we have y ∈ Pg,λ,1 \ C0. By Lemma 3.2(c), there is a τ ∈ k such that g−1yg = λy + τ (g − 1). Since λ 6= 1, we may assume τ = 0 after replacing y by y + τ 1−λ (g − 1). Let H′ be the Hopf subalgebra generated by g±1 and y. Since H is a domain, so is H′. Since H′ is noncommutative (as g−1yg = λy), GKdim H′ ≥ 2 by [GZ, Lemma 4.5]. It is easy to compute that GKdim H′ ≤ 2. Then H′ is isomorphic to A(1, λ) defined in Example 1.1. Since λ is a primitive pth root for some p > 1, yp is a skew primitive with µ(yp) = gp and γ(yp) = 1. In the Hopf algebra H′(∼= A(1, λ)), yp is not in its coradical, or equivalently, yp 6∈ k(gp − 1). Hence yp is not in the coradical of H. Therefore yp is a major skew primitive element. The consequence is clear. (b) Let Y∗ be the k-linear space spanned by all skew primitive elements y such that the commutator of y is not in √ . By [WZZ1, Theorem 3.9], dim Y∗/(Y∗ ∩ C0) ≤ GKdim H − GKdim C0 < 2 where the last inequality is the hypothesis. Since dim Y∗/(Y∗ ∩ C0) is an integer, it is at most 1. It is easy to see that Y∗/(Y∗ ∩ C0) ∼=Mg Mλ6∈√ P ′g,λ,∗ = P ′M . The assertions follow easily. (c) By part (b) it is suffices to show that P ′M 6= 0. Suppose that on contrary P ′M = 0. Since H is pointed and C0 6= H, there is a nontrivial skew primitive element y in some Pg,λ,∗ where g = µ(y). Since P ′M = 0, the commutator λ is in √ . By part (a), yp is a major skew primitive element which is not in C0. So P ′M 6= 0, a contradiction. Lemma 3.4. Let A be a locally PI domain. (cid:3) (a) A is an Ore domain and the quotient division ring Q(A) of A is locally PI. (b) For each nonzero scalar λ, there are no nonzero elements {g, α, β} in A such that αg = λgα + β and βg = λgβ. 16 D.-G. WANG, J.J. ZHANG AND G. ZHUANG Proof. (a) This is well-known. (b) By part (a) we may assume that A is a division algebra. First assume that λ = 1. Let f = αβ−1, then f g = gf + 1 by using the fact gβ = βg. Thus Q(A) contains the first Weyl algebra which is not (locally) PI, yielding a contradiction. Next assume that λ 6= 1. If λ is not a root of unity, then A contains a copy of kλ[g, β] (since every proper prime factor ring of kλ[g, β] has to kill either g or β). Since kλ[g, β] is not PI, a contradiction. The last case is when λ is a primitive pth root of unity for some p > 1. Let G = gp. Then we have αG = λpGα + nβgp−1 = Gα + β′ and β′G = Gβ′ where β′ = nβgp−1. The assertion follows from the case when λ = 1. (cid:3) Proposition 3.5. Let H be a Hopf domain that is either locally PI or having GKdim K < 3. Then Pg,λ,∗ = Pg,λ,1 for all pairs (g, λ). Proof. By Lemma 3.2(d) it suffices to show the assertion that Pg,λ,2 = Pg,λ,1, which is equivalent to the following claim: for any scalars λ, b, c ∈ k, there is no triple (g, y1, y2) with y1 ∈ Pg,λ,1\C0 and y2 ∈ Pg,λ,2\C0 such that g−1y1g = λy1 +b(g−1) and g−1y2g = λy2 + y1 + c(g − 1). Next we prove the claim. If λ = 0, then y1 ∈ C0 which yields a contradiction. Hence λ 6= 0. Without loss of generality, we may assume that H is generated by g±1, y1, y2, whence H is affine and pointed. We consider two cases. The first case is when H is (locally) PI. If λ = 1 and b 6= 0, then the statement follows from Lemma 3.4(b) by taking α = y1 and β = bg(g − 1). If λ = 1 and b = 0, then the statement follows from Lemma 3.4(b) by taking α = y2 and β = g(y1 + c(g − 1)). If λ 6= 1, then y1 and y2 can be modified so that b = c = 0. Then the statement follows from Lemma 3.4(b) by taking α = y2 and β = gy1. This finishes the first case. The second case is when H is not (locally) PI, and then GKdim H < 3. By Corollary 1.12, all non-PI affine pointed Hopf algebras of GKdim < 3 are classified, namely, algebras (IIb), (III) and (V) in Theorem 1.4. It is easy to verify the statement for all these Hopf algebras. (cid:3) Under the hypotheses of Proposition 3.5, there is an improved version of (I3.0.1) P ′T =Mg Mλ P ′g,λ,1. Theorem 3.7 below gives a bound for dim P ′g,λ,1. Lemma 3.6. Suppose that H is a Hopf domain such that dim P ′g,1,1 ≤ 1 for all g. If λ is a primitive pth root of unity for some p ≥ 2, then dim P ′g,λ,1 ≤ 1 for all g. Proof. Suppose by contrary that dim P ′g,λ,1 ≥ 2. Then there are two linearly inde- pendent (1, g)-primitive elements y1 and y2 such that g−1y1g = λy1 and g−1y2g = λy2. Let a and b be two noncommutative variables. Then (a + b)p = ap + M1(a, b) + M2(a, b) + ··· + Mp−1(a, b) + bp where Mi(a, b) denote the sum of all noncommutative monomials of a and b with total (a, b)-degree (i, p − i). For example, M1(a, b) = abp−1 + babp−2 + b2abp−3 + ··· + bp−1a. HOPF ALGEBRAS OF GK-DIMENSION TWO 17 Let ξ be a scalar in k. Then p (a + ξb)p = ap + ξM1(a, b) + ξiMi(a, b). Xi=2 For any ξ ∈ k let yξ := y1 + ξy2. Then yξ is a (1, g)-primitive and g−1yξg = λyξ. ξ is in Pgp,1,,1 \ C0 for any ξ. Let f = yp By the proof of Lemma 3.3(a), yp 1. Since dim P ′gp,1,,1 ≤ 1 by hypotheses, Pgp,1,,1 = kf + k(gp − 1). This implies that p Xi=0 ξiMi(y1, y2) = yp ξ ∈ kf + k(gp − 1). Since k is infinite, the above equation implies that Mi(y1, y2) ∈ kf + k(gp − 1) for all i. Let Mi(y1, y2) = aif + bi(gp − 1). Choose ξ so that Pp i=0 aiξi = 0. Then p p p biξi)(gp − 1) = b(gp − 1) yp ξ = Xi=0 ξiMi(y1, y2) = ( Xi=0 Xi=0 for some b ∈ k. Write b = −cp for some c ∈ k. Then yp ξ + (cg)p − cp = 0. aiξi)f + ( Since g−1yξg = λyξ, the above equation is equivalent to (yξ + cg)p − cp = 0, or Qp−1 p . Since H is a domain, y1+ξy2+cg−c′ = 0. n=0(yξ+cg−ηnc) = 0 where ηn = e Since g−1(y1 + ξy2)g = λ(y1 + ξy2), g−1(cg − c′)g = (cg − c′) and λ 6= 1, we have c′ = c = 0. This contradicts the fact y1 and y2 are linearly independent. We finish the proof. (cid:3) 2niπ Theorem 3.7. Suppose that H is a pointed Hopf domain with a commutative coradical C0 and that GKdim H < GKdim C0 + 2 < ∞. Then dim P ′g,λ,1 ≤ 1 for all pairs (g, λ). Proof. If λ is either 1 or not a root of unity, the assertion follows from Lemma 3.3(c). If λ is a primitive pth root of unity for p > 1, the assertion follows from Lemma 3.6. (cid:3) An important consequence of Theorem 3.7 is the following. Lemma 3.8. Let H be a Hopf algebra such that P ′g,λ,1 is 1-dimensional for some pair (g, λ). If G0 is an abelian subgroup of grouplike elements and it contains g, then there is a z ∈ Pg,λ,1 \ C0 such that either (a) h−1zh = χ(h)z for some character χ : G0 → k× (where χ(g) = λ), or (b) h−1zh = z + τ (h)(g − 1) for some additive character τ : G0 → k and λ = 1. In part (a), z is unique up to a scalar multiple. In part (b), z is unique up to an addition of k(g − 1). Proof. Let y be any element in Pg,λ,1 \ C0 and let V = k(g − 1) +Ph∈G0 k(h−1yh). Since G0 is abelian, h−1yh ∈ Pg,λ,1 \ C0. Hence V ⊂ Pg,λ,1. Thus dim V ≤ dim Pg,λ,1 = 2. The assertion follows from [WZZ1, Lemma 2.2(c)]. (cid:3) Finally we prove that the total space of skew primitive elements is finite dimen- sional. Theorem 3.9. Let H be a pointed Hopf domain of GKdim < 3 and suppose that C0 = kZ. Then P ′T is finite dimensional. 18 D.-G. WANG, J.J. ZHANG AND G. ZHUANG = 1. Denote this pair by (xM , ν). Proof. By Proposition 3.5 and Theorem 3.7, dim P ′g,λ,∗ = dim P ′g,λ,1 ≤ 1 for any pair (g, λ). It suffices to show that there are only finitely many pairs (g, λ) such that dim P ′g,λ,∗ 6= 0. By Lemma 3.3(c), there is exactly one pair (g, λ) such that λ 6∈ √ and dim P ′g,λ,∗ If there is another pair (g, λ) such that dim P ′g,λ,∗ 6= 0, then λ ∈ √ by Lemma 3.3(b). Write g = xn. Pick y ∈ Pg,λ,∗ \ C0 and we may assume that x−1yx = qy by Lemma 3.8(a). Then λ = qn and it is a primitive pth root of unity for some p > 1. Thus yp ∈ Pxnp,1,1\ C0 by the proof of Lemma 3.3(a). Thus P ′xnp,1,1 6= 0 and whence (xnp, 1) = (xM , ν). Since λ 6= 1, n 6= 0. Thus M 6= 0 and M = np. Since M is fixed and p > 1, there are only finitely many choices for n, or equivalently, finitely many choices for g = xn. For each fixed g = xn, λ is a primitive pth root of unity where p = M/n. Thus the possibilities for λ are also finite. Therefore there are only finitely many choices for pairs (g, λ) such that dim P ′g,λ,∗ 6= 0. (cid:3) We have an easy corollary. Corollary 3.10. Suppose H is a pointed Hopf domain of GKdim < 3. If H is finitely generated by grouplike and skew primitive elements, then P ′T is finite dimensional. Proof. If GKdim H < 2, then GKdim H ≤ 1 and such Hopf algebras are classified in [GZ, Section 2]. The assertion is easy to check. Suppose now GKdim H ≥ 2. Then all affine pointed Hopf algebras are classified except for that case when dim C0 = 1, see subsection 1.2. So assertions can be verified when GKdim C0 6= 1. The remaining case is when GKdim C0 = 1. Since H is a domain, so is C0. Then C0 = kΓ for an abelian torsionfree group Γ of rank 1 by [GZ, Section 2]. By Lemma 1.10, Γ is finitely generated. Thus Γ = Z. Now Theorem 3.9 applies. (cid:3) 4. A result of Heckenberger We need to use a result of Heckenberger [He] which concerns the classification of finite dimensional Nichols algebras of rank 2. Let G be a finite abelian group and let G GYD be the Yetter-Drinfel'd category. Let V be a Yetter-Drinfel'd module over kG of dimension 2 with left kG-action denoted by ∗ and the left kG-coaction denoted by δ : V → kG ⊗ V . Assume that V is of diagonal type, namely, there is a basis {v1, v2} such that (I4.0.1) for gi ∈ G, and (I4.0.2) where qij ∈ k×. The braiding on V is determined by σ(vi ⊗ vj) = qij vj ⊗ vi δ(vi) = gi ⊗ vi, i = 1, 2 gi ∗ vj = qij vj, i, j ∈ {1, 2} for all i, j ∈ {1, 2}. The Nichols algebra over V is denoted by B(V ). Heckenberger worked out the precise conditions on {qij} such that B(V ) is finite dimensional. To quote Heckenberger's result we need to introduce a few other notations. Following [He, p.118], let ∆+(B(V )) be the set of degrees of the (restricted) Poincar´e-Birkhoff- Witt generators counted with multiplicities. From this, dim B(V ) < ∞ if and only if ∆+(B(V )) is finite. Based on ∆+(B(V )), one can define ∆(B(V )), a subgroupoid Wχ,E, and an arithmetic root system (∆(B(V )), χ, E) (details are omitted). When HOPF ALGEBRAS OF GK-DIMENSION TWO 19 (∆(B(V )), χ, E) is an arithmetic root system, we implicitly assume that Wχ,E is full and finite. A very nice result of Heckenberger [He, Theorem 3] states that there is a one-to-one correspondence between finite ∆+(B(V )) and arithmetic root systems (∆, χ, E). Below is a re-statement of a part of a remarkable result [He, Theorem 7]. Recall that Rn is the set of primitive nth roots of unity. Lemma 4.1. [He] Let V be a 2-dimensional Yetter-Drinfel'd module over kG of diagonal type with structure coefficients (qij )2×2 defined in (I4.0.2). Suppose B(V ) is finite dimensional. Then, up to a permutation of {v1, v2}, one of the following is true. (3) q12q21 6= 1, q11q12q21 6= 1, q12q21q22 6= 1, q22 = −1, q11 ∈ R2 ∪ R3, and (4) q12q21 6= 1, q11q12q21 6= 1, q12q21q22 6= 1, q22 = −1, q11 6∈ R2 ∪ R3, and 11 or 11 or 12q3 12q2 21 6= 1 or (1) q12q21 = 1. (2) q12q21 6= 1, q12q21q22 = 1, and 21 6= 1 or 11 6= 1 or 21 6= 1 or (2.1) q11q12q21 = 1 or 12q2 (2.2) q11 = −1, q2 (2.3) q2 11q12q21 = 1 or (2.4) q3 11q12q21 = 1, q2 (2.5) q11 ∈ R3, q3 (2.6) q12q21 ∈ R8, q11 = (q12q21)2 or (2.7) q12q21 ∈ R24, q11 = (q12q21)6 or (2.8) q12q21 ∈ R30, q11 = (q12q21)12. (3.1) q11 = −1, q2 (3.2) q11 ∈ R3, q12q21 ∈ {q11,−q11} or (3.3) q0 := q11q12q21 ∈ R12, q11 = q4 0 or (3.4) q12q21 ∈ R12, q11 = −(q12q21)2 or (3.5) q12q21 ∈ R9, q11 = (q12q21)−3 or (3.6) q12q21 ∈ R24, q11 = −(q12q21)4 or (3.7) q12q21 ∈ R30, q11 = −(q12q21)5. (4.1) q12q21 = q−2 (4.2) q11 ∈ R5 ∪ R8 ∪ R12 ∪ R14 ∪ R20, q12q21 = q−3 (4.3) q11 ∈ R10 ∪ R18, q12q21 = q−4 11 or (4.4) q11 ∈ R14 ∪ R24, q12q21 = q−5 11 or (4.5) q12q21 ∈ R8, q11 = (q12q21)−2 or (4.6) q12q21 ∈ R12, q11 = (q12q21)−3 or (4.7) q12q21 ∈ R20, q11 = (q12q21)−4 or (4.8) q12q21 ∈ R30, q11 = (q12q21)−6. (5.1) q0 := q11q12q21 ∈ R12, q11 = q4 (5.2) q12q21 ∈ R12, q11 = q22 = −(q12q21)2 or (5.3) q12q21 ∈ R24, q11 = (q12q21)−6, q22 = (q12q21)−8 or (5.4) q11 ∈ R18, q12q21 = q−2 11 , q22 = −q3 11 or (5.5) q11 ∈ R30, q12q21 = q−3 11 , q22 = −q5 11. 0, q22 = −q2 0 or (5) q12q21 6= 1, q11q12q21 6= 1, q12q21q22 6= 1, q11 6= −1, q22 ∈ R3 and Proof. By definition, ∆+(B(V )) is finite if and only if B(V ) is finite dimensional, and by [He, Theorem 3], if and only if (∆(B(V )), χ, E) is an arithmetic root system. By the definition of an arithmetic root system, Wχ,E is full and finite. By [He, Page 131], Wχ,E is full and finite if and only if, up to a permutation of {v1, v2}, one of 20 D.-G. WANG, J.J. ZHANG AND G. ZHUANG the cases listed above is true (according to [He, page 131], this statement is also equivalent to [He, Theorem 7]). The assertion follows. (cid:3) Proposition 4.2. Retains the hypotheses of Lemma 4.1. Assume that (a) there are two scalars q1 and q2 and two positive integers n1 and n2 such that qij = qni j (b) gcd(n1, n2) = 1. for all i, j ∈ {1, 2}, and Let ǫ be a positive integer and let p1 = n2ǫ and p2 = n1ǫ. Further assume that (c) both q1 and q11 are primitive p1st roots of unity, and (d) both q2 and q22 are primitive p2nd roots of unity. Then, up to a permutation of {v1, v2}, one of the following holds. (I) q12q21 = 1, (II) n1 = n2 = 1, q1, q2 ∈ R3. (III) n1 = n2 = 1, q1, q2 ∈ R5. (IV) n1 = 1, n2 = 2, ǫ = 5, p1 = 10, p2 = 5 and q4 (V) n1 = n2 = 1, ǫ = 7, q1, q2 ∈ R7, q1q2 (VI) n1 = 1, n2 = 3, ǫ = 7, p1 = 21, p2 = 7, q3 1q2 = 1 and q2 1q3 2 = 1. 2 = 1 and q4 1q2 = 1. 1q4 2 = 1 and that q6 1q2 = 1. Proof. The proof is heavily dependent on Lemma 4.1. First of all, case (1) in Lemma 4.1 is just case (I) here. If ǫ = 1, then q21 = qn2 1 = 1 and similarly, q12 = 1. Thus case (I) occurs. 1 = qp1 Secondly, if ǫ = 2 then q21 = qn2 1 = −1 as q1 is a primitive (2n2)nd root of unity. Similarly, q12 = −1. Hence q12q21 = 1 and case (I) occurs. Thirdly, if q11 = −1 then p1 = 2. Thus ǫ is either 1 or 2. By the last two paragraphs, (I) occurs. Similarly if q22 = −1, then (I) occurs. Thus cases (2.2), (3.1-3.7), (4.1-4.8) in Lemma 4.1 can not happen under the extra hypotheses of Proposition 4.2. It remains to analyze cases (2.1), (2.3-2.8), (5.1-5.5). Below we are using the numbering in Lemma 4.1. Case (2): The equation q12q21q22 = 1 means that qn1 2 qn2 1 qn2 2 = 1. Then 1 = 1ǫ = (qn1 2 qn2 1 qn2 2 )ǫ = qp2 2 qp1 1 (qn2 2 )ǫ = (q22)ǫ Thus by hypothesis (d), p2 = ǫ which implies that n1 = 1. Below are subcases. Case (2.1): The equation q11q12q21 = 1 implies that n2 = 1 and q1q2 2 = 1 = q1q2 2 where the second equation follows from q12q21q22 = 1. These equations implies q1, q2 ∈ R3, and whence case (II) occurs. Case (2.3): The equation q2 1 = 1ǫ = (q2 11q12q21 = 1 implies that 11q12q21)ǫ = (q2 11)ǫqn1ǫ 2 qn2ǫ 1 = q2ǫ 11. Hence n2ǫ = p1 divides 2ǫ. So we have two subcases: either n2 = 1 or n2 = 2. When n2 = 1, the equation q12q21q22 = 1 is that q1q2 q2 11q12q21 = 1 is that q3 occurs. 2 = 1 and the equation 1q2 = 1. It is easy to see that q1, q2 ∈ R5. Then case (III) 2 = 1 and the equation 1q2 = 1. It is easy to see that q1 ∈ R10, q2 ∈ R5. Conse- 1q3 When n2 = 2, the equation q12q21q22 = 1 is that q2 q2 11q12q21 = 1 is that q4 quently, ǫ = 5. Then case (IV) occurs. Case (2.4): The equation q3 1 = 1ǫ = (q3 11q12q21 = 1 implies that 11q12q21)ǫ = (q3 11)ǫqn1ǫ 2 1 = q3ǫ qn2ǫ 11. Hence n2ǫ = p1 divides 3ǫ. So we have two subcases: either n2 = 1 or n2 = 3. HOPF ALGEBRAS OF GK-DIMENSION TWO 21 When n2 = 1, the equation q12q21q22 = 1 is that q1q2 When n2 = 3, the equation q12q21q22 = 1 is that q3 2 = 1 and the equation 1q2 = 1. It is easy to see that q1, q2 ∈ R7. Then case (V) 2 = 1 and the equation 2 = 1. Thus p2 = 7 = ǫ. So case 1q2 = 1. Consequently, q7 1q4 11q12q21 = 1 is that q4 q3 occurs. q3 11q12q21 = 1 is that q6 (VI) occurs. Case (2.5): Since q11 ∈ R3, p1 = 3. This implies that either ǫ = 1 or ǫ = 3. If ǫ = 1, then (I) occurs by the first paragraph. So we may assume ǫ = 3 and whence n2 = 1. Since n1 = 1 (see the beginning of case (2)), p2 = 3n1 = 3. This is case (II). Case (2.6): Since q11 = (q12q21)2, 1 = 1ǫ = (q12q21)ǫ = ((q12q21)ǫ)2 = qǫ 11 which implies that p1 = ǫ and n2 = 1. Since n1 = n2 = 1, q12q21 ∈ R8 means that q1q2 ∈ R8. The equation q12q21q22 = 1, see at the beginning of case (2), implies that q1q2 2 = q1q2 ∈ R8. Thus p1 = p2 = ǫ = 8. This contradicts the fact q1 = q−2 2 = 1 which is equivalent to q−1 2 ∈ R4. Case (2.7): Since q11 = (q12q21)6, 1 = 1ǫ = (q12q21)ǫ = ((q12q21)ǫ)6 = qǫ 11 which implies that p1 = ǫ and n2 = 1. Since n1 = n2 = 1, q12q21 ∈ R24 means that q1q2 ∈ R24. The equation q12q21q22 = 1 given at the beginning of case (2) implies that q1q2 2 = q1q2 ∈ R24. Thus p1 = p2 = ǫ = 24. This contradicts the fact q1 = q−2 2 = 1 which is equivalent to q−1 2 ∈ R12. Case (2.8): Since q11 = (q12q21)12, 1 = 1ǫ = (q12q21)ǫ = ((q12q21)ǫ)12 = qǫ 11 which implies that p1 = ǫ and n2 = 1. Since n1 = n2 = 1, q12q21 ∈ R30 means that q1q2 ∈ R30. The equation q12q21q22 = 1 implies that q1q2 2 = 1 which is equivalent to q−1 2 = q1q2 ∈ R30. Thus p1 = p2 = ǫ = 30. This contradicts the fact q1 = q−2 2 ∈ R15. Case (5): Since q22 ∈ R3, we have p2 = 3. Since ǫ p2, ǫ is either 1 or 3. If ǫ = 1, then (I) occurs, so a contradiction, by the first condition in case (5). Thus ǫ = 3 and consequently, n1 = p2/ǫ = 1. Below are subcases. 1q2)4 = q−4 Case (5.1): Since q0 = q11q12q21 = q1q2qn2 we have q1 ∈ R12. Thus p1 = 12, n2 = 4. The equation q11 = q4 q1 = (q5 q2 ∈ R3 and that q1 ∈ R12. R12 or R24. A contradiction. 1 q2, which implies that q2 = q5 Cases (5.2) and (5.3): Since (q12q21)ǫ = 1 and ǫ = 3, then q12q21 can not be in q2 ∈ R12 and q12q21 ∈ R3, 0 becomes 1. This contradicts the facts that 1 = q(1+n2) 1 Case (5.4): Since ǫ = 3, by q12q21 = q−2 11 , 1 = 13 = (q12q21)3 = q−6 11 which contradicts the fact q11 ∈ R18. Case (5.5): Since ǫ = 3, by q12q21 = q−3 11 , 1 = 13 = (q12q21)3 = q−9 11 which contradicts the fact q11 ∈ R30. This finishes the proof. (cid:3) 22 D.-G. WANG, J.J. ZHANG AND G. ZHUANG We are interested in the case when G = Z/(M ) for some integer M with a generator x and when V = kv1 ⊕ kv2 is a Yetter-Drinfel'd module over kG of diagonal type such that, for i = 1 and 2, δ(vi) = xni ⊗ vi, (I4.2.1) for some n1, n2 ∈ N and q1, q2 ∈ k×. Remark 4.3. Andruskiewitsch informed us that the Nichols algebra B(V ) is finite dimensional if V satisfies (I4.2.1) and {n1, n2, q1, q2} satisfies one of the following conditions: x ∗ vi = qivi (a) n1 = n2 = 1, q1 ∈ R5 and q2 = q2 1. (b) n1 = n2 = 1, q1 ∈ R7 and q2 = q3 1. (c) n1 = 1 and n2 = 2, q1 ∈ R10 and q2 = q6 1. (d) n1 = 1 and n2 = 3, q1 ∈ R21 and q2 = q15 1 . Recall that the weight commutator of a skew primitive element y is ω(y) := (µ(y), γ(y)). Definition 4.4. (a) Let N5 denote any Hopf domain of GK-dimension two that is generated by x±1, y1, y2 with ω(yi) = (xni , qni i ) for i = 1, 2 such that q2 = q2 1. n1 = n2 = 1, q1, q2 ∈ R5, (b) Let N10 denote any Hopf domain of GK-dimension two that is generated by x±1, y1, y2 with ω(yi) = (xni , qni i ) for i = 1, 2 such that n1 = 1, n2 = 2, q1 ∈ R10, q2 ∈ R5, q2 = q6 1. (c) Let N7 denote any Hopf domain of GK-dimension two that is generated by (e) Algebras N5, N10, N7, N21 (if exist) are called supplementary Hopf algebras of GK-dimension two. Remark 4.5. As far as we know, no example of supplementary Hopf algebras of GK-dimension two has been constructed. This leads to conjecture that algebras N5, N7, N10 and N21 do not exist. If this conjecture is true, then the hypothesis Ω′ in Theorem 0.1 and Corollary 0.2 can be removed and the hypothesis Ω defined in Section 6 is vacuous. 5. Analysis in the case s = 2 The proof of Theorem 0.1 requires some further analysis of skew primitive ele- ments. In this section we prove the following special case of Theorem 0.1. Theorem 5.1. Let H be a Hopf algebra satisfying the following conditions (a) H is a domain of GK-dimension two. (b) its coradical C0 is kZ with a generator x. (c) H is generated by x±1, y1 and y2 where yi ∈ P(xni ,λi,∗) for i = 1, 2, and H is not equal to the subalgebra generated by {x±1, y1}, or by {x±1, y2}. (d) Let N21 denote any Hopf domain of GK-dimension two that is generated x±1, y1, y2 with ω(yi) = (xni , qni n1 = n2 = 1, i ) for i = 1, 2 such that q1, q2 ∈ R7, q2 = q3 1. by x±1, y1, y2 with ω(yi) = (xni , qni n1 = 1, n2 = 3, i ) for i = 1, 2 such that q2 = q15 1 . q1 ∈ R21, q2 ∈ R7, HOPF ALGEBRAS OF GK-DIMENSION TWO 23 (d) gcd(n1, n2) = 1. Then H is isomorphic to either (1) B(1,{pi}2 (2) or one of N5, N7, N10 and N21. in this case, y1y2 = y2y1, 1, q,{αi}2 1) defined in Convention 2.5 where p1 = n2, p2 = n1, and The proof will be given at the end of the section and we start with the following lemma. Lemma 5.2. Retain the hypotheses of Theorem 5.1. Then, after choosing y1, y2 appropriately, the following hold. (a) There are two scalars q1, q2 such that yix = qixyi and λi = qni i (b) q1 and qn1 1 (= λ1) are both primitive p1st root of unity for some p1 > 1. And . yp1 1 is a major skew primitive element. (c) q2 and qn2 And yp2 2 2 (= λ2) are both primitive p2nd root of unity for some p2 > 1. is a major skew primitive element. (d) Replacing x by x−1 if necessary we may assume that n1 ≥ 0. Under this hypothesis, both n1 and n2 are positive integers and n1p1 = n2p2. The major weight is xn1p1 . y1 (respectively, the image of y2) is nonzero in H/(yp1 (e) There is a positive integer ǫ such that p1 = n2ǫ and p2 = n1ǫ. (f) yp1 1 and xn1p1 are central elements in H. As a consequence, H is PI. (g) yp1 1 (h) H/(yp1 is nonzero in the quotient Hopf algebra H′ := H/(xn1p1 − 1). 1 , (xn1p1 − 1)) is a finite dimensional Hopf algebra and the image of 1 , xn1p1 − 1). (i) Let y is a skew primitive element in H \ C0 with ω(y) = (xn3 , λ3) with gcd(n3, n1) = 1. Then there is a scalar q3 and positive integers n3 and p3 such that (i1) λ3 = qn3 3 , (i2) both q3 and λ3 are primitive p3rd root of unity, and (i3) n3p3 = n1p1 = n2p2. Proof. (a,d) By hypothesis (c) of Theorem 5.1, y1 and y2 are linearly independent in H/C0. By Theorem 3.7, (xn1 , λ1) 6= (xn2 , λ2). By Lemma 3.3(b), there is at least one λi which is in √ . By symmetry, we may assume that λ1 is a primitive p1st root of unity for some p1 > 1, and further we assume that n1 > 0 without loss of generality. Since λ1 6= 1, Lemma 3.8(b) can not happen, so by Lemma 3.8(a), there is a z ∈ P(xn1 ,λ1,1) \ C0 such that h−1zh = χ(h)z for all h ∈ Z. Replacing y1 by z, we may assume that y1x = q1xy1 where λ1 = χ(xn1 ) = (χ(x))n1 = qn1 1 . This also says that qn1 1 ∈ Pxn1p1 ,1,1 by 1 Lemma 3.3(a). By the proof of Lemma 3.3(a), Pxn1p1 ,1,1 6= k(xn1p1 − 1). Therefore xn1p1 is the major weight and 1 is the major commutator. Since H is not generated by x±1, y1 and since the major skew primitive elements 1 , (xn1p1 , 1) 6= (xn2 , λ2). By Lemma 3.3(b), λ2 ∈ √ . Say of H is generated by yp1 λ2 is a primitive p2nd root of unity for some p2 > 1. An argument similar to the above shows that is a primitive p1st root of unity. Therefore yp1 (i) y2x = q2xy2 and λ2 = qn2 2 ; (ii) xn2p2 is the major weight, and by the uniqueness of the major weight (iii) yp2 [Lemma 3.3(c)], we have n1p1 = n2p2; 2 ∈ Pxn2p2 ,1,1 = Pxn1p1 ,1,1, and whence yp2 in P ′xn2p2 ,1,1. 2 and yp1 1 are linearly dependent 24 D.-G. WANG, J.J. ZHANG AND G. ZHUANG Up to this point we have proved (a) and (d). Part (e) follows from part (d) and the fact gcd(n1, n2) = 1. (b,c) After replacing y1 by a scalar multiple, (iii) implies that (I5.2.1) yp1 1 = yp2 2 + α(xM − 1) where M = n1p1 = n2p2. Together with yix = qixyi for i = 1, 2, one derives that qp1 1 = qp2 2 when x is commuted with relation (I5.2.1). From this, qn2p1 1 = qn2p2 2 = 1 and, by definition, qn1p1 1 = 1. 1 = 1. Therefore both q1 and qn1 1 Since gcd(n1, n2) = 1, we obtain that qp1 are primitive p1st roots of unity. Hence (b) holds. Note that (c) is similar. As a consequence, yp1 1 and xM are central elements in H. (f) By the end of proof of (b,c), yp1 1 and xM are central elements in H. To finish (f) note that the subalgebra generated by yp1 1 and x±M is a Hopf subalgebra of H, which is central in H and is of GK-dimension two. The last assertion in (f) follows from [SmZ, Corollary 2]. 1 0 6= HK + 1 ]. Since yp1 (g) A result of Takeuchi [Ta1, Theorem 3.2] says that a Hopf algebra H is faithfully flat over its Hopf subalgebra if the coradical of H is cocommutative. Let K0 = k[x±M ] and K1 = k[x±M , yp1 is a nontrivial skew primitive element, K0 6= K1. By part (f) both K0 and K1 are two distinct central Hopf subalgebras of H. By [Mo, Proposition 3.4.3], HK + 1 6∈ H(xM−1). This means that yp1 1 is a nonzero primitive element in H′ := H/(xM−1). (h) Let K be the Hopf subalgebra of H generated by y1 and x±n1 . By the relations of K, one sees that K is a quotient Hopf algebra of A(1, qn1 1 ) [Example 1 ) are domains of GK-dimension two, K ∼= A(1, qn1 1.1]. Since both K and A(1, qn1 1 ). By part (f) xM is central in H and in K. Let K′ = K/(xM−1) and H′ = H/(xM−1) and recycle the letters for the elements in K′ and H′. By part (f) H is PI. Since any affine PI algebra is catenary, the principal ideal theorem implies that GKdim H′ = GKdim H − 1 = 1. Using this fact and (I5.2.1), one sees that y1 is nonzero in H′. Then the equation y1xn1 = qn1 1 x−n1 y1 implies that y1 is a nontrivial skew primitive element in H′. By part (g), yp1 is a nonzero (and whence nontrivial) primitive 1 element in H′. 1 . In particular, yp1 Let φ′ denote the natural Hopf map K′ → H′ which maps y1 ∈ K′ to a nontrivial skew primitive element y1 ∈ H′. Since the only nontrivial skew primitive elements in K′ are generated by y1 + k(xn1 − 1) and yp1 1 , the assertion proved in the last paragraph says that φ′ is injective when restricted to C1(K′). By [Mo, Theorem 5.3.1], φ′ is injective. By [Ta1, Theorem 3.2] H (respectively, H′) is faithfully flat over K (respectively, K′). Since yp1 is a nonzerodivisor of K′, it is also a 1 nonzerodivisor of H′. Thus GKdim H/(yp1 1 , xM − 1) ≤ GKdim H/(xM − 1) − 1 ≤ GKdim H − 2 = 0. 1 , xM − 1)) is affine, H/(yp1 Since H (and hence H(yp1 We proved the first part of (h). 1 , xM − 1) is finite dimensional. For the second part of (h), note that y1 is a nonzerodivisor of K′. So it is a 1 , xM − f ) = 0 nonzerodivisor of H′ since H′ is faithfully flat over K′. If y1 = 0 in H/(yp1 1 f for some f ∈ H′. Re-writing it as y1(1− yp1−1 1) = H′/(yp1 1 ), then y1 = yp1 1 HOPF ALGEBRAS OF GK-DIMENSION TWO 25 yields a contradiction with that fact y1 is a nonzerodivisor of H′. Therefore y1 is nonzero in H/(yp1 1 , xM − 1). By symmetry, y2 is nonzero in H/(yp1 1 , xM − 1). (i) The proof is similar to the proof of (b). (cid:3) The next result uses Proposition 4.2. Theorem 5.3. Retain the hypotheses of Theorem 5.1 and the notation of Lemma 5.2. Additionally assume that H is not isomorphic to any of N5, N7, N10 and N21. Let qij = qni j for i, j ∈ {1, 2}. Then q12q21 = 1. Proof. Let A = H/(yp1 1 , xM − 1). This is a quotient Hopf algebra of H. By Lemma 5.2(h), A is finite dimensional. Also by Lemma 5.2(h), the image of y1, which is still denoted by y1, (respectively, the image of y2) is nonzero in A. If y1 and y2 are linear dependent in A, then n1 = n2 = 1 and q1 = q2. This contradicts Theorem 3.7 since y1 and y2 are linearly independent in H/C0 by hypothesis (c) of Theorem 5.1. Therefore these two are linearly independent nontrivial skew primitive elements of A. Let B be the associated graded Hopf algebra of A with respect to its coradical filtration. Then B = C#G where G = Z/(M ) and C is a braided Hopf algebra. Let V be the subspace of B (also viewed as a subspace of C) spanned by y1 and y2. Then V is a Yetter-Drinfel'd module over G (recall that G = Z/(M ) = hx xM = 1i) with δ(yi) = xni ⊗ yi, xni ∗ yj = qni j yj for all i, j ∈ {1, 2}. The Nichols algebra over V , denoted by B(V ), is a subquotient of C. Therefore B(V ) is finite dimensional over k. By Proposition 4.2, up to a permutation, one of cases (I)-(VI) holds. We analyze these six case below. Case (I) is our assertion. Case (II): n1 = n2 = 1 and q1, q2 ∈ R3. Then either q2 = q1 or q2 = q−1 1 . 1 , or When q2 = q1, it yields a contradiction with Theorem 3.7. Therefore q2 = q−1 q1q2 = 1. Hence the assertion. Case (III): n1 = n2 = 1 and q1, q2 ∈ R5. Then q2 = qi Theorem 3.7, q2 = q1 is impossible. The case q2 = q4 to study q2 = q2 to q1 = q3 algebra H is N5 in Definition 4.4(a). 2, so we only consider the case when q2 = q2 1 and q2 = q3 1 for some 1 ≤ i ≤ 4. By 1 is our assertion. It remains 1 is equivalent 1. Under this condition, the 1. These two are equivalent since q2 = q2 Case (IV): n1 = 1, n2 = 2, ǫ = 5, p1 = 10, p2 = 5 and q4 1q2 = 1 and q2 1q3 2 = 1. This is the algebra N10 in Definition 4.4(b). algebra N7 in Definition 4.4(c). Case (V): n1 = n2 = 1, ǫ = 7, q1, q2 ∈ R7, q1q2 Case (VI): n1 = 1, n2 = 3, ǫ = 7, p1 = 21, p2 = 7, q3 2 = 1 and q4 1q2 = 1. This is the 1q4 2 = 1 and that q6 1q2 = 1. This is the algebra N21 in Definition 4.4(d). Combining all cases with the additional hypothesis, the assertion follows. (cid:3) Lemma 5.4. Retain the hypotheses of Theorem 5.1 and the notation of Lemma 5.2. Additionally assume that H is not isomorphic to any of N5, N7, N10 and N21. Then (a) n1 6= n2. (b) y2y1 − qn1 2 y1y2 = 0. Proof. First we prove the following: (c) If n1 = n2 = 1, then p1 = p2 > 2. 26 D.-G. WANG, J.J. ZHANG AND G. ZHUANG 2 y1y2 is a skew primitive element in C0. (d) y2y1 − qn1 (c) Clearly, p1 = n2ǫ = ǫ = n1ǫ = p2. If p1 = p2 = 1, then q1 = q2 = 1 and y1 and y2 are both major skew primitive, so y1 and y2 are not linearly independent in H/C0 by Theorem 3.7, which yields a contradiction by hypothesis (c) of Theorem 5.1. If p1 = p2 = 2, then q1 = q2 = −1. By (I5.2.1), 2 + a(x2 − 1) y2 1 = y2 for some a ∈ k. Let y3 = y1y2 + y2y1. Then y3 is a skew primitive of weight x2. (In general, if qn2 2 y2y1 is skew primitive of weight xn1+n2 ). Therefore 2 = 1, then y1y2 − qn1 1 qn1 for some b, c ∈ k. Pick some scalars α, β, γ satisfying the equations y1y2 + y2y1 = by2 2 + c(x2 − 1) 1 + α2 + αb = 0 a + αc + β2 = 0 −a − αc − γ2 = 0. Then (y1+αy2 + βx + γ)(y1 + αy2 + βx − γ) 2 + α(y1y2 + y2y1) + β(y1x + xy1) + αβ(y2x + xy2) + β2x2 1 + α2y2 + γ(y1 − y1) + αγ(y2 − y1) + βγ(x − x) − γ2 2 + a(x2 − 1) + α2y2 2 + α(by2 2 + c(x2 − 1)) + β2x2 − γ2 =y2 =y2 =0. 1 qn1 (d) By Theorem 5.3, qn2 2 = 1. Then y3 := y2y1 − qn1 Since H is a domain, either y1 + αy2 + βx + γ = 0 or y1 + αy2 + βx − γ = 0. Both leads to a contradiction with hypothesis (c) of Theorem 5.1. Therefore p1 = p2 > 2. 2 y1y2 is a skew primitive It is easy to by a direct computation. Suppose on the contrary that y3 6∈ C0. see that ω(y3) = (xn1+n2 , (q1q2)n1+n2 ). If xn1+n2 is a major weight (or y3 is a major skew primitive), then n1 + n2 = n1p1 = n2p2. Since gcd(n1, n2) = 1, we get n1 = n2 = 1 and p1 = p2 = 2. By part (c) this is impossible. Therefore xn1+n2 is not the major weight. Consequently, we have that n1 + n2 6= n1p1 = n2p2 and that y3 is not a major primitive element. Let n3 = n1 + n2. By Lemma 5.2(i), q3 := q1q2 and λ3 = qn3 3 are primitive p3rd roots of unity and n3p3 = n1p1 = n2p2. Since gcd(n1, n2) = 1, gcd(n3, n1) = gcd(n1 + n2, n1) = gcd(n2, n1) = 1. This implies that ni p3 for i = 1, 2. Since y1 and y3 are non-major skew primitives of different weight, then the Hopf subalgebra H′ generated by x±1, y1, y3 satisfies the hypotheses in Theorem 5.1(a-d). 1 = q5 Since n3 = n1 + n2 > 1, the Hopf domain H′ is isomorphic to neither N5 nor N7. If H′ is isomorphic to N10, then n3 = 2, n1 = 1, p1 = 10, p3 = 5 and q3 = q6 1. Hence q2 = q3q−1 1 has order 2, contradicting p2 = n1p1/n2 = p1 = 10. If H′ is isomorphic to N21, then n3 = 3, n1 = 1, p1 = 21, p3 = 7 and q3 = q15 1 . Hence n2 = 2, contradicting n2p2 = n1p1 = 21. Therefore the additional hypothesis in Theorem 5.3 holds for H′. Applying Theorem 5.3 to H′ we obtain that qn3 3 = 1. Now 1 qn1 q2n1 1 = q2n1 1 (qn2 1 qn1 2 ) = qn1+n2 1 (q1q2)n1 = qn3 1 qn1 3 = 1. HOPF ALGEBRAS OF GK-DIMENSION TWO 27 Since qn1 is a p1st primitive root of unity, p1 = 2. Since p1 = n3ǫ3 by Lemma 5.2(e) 1 for H′, ǫ3 = 1. This implies that p1 = n3 = 2 and p3 = n1 by Lemma 5.2(e). Since n1 + n2 = n3 = 2, we have n1 = n2 = 1 and q2 3 = 1. This means that y3 is a major skew primitive element, a contradiction. 1 = 1 and q1 Now we go back to prove the lemma. (a) Suppose n1 = n2. Then n1 = n2 = 1 as gcd(n1, n2) = 1 and q1q2 = 1 by Theorem 5.3. By part (c) p := p1 = p2 > 2. By part (d), y2y1 − q2y1y2 is a skew primitive in C0. Hence for some b ∈ k. By (I5.2.1), we have y2y1 − q2y1y2 = b(x2 − 1) yp 1 = yp 2 + a(xp − 1). Pick α and β such that Then and αβ(1 − q2) = −b, βp − αp = a. (y1 + αx)p = (y2 + βx)p − a (y2 + βx)(y1 + αx) − q2(y1 + αx)(y2 + βx) = −b. Thus the subalgebra Y generated by y1 + αx and y2 + βx has GK-dimension at most one. Since Y is a domain, it is commutative by [GZ, Lemma 4.5]. So (y2 + βx)(y1 + αx) = (q2 − 1)−1b or y2y1 + αq2xy2 + βxy1 + αβx2 = c where c = (q1 − 1)−1b. After applying ∆, we see that {y1, y2, x, 1} are linearly dependent. This contradicts hypothesis (c) of Theorem 5.1. 2 y1y2 is a skew primitive of weight xn1+n2 and in C0. (b) By part (d) y2y1 − qn1 So we have y2y1 − qn1 2 y1y2 = a(xn1+n2 − 1) for some a ∈ k. If a 6= 0, by commuting with x, the above equation implies that q1q2 = 1. By part (a), n1 6= n2, by symmetry, we may assume n2 > n1 ≥ 1. Then together with qn2 = 1. Since q1 is a p1st root of unity and p1 = n2ǫ, n2ǫ divides n2 − n1. Since gcd(n1, n2) = 1, we obtain n2 = 1, a contradiction. Therefore a = 0 and the assertion follows. 2 = 1 we have qn2−n1 1 qn1 (cid:3) 1 We are now ready to prove Theorem 5.1. Proof of Theorem 5.1. Suppose H is not isomorphic to any of N5, N7, N10 and N21. First we claim that qn2 2 = 1. By Lemma 5.4(b), 1 = qn1 y2y1 − qn1 2 y1y2 = 0. Now we see that all relations of K := K({p1, p2},{q1, q2},{α1, α2}, M ) (where M = p1n1) as listed (I2.1.1)-(I2.1.6) are satisfied by H. Then H is isomorphic to a quotient Hopf algebra of K. Since H is a domain, by Lemma 2.3(b), qni j = 1. Since qni j = 1 for all i 6= j, pj ni for i 6= j. Thus gcd(p1, p2) = gcd(n2, n1) = 1. Hence K is in fact the algebra B(1,{pi}2 1) by Proposition 2.4 and 1, q,{αi}2 Convention 2.5. So we have a surjective algebra map from B(1,{pi}2 1) to H between two Hopf domains of GK-dimension two. This map must be an isomorphism. (cid:3) 1, q,{αi}2 28 D.-G. WANG, J.J. ZHANG AND G. ZHUANG 6. Proof of Theorem 0.1 and Corollary 0.2 In this final section we put together a proof of Theorem 0.1 and Corollary 0.2. Definition 6.1. A Hopf algebra H is said to satisfy the hypothesis Ω if it does not contain any of N5, N7, N10 and N21 as a Hopf subalgebra. 1, q,{αi}s 1). Theorem 6.2. Let H be a pointed Hopf domain of GK-dimension strictly less than three. Suppose that H is finitely generated by grouplike and skew primitive elements and that H satisfies Ω. If the coradical C0 has GK-dimension one, then H is isomorphic to either kZ, or one of algebras in Theorem 1.4(III,IV,V) or the algebra B(n,{pi}s Proof. If H is not PI, the assertion follows from Corollary 1.12. So we may assume H is PI. Then GKdim H is an integer, either 1 or 2. If GKdim H = 1, the assertion follows from [GZ, Proposition 2.1] together with Lemmas 1.6 and 1.10. It remains to consider the case GKdim H = 2 and H is PI. By Lemma 1.10, C0 is affine. Since C0 = kΓ where Γ is an abelian torsionfree group rank 1, Γ is isomorphic to Z. Let x be a generator of C0. By Lemma 3.3(c), dim P ′M = 1. Note that H is generated by C0 and preimages of P ′T . Let s = dim P ′T − dim P ′M . By Theorem 3.9, s is finite. If s = 0, then only nontrivial skew primitive element z is in P ′M . So H is gener- ated by x±1, z. Suppose ω(z) = (xM , λ) and without loss of generality M ≥ 0. If λ 6= 1, then we may further assume x−1zx = qz by Lemma 3.8 and λ = qM . There- fore there is a Hopf surjective map A(M, q) → H. This is an isomorphism since both algebras are domains of GK-dimension two. If λ = 1, then either x−1zx = qz or x−1zx = z + (xM − 1). We have already dealt with the first case, and the second case forces H to isomorphic to C(M − 1) defined in Example 1.3, which is non-PI. Note that when s = 0, H is non-PI. Hence, under the hypothesis that H is PI (see the first paragraph of the proof), s > 0. If s = 1, let y be a nontrivial non-major skew primitive with ω(y) = (xn, λ) for some n > 0. Then λ is a pth primitive root of unity for some p ≥ 2 and yp is a nontrivial major skew primitive. So H is generated by x±1 and y. An argument similar to above shows that H ∼= A(n, q) for some q ∈ k×. Suppose now s ≥ 2. Pick any two linearly independent nontrivial non-major skew primitive elements, say y1, y2, with ω(yi) = (xni , λi). Let X = xn where n = gcd(n1, n2) and let K be the Hopf subalgebra generated by X±1, y1 and y2. Since λi 6= 1, K is noncommutative. Therefore GKdim K = 2 by [GZ, Lemma 4.5]. By Lemma 1.10 and its proof, the coradical C0(K) is generated by X±1. In K, ω(yi) = (X n′ i, λi) where n′i = ni/n. Thus gcd(n′1, n′2) = 1. Therefore K satisfies all hypotheses in Theorem 5.1. As a consequence, y1y2 = y2y1. Other consequences are (I6.2.1) any major skew primitive element is generated by C0 and y1, (I6.2.2) both n1 and n2 are positive, (I6.2.3) n1p1 = n2p2 and gcd(p1, p2) = 1 where pi is the order of λi, (I6.2.3) yp2 1 is a major skew primitive of weight xM . Re-cycling the notations ni etc. For each nontrivial non-major Pxni ,λi,∗, there is a yi ∈ Pxni ,λi,∗ \ C0, unique up to a scalar multiple by Lemma 3.8(a), such that (I6.2.4) Then λi = qni i . By (I6.2.1), H is generated by x±1, y1,··· , ys, and x−1yix = qiyi. 2 = yp1 1 + α2(xM − 1), where yp1 HOPF ALGEBRAS OF GK-DIMENSION TWO 29 1 = qpi i = yp1 j = 1. i = qni 1 + αi(xM − 1), where αi ∈ k. (a) every ni is positive, (b) n1p1 = n2p2 = ··· = nsps =: M where pi is the order of λi, (c) ypi By commuting x with the equation in (c) above, one sees that qp1 i . Note that ∆(yi) = yi ⊗ 1 + xni ⊗ yi. Since yiyj = yjyi, expanding ∆(yiyj) = ∆(yjyi) shows that qnj If all αi = 0, consider the subalgebra Y generated by yi, which is a commutative algebra k[y1,··· , ys]/(ypi 1 i ≥ 2). By the proof of [GZ, Construction 1.2], Y is isomorphic to a subalgebra of k[y, y−1] by identifying yi with ymi. Similar to the proof of Proposition 2.4, one see that qi = qmi and that H = Y [x, x−1; σ] for some graded algebra automorphism σ of Y . As a consequence, H/[H, H] = k[x±1]. By [GZ, Theorem 3.8(c)], (♮) holds. By Theorem 1.4, H is isomorphic to the algebra B(n, p0, . . . , ps, q) defined in Example 1.2. i = yp1 If some αi 6= 0 (and assume α1 = 0), then qp1 i = 1 for all i. Thus qi is also a pith primitive root of unity. Then all conditions (I2.0.1)-(I2.0.8) are verified and all relations (I2.0.9)-(I2.0.12) hold in H. By Proposition 2.4 and Convention 2.5, the algebra with relations (I2.0.9)-(I2.0.12) is B(n,{pi}s 1). Thus there is a Hopf algebra surjective map from B(n,{pi}s 1) → H which must be an isomorphism since both algebras are domains of GK-dimension two. Therefore the assertion follows. (cid:3) 1, q,{αi}s 1, q,{αi}s 1 = qpi Proof of Theorem 0.1 and Corollary 0.2. Since all algebras in Theorem 1.4 satisfy the condition Ext1 H (k, k) 6= 0, Theorem 0.1 and Corollary 0.2 are equivalent. We will basically prove Corollary 0.2. If GKdim C0 = 0, the assertion follows from Theorem 1.9. If GKdim C0 = 2, then the assertion follows from Theorem 1.7. As the discussion given in subsection 1.2, GKdim C0 is either 0, 1, or 2. It remains to deal with the case that GKdim C0 = 1. It is clear that the hy- pothesis Ω follows from the hypothesis Ω′. Hence we may apply Theorem 6.2. By Theorem 6.2 and by the fact GKdim H = 2, H is isomorphic to one of algebras in Theorem 1.4(III,IV,V) or the algebra B(n,{pi}s 1). When H is isomorphic to B(n,{pi}s H (k, k) = 0, Lemma 2.3(c) says that αi 6= αj for some i and j. This finishes the proof. 1, q,{αi}s 1, q,{αi}s 1) and Ext1 (cid:3) References [Be] G.M. Bergman, The diamond lemma for ring theory, Adv. in Math. 29 (2) (1978) 178 -- 218. [BZ] K.A. Brown and J.J. Zhang, Prime regular Hopf algebras of GK-dimension one, Proc. London Math. Soc. (3) 101 (2010) 260 -- 302. [GZ] K.R. Goodearl and J.J. Zhang, Noetherian Hopf algebra domains of Gelfand-Kirillov dimen- sion two, J. Algebra, 324 (2010) 3131-3168. [He] I. Heckenberger, Rank 2 Nichols algebras with finite arithmetic root system, Algebr. Repre- sent. Theory 11 (2008), no. 2, 115 -- 132. [KL] G.R. Krause and T.H. Lenagan, Growth of algebras and Gelfand-Kirillov dimension, Revised edition. Graduate Studies in Mathematics, 22. AMS, Providence, RI, 2000. [Li] G. Liu, On Noetherian affine prime regular Hopf algebras of Gelfand-Kirillov dimension 1, Proc. Amer. Math. Soc. 137 (2009), no. 3, 777 -- 785. [LL] M.E. Lorenz and M. Lorenz, On crossed products of Hopf algebras, Proc. Amer. Math. Soc. 123 (1995), no. 1, 33 -- 38. [LWZ] D.-M. Lu, Q.-S. Wu and J.J. Zhang, Homological integral of Hopf algebras, Trans. Amer. Math. Soc. 359 (2007), no. 10, 4945 -- 4975. 30 D.-G. WANG, J.J. ZHANG AND G. ZHUANG [MR] J. C. McConnell and J. C . Robson, Noncommutative Noetherian Rings, Wiley, Chichester, 1987. [Mo] S. Montgomery, Hopf Algebras and their Actions on Rings, CBMS Regional Conference Series in Mathematics, 82, Providence, RI, 1993. [Pa] D.S. Passman, The algebraic structure of group rings, Pure and Applied Mathematics. Wiley- Interscience, New York-London-Sydney, 1977. [SmZ] S.P. Smith and J.J. Zhang, A remark on Gelfand-Kirillov dimension. Proc. Amer. Math. Soc. 126 (1998), no. 2, 349 -- 352. [StZ] D. R. Stephenson and J. J. Zhang. Growth of graded noetherian rings. Proc. Amer. Math. Soc. 125 (1997), no.6, 1593-1605. [Ta1] M. Takeuchi, A correspondence between Hopf ideals and sub-Hopf algebras, Manuscripta Math., 7 (1972), 251 -- 270. [Ta2] M. Takeuchi, Free Hopf algebras generated by coalgebras, J. Math. Soc. Japan 23 (1971), 561-582. [WZZ1] D.-G. Wang, J.J. Zhang and G. Zhuang, Lower bounds of growth of Hopf algebras, preprint, (2011), arXiv:1101.1116, submitted for publication. [WZZ2] D.-G. Wang, J.J. Zhang and G. Zhuang, Primitive cohomology of Hopf algebras, in preparation (2011). [WZ] Q.-S. Wu and J.J. Zhang, Noetherian PI Hopf algebras are Gorenstein, Trans. Amer. Math. Soc. 355 (2003), no. 3, 1043 -- 1066. [Zh] G. Zhuang, Existence of Hopf subalgebras of GK-dimension two, J. Pure Appl. Algebra (to appear), preprint (2010), arXiv:1008.3604. Wang: School of Mathematical Sciences, Qufu Normal University, Qufu, Shandong 273165, P.R.China E-mail address: [email protected], [email protected] Zhang: Department of Mathematics, Box 354350, University of Washington, Seattle, Washington 98195, USA E-mail address: [email protected] Zhuang: Department of Mathematics, Box 354350, University of Washington, Seat- tle, Washington 98195, USA E-mail address: [email protected]
1606.00790	2	1606	2017-07-16T12:42:29	A classification of polynomial functions satisfying the Jacobi identity over integral domains	[ "math.RA" ]	The Jacobi identity is one of the properties that are used to define the concept of Lie algebra and in this context is closely related to associativity. In this paper we provide a complete description of all bivariate polynomials that satisfy the Jacobi identity over infinite integral domains. Although this description depends on the characteristic of the domain, it turns out that all these polynomials are of degree at most one in each indeterminate.	math.RA	math	A CLASSIFICATION OF POLYNOMIAL FUNCTIONS SATISFYING THE JACOBI IDENTITY OVER INTEGRAL DOMAINS JEAN-LUC MARICHAL AND PIERRE MATHONET Abstract. The Jacobi identity is one of the properties that are used to define the concept of Lie algebra and in this context is closely related to associativity. In this paper we provide a complete description of all bivariate polynomials that satisfy the Jacobi identity over infinite integral domains. Although this description depends on the characteristic of the domain, it turns out that all these polynomials are of degree at most one in each indeterminate. 1. Introduction Let R be an infinite integral domain with identity. In this paper we are interested in a classification of all bivariate polynomials P over R satisfying Jacobi's identity (1) P (P (x, y), z) + P (P (y, z), x) + P (P (z, x), y) = 0. To give a simple example, consider the set R = Z3[x] of univariate polynomials whose coefficients are in Z3. One can easily verify that the bivariate polynomial P over Z3[x] defined by P(A, B) = (1 − x2)AB +(x + 1)(1 − x2)(A + B) + x(x + 1)(1 − x − x2) satisfies Jacobi's identity (1). As it is well known, the Jacobi identity is one of the defining properties of Lie algebras. Recall that a Lie algebra (see, e.g., [4–6]) is a vector space g together with a binary map [⋅ , ⋅]∶ g × g → g, called Lie bracket, such that 1. [⋅ , ⋅] is bilinear, 2. [x, y] = −[y, x] for all x, y ∈ g. 3. [[x, y], z] +[[y, z], x] +[[z, x], y] = 0 for all x, y, z ∈ g. The second condition is usually called skew-symmetry while the third one is known as the Jacobi identity. By using a prefix notation for the Lie bracket, the Jacobi identity simply becomes the functional equation given in (1). The classical associativity property is closely connected to Lie algebras in the following way (see, e.g., [6, p. 6]). The Lie bracket defined by [x, y] = xy − yx on any associative algebra satisfies the three properties above, including Jacobi's identity. This is one of the reasons why "the Jacobi identity can be viewed as a substitute for associativity" [5, p. 54]. We now state our main result, which provides a complete description of the possible polynomial solutions over R of Jacobi's identity (1). Although the form Date: March 10, 2017. 2010 Mathematics Subject Classification. Primary 39B72; Secondary 13B25, 17B99. Key words and phrases. Jacobi's identity, polynomial, integral domain. 1 2 JEAN-LUC MARICHAL AND PIERRE MATHONET of these polynomials depends on the characteristic of R, they are all of degree at most one in each indeterminate. Main Theorem. Consider a bivariate polynomial P ∈ R[x, y]. ● If char(R) ≠ 3, then P satisfies Jacobi's identity iff there exist B, C ∈ R satisfying B2 + BC + C = 0 such that P(x, y) = Bx + Cy, ● If char(R) = 3, then P satisfies Jacobi's identity iff one of the following conditions holds: – there exist A, B, D ∈ R satisfying AD = B2 − B such that P(x, y) = Axy + B(x + y) + D, – there exist B, C, D ∈ R satisfying B2 + BC + C = 0 such that P(x, y) = Bx + Cy + D. The reader interested in possible generalizations of the Main Theorem might want to consider extensions of functional equation (1) to n-indeterminate polyno- mials, by analogy with n-ary generalizations of Lie algebras, where Jacobi's identity involves an n-linear bracket. In this direction we remark that a complete classifi- cation of n-ary associative polynomials over R can be found in [10] and that, in the special case when R is the complex plane C, this classification was recently generalized to n-ary associative formal power series in [3]. Remark. In the literature on Lie algebras the Jacobi identity is sometimes given in one of the following alternative forms (which are equivalent to the one above under bilinearity and skew-symmetry): (2) (3) (4) [x,[y, z]] +[y,[z, x]] +[z,[x, y]] = 0, [x,[y, z]] = [[x, y], z] +[y,[x, z]], [[x, y], z] = [x,[y, z]] +[[x, z], y]. It is however easy to see that P satisfies the functional equation corresponding to (2) iff the polynomial P ′ defined by P ′(x, y) = P(y, x) satisfies (1). As far as equations (3) and (4) are concerned, one can show that the corresponding functional equations have no nonzero solution. The proof of this latter observation is given in Appendix B. Note. The problem addressed in this paper was suggested by Jorg Tomaschek [11], who in turn was asked this question by Wolfgang Prager (University of Graz, Aus- tria) while the latter was studying local analytic solutions of the Bokov functional equation that appears in theoretical physics (see [1, 12]). 2. Technicalities and proof of the Main Theorem We use the following notation throughout this paper. For any integer m ≥ 1 and any prime p ≥ 2, we denote by sm(p) the set of positive integers expressible as sums of m powers of p, that is, integers n whose base p expansions n = ∑k i=0 ni pi (with 0 ≤ ni < p for i = 0, . . . , k) satisfy ∑k i=0 ni = m. We also use the Kronecker delta symbol: δi,j = 1, if i = j, and δi,j = 0, if i ≠ j. For any bivariate polynomial P = P(x, y), we let deg(P) denote the degree of P , that is, the highest degree of the homogeneous terms of P in both variables. We also let deg1(P) (resp. deg2(P)) A CLASSIFICATION OF POLYNOMIAL FUNCTIONS 3 denote the degree of P in its first (resp. second) variable. For any nonnegative component of degree k of P , that is, the polynomial obtained from P by considering integer k ≤ deg(P), unless otherwise stated we let Pk denote the homogeneous the terms of degree k only. For any monomial M of P , we let [M]P(x, y) denote the coefficient of M in P(x, y) (we let [M]P(x, y) = 0 if M is not a monomial of P ), and similarly for polynomials in more than two indeterminates. Finally, we define the following trivariate polynomial JP(x, y, z) = P(P(x, y), z) + P(P(y, z), x) + P(P(z, x), y). polynomials in n indeterminates over R with the ring of polynomial functions from Recall that the definition of R enables us to identify the ring R[x1, . . . , xn] of Rn to R. Recall also that if char(R) = p > 0, then p must be prime. In this case we have (x + y)p = xp + yp for any x, y ∈ R and this identity (often referred to as the freshman's dream) immediately extends to any sum of more than two terms. In this paper we will often make use of the following theorem, established in 1878 by E. Lucas [7–9]. For a more recent reference, see [2]. Theorem 1 (Lucas' theorem). For any integers n, m ≥ 0 and any prime p ≥ 2, the following congruence relation holds: n m ≡ i=0 nipi and m = ∑k k M (mod p), i=0 ni mi i=0 mipi are the base p expansions of n and m, b = 0 for any integers a, b such that where n = ∑k respectively. This uses the convention that a 0 ≤ a < b. Corollary 2. For any integer n > 1 and any prime p, the following two conditions are equivalent. (i) n ∈ s1(p). (ii) p divides n m for any integer m such that 0 < m < n. Moreover, if char(R) is a prime p, then any of these conditions holds iff (x + y)n = xn + yn for any x, y ∈ R. Proof. (i) ⇒ (ii). This implication immediately follows from Lucas' theorem. (ii) ⇒ (i). We prove this implication by contradiction. Suppose n ∉ s1(p). Let i=0 nipi be the base p expansion of n, let j ∈ {0, . . . , k} such that nj ≠ 0, n = ∑k and let m = n − pj. Then we have 0 < m < n and by Lucas' theorem we also have n m ≡ nj (mod p). This means that p does not divide n m, which is a contradiction. The second part of the corollary is straightforward. (cid:3) Corollary 3. Let n > 1 be an integer and let p be a prime. (a) If n = n1 + n2 ∈ s2(p) for some n1, n2 ∈ s1(p) (with n1, n2 distinct if p = 2), then ● p divides n m for any integer m ∈ {1, . . . , n − 1} ∖{n1, n2}. ● n n1 ≡ n n2 ≡ (1 + δn1,n2) (mod p). (b) If p divides n mm ℓ for any integers ℓ, m such that 0 < ℓ < m < n, then n ∈ s1(p) ∪ s2(p). Proof. Assertion (a) is a straightforward consequence of Lucas' theorem. To show that assertion (b) holds, we first proceed as in the proof of the implication (ii) ⇒ (i) 4 JEAN-LUC MARICHAL AND PIERRE MATHONET of Corollary 2. Suppose n ∉ s1(p)∪ s2(p). Let n = ∑k i=0 nipi be the base p expansion of n, let j ∈ {0, . . . , k} such that nj ≠ 0, and let m = n − pj. Then we have 0 < m < n and n m ~≡ 0 (mod p). Since n ∉ s2(p) we must have m ∉ s1(p) and we conclude the proof by applying Corollary 2. (cid:3) → R be a polynomial function satisfying Jacobi's identity (1), that is, such that JP = 0. We now prove the Main Theorem. Let P ∶ R2 Suppose first that deg2(P) = 0, that is, P(x, y) = P(x). Using Jacobi's identity, we obtain that P(P(x)) is a constant, and hence P is a constant C satisfying 3C = 0. Therefore, C can be any constant if char(R) = 3, and C = 0, otherwise. Thus, we shall henceforth assume that deg2(P) ≥ 1. Proposition 4. If P ∶ R2 deg2(P) ≥ 1, then deg1(P) ≤ 1. → R is a polynomial function satisfying JP = 0 and We prove Proposition 4 by contradiction. Thus we suppose that deg1(P) = d ≥ 2. Claim 1. We have deg2(P) = deg(P) = d. Moreover, the polynomial function P is of the form (5) P(x, y) = d Q k=0 k Q j=0 ck,j xjyk−j with cd,d cd,0 ≠ 0, cd d,d + cd d,0 = 0, and (6) Pd(x, y)d = cd d,d(xd2 − yd2). Proof of Claim 1. In this proof we use the notation [xk]xJP(x, y, z) to denote the coefficient of xk in the expansion of JP in powers of x. Set d2 = deg2(P) ≥ 1. Then there exist polynomial functions Rj∶ R → R (j = 0, . . . , d) and Sk∶ R → R (k = 0, . . . , d2), with Rd ≠ 0 and Sd2 ≠ 0, such that P(x, y) = d Q j=0 xj Rj(y) = d2 Q k=0 ykSk(x). We then have JP(x, y, z) = d Q j=0 d Q k=0 xkRk(y)j Rj(z)+ d2 Q k=0 xkSk(P(y, z))+ d Q j=0 d2 Q k=0 xkSk(z)j Rj(y). Now, if d > d2, then [xd2]xJP(x, y, z) = Rd(y)d Rd(z) from which we derive Rd = 0, a contradiction. Similarly, if d < d2, then [xd d2]xJP(x, y, z) = Sd2(z)d Rd(y) and hence we obtain Rd = 0 or Sd2 = 0, again a contradiction. Thus we have proved that d = d2. It then follows that [xd2]xJP(x, y, z) = Rd(y)Rd(y)d−1Rd(z) + Sd(z)d and hence (7) Rd(y)d−1Rd(z) + Sd(z)d = 0. A CLASSIFICATION OF POLYNOMIAL FUNCTIONS 5 Since d ≥ 2, from identity (7) it follows that both Rd and Sd are nonzero constant polynomial functions. Thus the polynomial function P is of the form P(x, y) = d Q j=0 d Q k=0 pj,k xjyk, 0,d = 0. with pd,0 p0,d ≠ 0 and pd,k = pj,d = 0 for j, k = 1, . . . , d. d,0 + pd pd Now, let r = deg(P) ≥ d and let M be an arbitrary monomial of JP of degree rd in (x, y) and degree 0 in z (e.g., M = xiyrd−i for some i ∈ {0, . . . , rd}). We then Identity (7) also implies have [M] P(P(x, y), z) = [M] d Q j=0 pj,0 P(x, y)j = [M] pd,0 Pr(x, y)d and [M] P(P(y, z), x) = [M] d Q j=0 = [M]pd,0P(y, 0)d + p0,d xd + d Q k=0 pj,k P(y, 0)jxk d−1 Q j=0 d−1 Q k=0 pj,k P(y, 0)jxk = [M] δr,d pd+1 d,0 yd2 Indeed, for j, k = 0, . . . , d−1, P(y, 0)jxk is of degree jd+k ≤ (d−1)d+(d−1) < d2 ≤ rd. We show similarly that [M] P(P(z, x), y) = [M] d Q j=0 Let us now show by contradiction that deg(P) = d. Suppose that r > d. By [M] JP(x, y, z) = [M] pd,0 Pr(x, y)d, a contradiction. We then have deg(P) = d. combining the latter three identities with JP = 0 we immediately obtain 0 = pj,k P(0, x)jyk = [M] δr,d pd,0 pd 0,d xd2 d Q k=0 . Using the same three identities for r = d, we obtain Pd(x, y)d + pd d,0 yd2 + pd 0,d xd2 = 0. Finally, since deg(P) = d, the polynomial function P must be of the form (5), with cd,d = pd,0 and cd,0 = p0,d. Therefore the identities cd,d cd,0 ≠ 0, cd d,0 = 0, and (6) hold. (cid:3) d,d + cd case. We now show that char(R) must be a prime number. This shows that a contra- diction is already reached if char(R) = 0, which then proves Proposition 4 in this Claim 2. The characteristic of R is a prime p and we have d ∈ s1(p). Moreover, we have Pd(x, y) = cd,d(xd − yd) and (x + y)d = xd + yd for any x, y ∈ R. Proof of Claim 2. By Claim 1 we have deg(P) = d and [yd]P(x, y) = cd,0 ≠ 0. Then we have (8) Pd(x, y) = cd,d xd + r Q j=0 cd,j xjyd−j for some integer 0 ≤ r ≤ d − 1, with cd,r ≠ 0. Equation (6) then becomes (9) d,d yd2 + cd d−1 Q k=0d k(cd,d xd)k r Q j=0 cd,j xjyd−jd−k = 0. 6 JEAN-LUC MARICHAL AND PIERRE MATHONET Clearly, the literal part of the monomial of highest degree in x in the left-hand side of (9) is xd(d−1)+ryd−r. Indeed, it corresponds to the values k = d − 1 and j = r in the sums and therefore has the coefficient d cd−1 d,d cd,r. Since cd,d cd,r ≠ 0, we must have d 1 = 0 (here the symbol 1 denotes the identity of R). If follows that the characteristic of R should be a prime p ≥ 2 that divides d. We now show by contradiction that d ∈ s1(p). Suppose that d ∉ s1(p). Then by k ~≡ 0 (mod p). Corollary 2 we can let m be the greatest k ∈ {1, . . . , d−1} such that d Equation (9) then reduces to (10) d,d yd2 + cd m Q k=0d k(cd,d xd)k r Q j=0 cd,j xjyd−jd−k = 0. The literal part of the monomial of highest degree in x in the left-hand side of (10) is xmd+r(d−m) y(d−r)(d−m). It corresponds to the values k = m and j = r and therefore has the coefficient d d,r ≠ 0, which leads to a contradiction. Therefore d ∈ s1(p) and hence by Corollary 2 we have (x + y)d = xd + yd for any x, y ∈ R. m cm d,d cd−m d,d + cd d,0 = 0 and hence cd,d + cd,0 = 0. Now, by Claim 1 we have (cd,d + cd,0)d = cd By Corollary 2 the identity (9) then reduces to d,d yd2 + cd r Q j=0 d,j xdjyd(d−j) = 0 , cd which implies r = 0. Using (8) we finally obtain Pd(x, y) = cd,d(xd − yd). if m ∈ s1(p) then am ≡ a (mod p) for every integer a. Remark. From now on we will often make an implicit use of Fermat's little theorem: (cid:3) We will now show (through Claims 3–6) that for every integer k such that 1 < k ≤ d the polynomial function Pk is of one of the following three forms. ● Type 0: Pk = 0. ● Type 1: Pk ≠ 0, k ∈ s1(p), and Pk(x, y) = ck,k(xk − yk). ● Type 2: Pk ≠ 0, k = k1 + k2 ∈ s2(p), with k1 ≥ k2 and k1, k2 ∈ s1(p), and Pk(x, y) = ck,k xk + ck,k1 1 + δk1,k2 (xk1 yk2 + xk2 yk1) + ck,0 yk. Note: This latter form simply means that ck,j = 0 whenever j ∉ {k, k1, k2, 0} and that ck,k1 = ck,k2 . For every real r ≥ 0 and every m ∈ {0, 1, 2} we let Sm,r = {k integer r < k ≤ d and Pk is of type m}. It is clear that the sets S0,r, S1,r, S2,r are pairwise disjoint. Moreover, if r ≤ r′ ≤ d, then we have Sm,r ⊇ Sm,r ′ ⊇ Sm,d = ∅. By Claim 2 we have d = sup S1,1. Regarding S2,1 we have two cases to consider. ● If S2,1 = ∅, then we set r0 = 1. ● If S2,1 ≠ ∅, then we set q = q1 + q2 = sup S2,1, with q1 ≥ q2 and q1, q2 ∈ s1(p). We also set r0 = q1 + q2q d . We then have 1 ≤ q1 < r0 < q < d. Note that r0 is 2(1 + q1~q2). But p does not divide (1 + q1~q2) an integer iff d divides q2q = q2 since q1~q2 ∈ s1(p). Hence r0 is an integer iff d divides q2 iff d ≤ q2 2. In this case we also have d ≤ q1q2 and hence d divides q1q2. 2, or equivalently, A CLASSIFICATION OF POLYNOMIAL FUNCTIONS 7 Note that if S2,r ≠ ∅ for some r ≥ 1, then clearly S2,1 ≠ ∅ and r < q. Remark. In all the equations that we will now consider, some expressions are asso- ciated with polynomial functions Pk for which k ∈ S2,1 (e.g., expressions involving q, q1, and q2). The proofs corresponding to those equations show that these ex- pressions are to be ignored when S2,1 = ∅. For every real r ≥ 1 we set αr = Q ck,k ck ℓ,ℓ , βr = Q k∈S2,r Q a,b∈S1,r a,b>q, ak1+bk2=rd k∈S1,r ∖{d} ℓ∈S1,r, kℓ=rd a,a ck2 ck,k ck1 b,b , γr = cr,r cr d,d + βr. If r is an integer, then we easily see that αr = 0 if r ∉ s1(p), and βr = 0 if r ∉ 2 ∈ s3(p) and hence s1(p) ∪ s2(p). If r = r0 is not an integer, then rd = q1d + q1q2 + q2 αr = βr = 0. Since cr,r is to be ignored in this case, we also have γr = 0. The proofs of the following two claims (Claims 3 and 4) are rather technical. For this reason we relegate them to Appendix A. (11) integers such that 1 ≤ i < u < r, then Claim 3. Let r ∈ {⌈r0⌉, . . . , d − 1} be such that {r + 1, . . . , d} ⊆ ⋃2 d }(i) cq2+1 i cr,ucu Here, χ{j,k}(i) = max{δi,j, δi,k}. Claim 4. Let r ∈ {⌈r0⌉, . . . , d − 1} ∪{r0} be such that {⌊r⌋ + 1, . . . , d} ⊆ ⋃2 (−1)u−iu d,d + δr,r0 δu, q,q1 = 0. { q1 q2 d , q2 2 χ qq2 d Then the following two conditions hold. m=0 Sm,r. If i, u are m=0 Sm,r. ● If either r, u are integers such that 1 ≤ u < r, or r = r0 is not an integer and u = q2q cd,d cd d , then r,u +(−1)r−ur + δr,r0(1 + δq1,q2)cq,q [xduyd(r−u)](Pd(x, y)q1 Pq(x, y)q2) = 0, uγr (12) where the first two summands are to be ignored when r = r0 is not an integer. ● If r is an integer, then (13) cd,d(cd d,d(cq,q − cq,0)q2 − δr,1cd,d = 0. m=0 Sm,1. r,0) +(1 +(−1)r)γr r,r + cd + δr,r0(1 + δq1,q2)cq,q1 cq1 Claim 5. We have {⌊r0⌋ + 1, . . . , d} ⊆ ⋃2 Proof of Claim 5. We prove by decreasing induction that any integer k ∈ {⌊r0⌋ + 1, . . . , d} is in ⋃2 m=0 Sm,1. This is true for k = d since d ∈ S1,1. Suppose that the result holds for k = r + 1, . . . , d for some integer r such that r0 < r < d and let us show that it holds for k = r. There are three mutually exclusive cases to consider. ● If r ∉ s1(p) ∪ s2(p), then βr = 0 and by Corollary 3(b) there exists integers i0, u0 satisfying 1 ≤ i0 < u0 < r such that r i0 ~≡ 0 (mod p). Using (11) u0u0 with i = i0 and u = u0 we immediately obtain cr,u0 = 0. Then, using (12) with u = u0 we obtain γr = 0, which implies cr,r = 0 (since βr = 0). Using 8 JEAN-LUC MARICHAL AND PIERRE MATHONET again (12) we obtain that cr,u = 0 for every integer u such that 1 ≤ u < r. Finally, by (13) we obtain cr,0 = 0 and hence Pr = 0, that is, r ∈ S0,1. ● If r ∈ s1(p), then by Corollary 2 we have r u ≡ 0 (mod p) for every integer u such that 1 ≤ u < r. Using (12) we then obtain cr,u = 0 for every integer In (13) we have (−1)r ≡ −1 (mod p) and hence u such that 1 ≤ u < r. r,0 = (cr,r + cr,0)d. Therefore Pr is of type 0 or 1, that is, r,r + cd 0 = cd r ∈ S0,1 ∪ S1,1. ● If r = r1 + r2 ∈ s2(p), with r1 ≥ r2 and r1, r2 ∈ s1(p), then by Corollary 3(a) we have r u ≡ 0 (mod p) for every integer u ∈ {1, . . . , r − 1} ∖{r1, r2}. Using (12) we obtain cr,u = 0 for every integer u ∈ {1, . . . , r − 1} ∖{r1, r2}. Now, if r1 ≠ r2, then using (12) for u = r1 and then for u = r2, we obtain cr,r1 = cr,r2. Therefore, Pr is of type 0 or 2, that is, r ∈ S0,1 ∪ S2,1. This completes the proof of the claim. (cid:3) We now show that {2, . . . , d} ⊆ S0,1∪S1,1 (i.e., Pk is of type 0 or 1 for k = 2, . . . , d). Claim 6. We have r0 = 1 (i.e., S2,1 = ∅). Proof of Claim 6. We proceed by contradiction. Suppose that r0 > 1, that is, S2,1 ≠ ∅ and r0 = q1 + q2q d . Using (12) with r = r0 and u = u0 = q2q d , we obtain (14) cd,d cd r0,u0 +(−1)q1r0 u0γr0 −(1 + δq1,q2) cq1 d,d cq2+1 q,q = 0. Setting r = q and u = q1 in (12) and (13), we obtain (15) and (16) cd,d cd q,q1 = (1 + δq1,q2) cq,q cq d,d cd,d(cd q,q + cd q,0) + 2cq,q cq d,d = 0. Indeed, δq,r0 = 0 and since S2,q = ∅ we have βq = 0 and hence γq = cq,qcq by Corollary 3(a) we have q q1 ≡ (1 + δq1,q2) (mod p). Now we have two cases to consider. d,d. Moreover, ● If r0 is not an integer, then the first two summands of (14) are to be ignored and hence we immediately derive cq,q = 0. Then from (15) and (16) we derive cq,0 = cq,q1 = 0, that is, Pq = 0 (i.e., q ∈ S0,1), a contradiction. ● If r0 is an integer (in which case d divides both q1q2 and q2 i ≡ ing (11) with r = r0, u = u0 = q2q (1 + δq1,q2) (mod p) by Corollary 3(a)) and then raising both sides of the (we note that u0 d , and i = q1q2 2), then us- d resulting equation to the power d we obtain (17) cd(q2+1) q,q1 = (1 + δq1,q2) cd r0,u0 cdu0 d,d . Raising both sides of (15) to the power (q2 + 1) and then combining the resulting equation with (17) we obtain d,d cq2+1 q,q . r0,u0 = (1 + δq1,q2) cq1−1 cd r0,u0 into (14) and observing by Lucas' theorem that r0 Substituting for cd 1 (mod p) we obtain γr0 = 0. Now, using (11) with r = r0, u = u0 = q1 + q1q2 u0 q1 cr0,u0 cu0 d,d = 0, and therefore cr0,u0 = 0 (since u0 u0 ≡ d , and i = q1, we obtain q1 ≡ 1 (mod p) by A CLASSIFICATION OF POLYNOMIAL FUNCTIONS 9 Corollary 3(a)). Using (12) with the same r = r0 and u = u0, we then obtain cq,q [xq1 d+q1q2 yq2 q,q1 cq1 2](Pd(x, y)q1 Pq(x, y)q2) = 0, that is, cq,q cq2 we obtain cq,q = cq,0 = cq,q1 = 0, that is, Pq = 0, a contradiction. d,d = 0. Combining this latter equation with (15) and (16) This completes the proof of the claim. (cid:3) Proof of Proposition 4. On the one hand, using (13) with r = r0 = 1 and the fact that S2,1 = ∅ (i.e., q does not exist), we obtain 0 = cd,d(cd 1,1 + cd 1,0 − 1d) = cd,d(c1,1 + c1,0 − 1)d, that is, (18) c1,1 + c1,0 = 1. On the other hand, by Claims 5 and 6 for any M ∈ {x, y, z} we have [M]P(P(x, y), z) = [M] Q = [M] Q Pk(P(x, y), z) + P1(P(x, y), z) + P0 ck,k(P(x, y)k − zk) + c1,1P(x, y) + c1,0z + c0,0. k∈S1,1 k∈S1,1 Clearly, the sum over k ∈ S1,1 above cannot contain monomials of degree 1. There- fore we have [M]P(P(x, y), z) = [M](c1,1P1(x, y) + c1,0z) = [M](c1,1(c1,1x + c1,0y) + c1,0z). Since the identity [x]JP(x, y, z) = 0 can be written as ∑M∈{x,y,z}[M]P(P(x, y), z) = 0, we have (19) c1,1(c1,1 + c1,0) + c1,0 = 0. Since the system (18)–(19) is inconsistent we immediately reach a contradiction. (cid:3) Proof of the Main Theorem. By Proposition 4, there exist two polynomial func- tions R∶ R → R and S∶ R → R such that We then have P(x, y) = x R(y) + S(y). JP(x, y, z) = x R(y)R(z) + S(y)R(z) + S(z) + y R(z)R(x) + S(z)R(x) + S(x) + z R(x)R(y) + S(x)R(y) + S(y). Suppose that deg(R) = r > 1 and set A = [yr]R(y). We can then readily see that [x yrzr]JP(x, y, z) = A2. We then have A = 0, a contradiction. Therefore R(y) = A1y + A0 for some A1, A0 ∈ R. Now, suppose that deg(S) = s > 1 and set B = [ys]S(y). It is then easy to see that [ys]JP(x, y, z) = (A0 + 1)B and [ysz]JP(x, y, z) = A1B. However, one can readily see that P cannot satisfy Jacobi's identity if A1 = 0 and A0 = −1. Thus we must have B = 0, again a contradiction. 10 JEAN-LUC MARICHAL AND PIERRE MATHONET Finally, the polynomial P must be of the form P(x, y) = Axy + Bx + Cy + D for some A, B, C, D ∈ R and we can immediately verify that this polynomial satisfies Jacobi's identity iff 3A2 = 3D(B + 1) = A(2B + C) = B2 + BC + C + AD = 0. The statement of the Main Theorem then follows straightforwardly. (cid:3) Appendix A. Proofs of Claims 3 and 4 Before providing the proofs of Claims 3 and 4, we first show that for any r ≥ r0 and any k = k1 + k2 ∈ S2,r, with k1 ≥ k2 and k1, k2 ∈ s1(p), the following conditions hold. (a) k1 = q1 and k2 ≤ q2. (b) d(r − k1) ≥ qk2. The equality holds iff r = r0 and k = q. (c) d(r − k2) ≥ qk1. The equality holds iff r = r0, k = q, and q1 = q2. (d) ak1 + bk2 ≤ rd for all a ≤ d and b ≤ q. The equality holds iff a = d, b = q, k2 = q2, and r = r0. (e) ak1 + bk2 ≤ rd for all a ≤ q and b ≤ d. The equality holds iff a = q, b = d, k2 = q2, q1 = q2, and r = r0. Proof. if S2,1 = ∅, then S2,r = ∅ for every r ≥ r0 = 1 and then there is nothing to prove. We therefore assume that S2,1 ≠ ∅. We then have r0 = q1 + q2q~d and q1 < r0 ≤ r < k ≤ q. (a) We have k1 = q1. Indeed, if we had k1 > q1, then we would have k > k1 ≥ pq1 ≥ 2q1 ≥ q1 + q2 = q, a contradiction. If we had k1 < q1, then we would have q1 ≥ pk1 ≥ 2k1 ≥ k1 + k2 = k, a contradiction. Finally, k ≤ q implies k2 ≤ q2. (b) We have d(r − k1) − qk2 ≥ d(r0 − q1) − qq2 = 0. (c) We have d(r − k2) − qk1 ≥ d(r0 − q2) − qq1 = (q1 − q2)(d − q) ≥ 0. (d) We have ak1 + bk2 ≤ dq1 + qq2 = r0d ≤ rd. (e) We have ak1 + bk2 ≤ qq1 + dq2 ≤ dq1 + qq2 = r0d ≤ rd, where the second (cid:3) inequality is equivalent to (q1 − q2)(d − q) ≥ 0. Proof of Claim 3. We consider the identity [M]JP(x, y, z) = 0 for M = xdiyd(u−i)zr−u. Since di ≥ d and d(u − i) ≥ d, we have [M]P(P(y, z), x) = 0 and [M]P(P(z, x), y) = 0. Also, we have [M]P(P(x, y), z) = [M] Q k∈S1,r Pk(P(x, y), z) +[M] Q k∈S2,r Pk(P(x, y), z) +[M]Pr(P(x, y), z) +[M] Q k<r Pk(P(x, y), z). Let us compute the latter four summands separately. ● We clearly have [M] Q k∈S1,r Pk(P(x, y), z) = [M] Q k∈S1,r ck,k(P(x, y)k − zk) = 0. A CLASSIFICATION OF POLYNOMIAL FUNCTIONS 11 ● Assuming that S2,r ≠ ∅ and setting M ′ = xdiyd(u−i), we obtain Pk(P(x, y), z) k∈S2,r [M] Q = [M] Q = [M ′] Q = [M ′] Q ck,k1 k∈S2,r k∈S2,r ck,k1 1 + δk1,k2 1 + δk1,k2 Pdu~k1(x, y)k1 zk2 + Pdu~k2(x, y)k2 zk1 Pdu~k1(x, y)k1 δk2,r−u + Pdu~k2(x, y)k2 δk1,r−u Pd(r−k2)~k1(x, y)k1 δk2,r−u + Pd(r−k1)~k2(x, y)k2 δk1,r−u. If d(r − k2)~k1 > q, then Pd(r−k2)~k1 is of type 0 or 1, so it does not contain any product terms and hence M ′ cannot appear in Pd(r−k2)~k1(x, y)k1 . We arrive at the same conclusion for Pd(r−k1)~k2 . Using conditions (b) and (c) above, we then obtain 1 + δk1,k2 ck,k1 k∈S2,r Pk(P(x, y), z) k∈S2,r [M] Q = [M ′] δr,r0 cq,q1 1 + δq1,q2 = [M ′] δr,r0 δu, = [M ′] δr,r0 δu, = δr,r0 δu, χ qq2 d qq2 d qq2 d δq1,q2 Pq(x, y)q1 δq2,r0−u + Pq(x, y)q2 δq1,r0−u cq,q1 Pq(x, y)q2 cq2+1 q,q1 2 + xq2 2 yq1q2 xq1q2 yq2 q,q1 . 1 + δq1,q2 d }(i) cq2+1 q2 2 { q1 q2 d , ● Since M is of degree du in (x, y), we have [M]Pr(P(x, y), z) = [M] r Q j=0 cr,j P(x, y)jzr−j = [M] cr,u P(x, y)uzr−u d,d(xd − yd)uzr−u = (−1)u−iu i cr,ucu d,d. = [M] cr,u cu ● Let us now compute [M] ∑k<r Pk(P(x, y), z). If k < r − u, the degree in z of Pk(P(x, y), z) cannot reach r − u. If r − u ≤ k < r, we have [M]Pk(P(x, y), z) = [M] ck,k−(r−u)P(x, y)k−(r−u)zr−u. This expression is 0 since the degree in (x, y) of P(x, y)k−(r−u) does not exceed d(k −(r − u)) < du. This completes the proof of the claim. (cid:3) Proof of Claim 4. We first consider the identity [M]JP(x, y, z) = 0 for the mono- mials M = xduyd(r−u) with 0 ≤ u ≤ r. These monomials are of degree rd in (x, y) and 0 in z. Thus we have [M]P(P(x, y), z) = [M] Q k∈S1,r ck,kP(x, y)k +[M] Q k∈S2,r ck,kP(x, y)k +[M] cr,rP(x, y)r +[M] Q k<r ck,kP(x, y)k. Let us compute the latter four summands separately. 12 JEAN-LUC MARICHAL AND PIERRE MATHONET ● We show that [M] ∑k∈S1,r ck,kP(x, y)k = [M](cd,dPr(x, y)d + αr(x − y)rd). Since k ∈ S1,r implies k ∈ s1(p), we have ck,kP(x, y)k = [M] Q k∈S1,r [M] Q k∈S1,r ck,kP rd k (x, y)k. We then observe that if d > k ∈ S1,r (hence d ≥ pk) and Pℓ ≠ 0, with ℓ = rd k , then necessarily ℓ ∈ S1,r. Indeed, since ℓ > r we must have ℓ ∈ S1,r ∪ S2,r by the hypotheses of the claim. If r0 = 1, then S2,r = ∅ and hence ℓ ∈ S1,r. If r0 > 1, then we have ℓ = rd k ≥ pr ≥ 2r0 > 2q1 ≥ q, and hence ℓ ∈ S1,r by definition of q. Therefore, we have [M] Q k∈S1,r ∖{d} ck,kP(x, y)k = [M] Q k∈S1,r ∖{d} ck,k Q ℓ∈S1,r, kℓ=rd ℓ,ℓ(x − y)rd ck = [M] αr(x − y)rd, which immediately gives the stated identity. ● Assuming that S2,r ≠ ∅, let us show that [M] Q k∈S2,r ck,kP(x, y)k = [M]βr(x−y)rd+δr,r0(1+δq1,q2)cq,q(Pd(x, y)q1 Pq(x, y)q2). Indeed, the left-hand side of this identity can be rewritten as [M] Q k∈S2,r d Q a,b=0 ck,kPa(x, y)k1 Pb(x, y)k2 . Since Pa(x, y)k1 Pb(x, y)k2 is a homogeneous polynomial function of degree ak1 + bk2, we can use conditions (d) and (e) above to analyze all the sum- mands corresponding to a ≤ q or b ≤ q. If a > q and b > q (hence a > r and b > r since q > r when S2,r ≠ ∅), then necessarily a, b ∈ S0,r ∪ S1,r and we obtain the stated identity. ● We have [M] cr,rP(x, y)r = [M] cr,rPd(x, y)r = [M] cr,rcr ● We have [M] ∑k<r ck,kP(x, y)k = 0 since the degree of P(x, y)k is bounded d,d(x − y)rd. by kd < rd. Summing up, we obtain [M]P(P(x, y), z) = [M]cd,dPr(x, y)d +(αr + γr)(x − y)rd + δr,r0(1 + δq1,q2)cq,q(Pd(x, y)q1 Pq(x, y)q2). (20) If r, u are integers such that 1 ≤ u < r, then M = xduyd(r−u) is a polynomial multiple of xdyd. Since no monomial in P(P(y, z), x) and P(P(z, x), y) is a poly- nomial multiple of xdyd we must have [M]JP(x, y, z) = [M]P(P(x, y), 0). We then observe that if αr ≠ 0, then r ∈ s1(p) and in this case we have [M](x − y)rd = 0 and d , then M = xq2 qyq1d and r < q. We then hence αr can be ignored in (20). We then immediately obtain (12). If r = r0 is not an integer and u = q2q have [M]P(P(y, z), x) = [M] Q k≤q Pk(P(y, z), x) +[M] Q k>q Pk(P(y, z), x), where the first summand is clearly zero. The second summand is also zero since k > q > r implies k ∈ S0,r ∪ S1,r. We show similarly that [M]P(P(z, x), y) = 0. A CLASSIFICATION OF POLYNOMIAL FUNCTIONS 13 therefore obtain (12), in which the first two summands are to be ignored. Moreover, the summands involving Pr and (αr + γr) are to be ignored in (20). We Let us now prove (13). We consider the monomial M = xrd and hence we have [M]JP(x, y, z) = [M]P(P(x, 0), 0) +[M]P(P(0, x), 0) +[M]P(P(0, 0), x). The first summand is exactly the right-hand side of (20) when u = r, that is cd,d cd r,r +(αr + γr) + δr,r0(1 + δq1,q2) cq,q cq1 d,d cq2 q,q. Similarly, the second summand is the right-hand side of (20) when u = 0, that is cd,d cd r,0 +(αr + γr)(−1)r − δr,r0(1 + δq1,q2) cq,q cq1 d,d cq2 q,0. The third summand is simply equal to δr,1cd,0 = −δr,1cd,d since d ∈ S1,r. We then conclude the proof by observing that (1 + (−1)r) αr = 0 since if αr ≠ 0 then r ∈ s1(p). (cid:3) Appendix B. Case of equations (3) and (4) The functional equations corresponding to (3) and (4) are respectively given by (21) (22) P(P(x, y), z) + P(y, P(x, z)) − P(x, P(y, z)) = 0, P(x, P(y, z)) + P(P(x, z), y) − P(P(x, y), z) = 0. It is then easy to see that P satisfies (22) iff the polynomial P ′ defined by P ′(x, y) = P(y, x) satisfies (21). → R be a polynomial function satisfying (21) and let us show Now, let P ∶ R2 Suppose that deg2(P) ≥ 1 and let us prove by contradiction that deg1(P) ≤ 1. Suppose that deg1(P) = d ≥ 2. By using the notation of the proof of Claim 1, we that necessarily P = 0. see that (21) can be rewritten as d Q j=0 d Q k=0 xkRk(y)j Rj(z) + d2 Q k=0 d Q j=0 xj Rj(z)k Sk(y) − d Q j=0 xjRj(P(y, z)) = 0. If d > d2 (resp. d < d2), then by equating the coefficients of xd2 (resp. xdd2) in the expansion in powers of x of each side of the latter equation, we obtain a contra- diction. Therefore, we have d = d2. By equating the coefficients of xd2 we then obtain Rd(y)d + Rd(z)d−1Sd(y) = 0, which shows that both Rd and Sd are nonzero constant polynomial functions. Now, by identifying x and y in (21), we obtain (23) or equivalently, P(P(x, x), z) = 0, d Q k=0 zkSk(P(x, x)) = 0. By equating the coefficients of zd in the latter equation we obtain Sd = 0, a contra- diction. Therefore we have deg1(P) ≤ 1 and hence we have P(x, y) = x R1(y) + R0(y). Substituting in (23), we then obtain (x R1(x) + R0(x)) R1(z) + R0(z) = 0. 14 JEAN-LUC MARICHAL AND PIERRE MATHONET If x R1(x) + R0(x) is nonconstant, then R1 = 0 and then also R0 = 0. Otherwise, if x R1(x) + R0(x) is a constant C, then R0(z) = −C R1(z) and hence C = x R1(x) + R0(x) = x R1(x) − C R1(x), from which we derive R1 = 0 and then also R0 = 0. Finally, P = 0, which contradicts the assumption that deg2(P) ≥ 1. Hence we have deg2(P) = 0, in which case we immediately see that P = 0. Acknowledgments This research is partly supported by the internal research project R-AGR-0500 of the University of Luxembourg. The authors thank Michel Rigo of the University of Li`ege for pointing out Lucas' theorem. They also thank Jorg Tomaschek of Deloitte Austria for bringing this problem to their attention. References [1] O. G. Bokov. A model of Lie fields and multiple-time retarded Greens functions of an elec- tromagnetic field in dielectric media. Nauchn. Tr. Novosib. Gos. Pedagog. Inst. 86:3–9, 1973. [2] N. Fine. Binomial coefficients modulo a prime. Amer. Math. Monthly 54:589–592, 1947. [3] H. Fripertinger. On n-associative formal power series. Aeq. Math. 90(2):449–467, 2016. [4] R. Gilmore. Lie groups, Lie algebras, and some of their applications. John Wiley and Sons, New York, 1974. [5] B. C. Hall. Lie groups, Lie algebras, and representations. An elementary introduction. Springer-Verlag, New York, 2003. [6] N. Jacobson. Lie algebras. Courier Dover Publications, 1979. [7] E. Lucas. Th´eorie des fonctions num´eriques simplement p´eriodiques. Am. J. Math. 1(2): 184–196, 1878. [8] E. Lucas. Th´eorie des fonctions num´eriques simplement p´eriodiques. Am. J. Math. 1(3): 197–240, 1878. [9] E. Lucas. Th´eorie des fonctions num´eriques simplement p´eriodiques. Am. J. Math. 1(4): 289–321, 1878. [10] J.-L. Marichal and P. Mathonet. A description of n-ary semigroups polynomial-derived from integral domains. Semigroup Forum 83:241–249, 2011 [11] Jorg Tomaschek. Deloitte Austria. Private communication. [12] A. V. Yagzhev. A functional equation from theoretical physics. Funct. Anal. Appl. 16(1):38– 44, 1982. Mathematics Research Unit, University of Luxembourg, Maison du Nombre, 6, avenue de la Fonte, L-4364 Esch-sur-Alzette, Luxembourg E-mail address: jean-luc.marichal[at]uni.lu University of Li`ege, Department of Mathematics, All´ee de la D´ecouverte 12 - B37, B-4000 Li`ege, Belgium E-mail address: p.mathonet[at]ulg.ac.be
1511.09172	1	1511	2015-11-30T06:15:04	Dimension and decomposition in modular upper-continuous lattices	[ "math.RA" ]	We translate notions and results of decomposition and dimension theories for module categories, into the lattice environment. In particular we translate dimension theory in module categories to complete modular upper-continuous lattices.	math.RA	math	DIMENSION AND DECOMPOSITION IN MODULAR UPPER-CONTINUOUS LATTICES JOS ´E R´IOS MONTES AND ANGEL ZALD´IVAR CORICHI v o N 0 3 ] . A R h t a m [ 1 v 2 7 1 9 0 . 1 1 5 1 : v i X r a ABSTRACT. We translate notions and results of decomposition and dimension theories for module categories, into the lattice environment. In particular we translate dimension theory in module categories to complete modular upper- continuous lattices. 1. INTRODUCTION The notions of dimension and decomposition for module categories have been extensively studied for many authors from different perspectives. Starting with the commutative case, the notions of primary decompositions and Krull dimension, have been extended to the non-commutative setting. Moreover, these construc- tions have also been extended to an arbitrary abelian category. Good accounts for these developments are [4, 7, 9, 10]. The book [2] organizes all the distinct de- compositions and dimensions in module categories and gives a general point of view for the treatment of these theories via radical functions, quasi-decomposition functions and quasi-dimension functions. Most of these treatments use lattice con- cepts for the particular case of the lattice of all torsion theories [3]. Later on, in [13] the author describes an analogue treatment of decompositions for modules of [2] for complete, modular, meet-continuous (upper-continuous) lattices via allo- cations. In the same setting as [2, 13] we develop the general dimension theory that is not developed in [2]. The organization of the paper is as follows: Section 2 gives the general background necessary for most of the paper. In Section 3 we develop the general setting of allocations and we introduce the concept of aspect for complete modular upper-continuous lattice . We also investigate some proper- ties and relations with allocations. Section 4 is an account based in some results of [13] with some generalizations. Section 5 describes the notion of dimension for complete modular upper-continuous lattices. We prove that this notion is exactly the analogue for module categories via filtrations of torsion theories. In Section 6 we prove that the concept of quasi-dimension function in a module category is intrinsically linked with the concept of aspect for the lattice of submodules of a given module M. 1 2 JOS ´E R´IOS MONTES AND ANGEL ZALD´IVAR CORICHI 2. PRELIMINARIES AND BACKGROUND MATERIAL An idiom (A, ≤,W, ∧, ¯1, 0) is a complete, upper-continuous, modular lattice, that is, A is a complete lattice that satisfies the following distributive laws: (IDL) holds for all a ∈ A and X ⊆ A directed, and a ∧ (_ X) = _ {a ∧ x x ∈ X} (ML) for all a, b, c ∈ A. These lattice were introduced in [11], and a more recent account (a ∨ c) ∧ b = a ∨ (c ∧ b) is in [14]. We also need the following class of idioms: A frame (A, ≤,W, ∧, ¯1, 0) is a complete lattice that satisfies (FDL) a ∧ (_ X) = _ {a ∧ x x ∈ X} for all a ∈ A and X ⊆ A any subset. Two fundamental examples are the following: Given a ring R and any left R-module M, the lattice SubR(M ) of all submodules of M is modular and upper-continuous, hence it is an idiom. Frames are the al- gebraic version of a topological space. Indeed, if S is a topological space then its topology, O(S) is a frame. The correspondence S 7→ O(S) has been extensively studied, for example see [6] and [8]. It is important to mention that frames are characterized by an implication. Recall that in any lattice A, an implication in A is an operation ( ≻ ) given by x ≤ (a ≻ b) ⇔ x ∧ b ≤ a, for all a, b ∈ A. For a proof of the following fact, see [12]. Proposition 2.1. A complete lattice A is a frame if and only if A has an implica- tion. We will use the following concepts. An inflator on an idiom A is a function d : A → A such that x ≤ d(x) and x ≤ y ⇒ d(x) ≤ d(y). A pre-nucleus d on A is an inflator such that d(x ∧ y) = d(x) ∧ d(y). A stable inflator on A is an inflator such that d(x)∧y ≤ d(x∧y) for all x, y ∈ A. Let I(A) denote the set of all inflators on A, P (A) the set of all prenuclei, and S(A) the set of all stable inflators. Clearly, P (A) ⊆ S(A) ⊆ I(A). A closure operator is an idempotent inflator c on A, that is, is an inflator such that c2 = c. Let C(A) the set of all closure operators in A. A nucleus on A is a idempotent pre-nucleus. Let N (A) be the set of all nuclei on A. All these sets are partially ordered by d ≤ f ⇔ d(a) ≤ f (a) for all a ∈ A. Note that the identity function idA and the constant function ¯d(a) = ¯1 for all a ∈ A (where ¯1 is the top of A) are inflators. These two inflators are the bottom and the top in all these partially ordered sets. Moreover, we can describe the infimum V I on A given by (V I)(a) = V {f (a)f ∈ I} for each a ∈ A. It is immediate that this function lies in L, and is the infimum of the family I. Therefore, each of these sets is a complete lattice. of any subset I ⊆ L, for L ∈ {I(A), P (A), S(A), C(A), N (A)}, as the function Inflators tell us something about the complexity of the idiom. Indeed, given an inflator d ∈ I(A), let d0 := idA, dα+1 := d ◦ dα for a non-limit ordinal α, and let dλ := W {dαα < λ} for a limit ordinal λ. These are inflators, ordered in a chain d ≤ d2 ≤ d3 ≤ . . . ≤ dα ≤ . . . . DIMENSION AND DECOMPOSITION IN MODULAR UPPER-CONTINUOUS LATTICES 3 By a cardinality argument, there exists an ordinal γ such that dα = dγ, for α ≥ γ. In fact, we can choose γ the least of these ordinals, say ∞. Thus, d∞ is an inflator such that d ≤ d∞, but more important this inflator satisfies d∞d∞ = d∞, that is, d∞ is a closure operator on A. We say that an idiom A has d-length if d∞(0) = ¯1, that is, the associated idempotent d∞ is just the top of I(A). In order to have a workable notion of dimension one has to work with the complete lattice on all nuclei: Theorem 2.2. For any idiom A, the complete lattice of all nuclei N (A) in A is a frame. A proof of this fact can be found in [11, 12, 14]. Another important fact about nuclei is that any element j ∈ N (A) gives a quotient of A, the set Aj of elements fixed by j. Even more, Aj is an idiom, and thus many properties of A are reflected in Aj. Since N (A) is an idiom, it has his own inflators, and we may consider any stable inflator S over N (A). Following Simmons we said that a nucleus j over A has S-dimension if S∞(j) = ¯d. In particular, for the nucleus idA of A, since AidA = A, if idA has S-dimension θ, then we say that the S-dimension of A is θ. This is actually the central idea of dimension: Given a property in the idiom A, this property gives a stable inflator S, and we want to measure how far A or some quotient Aj, with respect to the nucleus j, has the property; that measure is the ordinal θ. To organize all these, there is a frame, the base frame of the idiom A. Next, following Simmons [11], we review the construction of the base frame and other special frames used for the ranking and dimension of idioms. If A is an idiom and a, b ∈ A satisfy a ≤ b, the interval [a, b] is the set [a, b] = {x ∈ A a ≤ x ≤ b}. Denote by I(A) the set of all intervals of A. Given two intervals I, J, we say that I is a subinterval of J, denoted by I → J, if I ⊆ J, that is, if I = [a, b] and J = [a′, b′] with a′ ≤ a ≤ b ≤ b′ in A. We say that J and I are similar, denoted by J ∼ I, if there are l, r ∈ A with associated intervals L = [l, l ∨ r] [l ∧ r, r] = R where J = L and I = R or J = R and I = L. Clearly, this a reflexive and symmetric relation. Moreover, if A is modular, this relation is just the canonical lattice isomorphism between L and R. We say that a set of intervals A ⊆ I(A) is abstract if is not empty and it is closed under ∼, that is, J ∼ I ∈ A ⇒ J ∈ A. An abstract set B is a basic set of intervals if it is closed by subintervals, that is, J → I ∈ B ⇒ J ∈ B for all intervals I, J. A set of intervals C is a congruence set if it is basic and closed under abutting intervals, that is, for elements a, b, c ∈ A. A basic set of intervals B is a pre-division set if [a, b][b, c] ∈ C ⇒ [a, c] ∈ C ∀ x ∈ X h[a, x] ∈ B ⇒ [a,_ X] ∈ Bi 4 JOS ´E R´IOS MONTES AND ANGEL ZALD´IVAR CORICHI for each a ∈ A and X ⊆ [a, ¯1]. A set of intervals D is a division set if it is a congruence set and a pre-division set. Put D(A) ⊆ C(A) ⊆ B(A) ⊆ A(A) the set of all division, congruence, basic and abstract set of intervals in A. This gadgets can be understood like certain classes of modules in a module category R-Mod, that is, classes closed under isomorphism, subobjects, extensions and coproducts. From this point of view C(A) and D(A) are the idioms analogues of the Serre classes and the torsion (localizations) classes in module categories. Note that B(A) is closed under arbitrary intersections and unions, hence it is a frame. The top of this frame is I(A) and the bottom is the set of all trivial intervals of A, denoted by O(A) or simply by O. The frame B(A) is the base frame of the idiom A. The family C(A) is closed under arbitrary intersections, but suprema are not unions; to describe the suprema we take any basic set B and the least congruence set that contains it, this usual construction leads to a inflator over the base frame B(A) as follows: For each B ∈ B(A), let Cng(B) be the set of all intervals [a, b] which can be partitioned by B, that is, there is a finite chain a = x0 ≤ . . . ≤ xi ≤ . . . ≤ xm+1 = b such that [xi, xi+1] ∈ B for each 0 ≤ i ≤ m. Note that, by definition, B ⊆ Cng(B). As in Lemma 5.3 of [11] for each basic set B, Cng(B) is the least congruence set that includes B. Moreover, Cng( ) is a nucleus over the frame B(A) with fixed set C(A), that is, C(A) is a frame. An interval [a, b] is simple if there is no a < x < b that is [a, b] = {a, b}. Denote by Smp be the set of all simple intervals. An interval[a, b] of A is complemented if it is a complemented lattice, that is, for each a ≤ x ≤ b there exist a ≤ y ≤ b such that a = x ∧ y and b = x ∨ y. Let Cmp be the set of all complemented intervals. In fact, for every B we can define Smp(B) and Cmp(B): the former is the set of intervals that are B-simple, that is, the set of all [a, b] such that for each a ≤ x ≤ b, [a, x] ∈ B or [x, b] ∈ B, and the latter is the set of all intervals that are B-complemented, that is, [a, b] such that for every a ≤ x ≤ b exists a ≤ y ≤ b such that [a, x ∧ y] ∈ B and [x ∨ y, b] ∈ B. With this, we have that Smp = Smp(O) and Cmp = Cmp(O). There are others special sets of intervals: Given any B ∈ B(A) denote by Crt(B) the set of intervals such that for all a ≤ x ≤ b we have a = x or [x, b] ∈ B; this is the set of all B-critical intervals. Denote now by Fll(B) the set of all intervals [a, b] such that, for all a ≤ x ≤ b there exists a ≤ y ≤ b with a = x ∧ y and [x ∨ y, b] ∈ B; this is the set of all B-full intervals. Note that Smp(O) = Crt(O) and Cmp(O) = Fll(O). In [11] Simmons proves that for any B ∈ B(A), Crt(B) ≤ Smp(B), Fll(B) ≤ Cmp(B), Smp(B) ≤ Cmp(B) and Crt(B) ≤ Fll(B). Moreover, he shows that for any B ∈ B(A) the sets Smp(B), Cmp(B) and Crt(B), Fll(B) are basic. He also proves that Smp( ), Cmp( ) ∈ S(B(A)) and Crt( ), Fll( ) ∈ P (B(A)) For the set D(A) and for any B ∈ B(A) we can describe the least division set that contains it. Since D(A) is closed under arbitrary intersections, denote by Dvs(B) that division set that contains it. In [11] it is proved that Dvs( ) is a nucleus over B(A) and the quotient of this nucleus is D(A). In fact, there is a relation with this frame and the frame N (A): To describe this relation, take any DIMENSION AND DECOMPOSITION IN MODULAR UPPER-CONTINUOUS LATTICES 5 basic set B and a ∈ A; define B(a) = W X, where x ∈ X ⇔ [a, x] ∈ B. This produces the associated inflator of B. Moreover, if the basic set B is a congruence set, then B is a pre-nucleus in A, and if it is a division set, then B is a nucleus. In this way we have for every division set a nucleus. Now, given a nucleus j we can construct a division set [a, b] ∈ Dj ⇔ j(a) = j(b). These correspondences are bijections and moreover they define an isomorphism between D(A) and N (A), with this we have: Theorem 2.3. If A is an idiom, then there is an isomorphism of frames N (A) ←→ D(A) j ←→ D given by j 7−→ Dj [a, b] ∈ Dj ⇐⇒ b ≤ j(a) D 7−→ j = D The Dvs-construction can be described it in a useful way: Theorem 2.4. For every B ∈ B(A) we have [a, b] ∈ Dvs(B) ⇐⇒ (∀a ≤ x < b)(∃x < y ≤ b)[[x, y] ∈ B], for each interval [a, b]. Details are in [11], and a more recent account is given in [15] and [16]. Example 2.5. An interval [a, b] is simple if there is no a < x < b that is [a, b] = {a, b}. Denote by Smp be the set of all simple intervals. An interval[a, b] of A is complemented if it is a complemented lattice, that is, for each a ≤ x ≤ b there exist a ≤ y ≤ b such that a = x ∧ y and b = x ∨ y. Let Cmp be the set of all complemented intervals. In fact, for every B we can define Smp(B) and Cmp(B): the former is the set of intervals that are B-simple, that is, the set of all [a, b] such that for each a ≤ x ≤ b, [a, x] ∈ B or [x, b] ∈ B, and the latter is the set of all intervals that are B-complemented, that is, [a, b] such that for every a ≤ x ≤ b exists a ≤ y ≤ b such that [a, x ∧ y] ∈ B and [x ∨ y, b] ∈ B. With this, we have that Smp = Smp(O) and Cmp = Cmp(O). There are others special sets of intervals: Given any B ∈ B(A) denote by Crt(B) the set of intervals such that for all a ≤ x ≤ b we have a = x or [x, b] ∈ B; this is the set of all B-critical intervals. Denote now by Fll(B) the set of all intervals [a, b] such that, for all a ≤ x ≤ b there exists a ≤ y ≤ b with a = x ∧ y and [x ∨ y, b] ∈ B; this is the set of all B-full intervals. In [11] is proved that Note that Smp(O) = Crt(O) and Cmp(O) = Fll(O). for any B ∈ B(A), Crt(B) ≤ Smp(B), Fll(B) ≤ Cmp(B), Smp(B) ≤ Cmp(B) and Crt(B) ≤ Fll(B). Moreover, one can shows that for any B ∈ B(A) the sets Smp(B), Cmp(B) and Crt(B), Fll(B) are basic. Also it can be seen that Smp( ), Cmp( ) ∈ S(B(A)) and Crt( ), Fll( ) ∈ P (B(A)) 6 JOS ´E R´IOS MONTES AND ANGEL ZALD´IVAR CORICHI 3. Λ-ALLOCATIONS AND Λ-ASPECTS In [13] the author introduces the concept of Λ-allocation for an idiom A to study the decomposition of intervals on A. This concept can be understood as the id- iomatic version of the decomposition theory in [2]. Definition 3.1. If Λ is a complete lattice, for an idiom A a Λ-allocation is a func- tion ϕ : I(A) −→ Λ that satisfies the following: (1) ϕ(l ∧ r, r) = ϕ(l, l ∨ r) with r, l ∈ A. (2) ϕ(a, b) ≤ ϕ(a, c) for a ≤ c ≤ b. (3) ϕ(a, c) ∧ ϕ(c, b) ≤ ϕ(a, b) for a ≤ c ≤ b. (4) ϕ(a,W X) = V {ϕ(a, x) x ∈ X} for a ∈ A and X ⊆ [a, ¯1] directed. In item (4) of Definition 3.1 the subset X can be independent over a, this is pointed in 6.2 of [13]. In [13] Simmons shows that for any idiom A there is always a N (A)-allocation given by χ : I(A) → N (A), where χ(a, b) is the unique largest nucleus that satisfies χ(a, b)(a) ∧ b = a. Denote by Sit(A, Λ) = {ϕ ϕ is a Λ-allocation}. It is almost immediate that Sit(A, Λ) is a poset. Moreover, it is a complete lattice. Now, take any f : A −→ A′ idiom morphism and ϕ ∈ Sit(A′, Λ). Then, consider the induced poset morphism I(f ) : I(A) → I(A′) and the composition ϕ ◦ I(f ) : I(A) → Λ. From the definition and the fact that f is monotone we have that ϕ ◦ I(f ) ∈ Sit(A, Λ). Setting f ∗ : Sit(A′, Λ) → Sit(A, Λ) the following is straightforward. Proposition 3.2. Let be Λ a complete lattice. Then, Sit( , Λ) : ID −→ CL is a contravariant functor from the category of idioms ID to the category of complete latticesCL. (cid:3) Now, fixing the first component consider any morphism between two complete lattices, : Λ → Γ. Thus, for any Λ-allocation ϕ consider the composition ◦ ϕ : I(A) → Γ. It is easy to see that this function is a Γ-allocation, and thus: Proposition 3.3. If A is any idiom, then Sit(A, ) : CL −→ CL is a covariant endofunctor in the category of complete lattices and preserves monomorphisms. Proposition 3.4. Let be Λ a complete lattice and A an idiom. Then, there is a function Q : Sit(A, Λ) × Λ → C(A) given by [a, b] ∈ Q(ϕ, α) ⇔ α ≤ ϕ(x, b) for all x ∈ [a, b]. Moreover (cid:3) morphism in CL. (1) For every ϕ ∈ Sit(A, Λ), the function Q(ϕ, ) : Λop → C(A) is a V- (2) For every α ∈ Λ, the function Q( , α) : Sit(A, Λ) → C(A) is a V- morphism in CL. Proof. First we show that Q(ϕ, α) ∈ B(A) for every ϕ ∈ Sit(A, Λ) and every α ∈ Λ. From (1) of Definition 3.1 it follows that Q(ϕ, α) is abstract. Now, let [a, d] ∈ Q(ϕ, α) and take a ≤ b ≤ c ≤ d. For any b ≤ x ≤ c, we have that DIMENSION AND DECOMPOSITION IN MODULAR UPPER-CONTINUOUS LATTICES 7 ϕ(x, d) ≥ α. But from (2) of Definition 3.1 we have that ϕ(x, d) ≤ ϕ(x, c), that is, [b, c] ∈ Q(ϕ, α). Now consider [a, b], [b, c] ∈ Q(ϕ, α), and let x ∈ [a, c]. Thus, a ≤ b ∧ x ≤ b and b ≤ x ∨ b ≤ c. Hence, by the hypothesis, ϕ(x ∧ b, b) ≥ α and ϕ(b ∨ x, c) ≥ α. On the other hand, we have ϕ(x, c) ≥ ϕ(x, b ∨ x) ∧ ϕ(b ∨ x, c), and from modularity we deduce that [x ∧ b, b] ∼= [x, x ∨ b]. The last inequality, by (1) of Definition 3.1, is ϕ(x, b ∨ x) ∧ ϕ(b ∨ x, c) = ϕ(x ∧ b, b) ∧ ϕ(b ∨ x, c). But this infimum is above α, and thus ϕ(x, c) ≥ α, that is, Q(ϕ, α) ∈ C(A). Now, to prove part (1), note that if α ≤ α′ in Λ, then Q(ϕ, α′) ≤ Q(ϕ, α) by definition. Thus, if X ⊆ Λ we have that Q(ϕ,W X) ≤ T {Q(ϕ, α) α ∈ X}. But, for any [a, b] ∈ T {Q(ϕ, α) α ∈ X} we have that ϕ(x, b) ≥ α for ev- ery α ∈ X. Therefore, ϕ(x, b) ≥ W X, that is, [a, b] ∈ Q(ϕ,W X), and hence T {Q(ϕ, α) α ∈ X} = Q(ϕ,W X). For part (2), observe that if ϕ ≤ ϕ′, then Q(ϕ, α) ≤ Q(ϕ′, α). The result is (cid:3) immediate now. We know that Dvs( ) : B(A) → B(A) is a nucleus, so Dvs( ) : C(A) → D(A) is a frame morphism, in particular a ∧-morphism. Corollary 3.5. Let be Λ a complete lattice and A an idiom. Then, (1) For every ϕ ∈ Sit(A, Λ), the function Dvs( ) ◦ Q(ϕ, ) : Λop → D(A) is a ∧-morphism in CL. (2) For every α ∈ Λ, the function Dvs( ) ◦ Q( , α) : Sit(A, Λ) → D(A) is a ∧-morphism in CL. Proof. Direct from Proposition 3.4 and the previous observation. (cid:3) For every idiom A and any complete lattice Λ, we have a function S : Λ → Sit(A, Λ) defined by S(α)(a, b) = α. The following is straightforward. Proposition 3.6. Let Λ be a complete lattice and A an idiom. Then, the function S : Λ → Sit(A, Λ) defined by S(α)(a, b) = α, is an embedding in the category of complete lattices. Definition 3.7. Let be Λ a complete lattice. For an idiom A, a Λ-aspect is a func- tion ϕ : I(A) −→ Λ that satisfies the following: (1) ϕ(l ∧ r, r) = ϕ(l, l ∨ r), for r, l ∈ A. (2) ϕ(a, c) ∨ ϕ(c, b) = ϕ(a, b), for a ≤ c ≤ b. (3) ϕ(a,W X) = W {ϕ(a, x) x ∈ X}, for a ∈ A and X ⊆ [a, ¯1] directed. Denote by App(A, Λ) = {ϕ ϕ is a Λ-aspect}. It is immediate that App(A, Λ) is a poset and a complete lattice. Take any idiom morphism f : A −→ A′ and ϕ ∈ App(A′, Λ). Consider the induced poset morphism I(f ) : I(A) → I(A′). Then, ϕ ◦ I(f ) : I(A) → Λ, and from the definition and the fact that f is monotone we have ϕI(f ) ∈ App(A, Λ). For f ∗ : App(A′, Λ) → App(A, Λ) the following is immediate. Proposition 3.8. Let be Λ a complete lattice. Then, App( , Λ) : ID −→ CL is a contravariant functor from the category of idioms ID to the category of complete lattices CL. (cid:3) 8 JOS ´E R´IOS MONTES AND ANGEL ZALD´IVAR CORICHI Consider now any morphism between two complete lattices, : Λ → Γ. Thus, for any Λ-aspect ϕ, we have ◦ ϕ : I(A) → Γ, and it is easy to see that this function is a Γ-aspect. We have shown: Proposition 3.9. If A is any idiom then, App(A, ) : CL −→ CL is a covariant endofunctor in the category of complete lattices and preserves monomorphisms. Proposition 3.10. Let be Λ a complete lattice and A an idiom. Then, there is a function M : App(A, Λ) × Λ → C(A) given by [a, b] ∈ M(ϕ, α) ⇔ ϕ(a, b) ≤ α that satisfies (cid:3) in CL. (1) For every ϕ ∈ App(A, Λ), the function Mϕ : Λ → C(A) is aV-morphism (2) For every α ∈ Λ, the function Mα : App(A, Λ) → C(A) is aV-morphism in CL. Proof. Take (ϕ, α) ∈ App(A, Λ). By (1) of Definition 3.7 we have that M(ϕ, α) is abstract. Now, consider [a, d] ∈ M(ϕ, α) and a ≤ b ≤ c ≤ d. Then by (2) of Definition 3.7 we have that ϕ(a, c) ≤ ϕ(a, d) ≤ α, and again by (3) of Definition 3.7, ϕ(a, c) = ϕ(a, b) ∨ ϕ(b, c) ≤ α. Therefore, ϕ(b, c) ≤ α, and hence M(ϕ, α) is basic. Now, take [a, b], [b, c] ∈ M(ϕ, α). Then, α ≥ ϕ(a, b) ∨ ϕ(b, c) = ϕ(a, c), that is, [a, c] ∈ M(ϕ, α). Parts (1) and (2) are now straightforward. (cid:3) As in the case of a Λ-allocation, for any α ∈ Λ we have a function R(α) ∈ App(A, Λ) given by R(α)(a, b) = α. A direct calculation gives: Proposition 3.11. Let be Λ a complete lattice and A an idiom. Then, the function R : Λ → App(A, Λ) defined by R(α)(a, b) = α, is an embedding in the category of complete lattices. (cid:3) As an example of this, define ξ : I(A) → D(A) by ξ(a, b) = D(a, b), where D(a, b) is the least division set that contains the interval [a, b]. It is clear that ξ( ) is a D(A)-aspect of A. Theorem 3.12. Let A any idiom and Λ a complete lattice. Then, there is poset- morphism H : App(A, Λ) −→ Sit(A, Λ)op given for ψ ∈ App(A, Λ) by H(ψ)(a, b) = W {α ∈ Λ[a, b] ∈ Dvs(M(ψ, α))}. Proof. We must check that H(ψ) ∈ Sit(A, Λ). First observe that condition (1) of Definition 3.1 is clearly satisfied. For the other parts of Definition 3.1, consider any interval [a, c] on A and take a ≤ b ≤ c. From the definition of H(ψ) we have that H(ψ)(a, c) ≤ H(ψ)(a, b). Now, from the fact that ψ(a, b) ∨ ψ(b, c) = ψ(a, c) we deduce that H(ψ)(a, b) ∧ H(ψ)(b, c) ≤ H(ψ)(a, c). Now, let X ⊆ [a, ¯1] be a directed set. From the above paragraph we have that H(ψ)(a,W X) ≤ V {H(ψ)(a, x)x ∈ X}. For the other comparison, observe that since ψ is a Λ- aspect, we have ψ(a,W X) = W {ψ(a, x)x ∈ X}. Thus, ψ(a, x) ≤ ψ(a,W X). DIMENSION AND DECOMPOSITION IN MODULAR UPPER-CONTINUOUS LATTICES 9 Hence, ψ(a,W X) ≤ H(ψ)(a, x) for all x ∈ X, and therefore ψ(a,W X) ≤ V {H(ψ)(a, x)x ∈ X}. Thus, V {H(ψ)(a, x)x ∈ X} ≤ H(ψ)(a,W X). Lastly, consider any ψ ≤ ψ′ in App(A, Λ). Then, from the definition of H we have that H(ψ′) ≤ H(ψ), that is, H is a poset morphism. (cid:3) 4. SOME CONSTRUCTIONS FOR Sit(A, Γ) In this section we analyse how the elements of Sit(A, Γ) lead to decomposition theories for the idiom A. As in the case of categories of modules, the concept of radical function in idioms is natural in this context (see, for example [2] and [10]): Definition 4.1. Let be A an idiom and Ω a partial ordered set. A function ρ : I(A) −→ Ω is a radical function if: (1) ρ(l ∧ r, r) = ρ(l, l ∨ r) for any r, l ∈ A, and (2) ρ(a, c) ≤ ρ(a, b) for all a ≤ b ≤ c . Examples of these functions are, of course, any Γ-allocation for A. In particular, the N (A)-allocation χ. Other examples of these functions come from module the- ory: Denote by Λ(M ) the idiom of sub-modules of an R-module M, and consider any radical function in R-Mod in the sense of [2, Chapter 3]. Then, the restriction of any radical function to the intervals of Λ(M ) gives a radical function in our sense. If Ω is a lattice and ρ ∈ Rad(A, Ω), for l ∈ Ω define ρ′ : I(A) → Ω by ρ′(a, b) = ρ(a, b) ∧ l. Then, ρ′ is a radical function on A with values in the lattice Ω. Let Rad(A, Ω) = {ρ : I(A) → Ω ρ is radical }. We define a partial order in Rad(A, Ω) using the order of Ω as follows: ρ ≤ ⇔ ρ(a, b) ≤ (a, b) for all [a, b] ∈ I(A). As in the case of Sit(A, Γ), we have that Rad(A, ) : Pos −→ Pos is a covariant functor and Rad( , Ω) : ID −→ Pos is a contravariant functor. If Γ is a complete lattice, then Sit(A, Γ) is included in Rad(A, Γ). Definition 4.2. Let ρ ∈ Rad(A, Ω). An interval [a, b] is ρ-stable or ρ-inert if and only if a < b and ρ(a, b) = ρ(a, x) for all a < x ≤ b. For χ ∈ Sit(A, N(A)), the χ-stable intervals are precisely the inert intervals. In particular, any uniform interval is inert, [13]. In fact, in [13] the author describes in detail the decomposition theory generated by χ, and gives an application to geo- lattices. Now, for ρ ∈ Rad(A, Ω), the support of ρ is the set Σρ(a, b) = {ρ(a, x) a < x ≤ b is ρ-stable} . Proposition 4.3. The function Σρ : I(A) → P(Ω)op which assigns to each interval [a, b] the support of the radical function ρ, is a P(Ω)op-allocation. Proof. Let ρ ∈ Rad(A, Ω). By definition of radical function the first requirement to be a allocation is clearly satisfied, that is, Σρ(r ∧ l, r) = Σp(l, r ∨ l) for any l, r ∈ A. Now, consider any interval [a, c] and a ≤ b ≤ c. Then Σρ(a, b) ⊆ Σρ(a, c). Take any ρ(a, x) ∈ Σρ(a, c). If a < b ∧ x, then ρ(a, x) = ρ(a, b ∧ x) ∈ Σρ(a, b), and if a = b ∧ x we have ρ(a, x) = ρ(b ∧ x, x) = ρ(b, b ∨ x). Thus, from the above this last interval is ρ-stable and then ρ(a, x) ∈ Σρ(b, c), that is, 10 JOS ´E R´IOS MONTES AND ANGEL ZALD´IVAR CORICHI Σρ(a, c) ⊆ Σρ(a, b) ∪ Σρ(b, c). Now, for the last requirement take any X ⊆ [a, ¯1] directed, for some a ∈ A and consider ρ(a, y) ∈ Σρ(a,W X). Thus, from the idiom distributivity law we have y = y ∧ (W X) = W {y ∧ x x ∈ X}. Then, ρ(a, y ∧ x) ∈ Σρ(a, x). This proves that Σρ(a,W X) ⊆ S {Σρ(a, x) x ∈ X}. a < y ∧ x ≤ x for some x ∈ X, and from a < y ∧ x ≤ y we derive ρ(a, y) = The other comparison follows immediately from the first property. (cid:3) Assume now that Ω is a complete lattice and consider ϕ ∈ Sit(A, P(Ω)op). Define the function ϕ : I(A) → Ω by ϕ(a, b) = V ϕ(a, b). This function is clearly a radical function. Thus we have two functions Sit(A, P(Ω)op) Σ Rad(A, Ω) where Σ(ρ) = Σρ and (ϕ) = ϕ. Observe now that we have a diagram Sit(A, Ω) ι Rad(A, Ω) )❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ { }∗ Σ Sit(A, P(Ω)op) where { }∗ : Sit(A, Ω) → Sit(A, P(Ω)op) is the morphism induced by the inclu- sion Ω → P(Ω)op. Note that ◦ { }∗ = ι and so one triangle is commutative, but the other one does not need to commute. Definition 4.4. For ρ ∈ Rad(A, Ω), an interval [a, b] is ρ-atomic if Σρ(a, b) = {∗}. For ρ ∈ Rad(A, Ω), the idiom A is ρ-adequate if Σρ(a, b) is not empty for every non-trivial interval [a, b] of A. For ρ ∈ Rad(A, Ω) and a element p ∈ Ω, an interval [a, b] is p-inertial if ρ(a, b) = p and [a, b] is ρ-stable. From this observe that any ρ-stable interval [a, b] is a p = ρ(a, b)-inertial. We will use p-inertial intervals to generate a decomposition for the parent idiom. Using intervals p-inert with respect to some ϕ ∈ Sit(A, Ω) give us another look at allocations. For p ∈ Ω, consider the frame 2 with two elements 0 < 1, and define the function p : I(A) −→ 2 by p(a, b) = (cid:26) 1 if [a, b] is p-inertial 0 if not, for each interval [a, b]. Proposition 4.5. For each p ∈ Ω, the function p : I(A) −→ 2 is a 2-allocation, that is: (1) For l, r ∈ A, [l ∧ r, r] is p-inert ⇔ [l, l ∨ r] is p-inert. (2) For a ≤ b ≤ c, [a, c] p-inert ⇒ [a, b] is p-inert. (3) For a ≤ b ≤ c, [a, b] and [b, c] p-inert ⇒ [a, c] is p-inert. (4) For a ∈ A and X ⊆ [a, 1], [a,W X] is p-inert ⇔ (∀x ∈ X) [[a, x] is p-inert]. 3 3 s s / / ) Z Z DIMENSION AND DECOMPOSITION IN MODULAR UPPER-CONTINUOUS LATTICES 11 Proof. (1): For any l, r ∈ A, first suppose that [l ∧ r, r] is p-inert and consider l < x ≤ l ∨ r. Then, using the canonical isomorphism [l ∧ r, r] ∼= [l, l ∨ r] we have that x = y ∨ l for some y < l ∧ r ≤ r. Therefore, ϕ(l, x) = ϕ(l, y ∨ l) = ϕ(l ∧ r, y) = ϕ(l ∧ r, r) = p, where the second equality is because the axioms of allocations and the third one is by the hypothesis. The reverse implication is similar. (2): Let a ≤ b ≤ c with [a, c] p = ϕ(a, c)-inert. Then, for any a < x ≤ b we have ϕ(a, x) = ϕ(a, c) = ϕ(a, b). (4): Let a ∈ A and X ⊆ [a, ¯1] directed. It is enough to verify that (∀x ∈ (3): Given [a, b] and [b, c] p-inert intervals and any a < x ≤ c we have that p = ϕ(a, b) ∧ ϕ(b, c) ≤ ϕ(a, c) ≤ ϕ(a, x). Thus we only need to show that ϕ(a, x) ≤ p. First observe that a ≤ b ∧ x ≤ b. Now, if a = b ∧ x we have b < b ∨ x ≤ c, and therefore ϕ(a, x) = ϕ(b ∧ x, x) = ϕ(b, b ∨ x) = ϕ(b, c) = p, where the second equality comes from the fact that [b, b ∨ x] ∼= [b ∧ x, x]. X) [[a, x] is p-inert] implies [a,W X] is p-inert. To see this, note that sinceX is directed we have that ϕ(a,W X) = V {ϕ(a, x) x ∈ X} = p. Consider now a < y ≤ W X. Then, p = ϕ(a,W X) ≤ ϕ(a, y). To show the other comparison note that y = y ∧ (W X) = W {y ∧ x x ∈ X} using the idiom distributivity law. Then, for some x ∈ X we have that a ≤ y ∧ x ≤ y and a < y ∧ x ≤ x. It follows that ϕ(a, y) ≤ ϕ(a, y ∧ x) = ϕ(a, x) = p. (cid:3) Now, for ϕ ∈ Sit(A, Ω), let Dp = {[a, b] [a, b] is p-inert}. The last proposition says that Dp is a congruence set in A. Moreover, we know that for any congruence set C, if [a, x], [a, y] ∈ C then [a, x ∨ y], [a, x ∧ y] ∈ C. From this it is easily seen that C is closed under finite suprema. Corollary 4.6. The set Dp is a division set in A. Proof. Take any a ∈ A and X ⊂ [a, ¯1] with [a, x] p-inert for all x ∈ X. Let Y be the set of all elements of the form x1 ∨x2 . . .∨xn, with xi ∈ X for 0 ≤ i ≤ n. This is a directed set and [a, y] is p-inert. Using the same reasoning in the proof of (4) in Proposition 4.5, we have p = ϕ(a,W Y ) = V {ϕ(a, y) y ∈ Y } ≤ ϕ(a,W X) ≤ V {ϕ(a, x) x ∈ X} = p, and for any a < z ≤ W X ≤ W Y there is some y ∈ Y with a < z ∧ y ≤ y then ϕ(a, z) ≤ ϕ(a, z ∧ y) = ϕ(a, y) = p. The other comparison is clear. (cid:3) Definition 4.7. Let be [a, b] an interval over an idiom A. An element a ≤ x ≤ b is a p-inertial point or a p-stable point in [a, b], with p ∈ Ω, if [a, x] is p-inertial and if x ∧ y = a then [a, y] is not p-inertial for each a ≤ y ≤ b. The element x is an inertial point or a stable point in [a, b], if it is a p-inertial point for some p ∈ Ω. For the remaining part of this section we use concepts and results on independent sets on idioms as in [10]. We start by showing that there are enough inertial points in an idiom: Proposition 4.8. Let [a, b] be an interval of an idiom A, and ρ ∈ Sit(A, Ω) with Ω a complete lattice and consider p ∈ Σρ(a, b). Then, for each a ≤ z ≤ b with [a, z] p-inertial there is a p-inertial point z ≤ x ≤ b in [a, b]. 12 JOS ´E R´IOS MONTES AND ANGEL ZALD´IVAR CORICHI Proof. We use Zorn's lemma: Consider the family Π of subsets X ⊆ [a, b] satisfy- ing: (1) z ∈ X . (2) X is independent over a. (3) For each x ∈ X the interval [a, x] is p-inertial. By hypothesis, z is an inertial point and thus gives an element {z} in Π. Inclusion is a partial order in Π. Consider any chain Z of elements of Π, and its union S Z. Clearly, S Z ∈ Π and by Zorn's lemma there exists a maximal member X of Π. If x = W X, then a ≤ x ≤ b and by Proposition 4.5 it follows that [a, x] is p-inertial. Lastly, consider a ≤ y ≤ b with x ∧ y = a. Then, the family X ∪ {y} is independent over a and the maximality of X implies that [a, y] is not p-inertial. (cid:3) Lemma 4.9. Let be A an idiom and ϕ ∈ Sit(A, Ω). Suppose that A is ϕ-adequate. Then, for each interval [a, b] we have that χ(a, b) = ^ {χ(a, x) a < x ≤ b with [a, x] ϕ-stable} . Proof. Since χ is a N (A)-allocation, χ(a, b) ≤ ^ {χ(a, x) a < x ≤ b with [a, x] ϕ-stable} . For the other comparison, let Ξ = {χ(a, x) a < x ≤ b is ϕ-stable} and k = V Ξ. If a < k(a) ∧ b, by hypothesis there is a < x ≤ k(a) ∧ b with [a, x] ϕ-stable. Then, χ(a, x) ∈ Ξ and thus k ≤ χ(a, x). Hence, x ≤ k(a) ≤ χ(a, x)(a) and then x = χ(a, x)(a) ∧ x = a, which is a contradiction. (cid:3) The concept of p-inertial point is related to the concept of a large element: Lemma 4.10. Let be A an idiom, ϕ-adequate for some ϕ ∈ Sit(A, Ω) and suppose [a, b] is ϕ-atomic, that is, Σϕ(a, b) = {p}. Then, any p-inertial point in [a, b] is large in [a, b]. Proof. Suppose x is a p-inertial point in [a, b]. Then, [a, x] is p-inert in [a, b]. Consider any y ∈ [a, b] with a = x ∧ y and suppose a < y. Since A is ϕ-adequate, there is some a < z ≤ y with [a, z] ϕ-stable. Then a ≤ z ∧ x ≤ x ∧ y = a, which contradicts the p-point property of x. (cid:3) We can now extend the definition of decomposition for a interval [a, b] over an idiom A. Definition 4.11. Let A be an idiom and ϕ ∈ Sit(A, Ω). A ϕ-decomposition of an interval [a, b] of A is a family X = {xp p ∈ Σϕ(a, b)} of elements of [a, b] indexed by the support of ϕ such that: (1) X is independent over a. (2) W X is large in [a, b]. (3) The interval [a, xp] is p-inert for each p ∈ Σϕ(a, b). Theorem 4.12. For an idiom A and ϕ a Ω-allocation, the following are equivalent: (1) Each non-trivial interval of A has a ϕ-decomposition. (2) A is ϕ-adequate. DIMENSION AND DECOMPOSITION IN MODULAR UPPER-CONTINUOUS LATTICES 13 Proof. Assuming (1), every non-trivial interval [a, b] in A has a ϕ-decomposition X of [a, b], with W X large in [a, b]. Then, this element is not a and so X is not empty. Thus, Σϕ(a, b) is not empty. Now assume (2) and consider any non-trivial interval [a, b] of A. By Propo- sition 4.8 there is a family X = {xp p ∈ Σϕ(a, b)} ⊆ [a, b] such that xp is a p-inertial point in [a, b], and [a, xp] are p-inert intervals. To verify parts (1) and (2) of Definition 4.11 it is enough to prove that X is independent over a. Thus, we only need to check that every finite subset of X is independent over a. Let Y be a finite subset of X. We do induction on the cardinality of Y . Consider p, p1 . . . , pn distinct elements of Σϕ(a, b) such that Y = {p, p1 . . . , pn}. By in- duction hypothesis we know that xp1 . . . , xpn are independent over a. To show the independence of Y ∪ xp over a, let y = W Y . Then, Σϕ(a, y) = V {p1, . . . , pn} because Σϕ is a P(Ω)-allocation, and we also have that Σϕ(a, xp) = p. Then, Σϕ(a, y ∧ xp) = Σϕ(a, y) ∩ Σϕ(a, xp) = ∅ by (2) of Definition 3.1. Since A is ϕ-adequate, then x ∧ xp = a and thus Y ∪ xp is independent over a. To verify (2) of Definition 4.11 suppose that x = W X is not large in [a, b], that is, there exists a < y ≤ b with x ∧ y = a. By the hypothesis (2) we can assume that [a, y] is ϕ-inert with ϕ(a, y) = p ∈ Σϕ(a, b). Thus, the element xp is a p-inertial point in [a, b] and xp ∧ y ≤ x ∧ y = a, which is a contradiction. (cid:3) Theorem 4.12 is a bit more general than Theorem 8.2 in [13] which is the special case of an N A-allocation χ. In [13] the author applies this to geo-lattices and the decomposition theory generated by χ in connection with certain spatial properties of the corresponding idiom, that is, any χ-stable interval [a, b] gives a point (a ∧-irreducible element) of N (A). Thus, the resulting decomposition theory has a more module theoretic flavour. 5. SOME CONSTRUCTIONS IN App(A, Λ) The concept of dimension in an idiom can be stated in many ways, depending of the context, for example via inflators and nuclei as in [16]. In this section we will give the lattice theoretical constructions of these via App(A, Λ). these construc- tions are the idiomatic version of the one developed in [2]. Let Λ be a complete lattice with ⊤, ⊥ his top and his bottom elements, respec- tively. Denote by ∝ (Λ) the minimum of all cardinals ι such that ι > #Λ. Let ∞(Λ) = {κ κ is an ordinal and κ ≤∝ (Λ)}. Define seq(Λ) = {h : ∞(Λ) → Λ h is increasing and h(0) = ⊥, h(∝ (Λ)) = ⊤} . Note that seq(Λ) is a complete lattice in the usual way. Now, let h ∈ seq(A); then, there is an ordinal α <∝ (Λ) such that h(α) = h(α + 1) = · · · . The least of these ordinals will be denoted by Bnd(h). Let be A an idiom and Λ a complete lattice, for ψ ∈ App(A, Λ) and h ∈ seq(Λ). We dine dψ h : I(A) → ∞(Λ) inductively as follows: (i) dψ h (a) = 0 for all a ∈ A. 14 JOS ´E R´IOS MONTES AND ANGEL ZALD´IVAR CORICHI (ii) If [a, b] ∈ I(A) is non trivial, then dψ h (a, b) = inf{0 ≤ ι ≤∝ (Λ) ψ(a, b) ≤ h(ι)}. Proposition 5.1. Let be A an idiom and Λ a complete lattice. Then, the function dψ h is a ∞(Λ)-aspect for each ψ ∈ App(A, Λ) and h ∈ seq(Λ). Moreover: (1) For every ψ ∈ App(A, Λ), the function dψ : seq(Λ)op → App(A, ∞(Λ)) (2) For every h ∈ seq(Λ), the function dh : App(A, Λ) → App(A, ∞(Λ)) is is a W-morphism. a W-morphism. h (a, b), dψ h (b, c)}. Proof. Let ψ ∈ App(A, Λ) and h ∈ seq(Λ). The first requirement of Defini- tion 3.7 is clearly satisfied. Now consider any non-trivial interval [a, c] on A and take a ≤ b ≤ c. Then, ψ(a, c) = ψ(a, b) ∨ ψ(b, c) and so dψ h (a, c) = sup{dψ If X is a directed subset of [a, ¯1], then ψ(a,W X) = W{ψ(a, x) x ∈ X}. Therefore dψ Let H = {hj j ∈ J} be a family in seq(Λ), and h = V H. Consider any inter- val [a, b] on A, and let ι = dψ (a, b), and B = sup {B(j) j ∈ J}, for each j ∈ J. Then, ψ(a, b) ≤ h(ι) ≤ hj(ι) and thus ι ≥ B(j), for each j ∈ J. Hence ι ≥ B. For the other comparison, we have ψ(a, b) ≤ hj(B) for each j ∈ J. Then, ψ(a, b) ≤ h(B) and thus ι ≤ B, that is, ι = B. This proves assertion (1). h (a,W X) = sup{dψ h (a, b), B(j) = dψ hj h (a, x) x ∈ X}. h (a, b), B(j) = d Next, consider ψ = W {ψj j ∈ J} in App(A, Λ) and [a, b] an interval over A. ψj If ι = dψ h (a, b), and B = sup {B(j) j ∈ J}, for each j ∈ J, then ψj(a, b) ≤ h(B(j)) ≤ h(B) for each j ∈ J. Thus, ψ(a, b) ≤ h(B) and so ι ≤ B. Now, if this inequality is strict, then there exists a j ∈ J such that B(j) > ι. Hence, ψj(a, b) (cid:2) h(ι) and so ψ(a, b) (cid:2) h(ι), which is a contradiction. So we must have ι = B, and this proves assertion (2). (cid:3) Note that for every α ∈ Λ we have an embedding from Λ into seq(Λ) given by α 7→ hα, where hα(ι) = α. With this and the definitions above we obtain: Corollary 5.2. Let be A and idiom and Λ a complete lattice. Then, for any element α ∈ Λ the diagram: App(A, Λ)op dhα App(A, ∞(Λ)) &▼▼▼▼▼▼▼▼▼▼ M( ,α) x♣♣♣♣♣♣♣♣♣♣♣ M( ,0) C(A) conmutes. (cid:3) The method developed in Proposition 5.1 is the idiomatic version of the one described in [2]. Now remember from Section 2, that we can construct certain operations over the base frame B(A) for any idiom A, that is, certain inflators Opr : B(A) → B(A). With these we can define sequences hψ,α ∈ seq(Λ), for each ψ ∈ App(A, Λ) and α ∈ Λ, as follows: (1) hψ,α(0) = α. / / & x DIMENSION AND DECOMPOSITION IN MODULAR UPPER-CONTINUOUS LATTICES 15 (2) If 0 < ι <∝ (Λ), then hψ,α(ι) = hψ,α(ι − 1) ∨(cid:16)_ {ψ(a, b) [a, b] ∈ Opr(M(ψ, hψ,α(ι − 1)))}(cid:17). (3) If 0 < ι <∝ (Λ) is a limit ordinal, then hψ,α(ι) = W {hψ,α(λ) λ < ι}. This sequence will be called the Opr-filtration. From this we can apply d to ψ ∈ App(A, ∞(Λ)). This aspect will be called and the Opr-filtration to obtain dψ the (ψ, α)-dimension, or (ψ, α)-dim for short. hψ,α Now consider the particular case when Λ = D(A), and A is an idiom. Then, take the D(A)-aspect ξ( ) and the inflator Dvs ◦ Opr := Kpr, and define a (D, Kpr)-filtration of any D ∈ D(A) as follows: (1) Kpr0(D) = D. (2) Kprγ+1(D) = Kpr(Kprγ(D)). (3) Kprλ(D) = Dvs(S(cid:8)Kprβ(D) β < λ(cid:9)). For each ordinal γ and limit ordinal λ. Proposition 5.3. Let A be an idiom. With the above notations we have that the Opr-filtration and the (D, Kpr)-filtration are the same. Proof. We proceed by induction over the ordinals γ and limit ordinals λ. The case γ = 0 is trivial. For the induction step γ 7→ γ + 1 observe that by definition of the sequence hξ( ),D(γ) = hξ( ),D(γ−1)∨(cid:16)_(cid:8)ξ(a, b) [a, b] ∈ Opr(M(ξ( ), hξ( ),D(γ − 1)))(cid:9)(cid:17) we have that the congruence set M(ξ( ), hξ( ),D(γ − 1)) = M(ξ( ), Kprγ−1(D)). Recall that [a, b] ∈ M(ξ( ), hξ( ),D(γ−1)) precisely when ξ(a, b) ≤ hξ( ),D(γ−1), that is, ξ(a, b) ⊆ Kprγ−1(D). Thus, M(ξ( ), hξ( ),D(γ − 1)) = Kprγ−1(D), and from this and the induction hypothesis we obtain that hξ( ),D(γ) = Kprγ−1(D) ∨(cid:16)_(cid:8)ξ(a, b) [a, b] ∈ Opr(Kprγ−1(D))(cid:9)(cid:17) in D(A), that is, Dvs(Kprγ−1(D) ∪ Dvs([(cid:8)ξ(a, b) [a, b] ∈ Opr(Kprγ−1(D))(cid:9))) = = Dvs([(cid:8)ξ(a, b) [a, b] ∈ Opr(Kprγ−1(D))(cid:9)) because Kprγ−1(D) ⊆ Opr(Kprγ−1D). the Dvs( ) construction in Theorem 2.4 for any basic set of intervals, for an in- terval [a, b] ∈ B, there exists a proper subinterval [x, y] of [a, b] such that [x, y] ∈ Let B = Dvs(S(cid:8)ξ(a, b) [a, b] ∈ Opr(Kprγ−1(D))(cid:9)). By the description of S(cid:8)ξ(a, b) [a, b] ∈ Opr(Kprγ−1(D))(cid:9). Thus, [x, y] ∈ ξ(a′, b′) for some [a′, b′] ∈ Opr(Kprγ−1(D)). From this we can find a subinterval of [a′, b′] isomorphic to a subinterval of [x, y], say I, and this interval is in Opr(Kprγ−1(D)). This is the case when [a, b] ∈ Dvs(Opr(Kprγ−1(D))). Therefore B ⊆ Dvs(Opr(Kprγ−1(D))). The other inclusion is clear. Thus, from the definition of the D − Kpr-filtration we conclude that Kprγ+1(D) = Kpr(Kprγ(D)) = hξ( ),D(γ). 16 JOS ´E R´IOS MONTES AND ANGEL ZALD´IVAR CORICHI Now, for the limit case, we have hξ,D(λ) = _ {hξ,D(β) β < γ} = Dvs(cid:16)[ {hξ,D(β) β < γ}(cid:17) = Dvs(cid:16)[nKprβ(D) β < γo(cid:17) = Kprλ(D), where the second equality is by definition of suprema in D(A) and the induction hypothesis, the third equality is by the definition of the filtration on Kpr with respect to D in the limit case. (cid:3) Recall that for any basic set B on A, Crt(B) is the set of intervals such that for all a ≤ x ≤ b we have a = x or [x, b] ∈ B. This is the set of all B-critical intervals. Now, denote by Fll(B) the set of all intervals [a, b] such that, for all a ≤ x ≤ b there exists a ≤ y ≤ b with a = x∧ y and [x∨ y, b] ∈ B. This is the set of all B-full intervals. Then, we consider the Ctr-filtration and the Fll-filtration in seq(A). Recall that the Gabriel derivative is given by Gab( ) := Dvs ◦ Crt, and the Boyle derivative is Boy( ) := Dvs ◦ Fll on D(A). Then, we can construct the D − Gab-filtration and the D − Boy-filtration, for some division set D on A. With these and Proposition 5.3 the following is immediate. Corollary 5.4. If A is an idiom, then the Crt-filtration is exactly the D-Gabriel filtration and the Fll-filtration is exactly the D-Boy filtration. (cid:3) 6. DIMENSIONS IN CATEGORIES OF MODULES In [2] the following framework is introduced to deal with most dimensions in module theory. Recall some of that material. Fix a complete lattice Γ and consider a ring R, and the category of left modules R-Mod. Definition 6.1. A quasi-dimension function in R-Mod is a function R-Mod D−→ Γ that satisfies: (1) If 0 → N → M → K → 0 is a exact sequence in R-Mod then D(M ) = D(N ) ∨ D(K). (2) If M is a module which is a directed union of a directed family {Ni i ∈ Ω} of submodules of M, then D(M ) = W {D(Ni) i ∈ Ω}. If D(M ) = ⊥ ⇔ M = 0, we say that D is of pre-dimension. If the image of D is linearly ordered, we say that D is linear. A linear pre-dimension function will be called a dimension function. Let Q-dim(R, Γ) be the collection of all quasi-dimension functions in R-Mod with values in Γ. Let R-mod be the set of isomorphism classes of finitely gen- erated modules. It is easily seen that any quasi-dimension function is completely determined by its values in R-mod. Thus, Q-dim(R, Γ) is a set, and, in fact, a complete lattice. Denote by Λ(M ) the idiom of sub-modules of an R-module M. Observe that any D ∈ Q- dim(R, Γ) defines a Γ-aspect of each module as follows: Take any module M and let DM : I(Λ(M )) −→ Γ be defined by DM (K, L) = D(L/K). The definition of quasi-dimension function implies that DIMENSION AND DECOMPOSITION IN MODULAR UPPER-CONTINUOUS LATTICES 17 the function DM is a Γ-aspect for Λ(M ). In particular, DR defines a Γ-aspect for Λ(R). Lemma 6.2. For each R-module M we have a morphism of complete lattices [M ] : Q- dim(R, Γ) −→ App(M, Γ) given by [M ](D) = DM . Proof. Let D ⊆ Q-dim(R, Γ) be a family of quasi-dimensions functions. Then, [M ](W D) = W DM , and thus (cid:0)_ DM(cid:1)(K, L) = (cid:0)_ D(cid:1)(L/K) = _ {D(L/K) D ∈ D} = _ {DM (K, L) D ∈ D} = _ {[M ](D) D ∈ D} , as required. For [M ](V D) = V {[M ](D) D ∈ D} the proof is similar. Denote by B(R) the collection of all classes of modules closed under isomor- phism, sub-modules and quotients. This is the base frame of R-Mod thus as in [11] we can do the idiomatic constructions of C(R) and D(R). Note that this frames are the classes of all Serre classes and the frame of all hereditary torsion classes in R-Mod, in particular we can identify D(R) with R-tors. (cid:3) As an example, consider the quasi-dimension function ξ ∈ Q-dim(R, D(R)), take an left R-module M and the induced D(R)-aspect [M ](ξ) = ξM . The situations described by Proposition 5.3 and Corollary 5.4 are precisely the ones exposed in Golan lecture notes [2], but in our context. In the inflator context, for example, consider any basic class of modules B ∈ B(R), and let Crt(B) be the class of modules M such that for all N ⊆ M, N = 0 or M/N ∈ B. The Crt- filtration in D(R) is just the well-known Gabriel filtration given by the pre-nucleus Gab := Dvs ◦ Crt in D(R). Given a class of modules, B and a module M, as in [14] we define the slice of B by M as the set hM i(B) by [K, L] ∈ hM i(B) ⇔ L/K ∈ B. We know that (1) If B ∈ B(R), then hM i (B) ∈ B(Λ(M )). (2) If C ∈ C(R), then hM i (C) ∈ C(Λ(M )). (3) If D ∈ D(R), then hM i (D) ∈ D(Λ(M )). Slicing by a module M determines a morphism of frames hM i ( ) : B(R) −→ B(Λ(M )), in fact, we have the morphism hM i ( ) : D(R) −→ D(Λ(M )). Also from Proposition 3.10 we have, for the idiom Λ(M ), a morphism of com- plete lattices, Na : App(Λ(M), Γ) −→ D(Λ(M)), where Na := Dvs ◦ Ma for each a ∈ Γ. Now from [2], for each a ∈ Γ there is a morphism of complete lattices Na : Q−dim(R, Γ) −→ D(R)op given by M ∈ TN (D,a) if and only if D(M ) ≤ a. With all these we can state the following: 18 JOS ´E R´IOS MONTES AND ANGEL ZALD´IVAR CORICHI Theorem 6.3. Let Γ be a complete lattice, and R-Mod the category of modules over a ring R. For each a ∈ Γ and each R-module M the following square Q- dim(R, Γ) [M ]( ) App(M, Γ) Na D(R)op Na / D(Λ(M ))op hM i( ) commutes in the category of complete lattices. Proof. We already know that all morphisms in the diagram are morphisms in the category of complete lattices, hence we just have to check the commutativity. Con- sider any D ∈ Q- dim(R, Γ), and recall that Na([M ](D)) = Dvs(Ma(DM ). Then, [K, L] ∈ Na([M ](D)) = Dvs(Ma(DM ) if and only if for all K ≤ H < L there exists H < N ≤ M such that N/H ∈ Ma(DM ). That is, DM ([H, N ]) = D(N/H) ≤ a. But this is the condition for N/H belonging to TNa(D), and this happens precisely when [H, N ] ∈ hM i (Na(D)). Thus, [K, L] ∈ hM i (Na(D)). Therefore, Na([M ](D)) = hM i (Na(D)), for all D, and the diagram commutes. (cid:3) Corollary 6.4. Let Γ be a complete lattice, and consider the complete lattice ∞(Γ). Let R-Mod be the category of modules over a ring R. Then, for any a ∈ ∞(Γ) and any left R-module M, the following diagram commutes in the category of complete lattices: Q- dim(R, ∞(Γ)) [M ]( ) App(M, ∞(Γ)) Na D(R)op Na / D(Λ(M ))op. hM i( ) Example 6.5. Let Γ be a complete lattice and consider D ∈ Q- dim(R, Γ), a ∈ Γ. Following [2, Chapter 12] define a sequence kD,a, called the Gabriel filtration of (D, a), mimicking the one defined after our Corollary 5.2: (cid:3) (1) kD,a(0) = a. (2) If 0 < ι <∝ (Γ), then kD,a(ι) = kD,a(ι − 1) ∨ (_{D(M ) M is an N (D, kD,a(ι − 1)))-cocritical left (3) If 0 < ι <∝ (Γ) is a limit ordinal, then kD,a(ι) = W {kD,a(λ) λ < ι}. Then, taking this filtration and applying the construction in Proposition 5.1, in this context, we obtain the quasi-dimension function λ(D, kD,a) in Q- dim(R, ∞(Γ)). Now consider any left R-module M and [M ](λ(D, kD,a)) = λ(D, kD,a)M ∈ App(M, ∞(Γ)). Then, the ∞(Γ)-aspect for M is just the (DM , a)-dimension. module}). / / / / / / DIMENSION AND DECOMPOSITION IN MODULAR UPPER-CONTINUOUS LATTICES 19 REFERENCES [1] G. Gratzer, General Lattice Theory, Second edition. Birkhauser Verlag, Basel, (1998). [2] S. J. Golan, Decomposition and Dimensions in Module Categories, Marcel Dekker, New York, (1977). [3] S. J. Golan, Torsion Theories, Longman Scientific and Technical, New York, 29 (1986). [4] O. Goldman, "Elements of noncommutative arithmetic", J. Algebra35, 308-341(1075) [5] S. J. Golan and H. Simmons, Derivatives, nuclei and dimensions on the frame of torsion theo- ries, Longman Scientific and Technical, New York, (1988). [6] P. Johnstone, Stone Spaces, Cambridge studies in advanced mathematics 3, London, 1992. [7] G. Mitchler, "Goldman's primary decomposition and the tertiary decomposition ", J. Algebra, 16 129-137(1970). [8] J. Picado, A. Pultr, Frames and Locales: Topology without points, Birkhauser, Boston, 2010. [9] N. Popescu, Abelian categories with applications to rings and modules, London Mathematical Society Monographs No.3, Academic Press, New York, 1973. [10] H. Simmons, "Torsion theoretic points and spaces ", Proceedings of the Royal society of Edin- burgh, 96, 345-361 (1984). [11] H. Simmons, "Near Discreteness of modules and spaces measured by Gabriel and Cantor", J. Pure and Applied Algebra, 56 119-162 (1989). [12] H. Simmons, A collection of notes on frames , http://www.cs.man.ac.uk/∼hsimmons/FRAMES/frames.html. [13] H. Simmons, "A decomposition theory for complete modular meet-continuous lattices ", Alge- bra Universalis, 64, 349-377 (2011). [14] H. Simmons, An introduction to idioms, http://www.cs.man.ac.uk/∼hsimmons/00-IDSandMODS/ [15] H. Cantor-Bendixson, Simmons, socle and atomicity, http://www.cs.man.ac.uk/∼hsimmons/00-IDSandMODS/ (2) [16] H. Simmons, The Gabriel and the Boyle derivatives for a modular idiom,http://www.cs.man.ac.uk/∼hsimmons/00-IDSandMODS/ (4) [17] H. Simmons, Gabriel, Loewy, and Cantor-Bendixson sometimes agree http://www.cs.man.ac.uk/∼hsimmons/00-IDSandMODS/ (9) [18] B. Stenstrom, Rings of Quotients: An Introduction to Methods of Ring Theory, Springer Verlag, Berlin,1975. INSTITUTO DE MATEM ´ATICAS, CIUDAD UNIVERSITARIA, UNAM, M ´EXICO, D. F., 04510, M ´EXICO. E-mail address: [email protected], [email protected]
1905.08219	2	1905	2019-08-29T19:36:28	On the notion of Krull super-dimension	[ "math.RA" ]	We introduce the notion of Krull super-dimension of a super-commutative super-ring. This notion is used to describe regular super-rings and calculate Krull super-dimensions of completions of super-rings. Moreover, we use this notion to introduce the notion of super-dimension of any irreducible superscheme of finite type. Finally, we describe nonsingular superschemes in terms of sheaves of K\"{a}hler superdifferentials.	math.RA	math	ON THE NOTION OF KRULL SUPER-DIMENSION A.MASUOKA AND A. N. ZUBKOV Abstract. We introduce the notion of Krull super-dimension of a super- commutative super-ring. This notion is used to describe regular super-rings and calculate Krull super-dimensions of completions of super-rings. Moreover, we use this notion to introduce the notion of super-dimension of any irreducible superscheme of finite type. Finally, we describe nonsingular superschemes in terms of sheaves of Kahler superdifferentials. introduction The concept of Krull dimension is one of the most fundamental concepts of the theory of commutative rings. On the basis of this concept, one can define the dimension of an algebraic variety or, more generally, the dimension of a scheme. In contrast, so far no one has defined the concept of super-dimension of a su- perscheme, which would be naturally derived from the properties of its sheaf of "super-functions". Motivated by the above, we introduce the concept of Krull super-dimension of a (super-commutative) Noetherian super-ring. We develop some fragment of dimension theory of Noetherian super-rings. For example, we show that any finitely generated superalgebra A contains a polynomial super-subalgebra C of the same Krull super-dimension, such that A is a finitely generated C-supermodule. This result can be regarded as a super Noether Normalization Theorem. Further, we investigate local Noetherian super-rings and give a criteria when such a super-ring is regular (Proposition 5.3 and Theorem 5.5). We generalize this criteria for all (not necessary local) Noetherian super-rings (Proposition 5.8). Note that our definition of regularity is based on the notion of Krull super-dimension, and we do not use the notion of regular sequence as in [12], but our results coincide with the results therein (see Remark 5.9). Finally, we show how the super-dimension of a completion of Noetherian super-ring is determined by the super-dimensions of its localizations (Theorem 5.13). We also describe certain local Noetherian superalgebras in terms of Kahler su- perdifferentials (Theorem 5.16). This result is used to describe nonsingular irre- ducible superschemes of finite type over a perfect field. More precisely, such a su- perscheme X is nonsingular if and only if its sheaf of Kahler superdifferentials ΩX/K is a locally free sheaf of OX -supermodules of rank equal to the super-dimension of X (Theorem 6.3). In connection with the above result, one can note a new phenomena in the theory of superschemes, that does not take place in the theory of schemes. First, there are integral superschemes, which are singular at any point. Second, an integral superscheme X may contain a proper closed integral super-subscheme Y of the same super-dimension! Nevertheless, we show that if both X and Y are integral 1 2 A.MASUOKA AND A. N. ZUBKOV and generically nonsingular, then the coincidence of their super-dimensions implies X = Y (Theorem 6.8). The paper is organized as follows. In the first section we collect the necessary results on super-rings and supermodules. The second section is devoted to studying superschemes and sheaves of supermodules over them. In the third section we recall the definition of supermodule of relative differential forms and formulate some standard properties of it. These properties are generalized for sheaves of Kahler superdifferentials. The Krull super-dimension is introduced in the fourth section. As it has been mentioned above, we prove a super Noether Normalization Theorem and give a sim- ple algorithm how to calculate the odd Krull dimension of a given finitely generated superalgebra (Lemma 4.4). Using this algorithm, we construct a superalgebra A and its quotient A/I, such that the odd Krull dimension of A/I is greater than the odd Krull dimension of A! We also discuss the case of one relation superalgebra and calculate the super-dimension of a completion of a Noetherian super-ring (Theorem 5.13.) The fifth section is devoted to regular Noetherian super-ring (see above). In the sixth section we introduce the notion of super-dimension of an irreducible su- perscheme of finite type over a field. We characterize nonsingular superschemes, as well as their closed nonsingular super-subschemes, in terms of their sheaves of Kahler superdifferentials. In the seventh section we formulate some questions and open problems, those would stimulate the further progress in the dimension theory of super-rings and superschemes. 1. Super-rings and supermodules 1.1. Super-rings. A Z2-graded ring R (with unity) is called a super-ring. Let r 7→ r be a parity function on the set of non-zero homogeneous elements of R, i.e. r = i if and only if r ∈ Ri, i ∈ Z2. A homogeneous non-zero element r is called even, provided r = 0, otherwise r is called odd. In what follows, all homomorphisms of super-rings are supposed to be graded, unless otherwise stated. 1.2. Supermodules. Let R be a super-ring. A left Z2-graded R-module M is called an R-supermodule. The category of R-supermodules with graded morphisms (as well as the category of right R-supermodules), denoted by Rmod (respectively, by modR), is obviously abelian. If M is an R-supermodule, then let m 7→ m be a parity function on the set of m = i if and only if m ∈ Mi, i ∈ Z2. non-zero homogeneous elements of M , i.e. As above, a non-zero homogeneous element m is called even, provided m = 0, otherwise m is called odd. 1.3. Super-commutative super-rings. A super-ring R = R0 ⊕ R1 is said to be super-commutative, if the following are satisfied: (i) rs = sr, either if r, s ∈ R0, or if r ∈ R0 and s ∈ R1; (ii) s2 = 0, if s ∈ R1. These two conditions are equivalent to rs = (−1)rssr, provided 2 is not zero divisor in R. On the other hand, if 2 is a zero divisor in R, then R is super- commutative if and only if R is a commutative ring, which satisfies (ii). ON THE NOTION OF KRULL SUPER-DIMENSION 3 Example 1.1. Let B be a commutative ring, and let M be an B-module. We let ∧B(M ) denote the exterior B-algebra on M . Supposing that all elements in M are odd, we regard ∧B(M ) as a super-commutative B-superalgebra. This is indeed the quotient of the tensor R-algebra TB(M ) on M , which is an B-superalgebra, divided by the relation m2 = 0, m ∈ M . If M is finitely generated projective, then the canonical map M → ∧B(M ) is an injection, and ∧B(M ) is finitely generated projective as an B-module. If M has finite B-free basis Y1, . . . , Ys, we write B[Y1, . . . , Ys] for ∧B(M ), and call it the polynomial B-superalgebra in odd indeterminants Y1, . . . Ys. Example 1.2. More generally, for any superalgebra A one can define a polynomial A-superalgebra A[X1, . . . Xm Y1, . . . Yn] in m even indeterminants X1, . . . , Xm and n odd indeterminants Y1, . . . , Yn, as A ⊗ K[X1, . . . , Xm][Y1, . . . , Yn] . If it does not lead to confusion, we use the shorter notation A[X Y ]. If R is super-commutative, then any left R-supermodule M can be regarded as an right R-supermodule, by setting mr = (−1)rmrm, r ∈ R, m ∈ M , and vice versa. In other words, the categories Rmod and modR are naturally isomorphic. Moreover, the category Rmod is a tensor category with a braiding tM,N : M ⊗R N ≃ N ⊗R M, m ⊗ n 7→ (−1)mnn ⊗ m, m ∈ M, n ∈ N. From now on all super-rings are assumed to be super-commutative. Let R be a super-ring. It is obvious that any left super-ideal I of R is also right, hence two-sided. A super-ideal p is called prime (respectively, maximal ), provided R/p is an integral domain (respectively, a field). A localization of an R-supermodule M at a prime super-ideal p is defined as Mp = (R0 \ p0)−1M . Let IR denote the super-ideal RR1, and let R denote the quotient-ring R/IR. A super-ideal p is prime (respectively, maximal) if and only if IR ⊆ p and p/IR is prime (respectively, maximal) ideal of R. The intersection of all prime super-ideals of R coincides with the largest nil super-ideal of R, called a nil-radical of R and it is denoted by nil(R). It is obvious that IR ⊆ nil(R). A super-ring R with a unique maximal super-ideal is said to be local. For exam- ple, if p is a prime super-ideal of R, then its localizaion Rp is a local supe-ring with the maximal super-ideal Rpp. A morphism α : R → S between local super-rings is said to be local if α(m) ⊆ n, where m and n are the unique maximal super-ideals of R and S, respectively. 1.4. Super-vector spaces. If R is a field K, then an object V from Kmod is called a super-vector space over K, and the super-dimension sdimK(V ) of V is defined by sdimK(V ) = r s, where r = dimK(V0), s = dimK(V1). 1.5. Superdomains and superfields. A super-ring A is called reduced, if the ring A = A/IA is reduced, i.e. IA = nil(A). The following lemma is a folklore. Lemma 1.3. A super-ring A is reduced if and only if for any prime super-ideal p of A, the super-ring Ap is. 4 A.MASUOKA AND A. N. ZUBKOV We say that a super-ring A is an integral superdomain (or just superdomain), if IA is a prime super-ideal of A. If, additionally, the natural superalgebra morphism A → AIA is injective, then A is a strong superdomain. The additional condition is equivalent to that for any s ∈ A0 \ A2 1 and any a ∈ A the equality sa = 0 implies a = 0. This, in turn, is equivalent to the apparently stronger condition that for any s ∈ A \ IA and any a ∈ A the equality sa = 0 implies a = 0. Indeed, if sa = 0, then s0a = s1a, whence s0(s1a) = s2 1a = 0. The former condition ensures first s1a = 0, then s0a = 0, and finally a = 0. A superring A is said to be a superfield if any element a ∈ A \ IA is invertible, or equivalently, if any a ∈ A0 \ A2 1 is invertible. In other words, A is a superfield if and only if A is a local super-ring with the unique maximal super-ideal IA if and only if IA is a maximal super-ideal. Obviously, a superfield is a strong superdomain. For example, the polynomial superalgebra ∧K(Y1, . . . , Yr) over a field K in odd indeterminants Y1, . . . , Yr, is a superfield. If A is a superdomain, then AIA is obviously a superfield that is called a superfield of fractions of A and it is denoted by SQ(A). 1.6. Noetherian super-rings. Recall that a super-ring R is called Noetherian if the super-ideals of R satisfy ascending chain condition (ACC). As it has been proven in [10], R is Noetherian if and only if it is left or right Noetherian as a ring. Lemma 1.4. A super-ring R is Noetherian if and only if R0 is a Noetherian ring and the R0-module R1 is finitely generated. Proof. The "if" part. The assumptions imply that the R0-submodules of R satisfy the ACC, whence R is Noetherian. The "only if" part. Given an ascending chain r0 ⊂ r1 ⊂ . . . of ideals of R0 (respectively, of R0-submodules of R1), we have the ascending chain r0 ⊕ r0R1 ⊂ r1 ⊕ r1R1 ⊂ . . . (respectively, r0R1 ⊕ r0 ⊂ r1R1 ⊕ r1 ⊂ . . . ) of super-ideals of R. Therefore, if R is Noetherian, then R0 is Noetherian, and R1 is a Noetherian R0-module as well, hence finitely generated. (cid:3) 1.7. Completion. Let R be a super-ring and M be an R-supermodule. If I is a super-ideal of R, then M can be endowed with the I-adic topology so that a subset U ⊆ M is open if and only if U = ∪j∈J (mj + I kj M ) for some mj ∈ M, kj ∈ Z≥0. In other words, R turns into a topological super-ring and M turns into a topological R-supermodule with respect to their I-adic topologies. Similarly, M , and R as well, can be endowed with I0-adic topologies, being regarded as R0-modules. It is clear that the I0-adic topologies of both M and R coincide with their RI0-adic topologies. Lemma 1.5. These two topologies coincide. Proof. It suffices to prove that for any integer k > 0, I kM ⊆ I non-trivial first inclusion follows, by using I 2 1 ⊆ I0, as follows. [ k 2 ] 0 M ⊆ I [ k 2 ]M . The I k = X0≤i≤k I k−i 0 I i 1 ⊆ Xi even k− i I 2 0 + Xi odd k− i+1 I 0 2 I1 ⊆ I [ k 2 ] 0 I. (cid:3) ON THE NOTION OF KRULL SUPER-DIMENSION 5 The I-adic completion R/I n ←− bR = lim R/I k) form a base of neighborhhods of zero. of R is naturally a topological super-ring, in which the super-ideals bR(k) = ker(bR → We have the canonical map R → bR of topological super-rings. If it is an isomor- Finally, if R is a local super-ring with a maximal super-ideal m, then we just say phism, we say that R is complete with respect to its I-adic topology. that R is complete, provided R is complete with respect to its m-adic topology. Similarly, the I-adic completion M/I nM ←− cM = lim of an R-supermodule M has a natural structure of a topological bR-supermodule, in which the super-submodules cM (k) = ker(cM → M/I kM ) form a base of neigh- borhhods of zero. We also have the canonical map M → cM of topological R- supermodules. Remark 1.6. Lemma 1.5 implies that R/I n 0 R, ←− bR ≃ lim I0-adic completion of Ri. The similar statement holds for R-supermodules. R/ ∩n≥0 I n and M ′ = M/ ∩n≥0 I nM . Moreover, R′ and M ′ are Hausdorff spaces with respect to their I ′-adic topologies, where I ′ = I/ ∩n≥0 I n. Using this remark, one can easily superize Theorem 5 and Corollary 1, [14], chapter VIII, as follows. and, therefore, the homogeneous components bRi, i = 0, 1, are isomorphic to the Observe that bR and cM are naturally isomorphic to cR′ and cM ′, where R′ = Lemma 1.7. Let M be a finitely generated R-supermodule. Then cM = bRM . Corollary 1.8. Assume additionally that I is finitely generated. Then cM (k) = I kM = I kcM and bR(k) =cI k = I kbR =bI k for any k ≥ 0. Then the functor M → cM , that takes a finitely generated R-supermodule to its I-adic completion, is exact. Moreover, bR is a Noetherian super-ring, hence cM is Proposition 1.9. Let R be a Noetherian super-ring and I be a super-ideal of R. Proof. By Lemma 1.4 any finitely generated R-supermodule is finitely generated as an R0-module. It remains to combine Lemma 1.5, Remark 1.6 and Lemma 1.7 with Theorem 54 and 23(K), [9]. (cid:3) finitely generated whenever M is. [ Proposition 1.10. Let R be a Noetherian super-ring, I be a super-ideal of R, M be a finitely generated R-supermodule. Then ∩k≥0I kM consists of all elements m ∈ M such that there is x ∈ I0 with (1 − x)m = 0. In particular, if R is a local super-ring with a unique maximal super-ideal m, then ∩k≥0mkM = 0, that is M is a Hausdorff space with respect to its m-adic topology. Proof. As was already observed, M is a finitely generated R0-module and ∩k≥0I kM = ∩k≥0I k cludes the proof. 0 M . Then Krull intersection theorem (see [4], III, §3, Proposition 5) con- (cid:3) 6 A.MASUOKA AND A. N. ZUBKOV Lemma 1.11. Let R be a local Noetherian super-ring with a maximal super-ideal m. For its m-adic completion bR the following hold : (1) bR is a complete Noetherian local super-ring with maximal bm. (2) I bR coincides with cIR = bRR1. (3) bR = bR/I bR coincides with the m-adic completion bR of R, where m = m0/R2 Proof. The commutative ring bR0 is a m0-adic completion of the local ring R0, hence local. Moreover, the maximal ideal of bR0 coincides with the m0-adic completion of m0, that in turn coincides with bm0. Proposition 1.9 infers (1) and (3) as well. R, using Remark 1.6. It is cIR on one hand, and is I bR on the other hand. This Consider the image of the m0-adic completion of the canonical map R ⊗R0 R1 → proves (2). 1. (cid:3) 2. Superschemes For the details of the content of this section we refer to [5, 8, 10, 13, 15]. 2.1. Geometric superspaces. Recall that a geometric superspace (or local ringed superspace) X consists of a topological space X e and a sheaf of super-commutative super-rings OX such that all stalks OX,x, x ∈ X e, are local super-rings. A morphism of superspaces f : X → Y is a pair (f e, f ∗), where f e : X e → Y e is a morphism of topological spaces and f ∗ : OY → f e ∗ OX is a morphism of sheaves such that f ∗ x : OY,f (x) → OX,x is a local morphism for any x ∈ X e. Let V denote the category of geometric superspaces. Let X be a geometric superspace. If U is an open subset of X e, then (U, OXU ) is again a geometric superspace, which is called an open super-subspace of X. In what follows (U, OX U ) is denoted just by U . Let X be a geometric superspace. The sheafification of the pre-sheaf U → IOX (U) = OX (U )(OX )(U )1 is a sheaf of OX -super-ideals, that is denoted by IX . The purely even geometric superspace (X e, OX /IX ) is denoted by Xev, and by Xres, when it is regarded as an geometric space. Finally, with each geometric superspace X one can associate a purely even geo- metric superspace X0 = (X e, (OX )0), that can be also regarded as a geometric space. 2.2. Superschemes. Let R be a super-ring. An affine superscheme SSpec R can be defined as follows. The underlying topological space of SSpec R coincides with the prime spectrum of R, endowed with the Zariski topology. For any open sub- set U ⊆ (SSpec R)e the super-ring OSSpec R(U ) consists of all locally constant functions h : U → ⊔p∈U Rp such that h(p) ∈ Rp, p ∈ U . For any f ∈ R0 let D(f ) denote the open subset {p ∈ (SSpec R)e f 6∈ p}. As in the purely even case, D(f ) is isomorphic to SSpec Rf . Affine superschemes form a full subcategory of V, which is anti-equivalent to the category of super-rings. A superspace X is called a (geometric) superscheme if there is an open covering X e = ∪i∈I Ui, such that each open super-subspace Ui is isomorphic to an affine superscheme SSpec Ri. Superschemes form a full subcategory of V, denoted by SV. If X is a superscheme, then any its open super-subspace is a superscheme, called an open super-subscheme. A superscheme Z is a closed super-subscheme of X, if ON THE NOTION OF KRULL SUPER-DIMENSION 7 there is a closed embedding ι : Z e → X e such that the sheaf ι∗OZ is an epimorphic image of the sheaf OX . For example, Xev is a closed super-subscheme of X. Finally, a superscheme Z is said to be a super-subscheme of X, if Z is isomorphic to a closed super-subscheme of an open super-subscheme of X. It can be easily seen that Zres is an (open, closed) subscheme of Xres, provided Z is an (open, closed) super-subscheme of X. A superscheme X is called irreducible, if the topological space X e is irreducible. Thus it obviously follows that X is irreducible if and only if Xev is if and only if Xres is. A superscheme X is said to be Noetherian if X can be covered by finitely many open affine super-subschemes SSpec Ri with Ri to be Noetherian. Note that if a superscheme X is Noetherian, then Xres is a Noetherian scheme. The proof of the following lemma can be copied from the proof of Proposition II.3.2, [6]. Lemma 2.1. Let X be a superscheme. Then the following are equivalent : (a) X is Noetherian; (b) (1) X e is a quasi-compact topological space; (2) for any open affine super-subscheme U ≃ SSpec A of X, the super-ring A is Noetherian. Let K be a field. A superscheme X is said to be of finite type over K, if there is a finite open affine covering of X as above, such that each Ri is a finitely generated K-superalgebra. Similarly to the above lemma one can show that a superscheme X is of finite type over K if and only if X e is quasi-compact and for any open affine super-subscheme U ≃ SSpec A of X, A is a finitely generated K-superalgebra. 2.3. Integral superschemes. A superscheme X is called reduced, if Xres is a reduced scheme. Since OXres,x ≃ Ox, this property is local, i.e. X is reduced if and only if the superring Ox is reduced for any x ∈ X e (cf. [6], Exercise II.2.3(a)). In particular, an affine superscheme SSpec A is reduced if and only if the super- ring A is reduced. Moreover, a superscheme X is reduced if and only if any its open affine super-subscheme is. Obviously, if for any open subset U ⊆ X e the superring OX (U ) is reduced, then X is reduced. Nevertheless, we do not know whether the converse is true (compare with the definition in [6], II, §3). A superscheme X is called integral, if Xres is integral. If additionally each local superring Ox is a strong superdomain, then X is called strong integral. Proposition 2.2. A superscheme X is strong integral if and only if the following conditions hold : (1) X is irreducible and reduced; (2) for any its open affine super-subscheme U the superring O(U ) is a strong superdomain. Proof. By Proposition II.3.1, [6], Xres is integral if and only if Condition (1) holds. To prove "only if", assume that X is strong integral. Then Condition (1) holds. Therefore, for any open super-subscheme U ≃ SSpec A of X, A is a domain (cf. [6], Example II.3.0.1). Furthermore, Ap is a strong superdomain for any point p ∈ (SSpec A)e. This implies that the superdomain A is strong, ensuring Condition 8 A.MASUOKA AND A. N. ZUBKOV (2). Indeed, given s ∈ A0 \ A2 since it is so after localization at every p ∈ (SSpec A)e. 1, the multiplication a 7→ sa, A → A by s is injective Conversely, any local superring Ox can be identified with a local superring Ap of an open super-subscheme U ≃ SSpec A, where x = p ∈ U e. If A is a strong superdomain, then Ap is obviously a strong superdomain. This proves the "if" part. (cid:3) 2.4. Function superfield. The following lemma superizes Exercise II.3.6, [6]. Lemma 2.3. Let X be an integral superscheme and ξ ∈ X e be the generic point. Then Oξ is a superfield that is isomprphic to SQ(OX (U )) for any open affine super- subscheme U of X. Proof. Recall that a point ξ ∈ X e is generic if and only if {ξ} = X if and only if ξ belongs to any (not empty) open subset V ⊆ X e. Let U be an open affine super-subscheme of X. Then A = OX (U ) is a superdomain and the generic point ξ coincides with the smallest prime ideal IA. Thus our lemma obviously follows. (cid:3) Following [6] we call Oξ a function superfield of X and denote it by SK(X). 2.5. The functorial approach. There is an alternative way to define superspaces and superschemes as functors from the category of super-rings/superalgebras to the category of sets. Since we use this approach in Lemma 3.9 only, we will not introduce this stuff in a complete form. The interested reader can find all necessary notions/definitions in [10]. All we need to note is that the category SV is equivalent to a full subcategory, SF , of the category of the functors mentioned above (see [10, Theorem 5.14]). 2.6. Sheaves of OX -supermodules. Let R be a super-ring, and let M be an R-supermodule. Analogously to the purely even case, one can define an associ- constant functions h : U → ⊔p∈U Mp such that h(p) ∈ Mp, p ∈ U . A sheaf F of OX -supermodules is called quasi-coherent, if X can be covered by open affine super-subschemes Ui ≃ SSpec Ai, such that for each i there is an Ai- [6, 16]). If, additionally, each supermodule ated sheaf fM of OX -supermodules, where X = SSpec R. More precisely, for any open subset U ⊆ (SSpec R)e the OX (U )-supermodule fM (U ) consists of all locally supermodule Mi with FUi ≃ fMi (cf. Proposition 2.4. If X = SSpec A, then the functor M 7→ fM is an equivalence Mi is finitely generated, then the sheaf F is called coherent. of the category of A-supermodules and the category of quasi-coherent sheaves of X-supermodules (both with graded morphisms). Moreover, if A is a Noetherian superalgebra, then this functor is an equivalence of the category of finitely generated A-supermodules and the category of coherent sheaves of X-supermodules (both with graded morphisms). Proof. By Proposition 2.1, [15], this functor is full and faithful. Let F be a quasi- coherent sheaf of OX -supermodules. By Proposition 3.1, [15], and by Corollary II.5.5, [6] as well, FOX0 is Noetherian and F is coherent, then A0 is also Noetherian and M is a finitele generated A0-(super)module respectively. ≃ fM , where M is an A0-supermodule. Moreover, if A ON THE NOTION OF KRULL SUPER-DIMENSION 9 Let f denote the natural superscheme morphism X → X0 = SSpec A0, in- ) is a sheaf of OX - duced by the canonical embedding A0 → A. Then f ∗(FOX0 supermodules associated with the presheaf U 7→ OX (U ) ⊗OX0 (U) F (U ), U ⊆ X e 0 = X e. Moreover, the natural morphism f ∗(FOX0 induced by the morphism of presheaves (of OX -supermodules) ) → F of sheaves of OX -supermodules, OX (U ) ⊗OX0 (U) F (U ) → F (U ), recovers the structure of F as a sheaf of OX -supermodules. By Proposition 2.1 (5), [15], f ∗(FOX0 ) → F defines the ) ≃ ^A ⊗A0 M , hence the morphism f ∗(FOX0 structure of A-supermodule on M , such that F ≃ fM . Proposition is proven. The following proposition is a superization of Proposition II.5.9, [6]. (cid:3) Proposition 2.5. Let X be a superscheme. For any closed super-subscheme Y the super-ideal sheaf JY = ker(OX → i∗OY ) is quasi-coherent, where i is the corresponding closed embedding Y e → X e. If X is Noetherian, then JY is coherent. Conversely, any quasi-coherent super-ideal sheaf on X has the form JY for an uniquely defined closed super-subscheme Y of X. Proof. It is easy to see that Y0 is a closed subscheme of X0 with respect to the = i∗(OY OY0 same closed embedding i. Moreover, i∗OY OX0 ). By Proposition is a quasi-coherent sheaf of OX0 - 3.1, [15] and Proposition II.5.8(c), [6], i∗OY OX0 supermodules, hence, again by Proposition 3.1, [15], it is a quasi-coherent sheaf of OX -supermodules. Corollary 3.2, [15], infers that JY is quasi-coherent. By Proposition 2.4, the converse statement is proved just as proving Proposition II.5.9, [6]. For coherency of JY when X is Noetherian, copy verbatim the proof of Proposition II.5.9, [6]. (cid:3) Corollary 2.6. Combining Proposition 2.4 and Proposition 2.5, one sees that any closed super-subscheme of an affine superscheme SSpec A is isomorphic to SSpec A/I for a super-ideal I of A. The following lemma superizes Exercise II.5.7, [6]. Lemma 2.7. Let X be a Noetherian superscheme, and F be a coherent OX - supermodule. Then the following statements hold : (1) If the stalk Fx is a free Ox-supermodule for some point x ∈ X e, then there is a neighborhood U of x such that FU is a free OU -supermodule of the same rank; (2) F is a locally free OX -supermodule if and only if Fx is a free Ox-supermodule for any x ∈ X e. Proof. There are an open affine super-subscheme U ≃ SSpec A of X and a finitely generated A-supermodule M , such that x ∈ U and FU ≃ fM respectively. Then Fx ≃ Mp is a free Ap-supermodule, where p is a prime superideal of A, that corresponds to the point x. In other words, there are (homogeneous) elements m1, . . . , mt ∈ M , which form a basis of the free Ap-supermodule Mp. Let N denote a free A-supermodule with a basis n1, . . . , nt, such that the parity of each ni coincides with the parity of corresponding mi. Let u : N → M be a morphism of A-supermodules, induced by the map ni 7→ mi, 1 ≤ i ≤ t. Using 10 A.MASUOKA AND A. N. ZUBKOV Lemma 1.2, [16], and arguing as in [4], II, §5, Proposition 2, one can easily show that there is f ∈ A0 \ p0 such that (Coker)f = (Ker)f = 0, hence uf : Nf → Mf is an isomorphism. In particular, fMD(f ) ≃gMf is a free sheaf. The second statement is now obvious. 3. Kahler superdifferentials (cid:3) 3.1. Supermodules of relative differential forms. Let A be a superring, B be an A-superalgebra, and M be a left B-supermodule. Recall that M is also regarded as an right B-supermodule via mb = (−1)bmbm, b ∈ B, m ∈ M . An even or odd additive map d : B → M is called an A-superderivation if the following conditions hold : (1) d(ab) = (−1)adad(b) + d(a)b; (2) da = 0 for any a ∈ A. Let DerA(B, M ) denote a Z2-graded abelian group with DerA(B, M )i consisting of all superderivations of parity i = 0, 1. Observe that DerA(B, M ) has a natural structure of B-supermodule. The pairs (M, d), where M is a B-supermodule and d ∈ DerA(B, M ), form a category with (even) morphisms f : M → M ′ of B-supermodules such that d′ = f d. The proof of the following lemma is standard and we leave it for the reader. Lemma 3.1. (see [8], chapter 3, §1.8, or [6], II.8) There are B-supermodules ΩB/A,ev and ΩB/A,odd, two superderivations d0 : B → ΩB/A,ev and d1 : B → ΩB/A,odd, where d0 = 0, d1 = 1, such that for any B-supermodule M , composi- tions with d0 and d1 give isomorphisms (of abelian groups) DerA(B, M )0 ≃ HomB(ΩB/A,ev, M ) and DerA(B, M )1 ≃ HomB(ΩB/A,odd, M ) respectively. The B-supermodules ΩB/A,ev and ΩB/A,odd are called the even and odd super- modules of relative differential forms of B over A respectively. In what follows we denote ΩB/A,ev just by ΩB/A. Remark 3.2. If B is a finitely generated A-superalgebra, then ΩB/A is a finitely generated B-supermodule. Remark 3.3. Let B be a finitely generated A-superalgebra over a Noetherian superalgebra A. We have B ≃ A[X1, . . . , Xm Y1, . . . , Yn]/J, where the super-ideal J is generated by finitely many homogeneous elements, say f1, . . . , ft. One can easily show that ΩB/A ≃ F/N , where F is a free B-supermodule, freely generated by the elements d0X1, . . . , d0Xm, d0Y1, . . . , d0Yn, and N is a supersubmodule generated by d0f1, . . . , d0ft. Remark 3.4. Let Πd0 denote an odd superderivation B → ΠΩB/A,ev that takes b to (−1)bd0b, b ∈ B. Then (ΠΩB/A,ev, Πd0) ≃ (ΩB/A,odd, d1) with respect to the isomorphism ΠΩB/A,ev ≃ ΩB/A,odd of B-supermodules, induced by the map d0b 7→ (−1)bd1b. Lemma 3.1 infers that for any any B-supermodule M there is an isomorphism DerA(B, M )0 → DerA(B, M )1 of abelian groups, that takes (M, d) to (ΠM, Πd), where Πd(b) = (−1)bdb, b ∈ B. ON THE NOTION OF KRULL SUPER-DIMENSION 11 Proposition 3.5. Let f : B ⊗A B → B be the diagonal homomorphism b ⊗ b′ 7→ bb′, b, b′ ∈ B, and let I = ker f . Define a map d : B → I/I 2 by db = 1 ⊗ b − b ⊗ 1 (mod I 2). Then (I/I 2, d) ≃ (ΩB/A,ev, d0). In particular, (Π(I/I 2), Πd) ≃ (ΩB/A,odd, d1). Proof. The proof of the first statement can be copied from [9], Proposition (26.C). The second one follows by Remark 3.4. (cid:3) Lemma 3.6. (see [9], p.184) If A′ and B are A-superalgebras, let B′ = B ⊗A A′. Then ΩB′/A′ ≃ ΩB/A ⊗B B′. Furthermore, if S is a multiplicative system in B0, then ΩS−1B/A ≃ S−1ΩB/A. Proof. It is easy to see that the map B′ → ΩB/A ⊗B B′, defined by b ⊗ a′ 7→ d0b ⊗ a′, b ∈ B, a′ ∈ A′, is an A′-superderivation and it satisfies the property of universality. Similarly, the map S−1B → S−1ΩB/A, defined as b s 7→ sd0b − (d0s)b s2 , b ∈ B, s ∈ S, is an A-superderivation, that also satisfies the property of universality. (cid:3) In propositions below one finds some standard properties of supermodules of relative differential forms. Their proofs can be copied from [9], chapter 10, just verbatim. Proposition 3.7. (First Exact Sequence) Let A → B → C be superrings and superring morphisms. Then there is a natural sequence of C-supermodules ΩB/A ⊗B C → ΩC/A → ΩC/B → 0. Moreover, the map ΩB/A ⊗B C → ΩC/A has a left inverse if and only if any A- superderivation of B into any C-supermodule T can be extended to a superderivation of C into T . Proposition 3.8. (Second Exact Sequence) Let B be an A-superalgebra, let I be a super-ideal of B, let C = B/I and B′ = B/I 2. There is a natural exact sequence of C-supermodules I/I 2 → ΩB/A ⊗B C → ΩC/A → 0, where the first map takes b + I 2 to d0b ⊗ 1, b ∈ I. Moreover, ΩB/A ⊗B C ≃ ΩB′/A ⊗B′ C and the map I/I 2 → ΩB/A ⊗B C has a left inverse if and only if the extension 0 → I/I 2 → B′ → C → 0 is trivial. 3.2. Sheaves of Kahler superdifferentials. Lemma 3.9. Let f : X → Y be a superscheme morphism. For any open affine super-subschemes U ⊆ X and V ⊆ Y such that f (U ) ⊆ V , the induced morphism U ×V U → X ×Y X is an isomorphism onto an open super-subscheme of X ×Y X. Proof. By the remark after Proposition 5.12, [10], one needs to check the analogous statement in the category SF , which is obvious (see [7], I.1.7(3)). (cid:3) 12 A.MASUOKA AND A. N. ZUBKOV For the above morphism, let ∆ denote the diagonal morphism X → X ×Y X. Arguing as in [6, §8] and using Lemma 3.9, one can show that ∆ is an isomorphism of X onto a closed super-subscheme ∆(X) of an open super-subscheme W of X ×Y X, which is defined by a sheaf of super-ideals J ⊆ OW . Following [6, §8], we define the sheaf of Kahler superdifferentials of X over Y to be the OX -supermodule ΩX/Y = ∆∗(J /J 2). The isomorphism ∆ induces O∆(X) ≃ O∆(X) ≃ OW /J , through which we identify first OX with OW /J , and then ΩX/Y with OW /J - supermodule J /J 2. More precisely, let U = SSpec B be an affine open super-subscheme of X and V = SSpec A be an affine open super-subscheme of Y such that f (U ) ⊆ V , then U ×V U ≃ SSpec (B ⊗A B) and ∆(X) ∩ (U ×V U ) is a closed super-subscheme defined by the kernel of the diagonal homomorphism B ⊗A B → B. Proposition 3.5 implies that ΩU/V ≃ ^ΩB/A and by covering X and Y with U and V as above, one can define ΩX/Y by gluing the corresponding sheaves ^ΩB/A (cf. [6], Remark II.8.9.2). In particular, the OX -supermodule ΩX/Y is quasi-coherent. Moreover, if X is Noetherian and f is a morphism of finite type, then ΩX/Y is coherent. Finally, gluing superderivations d0 : B → ΩB/A one can construct a morphism OX → ΩX/Y of sheaves of superspaces, which is a superderivation of the local superrings at any point. Again, as in [6, §8] we formulate sheaf counterparts of the algebraic results of the previous subsection. Their proofs are standard and we leave them for the reader. Proposition 3.10. If f ′ : X ′ = X ×Y Y ′ → Y ′ is a base extension of f : X → Y with g : Y ′ → Y , then ΩX ′/Y ′ ≃ g′∗(ΩX/Y ), where g′ : X ′ → X is the first projection. Proposition 3.11. Let f : X → Y and g : Y → Z be superscheme morphisms. Then there is an exact sequence of OX -supermodules f ∗ΩY /Z → ΩX/Z → ΩX/Y → 0. Proposition 3.12. Let f : X → Y be a superscheme morphism, and let Z be a closed super-subscheme of X, defined by a superideal sheaf J . There is an exact sequence of OZ-supermodules J /J 2 → ΩX/Y ⊗OX OZ → ΩZ/Y → 0. 4. Krull super-dimension 4.1. Odd parameters. Let R be a Noetherian super-ring. Assume that the Krull dimension Kdim(R0) of R0 is finite. Let y1, . . . , ys be a sequence of odd elements in R1. For any subset I of the set s = {1, 2, . . . , s} we let yI =Yi∈I yi denote the product in R. This product can change only by sign, according to the order of the consisting elements. We may not and we will not refer to the order to discuss the product. Let AnnR0(yI ) = {r ∈ R0 ryI = 0} ON THE NOTION OF KRULL SUPER-DIMENSION 13 denote the ideal of R0 consisting of those elements which annihilate yI . We say that y1, . . . , ys form a system of odd parameters if Kdim(R0) = Kdim(R0/AnnR0 (ys)). In other words, the elements y1, . . . , ys form a system of odd parameters of R if and only if there is a longest prime chain of R0, say p0 ⊆ . . . ⊆ pn, n = Kdim(R0), such that AnnR0 (ys) ⊆ p0. Since for any I ⊆ s the ideal AnnR0 (yI ) is contained in AnnR0(ys), the elements yi, i ∈ I, form a system of odd parameters whenever y1, . . . , ys do. Let r = Kdim(R0). Let s be the largest number of those elements in R1 which form a system of odd parameters. The Krull super-dimension Ksdim(R) of R is defined by Ksdim(R) = r s. Moreover, the Krull dimension of R0, that is r, is called the even Krull dimension of R, and s is called the odd Krull dimension of R. They are denoted by Ksdim0(R) and Ksdim1(R), respectively. Finally, Ksdim1(R) = 0 if and only if for any y ∈ R1 and for any prime chain p0 ⊆ . . . ⊆ pn in R0 of length n = Kdim(R0) we have AnnR0 (y) 6⊆ p0. Moreover, since R1 is a finitely generated R0-module, the latter is equivalent to AnnR0 (R1) 6⊆ p0 for any prime p0 as above. Let B be a commutative ring and M be a B-module. Recall that dim(M ) denotes the dimension of M as a B-module, i.e. dim(B) = Kdim(B/AnnB(M )). Proposition 4.1. Let R be a Noetherian super-ring with Kdim(R0) < ∞. Choose a system of generators of R0-module R1, say y1, . . . , yd. Then the following conditions are equivalent : (1) There is a system of odd parameters of R of cardinality l ≥ 1; (2) For some 1 ≤ i1 < . . . < il ≤ d the elements yi1, . . . , yil form a system of odd parameters; (3) dim(Rl 1) = Kdim(R0). In particular, we have Ksdim1(R) = max{l dim(Rl 1) = Kdim(R0)}. Proof. For a given l ≤ Ksdim1(R), ters. Then AnnR0(Rl if dim(Rl where r = Kdim(R0), such that let z1, . . . , zl be a system of odd parame- 1) = Kdim(R0). Conversely, 1) = Kdim(R0), then there is a longest prime chain p0 ⊆ . . . ⊆ pr in R0, 1) ⊆ AnnR0(zl) implies dim(Rl ∩I⊆d,I=lAnnR0(yI ) = AnnR0 (Rl 1) ⊆ p0, and therefore AnnR0 (yI ) ⊆ p0 for some I. Proposition is proven. (cid:3) 4.2. Regular sequences. Recall that an odd element y ∈ R is called odd regular, if AnnR(y) = Ry (cf. [12, p.67]). Besides, a sequence y1, . . . , yt of odd elements of R is said to be odd regular if for each i the element yi is odd regular modulo Ry1 + . . . + Ryi−1. By [12, Corollary 3.1.2] the sequence y1, . . . , yt is odd regular if and only if AnnR(yt) = Ry1 + . . . + Ryt. Thus any odd regular sequence form a system of odd parameters. 14 A.MASUOKA AND A. N. ZUBKOV Lemma 4.2. Let R be a Noetherian super-ring of Krull super-dimension rs. Let y ∈ R1. The following hold : (a) If y is contained in a system of odd parameters of the largest length s, then Ksdim1(R/Ry) ≥ s − 1. (b) If y1, . . . , yt is an odd regular sequence, then t ≤ s and Ksdim(R/(Ry1 + . . . Ryt)) = r(s − t). (c) Any odd regular sequence can be extended to a system of odd parameters of the largest length s. Proof. (a) Let y1, . . . , ys be a system of odd parameters such that ys = y. Then y1, . . . , ys−1 form a system of odd parameters modulo Ry, which proves (a). Indeed, there obviously holds AnnR0/R1y(y(s−1)mod(Ry)) ⊆ AnnR0(ys)/R1y, and the latter is obviously included in the first of p0/R1y ⊆ . . . ⊆ pr/R1y, where (Ry1 ⊆)p0 ⊆ . . . ⊆ pr is some longest prime chain of R0. (b) To prove this when t = 1, assume that y is odd regular. Obviously we have Ksdim0(R/Ry) = r and s ≥ 1. If Ksdim1(R/Ry) ≥ s, then there is a system of odd parameters of length s modulo Ry, say y1, . . . , ys. Since rysy = 0 is equiv- alent to rys ∈ Ry, AnnR0/R1y(ysRy) coincides with AnnR0 (ysy)/R1y. Therefore, y1, . . . , ys, y form a system of odd parameters. This contradiction, combined with (a), proves the result when t = 1. For the general case use the obtained result repeatedly. (c) Assume that y1, . . . , yt is an odd regular sequence. Then we have t ≤ s by (b). Moreover, odd elements yt+1, . . . , ys can be chosen so that they form, modulo Ry1 + . . . + Ryt, a system of odd parameters. We see that y1, . . . , yt, . . . , ys is a desired system of odd parameters. (cid:3) 4.3. Noether Normalization Theorem for superalgebras. Let K be a field. Suppose that A is a finitely generated K-superalgebra. Then the K-algebra A0 is finitely generated. By the Noether Normalization Theorem [9, (14G)], A0 contains a polynomial subalgebra B = k[X1, . . . , Xr] over which A0 is integral, and so r = Ksdim0(A). The following proposition shows that odd parameters are nothing else but ele- ments algebraically independent over a subalgebra generated by "even" param- eters. Geometrically, this means that the even and odd components of the Krull super-dimension of A are equal to the number of even and odd degrees of free- dom of the superscheme SSpec A. More precisely, if Ksdim(A) = r s, then there is a certain (finite) superscheme morphism SSpec A → Ars = SSpec K[X1, . . . Xr Y1, . . . , Ys]. Proposition 4.3. For a sequence of odd elements y1, . . . , ys the following are equiv- alent: (a) y1, . . . , ys form a system of odd parameters; (b) For some/any polynomial subalgebra B ⊂ A0 as above, AnnB(ys) := B ∩ AnnA0(ys) equals 0; ON THE NOTION OF KRULL SUPER-DIMENSION 15 (c) For some/any polynomial subalgebra B ⊂ A0 as above, the B-superalgebra map ν : B[Y1, . . . , Ys] → A which is defined on the polynomial B-superalgebra in s odd variables by ν(Yi) = yi, 1 ≤ i ≤ s, is an injection. In particular, Ksdim1(A) = 0 if and only if AnnB(A1) 6= 0. Proof. There holds (a) ⇔ (b) ⇐ (c). Indeed, note from [9, Theorem 20] that Kdim(A0/AnnA0 (ys)) = Kdim(B/AnnB(ys)). Since B is an integral domain, any its non-zero ideal has a hight at least 1. Thus Kdim(B/AnnB(ys)) ≤ Kdim(B) − ht(AnnB(ys)) < Kdim(B) if and only if AnnB(ys) 6= 0. For (b) ⇒ (c), we wish to prove, assuming (b), that aI yI = 0, aI ∈ B, XI,I⊆s implies that all the coefficients aI equal 0. Assume the contrary. Choose aI 6= 0 with minimal I. Then, multiplying by the product ys\I , one obtains aI ys = 0, a contradiction. Since A1 is a finitely generated B-module, the last statement is now obvious. (cid:3) A ≃ A1/A3 Choose elements y1, . . . , yn ∈ A1, which form a system of generators of the B- module IA/I 2 1. Then the B-module IA is generated by the elements yI = yi1 . . . yik , where I = {i1 < . . . < ik} runs over all subsets of n = {1, 2, . . . , n}. For each I let JI denote the super-ideal AnnB(yI ). These super-ideals are par- tially ordered by JI ⊆ JI ′ whenever I ⊆ I ′. Define a set Γ = {I ⊆ n JI = 0}. By the above, for any I ∈ Γ the inclusion I ′ ⊆ I infers I ′ ∈ Γ. Let k denote max{I I ∈ Γ}, where I denotes the cardinality of I. Lemma 4.4. We have k = Ksdim1(A). Proof. Choose a set I ∈ Γ of maximal cardinality k. Proposition 4.3 (b) implies that the elements yi, i ∈ I, form a system of odd parameters. Assume that there is a system of odd parameters of cardinality k + 1, say z1, . . . , zk+1. Then and the product zk+1 is equal to I yI, b(j) b(j) zj =XI⊆n ( XI1⊔...⊔Ik+1=I XI⊆n I ∈ B, 1 ≤ j ≤ k + 1, ±b(1) I1 . . . b(k+1) Ik+1 )yI . Since the cardinality of each I from the above sum is at least k + 1, we have 0 6= QI≥k+1 JI ⊆ AnnB(zk+1), hence z1, . . . , zk+1 do not form a system of odd parameters. This contradiction concludes the proof. (cid:3) It is well known that the Krull dimension of a factor-ring A/I is at most the Krull dimension of a ring A. Surprisingly, the odd Krull dimension of a quotient of a super-ring A can be greater than the odd Krull dimension of A. Recall that a polynomial K-superalgebra K[X1, . . . , Xm Y1, . . . , Yn] is denoted, briefly, by K[X Y ]. 16 A.MASUOKA AND A. N. ZUBKOV Example 4.5. Let us consider a superalgebra A = K[X Y ]/J, where J = XI⊆n,I∩l6=∅ K[X]X1Y I , and n ≥ l ≥ 1, m ≥ 1. Then B = K[X] is isomorphically mapped onto a subalgebra of A, over which A0 is a finite module. Moreover, the residue classes of Y1, . . . , Yn form a system of generators of a B-module IA/I 2 A. Since AnnB(Y I ) 6= 0 if and only if I ∩ l 6= ∅, Lemma 4.4 implies that the residue classes of Yl+1, . . . , Yn form a system of odd parameters of the largest cardinality in A, i.e. Ksdim(A) = (mn − l). On the other hand, the superalgebra C = K[X Y ]/K[X Y ]X1 ≃ K[X2, . . . , Xm Y ] is a quotient of A. Moreover, Ksdim(C) = (m − 1 n), hence Ksdim1(C) > Ksdim1(A)! Lemma 4.6. Let A be a finitely generated K-superalgebra, and S be a multiplicative subset of A0. If nil(A) is a prime super-ideal, then Ksdim1(A) = Ksdim1(S−1A). Proof. Since nil(A0) is a smallest nilpotent prime ideal of A0, S−1nil(A0) is a small- , . . . , yt est nilpotent prime ideal of S−1A0 as well. Therefore, the elements y1 ∈ st s1 S−1A1 form a system of odd parameters if and only if AnnS−1B0 ( y1 . . . yt ) ⊆ st s1 S−1nil(A0). Using AnnS−1A0 ( y1 ) = S−1AnnA0(y1 . . . yt), one obtains the s1 required equality. (cid:3) . . . yt st 4.4. One relation superalgebras. Let A denote the polynomial superalgebra K[XY ] in r even and s odd free generators. For a nonzero homogeneous element f =XL⊆s cLY L, cL ∈ K[X], let B denote an one relation superalgebra A/Af . Below we discuss the problem of calculating of Ksdim1(B). To simplify our calculations we suppose that c∅ = 0, that is f ∈ IA. Since B ≃ A ≃ K[X], Lemma 4.6 infers that Ksdim1(B) = Ksdim1((K[X] \ 0)−1B). In other words, without loss of generality one can assume that A = K[Y ]. Further, by Lemma 4.4 the odd Krull dimension of B is equal to max{L Y L 6∈ Af }. The set Exp(f ) = {L cL 6= 0} is partially ordered by inclusion. Let L1, . . . , Lt be all pairwise different minimal elements of Exp(f ). The collection L1, . . . , Lt is called a basement of f . Observe that if Y L belongs to Af , then there is 1 ≤ i ≤ t, such that Li ⊆ L. Therefore, if L does not contain any Li, then Y L 6∈ Af . Since Y s is equal to Y s\L1f (up to a nonzero scalar multiple), there always holds Ksdim1(B) ≤ s − 1. A subset K ⊆ s of minimal cardinality, which meets each Li, is said to be If K is an extremal set of f , then k = K is called the the extremal set of f . index of f , and it is denoted by ind(f ). For example, ind(f ) = 1 if and only ∩1≤i≤tLi 6= ∅. Furthermore, ind(f ) ≤ t and ind(f ) = t if and only if Li ∩ Lj = ∅ for any 1 ≤ i 6= j ≤ t. ON THE NOTION OF KRULL SUPER-DIMENSION Lemma 4.7. We have Ksdim1(B) ≥ s − ind(f ). Proof. Let K be an extremal set of f . One easily sees that Y s\K 6∈ Af . 17 (cid:3) Lemma 4.7 implies that the worst lower bound for Ksdim1(B) is s − t. The following example shows this estimate is achievable. Example 4.8. Let ind(f ) = t and each Li has cardinality at least t. Then Ksdim1(B) = s − t. In fact, if L ≥ s − t + 1, then s \ L is not extremal, whence Li ⊆ L. Moreover, any Lj, j 6= i, meets L \ Li. Thus ±cLiY L = Y L\Lif . Lemma 4.9. For any 1 ≤ i ≤ t there is an element g = Y Li + h, such that the following conditions hold : (a) the minimal elements of Exp(h) are exactly Lj, j 6= i; (b) for any nonzero term dLY L of h, the exponent L does not contain Li; (c) Af = Ag. Proof. The polynomial f can be represented as f = Y Lip + h′, where p is an invertible element of A, such that the exponent of any its nonzero term does not meet Li. Furthermore, the minimal elements of Exp(h′) are exactly Lj, j 6= i, and the exponent of any its nonzero term does not contain Li. It is now obvious that g = p−1f is the required polynomial. (cid:3) An element g as in the above lemma is called a form of f reduced in Li. Corollary 4.10. If some Li is a singleton, then Ksdim1(B) = s − 1. Proof. Let g = Yj + h be a form of f reduced in Li = {j}. The map Yj 7→ g, Yk 7→ Yk, k 6= j, induces an automorphism of superalgebra A, which takes the odd regular element Yj to the odd regular element g. Lemma 4.9(c) and Lemma 4.2(b) imply Ksdim1(B) = s − 1. (cid:3) The following proposition shows that the lower bound in Lemma 4.7 is not always sharp. Proposition 4.11. If t = 2, then Ksdim1(B) = s − 2 if and only if L1 ∩ L2 = ∅ and both L1 and L2 are at least 2, otherwise Ksdim1(B) = s − 1. Proof. By Lemma 4.7 we have Ksdim1(B) ≥ s−2. Moreover, if Ksdim1(B) = s−2, then L1 ∩ L2 = ∅. Besides, Corollary 4.10 infers L1, L2 ≥ 2. Conversely, assume that L1 ∩ L2 = ∅ and both L1 and L2 have cardinalities at least two. Then any 1 ≤ i ≤ s does not belong either to L1, or to L2. In both cases Y s\i ∈ Af . For example, if i 6∈ L1, then (s \ (L1 ⊔ i)) ∩ L2 6= ∅ and therefore, Y s\(L1⊔i)f is equal to ±cL1Y s\i ± cL1⊔iY s. Since Y s belongs to Af , so Y s\i does. (cid:3) We do not know whether Ksdim1(B) is completely defined by the basement of f for any t ≥ 3, similarly to the above proposition. 18 A.MASUOKA AND A. N. ZUBKOV 5. Regular super-rings From now on all super-rings are supposed to be Noetherian, unless otherwise stated. Let A be a local super-ring with a maximal super-ideal m. Let K denote its residue field A/m. Observe also that Kdim(A0) = r < ∞ (cf. [1, Corollary 11.11]). Set Ksdim(A) = rs. Note that (1) m/m 2 = (m0/m 2 0 + A2 1) ⊕ (A1/m0A1). This is a super-vector space over K, which is finite-dimensional since m is finitely generated as an A0-module. Lemma 5.1. If y1, . . . , ys form a system of odd parameters consisting of s elements, we have s ≤ dimK((m/m 2)1). Proof. Suppose t = dimK((m/m2)1). Let z1, . . . , zt be elements of A1 which give 2)1 = A1/m0A1. By Nakayama's Lemma this is equivalent rise to a K-basis in (m/m to saying that z1, . . . , zt form a minimal system of generators of the A0-module A1. Therefore, each yi is an A0-linear combination of them. If s > t, it follows that ys = 0, whence y1, . . . , ys cannot form a system of odd parameters. (cid:3) Lemma 5.1 shows Ksdim(A) ≤ sdimK(m/m 2), or namely, r ≤ dimK((m/m2)0), s ≤ dimK((m/m2)1). We say that A is regular if Ksdim(A) = sdimK(m/m 2). Proposition 5.2. For a local super-ring (A, m), the following are equivalent: (a) A is regular; (b) (i) A is regular, and (ii) for some/any minimal system z1, . . . , zs of generators of the A0-module A1, we have AnnA0(zs) = A2 1. Proof. (a) ⇒ (b). Assume (a). Since (m/m2)0 = m/m we have (i) of (b). Let s = dimK((m/m odd parameters. Since A = A0/A2 arguing as in the proof of Proposition 4.3 we obtain 2 and Kdim(A0) = Kdim(A), 2)1). Then we have a system y1, . . . , ys of 1 is an integral domain (see [9, Theorem 48]), AnnA0(ys) ⊂ A2 1. Given an arbitrary minimal system of generators as in (ii), the presentation of each yi as an A0-linear combination of z1, . . . , zs gives ys = azs for some a ∈ A0, and so AnnA0(zs) ⊂ AnnA0(ys). Since one sees A1zs = 0, and so A2 1 ⊂ AnnA0 (zs), it follows that AnnA0(zs) = A2 1. (b) ⇒ (a). This is now easy. Note that any minimal system of generators as in (cid:3) (ii) of (b) form a system of odd parameters. ON THE NOTION OF KRULL SUPER-DIMENSION 19 5.1. Regular local super-rings. Let A be a (not necessary local) Noetherian super-ring. Recall that IA = AA1. This IA is nilpotent since A1 is now finitely generated as an A0-module. Given a positive integer n, we have A =(An+1 1 ⊕ An 1 ⊕ An+1 An 1 I n 1 n odd, n even, whence I n An A/I n+1 A = An 1 /An+2 1 1 is finitely generated as an A0-module. We define ; this is an A-module, which is finitely generated since grIA(A) :=Mn≥0 A/I n+1 I n A = A ⊕ IA/I 2 A ⊕ . . . . This is a graded superalgebra over A. We let λA : ∧A(IA/I 2 A) → grIA (A) denote the graded A-superalgebra map induced by the embedding IA/I 2 A → grIA (A). One sees that this is a surjection. If the A0-module A1 is generated by s elements, then As+1 A) = 0 = grIA(A)(n) for all n > s. 1 = 0, whence ∧n A (IA/I 2 In the remaining of this subsection we suppose that (A, m) is a local super-ring 1 denote the maximal ideal of with residue field K = A/m. We let m = m0/A2 A = A0/A2 1. Proposition 5.3. For a local super-ring A, the following are equivalent: (a) A is regular; (b) (i) The ring A is regular, (ii) the A-module IA/I 2 (iii) λA is an isomorphism. A is free (of rank Ksdim1(A)), and Proof. Assume (a). Then we have (i) and (ii) of (b) in Proposition 5.2; (i) is the same as (i) of (b) above. If we choose arbitrarily a minimal system y1, . . . , ys of generators of the A0-module A1, then (2) We have I s+1 A = 0. Moreover, for each 0 < t ≤ s, the products AnnA0(ys) = A2 1. yJ mod I t+1 A generate the A-module I t A , where J runs over all subsets of s of cardinality t. For (ii) and (iii) of (b) above, it remains to prove that the products of length t are A-linearly independent. By (2) this holds when t = s. Assume that an A-linear combination A/I t+1 XJ⊆s,J=t aJ yJ , aJ ∈ A0, belongs to I t+1 A . Multiplying by the product yL, L = s \ J, one obtains aJ ys = 0 for each subset J of cardinality t. This, together with the result proven above when t = s, show the desired result. 1 = IA/I 2 Assume (b). By (ii) we have elements z1, . . . , zs in A1 which give rise to an A-free A. The elements form a minimal system of generators of the A = 0, since they give rise to a K-basis of A1/m0A1 = 1. This together (cid:3) basis of A1/A3 A0-module A1, and so I s+1 (m/m2)1. By (iii) we have A ≃ Azs = I s with (i) show (a), in view of Proposition 5.2 . A, whence AnnA0 (zs) = A2 20 A.MASUOKA AND A. N. ZUBKOV Corollary 5.4. Given a regular local super-ring A, the localization Ap at any prime p of A is regular. Proof. Write p for p0. Assume (b) above. Note that the constructions relevant to defining λA commute with the localization. Therefore, (i) Ap = (A)p is regular, (ii) IAp /I 2 A)p is Ap-free, and (iii) λAp = (λA)p is an isomorphism. Thus, Ap satisfies (b). (cid:3) Ap = (IA/I 2 We define grm(A) :=Mn≥0 n/m n+1 = K ⊕ m/m 2 ⊕ . . . . m This is a graded superalgebra over K. Let SK((m/m2)0) denote the symmetric K-algebra on the even component of m/m2; this is a graded algebra. We let κA : SK((m/m 2)0) ⊗K ∧K((m/m 2)1) → grm(A) denote the graded K-superalgebra map induced by the embedding m/m One sees that this is a surjection. 2 → grm(A). The following theorem generalizes [9, Theorem 35] (see also [1, Theorem 11.22]). Theorem 5.5. A local super-ring A is regular if and only if κA is an isomorphism. Proof. The part "if". Assume that κA is an isomorphism. Let x1, . . . , xr and y1, . . . , ys be even and odd elements, respectively, of m which all together give rise to a K-basis of m/m2. We have As+1 1 = 0, as was seen before. We wish to show that the graded (polynomial) subalgebra P of grm(A) which is freely generated by xi mod(m2 1), 1 ≤ i ≤ r, maps isomorphically onto grm(A); this implies that A is regular by [9, Theorem 35]. Fix t > 0. It suffices to show that if an A0-linear combination 0 + A2 (3) X0≤i1≤···≤it≤r ai1...it xi1 . . . xit t+1 0 +A2 is in m and so the assumption implies 1, then all the coefficients ai1...it are in m0. This follows since A2 1ys = 0, X0≤i1≤···≤it≤r ai1...it xi1 . . . xit ys ∈ m t+s+1. by Proposition 5.2 that A is regular. The result just proven implies that in grm(A), no non-zero element in P anni- hilates ys mod ms+1. Therefore, the set AnnA(ys) of those elements in A which 1. It follows i = 0, and so AnnA0 (ys) = A2 The part "only if". First, we wish to see from the results in Section 1.7 that A, and hence A0 as well as A, with all replaced by their completions, may be assumed annihilate ys on A is included inTi≥0 m to be complete. Indeed, we see m/m2 = bm/bm2, grm(A) = gr bm(bA), and that κA is identified with κ bA. In view of Lemma 1.11 it remains to prove that bA is regular, assuming that A is regular. By [9, (24D)] bA0 is regular. The desired result follows 1 = (bA1)2 as the by Proposition 5.2, since one sees, using Proposition 1.9, that if ys : A0 → A, the multiplication by ys, has A2 kernel. 1 as the kernel, then ys : bA0 → bA has cA2 ON THE NOTION OF KRULL SUPER-DIMENSION 21 We may now assume that A, A0 and A are complete, and satisfy (i) -- (iii) of (b) in Proposition 5.3. Choose the even and odd elements x1, . . . , xr and y1, . . . , ys as above. They are the topological generators of A, regarded as a topological ring, i.e. each element a ∈ A is equal to a (not necessary unique) series Xα∈Zs ≥0,I⊆s aα,I xαyI , 1 . . . xαr where xα = xα1 r , provided α = (α1, . . . , αr), and aα,I ∈ A0 \ m0, whenever aα,I 6= 0. Such a series is said to be formally nonzero, if at least one coefficient aα,I 6= 0. To complete the proof one has to show that any formally nonzero series represents a nonzero element of A. Observe that A is topologically generated by the elements xi + A2 1, 1 ≤ i ≤ r, that is each element a of A is equal to a series Xα∈Zs ≥0 aα,Ixα (mod A2 1), and [9, Theorem 35] says that a 6= 0 if and only if a is represented by a formally nonzero series in x-s. Assume that there is a formally nonzero series Xα∈Zs ≥0,I⊆s aα,I xαyI , that represents a zero in A. This series can be represented as a sum PI⊆s fI yI, where among all coefficients fI ∈ A0 there is at least one, say fJ , which is formally nonzero modulo A2 1. We call J a good exponent. Choose a good exponent J with minimal J. Multiplying by ys\J , one obtains an equation which obviously contradicts the fact that I s ated by ys. 1 is a free A-module, freely gener- (cid:3) fJ ys = fJ ys = 0, A = As Given a filed K and non-negative integers r and s, we have the K-superalgebra K[[X1, . . . , Xr]] ⊗K ∧K(Y1, . . . , Ys), as above; this is called the formal power series K-superalgebra in even and odd variables, X1, . . . , Xr and Y1, . . . , Ys. In fact, this is a complete regular local super- ring by Theorem 5.5, since the kappa map is an isomorphism; it is indeed the identity map on K[X1, . . . , Xr] ⊗K ∧K (Y1, . . . , Ys). Remark 5.6. Theorem 5.5 coincides with Theorem 3.3 from [12], but our proof is quite different and seems more elementary. Corollary 5.7. (see [12], or [11], A.3) Let (A, m) be a local super-ring. Assume that A0 includes a field. Then the following are equivalent: (a) A is regular; (b) bA is regular; (c) bA is isomorphic to K[[X1, . . . , Xr]] ⊗K ∧K (Y1, . . . , Ys), where K = A/m and r s = sdimK(m/m 2). 22 A.MASUOKA AND A. N. ZUBKOV Proof. (c) ⇒ (b). This was just seen above. (b) ⇒ (a). This follows by Theorem 5.5, since κA is identified with κ bA. (a) ⇒ (c). As above, one can assume that A is complete. Since A0 includes a field, Cohen's theorem implies that A has a coefficient field (or field of representa- tives), isomorphic to K. Moreover, A ≃ K[[X1, . . . , Xr]] is formally smooth (see [9, Example 3., p.200; Corollary 2, p.206]). Thus the epimorphism A0 → A splits and (c) follows by Proposition 5.3(b). (cid:3) 5.2. Regular super-rings. Let A be a super-ring which we continue to assume to be Noetherian. The super-ring A is said to be regular if for every prime p of A, the local super-ring Ap is regular. By Corollary 5.4 this is equivalent to saying that for every maximal m of A, the local super-ring Am is regular. Proposition 5.8. For a super-ring A, the following are equivalent: (a) A is regular; (b) (i) The ring A is regular, (ii) the A-module IA/I 2 (iii) λA is an isomorphism. A is projective, and Proof. This follows from Proposition 5.3, since the conditions above admit "local criterion" in the sense that each of them is satisfied if and only if the condition naturally obtained by localization at every prime/maximal is satisfied. (cid:3) Remark 5.9. Proposition 5.8 shows that our definition of regularity is equivalent to the definition of regularity introduced in [12, 3.3, p.79]. For example, if A is an regular super-ring, no matter local or not, then any minimal system of generators of the A0-module IA/I 2 A is an odd regular sequence of maximal length. Remark 5.10. The converse of Lemma 4.2(b) is not true, even if R is regular. In fact, let R = K[Y ] be a polynomial superalgebra in s odd free generators as in subsection 4.4. Set f = Y L1 + Y L2, where L1 6⊆ L2, L2 6⊆ L1, L1 ∩ L2 6= ∅, and L1 ≡ L2 (mod 2). Proposition 4.11 implies Ksdim1(R/Rf ) = s − 1, but Y L1∩L2 ∈ AnnR(f ) \ Rf . 5.3. Krull super-dimension of completions of super-rings. We still assume that all super-rings are Noetherian. Lemma 5.11. Let B be a Noetherian super-ring with Kdim(B) < ∞. If p is a prime super-ideal of B such that Ksdim0(B) = Ksdim0(Bp), then Ksdim1(Bp) ≤ Ksdim1(B). Proof. The condition Ksdim0(B) = Ksdim0(Bp) is equivalent to ht(p0) = Kdim(B0). In other words, any longest prime chain q0 ⊆ . . . ⊆ qr = p0 is a longest prime chain in B0 as well. Thus the elements z1 , where z1, . . . , zt ∈ B1, s1, . . . , st ∈ S = s1 B0 \ p0, form a system of odd parameters in Bp if and only if , . . . , zt st AnnS−1B0 ( z1 s1 . . . zt st ) = S−1AnnB0 (z1 . . . st) ⊆ S−1 q0 if and only if AnnB0 (z1 . . . st) ⊆ q0 for an ideal q0 as above. Now the statement is obvious. (cid:3) Theorem 5.12. Let A be a local Noetherian super-ring with maximal super-ideal m. Let bA denote its m-adic completion. Then Ksdim(A) = Ksdim(bA). ON THE NOTION OF KRULL SUPER-DIMENSION 23 Proof. Since bA0 coincides with the m0-adic completion of A0, the equality Ksdim0(A) = Ksdim0(bA) follows by (24.D)(i), [9]. Fix a generating set y1, . . . , yd of A0-module Let b be a homogeneous element from A. Arguing as in Theorem 5.5, we obtain A1. an exact sequence 0 → \AnnA0(b) → bA0 → bA, (b) = pletion of A0/AnnA0 (b), that in turn infers (b) is isomorphic to the m0/AnnA0(b)-adic com- where bA0 → bA is defined as a 7→ ab, a ∈ bA0. In other words, we have Ann bA0 \AnnA0 (b). Furthermore, bA0/Ann bA0 Kdim(bA0/Ann bA0 By Theorem 55, [9], bA1 is generated by the same elements y1, . . . , yd as an bA0- module, and Proposition 4.1 completes the proof. (b)) = Kdim(A0/AnnA0(b)). The set of super-dimension vectors can be lexicographically ordered as ab < cd, provided a < b or a = b and then c < d. The following theorem superizes (24.D)(i′), [9]. (cid:3) Theorem 5.13. Let A be a Noetherian super-ring with Ksdim0(A) < ∞. Let I be a super-ideal of A and let bA be its I-adic completion. Then Ksdim(Ap). sup p∈SSpec(A),I⊆p Proof. Using Lemma 1.5 and (24.D)(i′), [9], we have Ksdim(bA) = sup q∈Spec(A0),I0⊆q Kdim((A0)q) = r = Ksdim0(bA) = As above, bA0-module bA1 is generated by some y1, . . . , yd ∈ A1. By Proposition 4.1 there is a system of odd parameters yi1, . . . , yis of bA of largest cardinality s = Ksdim1(bA). Let J denote the ideal AnnA0(yi1 . . . yis). Arguing as in Theorem 5.12, we have Ksdim0(Ap). p∈SSpec(A),I⊆p sup (yi1 . . . yis) ≃ \A0/J, where \A0/J denotes the completion of A0/J in the (I + J)/J-adic topology. Thus bA0/Ann bA0 Using (24.D)(i′), [9], one sees Kdim(bA0) = Kdim(bA0/Ann bA0 (yi1 . . . yis)) = Kdim(\A0/J). r = sup Kdim((A0)q) = sup q∈Spec(A0),I0⊆q q∈Spec(A0),I0+J⊆q Kdim((A0/J)q/J ). There are two prime ideals q and q ′ such that I0 ⊆ q, I0 + J ⊆ q ′, and r = Kdim((A0)q) = Kdim((A0/J)q′/J ). On the other hand, we have Kdim((A0)q) ≥ Kdim((A0)q′ ) ≥ Kdim((A0)q′ /Jq′) = Kdim((A0/J)q′/J ), hence r = Kdim((A0)q) = Kdim((A0)q′ ) = Kdim((A0)q′ /Jq′) 24 A.MASUOKA AND A. N. ZUBKOV and the elements yi1 , . . . , yis form a system of odd parameters in (A0)q′ . The latter infers the inequality and therefore, the inequality Ksdim1(bA) ≤ Ksdim1((A)q′+A1 ), Ksdim(bA) ≤ p∈SSpec(A),I⊆p sup Ksdim(Ap). Let p be a prime super-ideal such that I ⊆ p and Ksdim0(bA) = Ksdim0(Ap). Arguing as in (24D), [9], one can show that Ap and bAbp are analytically isomorphic, that is their completions (as local super-rings) are isomorphic. Combining Theorem 5.12 and Lemma 5.11, one obtains hence Theorem is proven. Ksdim(Ap) = Ksdim(bAbp) ≤ Ksdim(bA), Ksdim(Ap) ≤ Ksdim(bA). p∈SSpec(A),I⊆p sup (cid:3) 5.4. Kahler superdifferentials and regularity. Lemma 5.14. Let B be a local K-superalgebra with the maximal super-ideal m. Assume that its residue field B/m = K(B) is a separably generated extension of K. Then ΩB/K ⊗B K(B) ≃ m/m 2 ⊕ ΩK(B)/K. In particular, if B-supermodule ΩB/K is finitely generated, then a minimal system of generators of ΩB/K consists of dimK(B)((m/m 2)0) + tr.degK(K(B)) even generators and odd generators respectively. dimK(B)((m/m 2)1) Proof. Applying [9, p.205] to a complete local ring (B/m2)0, one can choose its field of representatives, say L ≃ K(B), so that the exact sequence 0 → m/m 2 → B/m 2 → K(B) → 0 is split (on the right). Proposition 3.8 and [9, Theorem 59] conclude the proof. (cid:3) Lemma 5.15. Let A be a Noetherian local superdomain with residue field K and superfield of fractions F . Let M be a finite A-supermodule such that the F - supermodule MIA = M ⊗A F is free of rank pq. If sdimKM ⊗A K = pq, then M is a free A-supermodule of rank pq. Proof. Since F is a flat A-supermodule (cf. superize the proof of Lemma II.8.9, [6]. [10], Lemma 1.2(i)), one can easily (cid:3) Theorem 5.16. Assume that B is a K-superalgebra as above. Assume also that K is perfect and B is a localization of a finitely generated K-superalgebra. Then B is regular if and only if ΩB/K is a free B-supermodule of rank equal to Ksdim(B) + tr.degK(K(B))0. ON THE NOTION OF KRULL SUPER-DIMENSION 25 Proof. First of all, Remark 3.2 and Lemma 3.6 imply that ΩB/K is a finitely gen- erated B-supermodule. Set sdimK(B)(m/m2) = mn and Ksdim(B) + 0tr.degK(K(B)) = pq. Note that K(B) is a finitely generated extension of K, hence it is also separably generated If ΩB/K is free of rank pq, then Lemma 5.14 implies over K (see [9], p.194). Ksdim(B) = mn, whence B is regular. Conversely, assume that B is regular, i.e. the Krull superdimension of B is equal to mn. Again, by Lemma 5.14 the K(B)-superspace ΩB/K ⊗B K(B) has superdimension pq. Proposition 5.3 implies that B is a strong superdomain. The superfield F = SQ(B) is a local Noetherian super-ring with the maximal super-ideal IF . Moreover, by Corollary 5.4 the super-ring F is regular. The residue field of F , K(F ), is isomorphic to the quotient field of B. Further, the maximal super-ideal IF is nilpotent, hence F is complete. Since Ksdim1(F ) = Ksdim1(B) = n (see the proof of Proposition 6.1 below), Corollary 5.7 shows that F ≃ K(F )[Y1, . . . , Yn]. Observe that any K-superderivation of K(F ) into a F -supermodule T can be extended to a K-superderivation F → T . Therefore, Proposition 3.7 infers ΩF/K ≃ (ΩK(F )/K ⊗K(F ) F ) ⊕ ΩF/K(F ). By Remark 3.3, ΩF/K(F ) is a free F -supermodule of rank 0n. On the other hand, Lemma 3.6 implies ΩK(F )/K ≃ ΩB/K ⊗B K(F ). Since ΩB/K is a free B-module of rank m + tr.degK(K(B)) (see [6, Exercise II.8.1(b)]), ΩF/K is a free F -supermodule of rank pq. Again, by Lemma 3.6 there is ΩB/K ⊗B F ≃ ΩF/K , and Lemma 5.15 concludes the proof. (cid:3) 6. Dimension theory of superschemes 6.1. Super-dimension of a superscheme. From now on all superschemes are assumed to be of finite type over a field K, unless otherwise stated. Proposition 6.1. Let X be an irreducible superscheme. If U and V are non-empty open affine super-subschemes of X, then Ksdim(O(U )) = Ksdim(O(V )). Proof. Since U e ∩ V e is not empty, all we need is to consider the case U ⊆ V . Set U ≃ SSpec A and V ≃ SSpec B. Then the natural open immersion U → V coincides with SSpec φ for some super-ring morphism φ : B → A . By Lemma 3.5, [10], there are b1, . . . , bt ∈ B0 such thatP1≤i≤t A0φ(bi) = A0 and the induced morphisms Bbi → Aφ(bi) are isomorphisms. Thus the general case can be reduced to V = SSpec B and U = SSpec Bb, b ∈ B0 \ nil(B0). Observe that nil((B0)b) = nil(B0)b. Since B0/nil(B0) is a domain, Ksdim0(B) = Kdim(B0/nil(B0)) coincides with the transcendence degree of its field of fractions (cf. [1], XI), whence Ksdim0(B) = Ksdim0(Bb). Lemma 4.6 concludes the proof. (cid:3) Proposition 6.1 allows to define a super-dimension of any irreducible superscheme X as sdim(X) = Ksdim(O(U )), where U is any (non-empty) open affine super- subscheme of X. 26 A.MASUOKA AND A. N. ZUBKOV Remark 6.2. Let X be an irreducible superscheme. The same arguments as in Proposition 6.1 show that Ksdim1(Ox) = sdim1(X) for any point x ∈ X e. In particular, if x is a closed point of X, then Ksdim(Ox) = sdim(X). Besides, we have sdim(U ) = sdim(X) for (nonempty) open super-subscheme U of X. 6.2. Nonsingular superschemes. Let X be an irreducible superscheme. Then X is said to be nonsingular, if for any x ∈ X e the super-ring Ox is regular. Theorem 6.3. Let X be as above and assume that K is perfect. Then X is nonsingular if and only if the sheaf ΩX/K is locally free of rank sdim(X). Proof. Without loss of generality one can assume that X is affine, say X = SSpec R. Then Ksdim(R) = sdim(X) and all one need to prove is that Rp is regular if and only if (ΩR/K)p ≃ ΩRp/K is a free Rp-supermodule of rank Ksdim(R), for each point p ∈ (SSpec R)e. On the other hand, by Theorem 5.16 the local super-ring Rp is regular if and only if ΩRp/K is a free Rp-supermodule of rank Ksdim(Rp) + tr.degK(K(Rp))0. Since Ksdim1(R) = Ksdim1(Rp), the equality Ksdim(R) = Ksdim(Rp) + tr.degK(K(Rp))0 holds if and only if Ksdim0(R) = Ksdim0(Rp) + tr.degK(K(Rp)) does. Note that Ksdim0(R) = Kdim(R), Ksdim0(Rp) = Kdim(Rp). Moreover, since K(Rp) = K(Rp), we have tr.degK(K(Rp)) = tr.degK(K(Rp)). Thus any of the above equalities holds if and only if Rp is regular (see [6], Exercise II.8.1.c). Theorem obviously follows. (cid:3) 6.3. Generically nonsingular superschemes. Oppositely to the purely even case, there are (even strong) integral superschemes, which are singular at any point (compare with [6], Exercise II.8.1(d)). Example 6.4. In fact, let A be a finitely generated K-superalgebra such that A0 is a domain and A2 1 = 0. Assume also that A1 is a free A0-module of rank t > 1. Set X = SSpec A. Since A is a strong superdomain, X is strong integral. Further, (Ap)1 is a free (Ap)0-module of the same rank t, for each p ∈ X e. Proposition 5.2(b) immediately shows that Ap is not regular. A superscheme X is called generically nonsingular, provided X contains an nonempty open nonsingular super-subscheme. Lemma 6.5. Let X be an irreducible superscheme over a perfect field K. Then X is generically nonsingular if and only if there is x ∈ X e such that Ox is regular. Proof. As it has been proven in Theorem 6.3, a local super-ring Ox is regular if and only if (ΩX/K)x is a free Ox-supermodule of rank sdim(X). Thus our statement follows by Lemma 2.7 and Remark 6.2. (cid:3) ON THE NOTION OF KRULL SUPER-DIMENSION 27 Proposition 6.6. Let X be an integral superscheme over a perfect field K. Let ξ be the generic point of X. The following conditions are equivalent : (a) X is generically nonsingular; (b) Oξ is regular; (c) there is a nonempty open super-subscheme U of X such that Uev is regularly immersed into U (cf. [12, 4.5]). Proof. Use Lemma 6.5 to prove (a)⇔ (b). Without loss of generality one can assume that X is affine, say X ≃ SSpec A. Recall that Oξ is a superfield with the maximal superieal mξ = IOξ . Thus Oξ is regular if and only if λOξ : ∧Oξ ξ ) → grIξ (Oξ) is an isomorphism if and only IOξ is an regular superideal (see [12, 3.2]). (Iξ/I 2 Since both ∧A(IA/I 2 A) and grIA(A) are finitely generated A-(super)modules, ar- guing as in Lemma 2.7 one can show that λOξ is an isomorphism if and only if there is an open subset U ⊆ X e, such that for any p ∈ U the following hold : (1) λOp is an isomorphism; (2) Ip/I 2 p is a free Op-supermodule, whence (b)⇔(c). (cid:3) Recall that if X is an integral scheme and Y is a closed integral subscheme of X, such that dim(X) = dim(Y ), then X = Y . Surprisingly, a naive analog of this statement is no longer true in the category of superschemes. Example 6.7. Let X = SSpec A be the everywhere singular superscheme from Example 6.4. Let b ∈ A1 is a free generator of A0-module A1. Set Y = SSpec A/Ab. Then Y is isomorphic to a proper (strong integral as well) closed super-subscheme of X, but sdim(Y ) = sdim(X). Nevertheless, a weaker super-analog of the above statement takes place. Theorem 6.8. Let X be a generically nonsingular integral superscheme over a per- fect field K. If Y is a closed super-subscheme of X, which is generically nonsingular and integral as well, then sdim(X) = sdim(Y ) implies X = Y . Proof. The purely even version of our theorem infers that Xres = Yres, that is JY ⊆ IX . Thus ξ ∈ Y e, where ξ is a generic point of X. Since both X and Y are irreducible, Proposition 6.6 allows to assume that both X and Y are nonsingular. Moreover, one can also assume that both X and Y are affine, say X = SSpec A and Y ≃ SSpec A/I, where I ⊆ IA and both A and A/I are regular. Let B denote the superalgebra A/I. Note that any prime super-ideal of B has a form p = p/I, where p is a prime super-ideal of A. Thus the residue fields of both Ap and Bp are isomorphic, say, to a field L. For any prime super-ideal p of A we have sdim1(X) = Ksdim1(Ap) = sdim1(Y ) = Ksdim1(Bp), hence both A ≃ B-modules IA/I 2 B are projective modules of the same rank, hence isomorphic. By Proposition 5.8(b), the epimorphism A → B induces an isomorphism grIA (A) ≃ grIB (B), hence A → B is an isomorphism and X = Y . (cid:3) A and IB/I 2 28 A.MASUOKA AND A. N. ZUBKOV 6.4. Closed super-subschemes of generically nonsingular superschemes. The following theorem superizes Theorem II.8.17, [6]. Theorem 6.9. Let X be a nonsingular irreducible superscheme of finite type over a perfect field K. Let Y be an irreducible closed super-subscheme of X defined by a sheaf of superideals J . Then Y is nonsingular if and only if the following conditions hold : (1) ΩY /K is locally free; (2) The sequence 0 → J /J 2 → ΩX/K ⊗OX OY → ΩY /K → 0 is exact. Moreover, the sheaf J /J 2 is locally free of rank sdimX − sdimY . Proof. Assume that both (1) and (2) hold. Let sdim(X) = mn. For any point y ∈ Y e let A and B denote the local superalgebras OX,y and OY,y respectively. The stalk Jy is naturally identified with the kernel of the local morphism A → B, say J, so that there is an exact sequence 0 → J/J 2 → ΩA/K ⊗A B → ΩB/K → 0, where A-supermodule ΩA/K is free of rank mn and B-supermodule is free of rank pq. The latter implies that the above sequence splits, hence J/J 2 is a free B- supermodule of rank (m − p)(n − q). By [12, Theorem 3.5] the super-ideal J is generated by a regular sequence consisting of m − p even and n − q odd elements. Combining [9, (15.F), Lemma 4] with Lemma 4.2, one obtains Ksdim(B) = Ksdim(A) − (m − p)(n − q). Since X is nonsingular and K(B) is an extension of K(A), there hold (mn) = Ksdim(A) + tr.degK(K(A)) and tr.degK(K(A)) ≤ tr.degK(K(B)). Combining with Lemma 5.14 and Lemma 5.1, one derives pq ≥ Ksdim(B) + tr.degK(K(B)) ≥ Ksdim(A) − (m − p)(n − q) + tr.degK(K(A)) = pq, and by Theorem 5.16 one derives that B is regular. Conversely, assume that Y is nonsingular of superdimension pq. Arguing as in Theorem II.8.17, [6], one can construct a closed (nonsingular) super-subscheme Y ′ of X, such that Y ⊆ Y ′ and Y ′ has the same superdimension pq. Theorem 6.8 concludes the proof. (cid:3) Let X be a generically nonsingular irreducible superscheme of finite type over a perfect field K. Let U denote the open subset {x ∈ X e Ox is regular}. It is clear that U is the largest open nonsingular super-subscheme of X. Corollary 6.10. Let X and U be as above. Let Y be a closed irreducible super- subscheme of X with Y e ∩ U e 6= ∅. Then Y is generically nonsingular if and only there is a point y ∈ Y e such that (1) ΩOY,y/K is free; ON THE NOTION OF KRULL SUPER-DIMENSION 29 (2) The sequence 0 → Jy/J 2 y → ΩOX,y/K ⊗OX,y OY,y → ΩOY,y/K → 0 is exact. Proof. Use Lemma 2.7 and the above theorem. (cid:3) 7. Some questions and open problems 7.1. One relation superalgebras. Let A = K[XY ] and f ∈ IA. Let B denote the one relation superalgebra A/Af , as in subsection 4.4. The discussion therein rises the following natural question. Question 7.1. Is the odd Krull dimension of one relation superalgebra A/Af is determined by the basement of f ? 7.2. Grobner-Shirshov basis method. One of the most powerful tools in the theory of polynomial ideals is the Grobner-Shirshov basis method (cf. [2, 3]). The following question is also motivated by the discussion in subsection 4.4. Question 7.2. Does any super-version of Grobner-Shirshov basis method exist? 7.3. Dimension theory of supermodules. Let R be a Noetherian super-ring and M be a finite R-supermodule. One can define a super-dimension of M as sdim(M ) = Ksdim(R/Ann(M )). Problem 7.3. Develop a dimension theory of supermodules over Noetherian super- rings. References [1] V.F.Atiyah, I.G.Macdonald, Introduction to Commutative Algebra, Addison-Wesley, Read- ing, Mass., 1969. [2] W. Adams and P. Loustaunau, An Introduction to Grobner Bases, Graduate Studies in Mathematics, AMS, Providence, 1994. [3] B. Buchberger, Grobner bases: An algorithmic method in polynomial ideal theory, In N. K. Bose, editor, Multidimensional Systems Theory, chapter 6, pages 184 -- 232. Reidel, Dordrecht, 1985. [4] N. Bourbaki, Commutative Algebra, Springer, 1989. [5] C. Carmeli, L. Caston, R. Fioresi, Mathematical foundations of supersymmetry, EMS Ser. Lect. Math., European Math. Soc., Zurich, 2011. [6] R.Hartshorne, Algebraic Geometry, Springer, 1977. [7] J.C.Jantzen, Representations of algebraic groups. Second edition. Mathematical Surveys and Monographs, 107. American Mathematical Society, Providence, RI, 2003. [8] Y.I.Manin, Gauge Field Theory and Complex Geometry, Springer, Berlin, 1988. Translated by N. Koblitz and J. R. King. [9] H. Matsumura, Commutative algebra, 2nd ed., Math. Lec. Notes Series 56, Ben- jamin/Cunning Publishing Company, 1980. [10] A. Masuoka and A. N. Zubkov, Quotient sheaves of algebraic supergroups are superschemes. J. Algebra, 348 (2011), 135 -- 170. [11] A.Masuoka, A. N. Zubkov, Solvability and nilpotency for algebraic supergroups, Journal of Pure and Applied Algebra, 221(2017), 339 -- 365. [12] T. Schmitt, Regular sequences in Z2-graded commutative algebra, J.Algebra, 124 (1989), 60 -- 118. [13] V.S. Varadarajan, Supersymmetry for mathematicians: an introduction, Courant Lecture Notes 11, American Mathematical Society 2004. [14] O.Zariski , P.Samuel, Commutative algebra. Vol. II. Reprint of the 1960 edition. Graduate Texts in Mathematics, Vol. 29. Springer-Verlag, New York-Heidelberg, 1975. x+414 pp. 30 A.MASUOKA AND A. N. ZUBKOV [15] A.N.Zubkov, Some properties of Noetherian superschemes, Algebra and Logic, Vol. 57 (2018), No. 2, 130 -- 140. (Russian Original Vol. 57, No. 2, March-April, 2018). [16] A.N.Zubkov, Affine quotients of supergroups, Transformation Groups, 14(2009), No.3, 713- 745. Institute of Mathematics, University of Tsukuba, Ibaraki 305-8571, Japan E-mail address: [email protected] Department of Mathematical Science, UAEU, Al-Ain, United Arabic Emirates; Sobolev Institute of Mathematics, Omsk Branch, Pevtzova 13, 644043 Omsk, Russian Federation E-mail address: [email protected]
1810.01282	1	1810	2018-10-01T05:45:12	Weak Nil Clean Ideal	[ "math.RA" ]	As a generalization of nil clean ideal, we define weak nil clean ideal of a ring. An ideal $I$ of a ring $R$ is weak nil clean ideal if for any $x\in I$, either $x=e+n$ or $x=-e+n$, where $n$ is a nilpotent element and $e$ is an idempotent element of $R$. Some interesting properties of weak nil clean ideal and its relation with weak nil clean ring have been discussed.	math.RA	math	WEAK NIL CLEAN IDEAL DHIREN KUMAR BASNET* AND AJAY SHARMA Abstract. As a generalization of nil clean ideal, we define weak nil clean ideal of a ring. An ideal I of a ring R is weak nil clean ideal if for any x ∈ I, either x = e + n or x = −e + n, where n is a nilpotent element and e is an idempotent element of R. Some interesting properties of weak nil clean ideal and its relation with weak nil clean ring have been discussed. 1. Introduction In this article, rings are associative with unity. The Jacobson radical, set of units, set of idempotents, set of nilpotent elements and centre of a ring R are denoted by J(R), U(R), Idem(R), Nil(R) and C(R) respectively. Nicholson [6] defined an element x of a ring R to be a clean element, if x = e + u for some e ∈ Idem(R), u ∈ U(R) and called the ring R as clean ring if all its elements are clean. Diesl [5] defined a ring R to be nil clean ring if every element of R can be written as a sum of an idempotent and a nilpotent element of R. Weakening the condition of clean ring, Ahn and Anderson [1] defined a ring R to be weakly clean, if every x ∈ R can be expressed as x = u + e or x = u − e, where u ∈ U(R), e ∈ Idem(R). Also Basnet and Bhattacharyya [2] defined a ring R to be weak nil clean, if every element x ∈ R can be written as x = n + e or x = n − e, where n ∈ Nil(R) and e ∈ Idem(R). H. Chen and M. Chen [3] defined an ideal I of a ring R to be clean ideal, if for any x ∈ I, x = u + e, for some u ∈ U(R) and e ∈ Idem(R). They proved that every ideal having stable range one of a regular ring is clean. Following the idea of clean ideal, Sharma and Basnet [7] defined an ideal I of a ring R to be nil clean ideal, if for any x ∈ I, x = n + e, where n ∈ Nil(R) and e ∈ Idem(R). They proved that for a nil clean expression of an element of a nil clean ideal of a ring R, the nilpotent and idempotent elements are actually elements of the ideal. Also they characterized a nil clean ring with its nil clean ideals. As a generalization of clean ideal, Sharma and Basnet [8] also introduced weakly clean ideal of a ring. An ideal I of a ring R is said to be weakly clean ideal, if for any x ∈ I, x = u + e or x = u − e, where u ∈ U(R) and e ∈ Idem(R). In this article we introduce the notion of weak nil clean ideal as a generalization of nil clean ideal. An ideal I of a ring R is called weak nil clean ideal if for each a ∈ I, either a = e + n or a = −e + n, where e ∈ Idem(R) and n ∈ Nil(R). Here also we 2010 Mathematics Subject Classification. 16N40, 16U99. Key words and phrases. Clean ideals, weakly clean ideals, uniquely clean ideal, weakly uniquely clean ideal. * Corresponding Author . 1 2 D. K. BASNET AND A. SHARMA proved that for a weak nil clean expression of an element of a weak nil clean ideal of a ring R, the nilpotent and idempotent elements are actually elements of the ideal. Further we characterized a weak nil clean ring with its weak nil clean ideal and nil clean ideal of R. Also we discuss some interesting properties of weak nil clean ideals. 2. Weak nil clean ideal Definition 2.1. An ideal I of a ring R is called weak nil clean ideal of R if for any x ∈ I, there exist e ∈ Idem(R) and n ∈ Nil(R) such that x = e + n or x = −e + n. Also I is called uniquely weak nil clean ideal of R if for any x ∈ I there exists a unique e ∈ Idem(R) such that x − e ∈ Nil(R) or x + e ∈ Nil(R). Clear from the definition that every ideal of a weak nil clean ring is weak nil clean ideal. But there are non weak nil clean rings which contains some weak nil clean ideals, for example the ring Zpn, where n > 1 and p > 5, a prime number, is not weak nil clean ring but every proper ideal of Zpn is weak nil clean ideal. Another such example is given below. Example 2.2. Let R1 be weak nil clean ring and R2 be non weak nil clean ring. Then R = R1 ⊕ R2 is not a weak nil clean ring. But clearly I = R1 ⊕ 0 is weak nil clean ideal of R. Observe that every nil clean ideal is weak nil clean ideal but the converse is not true as {0, 2, 4} is weak nil clean ideal of Z6 but not nil clean ideal of Z6. Lemma 2.3. Every weak nil clean ideal of a ring R is weakly clean ideal of R. Proof. Let I be a weak nil clean ideal of R. For x ∈ I, either x = e + n or x = −e + n, where e ∈ Idem(R) and n ∈ Nil(R). If x = e + n then x = (1 − e) + (2e − 1 + n) and if x = −e + n then x = (1 − e) + (−1 + n), where 1 − e ∈ Idem(R) and −1 + n, 2e − 1 + n ∈ U(R). (cid:3) The converse of Lemma 2.3 is not true as ideal {0, 3, 6, 9, 12} of Z15 is weakly clean ideal but not weak nil clean ideal of Z15. Proposition 2.4. If I is a weak nil clean ideal of a ring R then I ∩ J(R) is a nil ideal of R. Proof. Let x ∈ I ∩ J(R), so either x = e + n or x = −e + n, where e ∈ Idem(R) and n ∈ Nil(R). If x = e + n, then by Proposition 2.4 [7], x = n. If x = −e + n then t,r∈R txr + ek = 0 ⇒ t,r∈R txr ∈ J(R) as x ∈ J(R). So 1 − e ∈ Idem(R) ∩ U(R) = {1}, hence (cid:3) there exists k ∈ N such that nk = 0. Now nk = (x + e)k = Pf inite e = −Pf inite 1 − e = 1 ⇒ e = 0 ⇒ x = n. Thus the result follows. Corollary 2.5. If R is a weak nil clean ring then J(R) ⊆ N(R). In particular for is a commutative ring R, J(R) = N(R). Let R be a ring. An element a ∈ R is called weakly clean element of type-I if a = e+u and called weakly clean element of type-II if a = −e+u, where e ∈ Idem(R) and u ∈ U(R). WEAK NIL CLEAN IDEAL 3 Definition 2.6. An ideal I of a ring R is said to be strongly weak nil clean ideal if for any x ∈ I, there exist e ∈ Idem(R) and n ∈ Nil(R) such that x = e + n or x = −e + n and en = ne. Also I is called strongly weakly clean ideal if for any x ∈ I, there exist e ∈ Idem(R) and u ∈ U(R) such that x = e + u or x = −e + u and eu = ue. Theorem 2.7. Let I be an ideal of R then (1) If I is strongly weak nil clean ideal then it is strongly weakly clean ideal and a − a2 or a + a2 is nilpotent. (2) If I is strongly weakly clean ideal and a − a2 or a + a2 is nilpotent provided a is of type-I or type-II weakly clean element of R respectively, then I is strongly weak nil clean ideal. Proof. (1) Let I be a strongly weak nil clean ideal and a ∈ R. Then either a = e+n or a = −e+n, where e ∈ Idem(R), n ∈ Nil(R) and en = ne. If a = e+n then a = (1 − e) + (2e − 1 + n) is strongly weakly clean decomposition of a and also a−a2 = (1−2e−n)n is nilpotent. If a = −e+n, then a = −(1−e)+(1−2e+n) is strongly weakly clean decomposition of a and also a + a2 = (1 − 2e + n)n is nilpotent. (2) Let a ∈ I, so either a = e+u or a = −e+u, where e ∈ Idem(R) and u ∈ U(R). If a = e + u, then a − a2 is nilpotent, which implies 1 − 2e − u is nilpotent and a = (1 − e) + (−1 + 2e + u), a strongly weak nil clean expression of a. If a = −e + u, then a + a2 is nilpotent, which implies 1 − 2e + u is nilpotent and a = −(1 − e) + (1 − 2e + u), a strongly weak nil clean expression of a. (cid:3) Lemma 2.8. Every idempotent in a uniquely weak nil clean ideal is a central idem- potent. Proof. Let I be a uniquely weak nil clean ideal of a ring R and e be any idempotent in I. For any x ∈ R, since e = (e+ ex(1 −e)) + (−ex(1 −e)) = e+ 0, so e+ ex(1 −e) = e ⇒ ex = exe, as e + ex(1 − e) ∈ Idem(R). Similarly we can show that xe = exe. Hence xe = ex. (cid:3) The following theorem shows that, for a weak nil clean expression of an element of a weak nil clean ideal of a ring R, the nilpotent and idempotent elements are actually elements of the ideal. Theorem 2.9. An ideal I of a ring R is weak nil clean ideal if and only if for any x ∈ I, either x = e + n or x = −e + n, where e ∈ Idem(I) and n ∈ Nil(I). Proof. Let I be a weak nil clean ideal of R and x ∈ I. There exist n ∈ Nil(R) and e ∈ Idem(R) such that either x = e + n or x = −e + n. So nk = 0, for some i=1 qixpi, for some pi, qi ∈ R, i=1 qixpi ∈ I, so ek = e ∈ I. Similarly if x = −e + n, then also we get (cid:3) k ∈ N. If x = e + n, then (x − n)k = (−1)knk + Ps (x − n)k = Ps e ∈ I. Hence n ∈ I, as required. The following corollary is immediate. 4 D. K. BASNET AND A. SHARMA Corollary 2.10. If R is a local ring, then every proper weak nil clean ideal of R is nil ideal. In fact if R has no non trivial idempotents, then every proper weak nil clean ideal of R is nil ideal. In the following Theorem, we characterize weak nil clean ring R by weak nil clean ideal and nil clean ideal of R. Theorem 2.11. R is a weak nil clean ring if and only if there exists a central idem- potent e in R such that ideals generated by e and 1 − e are both weak nil clean ideals of R and one of them is nil clean ideal of R. Proof. If R is weak nil clean ring, then e = 1 works. Conversely, without loss of generality assume that < e > is weak nil clean ideal and < 1 − e > is nil clean ideal of R. For x ∈ R, since R =< e > + < 1 − e >, so x = a + b, where a ∈< e > and b ∈< 1 − e >. There exist f1 ∈ Idem(< e >) and n1 ∈ Nil(< e >), such that either a = f1 + n1 or a = −f1 + n1. If a = f1 + n1, then we set b = f2 + n2, where f2 ∈ Idem(< 1 − e >) and n2 ∈ Nil(< 1 − e >), then x = (f1 + f2) + (n1 + n2) is a nil clean expression of x in R. Also if a = −f1 + n1 then we set b = −f2 + n2, where f2 ∈ Idem(< 1 − e >) and n2 ∈ Nil(< 1 − e >), so x = −(f1 + f2) + (n1 + n2) is a weak nil clean expression of x in R. Hence R is a weak nil clean ring. (cid:3) A finite orthogonal set of idempotents e1, · · ·, en in a ring R, is said to be complete set if e1 + · · · + en = 1. Now we generalize the above result in terms of complete set of idempotents. Theorem 2.12. Ring R is weak nil clean if and only if there exists a complete set of central idempotents e1, · · ·, en in R, such that ideal generated by ei is weak nil clean ideal of R for all i and at most one < ei > is not nil clean ideal. Proof. (⇒) Taking e = 1. (⇐) Clearly < e1 > + < e2 > + · · · + < en >= R, so similar to the proof of Theorem 2.11, we can show that R is weak nil clean ring. (cid:3) Proposition 2.13. Let I be an ideal of a ring R. Then the following are equivalent: (1) I is weak nil clean ideal of R. (2) There exists a complete set of central idempotents e1, · · ·, en such that eiI is a weak nil clean ideal of eiR, for all i and at most one eiI is not nil clean ideal of eiR. Proof. (1)⇒ (2) Taking e = 1. (2)⇒(1) Let e1, · · ·, en be a complete set of idempotents in R such that eiI is a weak nil clean ideal of eiR, for all i and at most one eiI is not nil clean ideal of eiR. It is enough to show the result for n = 2. Clearly I = e1I ⊕e2I. Without loss of generality assume e1I is a nil clean ideal of e1R. Let x ∈ I, then x = a + b, where a ∈ e1I and b ∈ e2I, so there exist f2 ∈ Idem(e2I) and m2 ∈ Nil(e2I) such that either b = f2 +m2 or b = −f2 + m2. If b = f2 + m2, then we set a = f1 + m1, where f1 ∈ Idem(e1I) and m1 ∈ Nil(e1I) and we get x = (f1 +f2)+(m1 +m2) is a weak nil clean expression of x. If b = −f2 + m2, then we set a = −f1 + m1, where f1 ∈ Idem(e1I) and m1 ∈ Nil(e1I) and we get x = −(f1 + f2) + (m1 + m2) is a weak nil clean expression of x. (cid:3) WEAK NIL CLEAN IDEAL 5 Theorem 2.14. Let R be a ring and I1 be an ideal containing the nil ideal I. Then I1 is weak nil clean ideal of R if and only if I1/I is weak nil clean ideal of R/I. Proof. If I1 is weak nil clean ideal of R, then clearly I1/I is weak nil clean ideal of R/I. Conversely, let I1/I be weak nil clean ideal of R/I and x ∈ I1. Then either x = e+n or x = −e+n, where e ∈ Idem(I1/I) and n ∈ Nil(I1/I). Since idempotents can be lifted modulo nil ideal, so lift e to e ∈ I1. Then x − e or x + e is nilpotent in I1, modulo I and hence x − e or x + e is nilpotent in I1. (cid:3) Theorem 2.15. Every homomorphic image of weak nil clean ideal of a ring is also weak nil clean ideal. Theorem 2.16. Let {Ri}m i=1 be a family of rings and I ′ is are ideals of Ri, then the i=1 Ri is weak nil clean ideal if and only if each Ii is weak ideal I = Qm i=1 Ii of R = Qm nil clean ideal of Ri and at most one Ii is not nil clean ideal. Proof. (⇒) Let I be weak nil clean ideal of R. Then being homomorphic image of I each Iα is weak nil clean ideal of Rα. Suppose Iα1 and Iα2 are not nil clean ideals, where α1 6= α2. Since Iα1 is not nil clean ideal, so not all elements x ∈ Iα1 are of the form x = n − e, where n ∈ Nil(Rα1) and e ∈ Idem(Rα1). As Iα1 is weak nil clean ideal of Rα1, so there exists xα1 ∈ Iα1 with xα1 = nα1 + eα1, where nα1 ∈ Nil(Rα1) and eα1 ∈ Idem(Rα1), but xα1 6= n − e, for any n ∈ Nil(Rα1) and e ∈ Idem(Rα1). Similarly there exists xα2 ∈ Iα2 with xα2 = nα2 − eα2, where nα2 ∈ Nil(Rα2) and eα2 ∈ Idem(Rα2), but xα2 6= n + e, for any n ∈ Nil(Rα2 ) and e ∈ Idem(Rα2). Define x = (xα) ∈ I by xα = xα = 0 if α ∈ {α1, α2} if α /∈ {α1, α2} Then clearly x 6= n ± e, for any n ∈ Nil(R) and e ∈ Idem(R). Hence at most one Iα is not nil clean ideal. (⇐) If each Iα is nil clean ideal of Rα then I = Q Iα is nil clean ideal of R by Theorem 2.20 [7] and hence weak nil clean ideal of R. Assume Iα0 is weak nil clean ideal but not nil clean ideal of Rα0 and that all other Iα's are nil clean ideals of Rα. If x = (xα) ∈ I, then in Iα0, we can write xα0 = nα0 + eα0 or xα0 = nα0 − eα0, where nα0 ∈ Nil(Rα0) and eα0 ∈ Idem(Rα0). If xα0 = nα0 + eα0, then for α 6= α0 we set, xα = nα + eα, where nα ∈ Nil(Iα) and eα ∈ Idem(Iα). If xα0 = nα0 − eα0, then for α 6= α0 we set, xα = nα − eα, where nα ∈ Nil(Iα) and eα ∈ Idem(Iα); then n = (nα) ∈ Nil(R) and e = (eα) ∈ Idem(R), such that x = n + e or x = n − e and hence I is weak nil clean ideal of R. (cid:3) If the collection of rings is infinite then Theorem 2.16 is not true as shown by the following example. Example 2.17. Consider the ring R = Z3 × Z22 × Z23 × · · ·, clearly for any n ∈ N, Z2n is weak nil clean ring and hence ideal generated by 2 in Z2n say < 2 >n is also weak nil clean ideal of Z2n. But the ideal I =< 3 > × < 2 >2 × < 2 >3 × · ·· is not weak nil clean ideal of R as (3, 2, 2, · · ·) ∈ I can not be written as a sum of an idempotent and a nilpotent element of R. 6 D. K. BASNET AND A. SHARMA Next we study the relationship between weak nil clean ideal of a given ring R and weak nil clean ideal of upper triangular matrix ring Tn(R). Here given a matrix X, Xij denotes the (i, j)th entry of X. Lemma 2.18. For E, N ∈ Tn(R) the following hold: (1) If E 2 = E, then (Eii)2 = Eii for 1 ≤ i ≤ n. (2) N is nilpotent if and only if Nii is nilpotent for 1 ≤ i ≤ n. Proof. See Lemma 2.1.1 [4]. (cid:3) Theorem 2.19. Let T be a 2 × 2 upper triangular matrix ring over R. Then an ideal S = (cid:18) I R 0 J (cid:19) of T is weak nil clean ideal if and only if I and J are weak nil clean ideals of R and one of them is nil clean ideal. Proof. Without loss of generality, assume that I and J are respectively nil clean and 0 b (cid:19) ∈ (cid:18) I R b = −f + n1, where f ∈ Idem(J) and n1 ∈ Nil(J). If b = f + n1, set a = e + n, weak nil clean ideals of R. Let x = (cid:18) a r where e ∈ Idem(I) and n ∈ Nil(I). Then x = (cid:18) e 0 (cid:18) e 0 0 f (cid:19) ∈ Idem(T ) and (cid:18) n r where e ∈ Idem(I) and n ∈ Nil(I). Then x = −(cid:18) e 0 (cid:18) n r 0 J (cid:19). So either b = f + n1 or 0 n1 (cid:19), where 0 n1 (cid:19) ∈ Nil(T ). If b = −f + n1, set a = −e + n, 0 n1 (cid:19), where 0 n1 (cid:19) ∈ Nil(T ) and (cid:18) e 0 0 f (cid:19) ∈ Idem(T ) by Lemma 2.18. 0 f (cid:19) + (cid:18) n r 0 f (cid:19) + (cid:18) n r For the converse, clearly I and J are weak nil clean ideals of R. Suppose both are not nil clean ideals of R. As I is not weak nil clean ideal of R, so there exists x ∈ I such that x = e1 + n1, where e1 ∈ Idem(I) and n1 ∈ Nil(I) but x 6= n − e, for all n ∈ Nil(I) and e ∈ Idem(I). Similarly there exists y ∈ J such that y = −e2 + n2, where e2 ∈ Idem(J) and n2 ∈ Nil(J) but y 6= n + e, for all n ∈ Nil(J) and e ∈ Idem(J). Then it is easy to see that (cid:18) x 0 0 y (cid:19) is not weak nil clean element of T . (cid:3) Let R be a commutative ring and M be a R-module. Then the idealization of R and M is the ring R(M) with underlying set R × M under coordinatewise addition and multiplication given by (r, m)(r′, m′) = (rr′, rm′ + r′m), for all r, r′ ∈ R and m, m′ ∈ M. It is obvious that if I is an ideal of R then for any submodule N of M, I(N) = {(r, n) : r ∈ I , n ∈ N} is an ideal of R(M). First we mention basic existing results about idempotents and nilpotent elements in R(M) and study the nil clean ideals of the idealization R(M) of R and R-module M. Lemma 2.20. Let R be a commutative ring and R(M) be the idealization of R and R-module M . Then the following hold: (1) (r, m) ∈ Idem(R(M)) if and only if r ∈ Idem(R) and m = 0. WEAK NIL CLEAN IDEAL 7 (2) (r, m) ∈ Nil(R(M)) if and only if r ∈ Nil(R). Proof. (1) is obvious and (2) follows from the fact that (r, m)n = (rn, nrn−1m), for any r ∈ R and m ∈ M. (cid:3) Proposition 2.21. Let R be a commutative ring and R(M), the idealization of R and R-module M . Then an ideal I of R is weak nil clean ideal of R if and only if I(N) is weak nil clean ideal of R(M), for any submodule N of M . Proof. (⇒) Let I be weak nil clean ideal of R. Consider an Ideal I(N) of R(M), for some submodule N of M. Let (x, m) ∈ I(N). Then either x = n + e or x = −e + n, for some e ∈ Idem(R) and n ∈ Nil(R). So either (x, m) = (e, 0) + (n, m) or (x, m) = −(e, 0) + (n, m), where (e, 0) ∈ Idem(R(M)) and (n, m) ∈ Nil(R(M)) by Lemma 2.20. (⇐) Let I(N) be a weak nil clean ideal of R(M) and r ∈ I. For (r, 0) ∈ I(N), either (r, 0) = (e, 0) + (n, 0) or (r, 0) = −(e, 0) + (n, 0), for some (e, 0) ∈ Idem(R(M)) and (n, 0) ∈ Nil(R(M)). By Lemma 2.20, we conclude that either r = e+n or r = −e+n, where e ∈ Idem(R) and n ∈ Nil(R). (cid:3) In the following proposition we study about weak nil clean element of a corner ring. Proposition 2.22. Let R be a ring and f ∈ Idem(R). An element a ∈ f Rf is strongly weak nil clean element of R if and only if a ∈ f Rf is strongly weak nil clean element of f Rf . Proof. As f Rf is a left ideal of f R and f R is a right ideal of R. Hence the result follows from Theorem 2.9. (cid:3) Corollary 2.23. If R is strongly weak nil clean ring and f ∈ R is any idempotent, then the corner ring f Rf is strongly weak nil clean. A Morita context denoted by (R, S, M, N, ψ, φ) consists of two rings R and S, two bimodules RNS and SMR and a pair of bimodule homomorphisms (called pairings) ψ : N ⊗S M → R and φ : M ⊗R N → S, which satisfy the following associativity: ψ(n ⊗ m)n′ = nφ(m ⊗ n′) and φ(m ⊗ n)m′ = mψ(n ⊗ m′), for any m, m′ ∈ M r ∈ R, s ∈ S, m ∈ M and n ∈ N forms a ring denoted by T , called the ring and n, n′ ∈ N. These conditions ensure that the set of matrices (cid:18) r n of the context. For any subset I of T , define pR(I) = {a ∈ R : (cid:18) a x pM (I) = {y ∈ M : (cid:18) a x {x ∈ N : (cid:18) a x y b (cid:19) ∈ I} A morita context R = (cid:18) A M m s (cid:19), where y b (cid:19) ∈ I}, y b (cid:19) ∈ I} and pN (I) = y b (cid:19) ∈ I}, pS(I) = {b ∈ S : (cid:18) a x products MN = 0 and NM = 0. N B (cid:19) is called morita context of zero pairing if context 8 D. K. BASNET AND A. SHARMA Lemma 2.24. Let T = (cid:18) A M and only if I = (cid:18) A1 M1 N B (cid:19) be a morita context. Then I is an ideal of T if N1 B1 (cid:19), where A1 and B1 are ideals of A and B respectively, M1 and N1 are submodules of AMB and BNA respectively, with M1N ⊆ A1, N1M ⊆ B1, A1M ⊆ M1, B1N ⊆ N1, MN1 ⊆ A1, NM1 ⊆ B1, MB1 ⊆ M1 and NA1 ⊆ N1. In this case A1 = pA(I), B1 = pB(I), M1 = pM (I) and N1 = pN (I). Proof. See Lemma 2.1 [9]. (cid:3) Theorem 2.25. Let R = (cid:18) A M N B (cid:19) be a morita context and I = (cid:18) A1 M1 N1 B1 (cid:19) be a strongly weak nil clean ideal of R. Then A1 and B1 are strongly weak nil clean ideals of A and B respectively. Proof. The proof follows from Corollary 2.23 and Theorem 2.9. (cid:3) B1 are weak nil clean ideals of A and B respectively, where at least one of them is N B (cid:19) be a morita context of zero pairing. If A1 and Theorem 2.26. Let R = (cid:18) A M strongly nil clean ideal then I = (cid:18) A1 M1 N1 B1 (cid:19) is a weak nil clean ideal of R. Proof. Let A1 be strongly nil clean ideal of A and B1 be strongly weak nil clean ideal q ∈ Nil(B1) such that either b = e + q or b = −e + q. If b = e + q, then set a = f + p, of B respectively. Let x = (cid:18) a m where f ∈ Idem(A1) and p ∈ Nil(A1) and we get, x = (cid:18) f 0 pk = 0 and qk = 0, for some k ∈ N, where (cid:18) f 0 2.28 [7], we conclude that (cid:18) p m n b (cid:19) ∈ I. Then there exist e ∈ Idem(B1) and n q (cid:19) and 0 e (cid:19) ∈ Idem(R). Now from Theorem n q (cid:19) ∈ Nil(R). Also if b = −e+q then set a = −f +p 0 e (cid:19) + (cid:18) p m and similar as above we can show that x is weak nil clean element of R. (cid:3) References [1] Ahn M. S. and Anddreson D. D., Weakly clean rings and almost clean rings, Rocky mountain J. math., 36 (2006), 783-798. [2] Basnet D. K. and Bhattacharyya J., Weak nil clean rings, Preprint arXiv:1510.07440, (2015). [3] Chen H. and Chen M., On clean ideals, Int. J. Math. Math. Sci. 62 (2003), 3949-3956. [4] Diesl A. J., Classes of Strongly Clean Rings, Ph. D. Thesis, University of California, Berkeley, (2016). [5] Diesl A. J., Nil clean rings, J. Algebra 383 (2013), 197-211. [6] Nicholson W. K., Lifting idempotents and exchange rings, Trans. Amer. Math. Soc., 229 (1977), 269-278. [7] Sharma A. and Basnet D. K., Nil Clean Ideal, Preprint arXiv:1709.02065, (2017). [8] Sharma A. and Basnet D. K., Weakly Clean Ideal, Preprint arXiv:1704.08814, (2017). [9] Tang G., Li C. and Zhou Y., Study of Morita contexts, Comm. Algebra, 42(4) (2014), 1668- 1681. WEAK NIL CLEAN IDEAL 9 D. K. Basnet and A. Sharma Department of Mathematical Sciences, Tezpur University, Napaam-784028, Sonitpur, Assam, India. E-mail address: [email protected], [email protected]
1506.05080	1	1506	2015-06-16T19:12:41	Behaviour of injective dimension with respect to regradings	[ "math.RA" ]	Given a left noetherian k-algebra A graded by a group G, an injective object I in the category of G-graded A-modules and a morphism from G to another group G', we provide bounds for the injective dimension of I as a G'-graded A-module. For this, we use three change of grading functors. Most of the constructions concerning these functors work in the context of H-comodule algebras, where H is a Hopf algebra, so we develop them in this general context.	math.RA	math	BEHAVIOUR OF INJECTIVE DIMENSION WITH RESPECT TO REGRADINGS ANDREA SOLOTAR AND PABLO ZADUNAISKY ABSTRACT. Given a left noetherian k-algebra A graded by a group G, an injective object I in the category of G-graded A-modules and a morphism from G to another group G ′, we provide bounds for the injective dimension of I as a G ′-graded A-module. For this, we use three change of grading functors. Most of the constructions concerning these functors work in the context of H-comodule algebras, where H is a Hopf algebra, so we develop them in this general context. 2010 MSC: 16W50, 16D50, 16E10, 16T15. Keywords: injective dimension, gradings. 1. INTRODUCTION In [VdB97], M. Van den Bergh states that if A is a left noetherian N-graded connected algebra over a field k and I is an injective object in the category of Z-graded A-modules, then I has injective dimension at most one when considered as an A-module. He leaves the proof of this fact as "a pleasant exercise in homological algebra". In this note we prove a general version of this result: we provide bounds for the injective dimension of a graded injective module when the grading changes. This is useful for example when one wants to study how A-modules behave with respect to a property which is valid for graded A-modules. Notice that B. Fossum and H. Foxby had already proven in [FF74, Theorem 4.10] the statement in Van den Bergh'a article for A commutative, using localization techniques. Also, A. Yekutieli gave in [Yek14] a detailed proof of Van den Bergh's statement. Our proof is completely unrelated to the ones found in either of these references and it deals with arbitrary grading groups and change of gradings in the non necessarily commutative case. Just to get an idea of the situation, let us start by looking at a possible solution to Van den Bergh's exercise. Let A be a Z-graded algebra and let N be a left A-module. We can turn N ⊗k k[t, t−1] into a Z-graded left A-module, with the action of an element a ∈ A of degree l ∈ Z given by a · (n ⊗ tr) = an ⊗ tr+l for all n ∈ N and r ∈ Z. There is an A-linear surjective map N ⊗k k[t, t−1] −→ N, induced by the projection k[t, t−1] −→ k[t, t−1]/(t − 1) ∼= k, with kernel N ⊗k (t − 1)k[t, t−1]. Thus we obtain an IMAS AND DTO. DE MATEM ´ATICA, FACULTAD DE CIENCIAS EXACTAS Y NATURALES, UNIVERSIDAD DE BUENOS AIRES, CIUDAD UNIVERSITARIA, PABELL ´ON 1, 1428) BUENOS AIRES, ARGENTINA E-mail address: [email protected], [email protected]. This work has been supported by the projects UBACYT 20020130100533BA, PIP-CONICET 112 -- 201101 -- 00617, PICT 2011 -- 1510 and MATHAMSUD-GR2HOPF. The first author is a research member of CONICET (Argentina). 1 exact sequence of A-modules 0 / N ⊗k k[t, t−1] ·(t−1) / N ⊗k k[t, t−1] / N / 0. As a sequence of k-modules, it is the tensor product of N with the minimal projective resolution of k ∼= k[t, t−1]/(t − 1) as a k[t, t−1]-module. Let I be an injective object in the category of Z-graded left A-modules. Since N ⊗k k[t, t−1] is a Z-graded left A-module, it is natural to ask whether the fact that I is graded injective implies that Exti A(N ⊗k k[t, t−1], I) = 0 for all i > 0; in Prop. 3.2, we prove that this holds if A is noetherian. Thus, in this case we have a resolution of length 1 of N by left A-modules which are acyclic with respect to the functor HomA(−, I); this proves that Ext2 A(N, I) = 0. Since N is arbitrary, we deduce that the injective dimension of I in the category of A-modules is at most 1. Motivated at first by the example of change of grading considered in [RZ], the ob- jective of this note is to put this result in a more general perspective, showing how injective dimension changes when we change the grading group over a fixed algebra. We show that the general case follows the same pattern as the Z-graded case, and re- quires little more than general homological algebra. For this, we define three change of grading functors. These functors arise in the more general situation of A-modules endowed with two different comodule structures over two Hopf algebras related by a morphism. We prove the following result. Theorem: Let A be a noetherian G-graded k-algebra and let ϕ : G −→ G ′ be a morphism of groups. Let n be the projective dimension of kG ′ as a G ′-graded left kG- module. Given a G-graded injective A-module I, the injective dimension of I when considered as a G ′-graded A-module through ϕ is less than or equal to n. In Section 2 we first recall some definitions and fix notations. Afterwards we define the change of grading functors in the Hopf algebra setting and prove useful properties about them. In Section 3 we specialize to the group-graded situation and we prove our main result. In Section 4 we obtain bounds for the injective dimension of a graded module when the grading changes through a group morphism. Throughout this work k denotes a commutative ring with unit and A is a k-algebra, projective as k-module. All unadorned tensor products are over k, and all Hopf alge- bras will also be projective as k-modules. We thank Mariano Su´arez-Alvarez for a careful reading of a previous version of this article. 2. GRADED ALGEBRAS AND MODULES 2.1. Some definitions and notations. Let A be a k-algebra and let G be a group. The algebra A is G-graded if it can be decomposed as the direct sum of sub-k-modules Ag indexed by G such that AgAg ′ ⊆ Agg ′. A G-graded module M over A is an A-module with a decomposition as a direct sum of sub-k-modules Mg such that AgMg ′ ⊆ Mgg ′. 2 / / / / The k-module Mg is called the homogeneous component of M of degree g. We consider k to be a G-graded algebra with k1G = k and all other components equal to 0. An A-linear morphism f : N −→ M between G-graded left A-modules is said to be homogeneous of degree g if f(Ng ′ ) ⊆ Mg ′g for all g ′ in G. A G-grading on A is equiv- alent to a right kG-comodule algebra structure on A, and a G-graded left A-module is a left A-module with a compatible structure of kG-comodule, that is, the following diagram commutes: µM A ⊗ M ρA⊗M M ρM A ⊗ M ⊗ kG µm⊗Id / M ⊗ kG where µM denotes the action of A on M and ρM, ρa⊗M denote the respective kG coactions. Given a Hopf algebra H, we will denote by A ModH the category of left A- modules with compatible structure of right H-comodules. The morphisms in A ModkG, which we shall denote HomkG A−(−, −) are homogeneous of trivial degree. For any f ∈ HomkG A−(N, M) and g ∈ G we write fg for the k-linear map fg : Ng −→ Mg obtained by restriction and corestriction of f, and we call it the homogeneous component of f in degree g. Notice that f is determined by its homogeneous components. Given a G-graded left A-module M and g ∈ G, we denote by M(g) the A-module M with a new G-grading, whose homogeneous components are given by M(g)g ′ = Mg ′g. If f ∈ HomkG underlying function is the same as that of f. It is easy to see that f(g) is homogeneous A−(M, M ′), we define f(g) : M(g) −→ M ′(g) to be the morphism whose of trivial degree. Thus we obtain the g-shift functor, denoted by Σg : AModkG −→ AModkG. One can check that Σg ◦ Σg−1 = Id, so the g-shift functor is an automorphism of AModkG. An obvious example of a G-graded algebra is the group algebra kG. We will write ug for the canonical generator of the homogeneous component of kG of degree g, that is (kG)g = kug. 2.2. Change of grading functors. Let H and H ′ be two Hopf algebras and let ϕ : H −→ H ′ be a morphism of Hopf algebras. Suppose we are given a right H-comodule algebra A with structure morphism ρH A : A −→ A ⊗ H ′ turns A into an H ′-comodule algebra. Of course the same idea works with left H-comodule algebras; in particular, we consider H as a left H ′-comodule algebra with structure map ϕ ⊗ 1 ◦ ∆H. A : A −→ A ⊗ H. The composition 1 ⊗ ϕ ◦ ρH The morphism ϕ induces two functors ϕ! : A ModH −→ A ModH ′ which we now define. ϕ∗ : A ModH ′ Given an object M of A ModH with structure morphism ρH −→ A ModH M : M −→ M ⊗ H, the ϕ!(M) = M : M −→ M ⊗ H ′. Furthermore, given a morphism f : M −→ M ′ in A ModH, object ϕ!(M) coincides with M as A-module, while its structure morphism is ρH ′ 1 ⊗ ϕ ◦ ρH 3 / / / we set ϕ!(f) = f. It is easy to check that the functor ϕ! is well defined. We say that ϕ! is obtained by restriction along ϕ. , we define ϕ∗(N) as N(cid:3) H ′ H, where (cid:3) Now, given an object N of A ModH ′ H ′ is the cotensor product of H ′-comodules; here we consider H as a left H ′-comodule through ϕ. Notice that N ⊗ H has a left A ⊗ H-module structure. Since ρH A : A −→ A ⊗ H is a ring homomorphism, it induces a left A-module structure on N ⊗ H. The subspace ϕ∗(N) = N(cid:3) H ′H ⊆ N ⊗ H is an A-submodule of N ⊗ H. For any morphism f : N −→ N ′ in A ModH ′ , we set ϕ∗(f) : ϕ∗(N) −→ ϕ∗(N ′) to be f ⊗ Id : N(cid:3) H ′H, that is the restriction and corestriction of f ⊗ Id. H ′ H −→ N ′(cid:3) The cotensor product has been defined originally in [EM66]. The functors ϕ! and ϕ∗ are A-equivariant versions of those introduced in [Doi81, 1.2]. Proposition 2.1 below corresponds to Proposition 6 of that article. Proposition 2.1. The functor ϕ∗ is right adjoint to ϕ!. Proof. The image of the coaction map ρH hence induces a map ιM : M −→ ϕ∗(ϕ!(M)), which is natural in M and compatible with the action of A. Also, there is a natural transformation εN : ϕ!(ϕ∗(N)) −→ N H ′ H ′ ∼= N, given by the composition of Id(cid:3) which is also A-linear. These are respectively the unit and the counit of the adjoint pair (ϕ!, ϕ∗). (cid:3) M : M −→ M ⊗ H is contained in M(cid:3) H ′ϕ with the canonical isomorphism N(cid:3) H ′H, and By definition the functor ϕ! is exact and reflects exactness, meaning that a complex in A ModH is exact if and only if its image by ϕ! is exact. As shown in [Doi81, Proposition 5], the functor ϕ∗ is exact if H is an injective H ′-comodule, for example if it is free. By standard properties of adjoint functors, we obtain the following corollary. For the proof of the second statement, see [Wei94, Prop. 2.3.10]. Corollary 2.2. The functor ϕ∗ sends injective objects to injective objects. Furthermore, if H is an injective H ′-comodule, the functor ϕ! sends projective objects to projective objects. The hypothesis of the last part of the corollary is satisfied for example when H ′ is a group algebra over a field, since in this case H ′ is cosemisimple and thus every comod- ule is injective. For the rest of this subsection, k will be a field. Let L = {x ∈ H : ρH ′ H (x) = x ⊗ 1} be the subalgebra of coinvariants of H as left H ′-comodule. Recall that L ⊆ H is called cleft if there exists a convolution-invertible left H ′-comodule morphism γ : H ′ −→ H and in that case there is a left H ′-colinear, right L-linear isomorphism of Hopf algebras σ#L and the extension L ⊆ H is H ′-Galois; see [Mon93, Theorem 7.2.2] for H −→ H ′ By [Wis00, 5.1], the category A ModH ′ is a Grothendieck category. If the extension L ⊆ H is cleft, then H is a free H ′-comodule, so it is injective and in particular ϕ∗ : A ModH ′ −→ A ModH is an exact functor, hence preserves colimits. Thus, by Freyd's Adjoint Functor Theorem, ϕ∗ has a right adjoint, which we denote by ϕ∗. details. The following corollary is also a consequence of adjointness. Corollary 2.3. If the extension L ⊆ H is H ′-cleft, then the functors ϕ! and ϕ∗ send projective objects to projective objects. 4 We set some more notation. Given a Hopf algebra H and an object N of A ModH we denote by Add(N) the full subcategory of A ModH whose objects are direct summands in A ModH of coproducts of copies of N. In other words Add(N) is the smallest full subcategory of A ModH containing N and closed by direct sums and direct summands. Proposition 2.4. Suppose L ⊆ H is cleft and H ′ is cocomutative. If N is an object of A ModH ′ then there exists a resolution S• −→ N, with Si in Add(ϕ!(ϕ∗(N))) for all i ≥ 0. Moreover, if n is the projective dimension of H ′ in H ModH ′ Proof. Since H ′ is cocommutative, given T ∈ H ModH ′ H ′T is an H ′- comodule and it has an A-module structure given by a(n ⊗ t) = a0n ⊗ a1t, compatible H ′− as a functor from H ModH ′ with the H ′-comodule structure. We can thus view N(cid:3) to A ModH ′ , then Si = 0 for all i > n. , the k-module N(cid:3) . , given by Θ(T ) = H ⊗L T , so H ModH ′ By [Mon93, Theorem 8.5.6] and originally [Sch90], if the antipode of H ′ is bijective -and this is the case here-, then there is an equivalence of categories Θ : L Mod −→ Mod ∼= H ModH ′ resolution P• −→ H ′ in H ModH ′ H ModH ′ the complex N(cid:3) isomorphic to N(cid:3) H ′ has enough projectives and H each projective object is a direct summand of a direct sum of copies of H. Fix a projective . Since each Pi is a direct summand of a free module in and H is free as an H ′-comodule, this is a flat resolution of H ′ so, if we consider H ′ P•, it is acyclic in positive degrees and its homology in degree zero is H ′ H ′ ∼= N. For each Pi choose Qi in H ModH ′ such that Pi ⊕ Qi is free of H ′(Pi ⊕ Qi), which is isomorphic as H ′P• (cid:3) to the direct sum of ni copies of ϕ!(ϕ∗(N)). Thus setting S• = N(cid:3) rank ni ∈ N ∪ {∞}, so (N(cid:3) object of A ModH ′ the proof is complete. H ′ Qi) ∼= N(cid:3) H ′ Pi) ⊕ (N(cid:3) If M ∈ A ModH, there is an A-linear map ζ : M ⊗ L −→ ϕ∗(ϕ!(M)) such that m ⊗ l 7→ m0 ⊗ m1l. This map is injective since the corestriction of η : ϕ∗(ϕ!(M)) −→ M ⊗ H defined as m ⊗ g 7→ m0 ⊗ S(m1)g is a left inverse of ζ. Clearly ζ is an isomorphism if H and M are finite-dimensional, and it is easily checked that this also holds if H and H ′ are group algebras. We do not know if this holds in more general situations. 3. THE GROUP ALGEBRA CASE We now focus on the case in which H = kG and H ′ = kG ′ are group algebras and ϕ : H −→ H ′ is induced by a group morphism ^ϕ : G −→ G ′. In this case, we have L = k^L with ^L = ker ^ϕ. As a particular case of the previous constructions, the morphism ^ϕ : G −→ G ′ induces a G ′-grading on A via the functor ϕ!, the homogeneous G ′- components are Ah = M g∈ϕ−1(h) Ag for h ∈ G ′. Lemma 3.1. Let M be an object of A ModH. The morphism Ll∈^L M[l] −→ ϕ∗(ϕ!(M)) send- ing m ∈ M[l]g to m ⊗ g ∈ ϕ∗(ϕ!(M)) is an isomorphism in A ModH, which is natural. 5 Proof. In this case the map η : ϕ∗(ϕ!(M)) −→ M ⊗ H defined at the end of the pre- vious subsection is a two-sided inverse of ζ, so ϕ∗(ϕ!(M)) ∼= M ⊗ L. Now the map Ll∈^L M[l] −→ M ⊗ L given by m ∈ M[l] 7→ m ⊗ l−1 ∈ M ⊗ L is an isomorphism in A ModH, and the morphism in the statement is the composition of this map with ζ. (cid:3) In the case where H = kG and H ′ = kG ′ are group algebras, the functor ϕ∗ has the following concrete description. Explicitly, if M is a G-graded left A-module and h ∈ G ′, the homogeneous component of degree h of ϕ∗(M) is ϕ∗(M)h = Y Mg. g∈ϕ−1(h) We now define the left action of A on ϕ∗(M) in the group algebra case; notice that it is enough to define the action of a G-homogeneous element of A over a G ′-homogeneous If a ∈ Ag and (mg)g∈ϕ−1(h) ∈ ϕ∗(M)h, then a · (mg)g∈ϕ−1(h) = element of ϕ∗(M). (amg)g∈ϕ−1(h). Given a morphism f in A ModkG, the homogeneous component in degree h of ϕ∗(f) is given by ϕ∗(f)h = Q fg, which is easily seen to be A-linear. g∈ϕ−1(h) We refer to the functors ϕ∗, ϕ∗, ϕ! defined in the previous paragraphs as the change of grading functors. As we have seen, they form an adjoint triple, i.e. ϕ∗ is right adjoint to ϕ! and left adjoint to ϕ∗. These functors are not new. They appear in different guises in [Doi81], [NVO04], [PP11], and probably many other places. However, as far as we know this is the first time the functor ϕ∗ is discussed in the context of graded modules over an algebra. Remark 3.1.1. If M is an object of A ModkG, there is a natural isomorphism ϕ∗ ◦ ϕ∗(M) ∼= Y M[l]. l∈ker ϕ In our setting, the change of grading functors are exact, and also ϕ∗ reflects exactness; this does not hold for ϕ∗ unless ϕ is surjective. Given I, M of A ModH, we recall that M is I-acyclic if Ri HomH A−(M, I) = 0 for all i > 0. Proposition 3.2. Assume A is a noetherian G-graded k-algebra and let I be an injective object in A ModkG. (1) If M is an object of A ModkG, then ϕ!(M) is ϕ!(I)-acyclic. (2) Let L = HcoH ′ Mod ∼= H ModH ′ be the equivalence of categories such that Θ(T ) = H ⊗L T . If P is a projective L-module and N is any G ′-graded A- module, then N(cid:3)kG ′ Θ(P) is ϕ!(I)-acyclic. and Θ : L Mod −→ H ′ H Proof. By Remark 3.1.1 and the adjunctions between the change of grading functors, there are natural isomorphisms HomkG ′ A− (ϕ!(−), ϕ!(I)) ∼= HomkG A−(−, ϕ∗(ϕ!(I)) ∼= HomkG A−(−, M l∈^L I[l]). Since shift functors are autoequivalences of the category AModkG, they preserve in- jectives. Since A is noetherian, the G ′-graded A-module ϕ∗(ϕ!(I)) is injective. This follows from the graded analogue of the Bass-Papp Theorem (see for example the 6 proof in [GW04, Theorem 5.23], which adapts easily to the graded case). Therefore the functor HomkG ′ A− (ϕ!(−), ϕ!(I)) is exact. On the other hand, ϕ! is an exact functor that sends projective objects to projective objects, so there are natural isomorphisms (Ri HomkG ′ A− (ϕ!(−), ϕ!(I))), and, since the last functor is identically zero, this proves the first statement in the proposition. A− )(ϕ!(−), ϕ!(I)) ∼= Ri(HomkG ′ Now, if P is a projective L-module, then it is a direct summand of a free L-module F, so N(cid:3)kG ′ Θ(P) is a direct summand of N(cid:3)kG ′Θ(F), which in turn is isomorphic to a direct sum of copies of ϕ!(ϕ∗(N)). Thus N(cid:3)kG ′Θ(F), being a direct sum of ϕ!(I)-acyclic modules, is itself ϕ!(I)-acyclic and so are its direct summands. This proves item 2. (cid:3) We write Ext kG,i A− (−, M) for the ith right derived functor of HomkG A−(−, M). Theorem 3.3. Let A be a noetherian G-graded k-algebra. Let n be the projective dimension of kG ′ in the category kG ModkG ′ . If I is an injective object in A ModkG, then the injective dimension of ϕ!(I) is less than or equal to n. Proof. By Proposition 2.4 every object N of A ModkG ′ has a resolution S• by objects of Add(ϕ!(ϕ∗(N))). Moreover, this resolution can be chosen of length smaller than or equal to n. By item 2 of Proposition 3.2, each object of Add(ϕ!(ϕ∗(N))) is acyclic for the functor HomkG ′ A− (−, ϕ!(I)). This fact implies that for all i ≥ 0 there is an isomorphism kG ′,i A− (N, ϕ!(I)) = Ext 0 for any N and all i > n, and the result follows immediately. (cid:3) A (S•, ϕ!(I))) for all i ≥ 0. In particular Ext kG ′,i A− (N, ϕ!(I)) ∼= Hi(HomkG 4. INJECTIVE DIMENSION OF GRADED MODULES The results from the previous section have an easy consequence that we prove next. Let ^L be, as before, the kernel of ^ϕ : G −→ G ′. Recall that the cohomological dimen- sion of ^L over k coincides with the projective dimension of the trivial L-module k. We denote by idG A M the injective dimension of an object M in A ModkG. Proposition 4.1. If M is an object of A ModkG and n is the cohomological dimension of ^L, then: idG A M ≤ idG ′ A ϕ!(M) ≤ idG A M + n. Proof. Since ϕ∗ preserves injectives, it follows that idG A ϕ!(M). On the other hand, we already know that there is a natural isomorphism ϕ∗(ϕ!(M)) ∼= Ll∈^L M[l]. The shift functors are autoequivalences of the category A ModkG, therefore: A ϕ∗(ϕ!(M)) ≤ idG ′ idG A  M l∈^L M[l]  = sup{idG A M[l] l ∈ ^L} = idG A M, and the first inequality follows. The second one is obvious if idG A M is infinite, so we assume it is finite and proceed by induction on d = idG A M. The case d = 0 is Theorem 3.3. Assume d > 0 and let M −→ I be a monomorphism into an injective G-graded A-module; let ΩM be its 7 cokernel, which has injective dimension d − 1 in A ModkG. We get a short exact sequence of G ′-graded modules 0 −→ ϕ!(M) −→ ϕ!(I) −→ ϕ!(ΩM) −→ 0. A ϕ!(I), idkG Using a standard argument, idG ′ A ϕ!(ΩM)} + 1, which by the inductive hypothesis is smaller than or equal to d. This proves the second inequal- ity. (cid:3) A ϕ!(M) ≤ max{idkG The inequalities in the statement of Corollary 4.1 are sharp, as the following examples show. For the first inequality, take G = Z, G ′ = {0} and, of course, ϕ : Z −→ {0} the trivial morphism. Set A = k[t] with the obvious Z-grading. We see that idZ A A = idA A = 1, thus in this case, for M = A, the first inequality is in fact an equality. On the other hand, idZ k[t, t−1] = 0, but idA k[t, t−1] = 1, so in this the case the second A inequality is an equality. Incidentally, the case where G = Z and G ′ = {0} was already studied by E. Ekstr om, see [Eks89, Theorem 0.2]. Levasseur has proved in [Lev92, 3.3] that if A is noetherian and N-graded, its injec- tive dimension and its graded injective dimension are equal. The proof of this result uses a spectral sequence which is not available if the grading group is not Z. It would be interesting to find a different proof using the change of grading functors, but we have been unable to do so. We can, however, prove the following result, which holds even if the algebra A is not noetherian and thus provides a generalization of Levasseur's result. Recall that an Nn-graded algebra is a Zn-graded algebra such that the support {ξ ∈ Zn Aξ 6= 0} is contained in Nn. Proposition 4.2. Suppose A is an Nn-graded algebra and let ϕ : Zn −→ Z be the morphism defined by ϕ(a1, . . . , an) = a1 + · · · + an. The injective dimension of A in ModZn A is equal to its injective dimension in ModZ A. Proof. Since we already know that idZn prove the opposite inequality. For every n ∈ N the set ϕ−1(n) ∩ Nn is finite, so A A by Proposition 4.1, we only need to A A ≤ idZ M ξ∈ϕ−1(n) Aξ = Y Aξ. ξ∈ϕ−1(n) It follows from this that ϕ!(A) = ϕ∗(A) and, since ϕ∗ is right adjoint to the exact functor ϕ∗, it preserves injectives; in particular, idZ A A, which completes the proof. (cid:3) A ϕ!(A) is at most idZn A ϕ∗(A) = idZ REFERENCES [Doi81] Y. Doi, Homological coalgebra, J. Math. Soc. Japan 33 (1981), no. 1, 31 -- 50. [EM66] S. Eilenberg and J. C. Moore, Homology and fibrations. I. Coalgebras, cotensor product and its derived functors, Comment. Math. Helv. 40 (1966), 199 -- 236. [Eks89] E. K. Ekstr om, The Auslander condition on graded and filtered Noetherian rings, Ann´ee (Paris, 1987/1988), Lecture Notes in Math., vol. 1404, Springer, Berlin, 1989, pp. 220 -- 245. [FF74] R. Fossum and H.-B. Foxby, The category of graded modules, Math. Scand. 35 (1974), 288 -- 300. [GW04] K. R. Goodearl and R. B. Warfield Jr., An introduction to noncommutative Noetherian rings, 2nd ed., London Mathematical Society Student Texts, vol. 61, Cambridge University Press, Cambridge, 2004. 8 [Lev92] T. Levasseur, Some properties of noncommutative regular graded rings, Glasgow Math. J. 34 (1992), no. 3, 277 -- 300. [Mon93] S. Montgomery, Hopf algebras and their actions on rings, CBMS Regional Conference Series in Mathematics, vol. 82, Published for the Conference Board of the Mathematical Sciences, Wash- ington, DC, 1993. [NVO04] C. Nastasescu and F. Van Oystaeyen, Methods of graded rings, Lecture Notes in Mathematics, vol. 1836, Springer-Verlag, Berlin, 2004. [PP11] A. Polishchuk and L. Positselski, Hochschild (co)homology of the second kind I (2011). Available at arixv.org/abs/1010.0982. [RZ] L. Rigal and P. Zadunaisky, Twisted Semigroup Algebras, Algebra Represent. Th. Available at http://arxiv.org/abs/1406.2985. [Sch90] H.-J. Schneider, Representation theory of Hopf Galois extensions, Israel J. Math. 72 (1990), no. 1-2, 196 -- 231. [VdB97] M. Van den Bergh, Existence theorems for dualizing complexes over non-commutative graded and fil- tered rings, J. Algebra 195 (1997), no. 2, 662 -- 679. [Wei94] C. A. Weibel, An introduction to homological algebra, Cambridge Studies in Advanced Mathemat- ics, vol. 38, Cambridge University Press, Cambridge, 1994. [Wis00] R. Wisbauer, Module and comodule categories -- a survey, Proceedings of the Mathematics Confer- ence (Birzeit/Nablus, 1998), World Sci. Publ., River Edge, NJ, 2000, pp. 277 -- 304. [Yek14] A. Yekutieli, Another proof of a theorem of Van den Bergh about graded-injective modules (2014). Avail- able at http://arxiv.org/abs/1407.5916. 9
1505.02238	2	1505	2017-01-02T18:00:02	2-D Skew Constacyclic Codes Over $R[x,y,\rho,\theta]$	[ "math.RA" ]	In this paper, we give some characterizations of the ring $R[x,y,\rho,\theta]$ where R is a commutative ring. We investigate 2-D skew $(\lambda_1,\lambda_2)$-constacyclic codes.	math.RA	math	2-D SKEW CONSTACYCLIC CODES OVER R[x, y; ρ, θ] H. MOSTAFANASAB Abstract. For a finite field Fq, the bivariate skew polynomial ring Fq[x, y; ρ, θ] has been used to study codes [25]. In this paper, we give some characterizations of the ring R[x, y; ρ, θ] where R is a commutative ring. We investigate 2-D skew (λ1, λ2)-constacyclic codes in the ring R[x, y; ρ, θ]/hxl − λ1, ys − λ2il . Also, the dual of 2-D skew (λ1, λ2)-constacyclic codes is investigated. 1. Introduction Cyclic codes are amongst the most studied algebraic codes. Their structure is well known over finite fields [17]. Recently codes over rings have generated a lot of interest after a breakthrough paper by Ham- mons et al. [12] showed that some well known binary non-linear codes are actually images of some linear codes over Z4 under the Gray map. Constacyclic codes over finite fields form a remarkable class of linear codes, as they include the important family of cyclic codes. Consta- cyclic codes also have practical applications as they can be efficiently encoded using simple shift registers. They have rich algebraic struc- tures for efficient error detection and correction, which explains their preferred role in engineering. In general, due to their rich algebraic structure, constacyclic codes have been studied over various finite chain rings (see [1, 2, 7, 8, 9, 10, 18, 24]). Two-dimensional (2-D) cyclic codes are generalizations of usual cyclic codes which were introduced by Ikai [13] and Imai [14]. Guneri and Ozbudak studied the relations et al. between quasi-cyclic codes and 2-D cyclic codes [11]. Decoding prob- lem for 2-D cyclic codes was studied by some authors [20, 21, 22, 23]. Polynomial rings and their ideals are essential to the construction and understanding of cyclic codes. For the first time in [3] non-commutative skew polynomial rings have been used to construct (a generalization of) cyclic codes. Skew-cyclic codes were introduced by Boucher et al. [5, 6]. They considered the skew-cyclic codes as ideals or submodules over skew polynomial rings and studied dual skew-cyclic codes. Skew MSC(2010): Primary: 00A69; Secondary: 12E20; 94B05 Keywords: Cyclic codes, Skew polynomial rings, 2-D skew constacyclic codes. Published in Journal of Algebra and Related Topics, Vol. 4, No 2, (2016), pp 49-63. 1 2 MOSTAFANASAB in [4] constacyclic codes have been investigated by by Boucher et al. and Jitman et al. in [15]. Xiuli and Hongyan [25] generalized the 2- D cyclic codes to 2-D skew-cyclic codes. They studied the structures and properties of 2-D skew-cyclic codes. Also, they built relationships between 2-D skew-cyclic codes and other known codes. Throughout this paper, let R be a commutative ring. For two given automorphisms ρ and θ of R, we consider the set of formal bivariate polynomials R[x, y; ρ, θ] =n ai,jxiyjai,j ∈ R and k, t ∈ N0o t k Xj=0 Xi=0 which forms a ring under the usual addition of polynomials and where the multiplication is defined using the rule axiyj ⋆ bxrys = aρiθj(b)xi+ryj+s, and extended to all elements of R[x, y; ρ, θ] by associativity and dis- tributivity. The ring R[x, y; ρ, θ] is called a bivariate skew polynomial ring over R and an element in R[x, y; ρ, θ] is called a bivariate skew polynomial. It is easy to see that R[x, y; ρ, θ] is a non-commutative ring unless ρ and θ are indentity automorphisms on R. For a bivariate skew polynomial f (x, y) in R[x, y; ρ, θ], let hf (x, y)il denote the left ideal of R[x, y; ρ, θ] generated by f (x, y). Note that hf (x, y)il does not need to be two-sided. In section 2, we give some characterizations of the ring R[x, y; ρ, θ]. In section 3, we introduce and investigate 2-D skew (λ1, λ2)-constacyclic codes in R[x, y; ρ, θ]/hxl − λ1, ys − λ2il . Also, the dual of 2-D skew (λ1, λ2)-constacyclic codes is investigated. Let N2 0 = N0 ×N0. Then N2 0 is a partial ordered set with (i, j) ≥ (k, l) if and only if i ≥ k and j ≥ l. Moreover, N2 0 can be also totally ordered by a kind of lexicographic order "⇒", where (i, j) ⇒ (k, l) if and only if j > l or both j = l and i ≥ k. Otherwise, (i, j) → (k, l) means j > l or both j = l and i > k. Notice that (i, j) ≥ (k, l) implies that (i, j) ⇒ (k, l), but the converse does not necessarily hold. A nonzero bivariate polynomial f (x, y) ∈ R[x, y; ρ, θ] is said to have quasi-degree deg(f (x, y)) = (k, t) if f (x, y) has a nonzero term ak,txkyt but does not have any nonzero term ai,jxiyj such that (i, j) → (k, t) holds. In this case, ak,t is called the leading coefficient of f (x, y). A bivariate skew polynomial is called monic provided its leading coefficient is 1. Let Fq be the Galois field with q elements. For any polynomials f (x, y) and g(x, y) in Fq[x, y; ρ, θ] we have deg(f (x, y) ⋆ g(x, y)) = deg(f (x, y)) + deg(g(x, y)). 2-D SKEW CONSTACYCLIC CODES 3 It is straightforward to see that Fq[x, y; ρ, θ] has no nonzero zero-divisors. 2. Basic properties of R[x, y; ρ, θ] In this paper we denote by Rρ,θ (resp. Rρθ) the subring of R that is fixed by ρ, θ (resp. ρθ). Let R be a commutative ring and f (x) ∈ R[x]. In [16], McCoy observed that if 0 6= g(x) ∈ R[x] be such that f (x)g(x) = 0, then there exists a nonzero element r of R such that f (x)r = 0. Now, we state the bivariate skew version of the McCoy condition. Theorem 2.1. Let f (x, y) ∈ Rρ,θ[x, y; ρ, θ]. If f (x, y) ⋆ g(x, y) = 0 for some 0 6= g(x, y) ∈ R[x, y; ρ, θ], then there exists 0 6= r ∈ R such that f (x, y) ⋆ r = 0. fi,jxiyj. We can assume that Proof. Suppose that f (x, y) = P(i,j)∈Λ g(x, y) is of the minimal degree deg(g(x, y)) = (u, v)(cid:0) → (0, 0)(cid:1) (with respect to "⇒"), and g(x, y) has the leading coefficient gu,v. By the minimality of deg(g(x, y)) we have f (x, y) ⋆ gu,v 6= 0. If fi,jg(x, y) = 0 for every (i, j) ∈ Λ, then fi,jgu,v = 0 for every (i, j) ∈ Λ. Now, since f (x, y) ∈ Rρ,θ[x, y; ρ, θ], then f (x, y) ⋆ gu,v = 0 which is a con- tradiction. Therefore fi,jg(x, y) 6= 0 for some (i, j) ∈ Λ and so as- sume that (k, t) is the largest pair with this property (with respect to "⇒"). Hence 0 = f (x, y) ⋆ g(x, y) = ( fi,jxiyj) ⋆ g(x, y) which t k Pj=0 Pi=0 implies that ρiθj(fk,tgu,v) = fk,tρiθj(gu,v) = 0. Then fk,tgu,v = 0. Thus deg(g(x, y)) → deg(fk,tg(x, y)). On the other hand f (x, y) ⋆ (fk,tg(x, y)) = f (x, y) ⋆ (g(x, y) ⋆ fk,t) = 0 which contradics the mini- mality of deg(g(x, y)). (cid:3) The center of a ring S, denoted by Z(S), is the subset of S consisting of all those elements in S that commute with every element in S. Proposition 2.2. Let g(x, y) ∈ Z(R[x, y; ρ, θ]). Then hxkyt ⋆ g(x, y)il is a two-sided ideal in R[x, y; ρ, θ] for every k, t ∈ N0. 4 MOSTAFANASAB Proof. Let g(x, y) ∈ Z(R[x, y; ρ, θ]). Assume that f (x, y) = ∈ R[x, y; ρ, θ]. Then ai,jxiyj n Pj=0 m Pi=0 n m (cid:0)xkyt ⋆ g(x, y)(cid:1) ⋆ f (x, y) = (cid:0)xkyt ⋆ g(x, y)(cid:1) ⋆ Xi=0 Xj=0 m n ai,jxiyj Xi=0 Xj=0 ai,jxiyj(cid:1)(cid:17) ⋆ g(x, y) m (since g(x, y) is central) = (cid:16)xkyt ⋆(cid:0) Xi=0 Xj=0 ρkθt(ai,j)xk+iyt+j(cid:1) ⋆ g(x, y) = (cid:0) = (cid:16)(cid:0) ρkθt(ai,j)xiyj(cid:1) ⋆ xkyt(cid:17) ⋆ g(x, y) Xi=0 Xj=0 Xj=0 Xi=0 ρkθt(ai,j)xiyj(cid:1) ⋆(cid:0)xkyt ⋆ g(x, y)(cid:1), = (cid:0) m m n n n which belongs to hxkytg(x, y)il. Hence, the result follows. (cid:3) Proposition 2.3. Let hρi l and hθi s. Then Rρ,θ[xl, ys; ρ, θ] ⊆ Z(R[x, y; ρ, θ]). Proof. Let a ∈ Rρ,θ, b ∈ R and i, j, k, t ∈ N0. Therefore (axliysj) ⋆ (bxkyt) = aρliθsj(b)xli+kysj+t = baxk+liyt+sj = (bxkyt) ⋆(cid:0)θ−tρ−k(a)xliysj(cid:1) = (bxkyt) ⋆ (axliysj) (since a ∈ Rρ−1,θ−1). Consequently Rρ,θ[xl, ys; ρ, θ] ⊆ Z(R[x, y; ρ, θ]). (cid:3) Proposition 2.4. Let λ be a unit in R. The following conditions hold: (1) If hxl − λil is a two-sided ideal of R[x, y; ρ, θ], then hρi l and ρ(λ) = λ. (2) If hys − λil is a two-sided ideal of R[x, y; ρ, θ], then hθi s and θ(λ) = λ. Proof. The proof is similar to that of [15, Proposition 2.2]. (cid:3) Proposition 2.5. Let λ1, λ2 be units in R. The following conditions hold: 2-D SKEW CONSTACYCLIC CODES 5 (1) If hθi s and hxl−λ1, ys−λ2il is a two-sided ideal of R[x, y; ρ, θ], then hρi l and ρ(λ1) = λ1. (2) If hρi l and hxl−λ1, ys−λ2il is a two-sided ideal of R[x, y; ρ, θ], then hθi s and θ(λ2) = λ2. Proof. Suppose that hθi s and hxl − λ1, ys − λ2il is a two-sided ideal. Let r ∈ R. Then rxl − rλ1 = r(xl − λ1) = (xl − λ1)r′ for some r′ ∈ R. The remainder is similar to the proof of [15, Proposition 2.2]. (cid:3) Theorem 2.6. Let λ ∈ R. The following conditions hold: (1) If hρi l and λ ∈ Rρ,θ, then xl − λ is central in R[x, y; ρ, θ]. (2) If hθi s and λ ∈ Rρ,θ, then ys − λ is central in R[x, y; ρ, θ]. (3) If hρi l, hθi s and λ1, λ2 ∈ Rρ,θ, then hxl − λ1, ys − λ2il is a two-sided ideal of R[x, y; ρ, θ]. Proof. (1) Suppose that hρi l, ρ(λ) = λ and θ(λ) = λ. Let a ∈ R and k, t ∈ N0. Thus (axkyt) ⋆ (xl − λ) = axk+lyt − axkytλ = axl+kyt − ρkθt(λ)axkyt = (cid:0)xl ⋆ ρ−l(a)xkyt(cid:1) − λaxkyt = (cid:0)xl ⋆ axkyt(cid:1) − λaxkyt = (xl − λ) ⋆ (axkyt). Consequently xl − λ ∈ Z(R[x, y; ρ, θ]). (2) The proof if similar to that of (1). (3) Is a direct consequence of parts (1) and (2). (cid:3) Corollary 2.7. Let R be a ring. The following conditions hold: (1) Let θ(λ) = λ. Then hρi l and ρ(λ) = λ if and only if xl − λ is central in R[x, y; ρ, θ]. (2) Let ρ(λ) = λ. Then hθi s and θ(λ) = λ if and only if ys − λ is central in R[x, y; ρ, θ]. Proof. By Proposition 2.4 and Theorem 2.6. (cid:3) Proposition 2.8. Let f ⋆g be a monic central bivariat skew polynomial in R[x, y; ρ, θ] for some f, g ∈ R[x, y; ρ, θ]. Then f ⋆ g = g ⋆ f . Proof. Since f ⋆ g is central, then we have (g ⋆ f ) ⋆ g = g ⋆ (f ⋆ g) = (f ⋆ g) ⋆ g. Therefore (g ⋆ f − f ⋆ g) ⋆ g = 0. On the other hand f ⋆ g is monic, so the leading coefficient of g is unit. Hence g is not a zero-divisor. Consequently f ⋆ g = g ⋆ f in R[x, y; ρ, θ]. (cid:3) Corollary 2.9. Let hρi l, hθi s and λ1, λ2 ∈ Rρ,θ. If (xl − λ1) ⋆ (ys − λ2) = f ⋆ g for some f, g ∈ R[x, y; ρ, θ], then f ⋆ g = g ⋆ f . 6 MOSTAFANASAB Proof. By Theorem 2.6 and Proposition 2.8. (cid:3) 3. 2-D skew (λ1, λ2)-constacyclic codes Definition 3.1. Let C be a linear code over R of length ls whose codewords are viewed as l × s arrays, i.e., c ∈ C is written as c0,0 c1,0 ... cl−1,0 c =  c0,1 c1,1 ... cl−1,1 . . . . . . . . . . . . c0,s−1 c1,s−1 ... cl−1,s−1 .   We say that C is a column skew λ1-constacyclic code if for every l × s array c = (ci,j) ∈ C we have that λ1ρ(cl−1,0) λ1ρ(cl−1,1) ρ(c0,0) ... ρ(c0,1) ... ρ(cl−2,0) ρ(cl−2,1)   ρ(c0,s−1) . . . λ1ρ(cl−1,s−1) . . . . . . . . . ρ(cl−2,s−1) ... ∈ C.   Also, we say that C is a row skew λ2-constacyclic code if for every l × s array c = (ci,j) ∈ C we have that λ2θ(c0,s−1) λ2θ(c1,s−1) ... θ(c0,0) θ(c1,0) ... λ2θ(cl−1,s−1) θ(cl−1,0) . . . . . . . . . . . . θ(c0,s−2) θ(c1,s−2) ... θ(cl−1,s−2)   ∈ C.   If C is both column skew λ1-constacyclic and row skew λ2-constacyclic, then we call C a 2-D skew (λ1, λ2)-constacyclic code. Define R◦ := R[x, y; ρ, θ]/hxl − λ1, ys − λ2il . Consider the R-module isomorphism Rl×s → R◦ defined by (ai,j) 7→ ai,jxiyj where Rl×s s−1 Pj=0 l−1 Pi=0 denotes the set of all l×s arrays. Then a codeword c ∈ C can be denoted by bivariate skew polynomial c(x, y) under above isomorphism. Theorem 3.2. A code C in R◦ is a 2-D skew (λ1, λ2)-constacyclic code if and only if C is a left R[x, y; ρ, θ]-submodule of the left R[x, y; ρ, θ]- module R◦. Proof. Similar to the proof of [25, Theorem 3.2]. (cid:3) As a direct consequence of Theorem 2.6(3) and Theorem 3.2 we have the next result. 2-D SKEW CONSTACYCLIC CODES 7 Corollary 3.3. Let hρi l, hθi s and λ1, λ2 ∈ Rρ,θ. A code C in R◦ is a 2-D skew (λ1, λ2)-constacyclic code if and only if C is a left ideal of R◦. Theorem 3.4. Let f1(x, y), f2(x, y) ∈ R[x, y; ρ, θ] be two nonzero bi- variate polynomials where f2(x, y) is monic. Provided that deg(f1(x, y)) ≥ deg(f2(x, y)), there exists a pair of polynomials h(x, y)(6= 0), g(x, y) ∈ R[x, y; ρ, θ] which satisfy f1(x, y) = h(x, y) ⋆ f2(x, y) + g(x, y) such that g(x, y) = 0 or deg(f2(x, y)) (cid:2) deg(g(x, y))(cid:0) ← deg(f1(x, y))(cid:1). Proof. The proof is similar to that of [25, Theorem 2.1]. Lemma 3.5. Let C be a 2-D skew (λ1, λ2)-constacyclic code in R◦ and g(x, y) be a monic polynomial in R[x, y; ρ, θ]. Then g(x, y) is of the minimum degree (with respect to "≤") in C if and only if C = hg(x, y)il. (cid:3) Proof. The "if" part is obvious. Let g(x, y) be a monic polynomial of the minimum degree (with respect to "≤") in C. If c(x, y) ∈ C, then by the Division Algorithm in R[x, y; ρ, θ], there exists a pair of polynomials h(x, y)(6= 0), r(x, y) ∈ R[x, y; ρ, θ] which satisfy c(x, y) = h(x, y) ⋆ g(x, y) + r(x, y), where either r(x, y) = 0 or deg(g(x, y)) (cid:2) deg(r(x, y))(cid:0) ←− deg(c(x, y))(cid:1). As C is an R[x, y; ρ, θ]-module, r(x, y) ∈ C and the minimality of the degree of g(x, y) implies r(x, y) = 0. Theorem 3.6. Let C be a 2-D skew (λ1, λ2)-constacyclic code in F◦ q and g(x, y) be a monic generator polynomial of C in Fq[x, y; ρ, θ]. Then the following conditions hold: (cid:3) (1) g(x, y) divides (xl − λ1) ⋆ (ys − λ2). (2) Suppose that deg(g(x, y)) = (l − k, s − t). Then B =  g(x, y), x ⋆ g(x, y), ... y ⋆ g(x, y), xy ⋆ g(x, y), ... xk−1 ⋆ g(x, y), xk−1y ⋆ g(x, y), . . . . . . . . . . . . is a basis for C over Fq and so C = qkt. , yt−1 ⋆ g(x, y) , xyt−1 ⋆ g(x, y) ... , xk−1yt−1 ⋆ g(x, y)   Proof. (1) Assume that g(x, y) does not divide (xl − λ1) ⋆ (ys − λ2). By the Division Algorithm, there exist f (x, y)(6= 0), r(x, y) ∈ Fq[x, y; ρ, θ] which satisfy (xl − λ1) ⋆ (ys − λ2) = f (x, y) ⋆ g(x, y) + r(x, y) such that either r(x, y) = 0 or deg(g(x, y)) (cid:2) deg(r(x, y)). If deg(g(x, y)) (cid:2) deg(r(x, y)), then r(x, y) = (xl − λ1) ⋆ (ys − λ2) − f (x, y) ⋆ g(x, y) ≡ −f (x, y) ⋆ g(x, y) (cid:0)mod hxl − λ1, ys − λ2il(cid:1) 8 MOSTAFANASAB is also in C which contradicts the minimality of deg(g(x, y)) with re- spect to "≤". Consequently r(x, y) = 0 and so we are done. (2) Let 0 6= f (x, y) ∈ C = hg(x, y)il. Then there exists q(x, y) ∈ Fq[x, y; ρ, θ] such that f (x, y) = q(x, y)⋆g(x, y). Notice that deg(f (x, y)) ≤ (l − 1, s − 1). Hence deg(q(x, y)) = deg(f (x, y)) − deg(g(x, y)) ≤ (l − 1, s − 1) − (l − k, s − t) = (k − 1, t − 1). Therefore, B is a generating set for C. Assume that {ai,j0 ≤ i ≤ k − 1, 0 ≤ j ≤ t − 1} be a subset of R such that Given r(x, y) := ai,jxiyj ⋆ g(x, y) = 0. k−1 Xi=0 ai,jxiyj such that r(x, y) 6= 0, we have that t−1 Xj=0 Pi=0 k−1 t−1 Pj=0 deg(r(x, y)) ≤ (k − 1, t − 1) and r(x, y) ⋆ g(x, y) = 0. Therefore, there exist q1(x, y) and q2(x, y) in Fq[x, y; ρ, θ] such that r(x, y) ⋆ g(x, y) = q1(x, y) ⋆ (xl − λ1) + q2(x, y) ⋆ (ys − λ2). Hence deg(cid:0)q1(x, y) ⋆ (xl − λ1) + q2(x, y) ⋆ (ys − λ2)(cid:1) = deg(r(x, y)) + deg(g(x, y)) ≤ (l − 1, s − 1), which is a contradiction. Consequently r(x, y) = 0 that implies ai,j = 0 for all 0 ≤ i ≤ k − 1 and 0 ≤ j ≤ t − 1. (cid:3) Proposition 3.7. Let R be a domain and λ1, λ2 be units in R. Suppose that C is a 2-D skew (λ1, λ2)-constacyclic code in R◦ that is generated by s−t−1 l−k−1 s−t l−k−1 g(x, y) = gi,jxiyj + xl−kyj + xiys−t. Xj=0 Xi=0 Xj=0 Xi=0 Then g(x, y) ⋆ xy ∈ C if and only if g(x, y) ∈ Rρθ[x, y; ρ, θ]. Proof. Clearly xy ⋆g(x, y) ∈ C. Assume that g(x, y)⋆xy ∈ C. Therefore s−t−1 l−k−1 xy ⋆ g(x, y) − g(x, y)xy = (ρθ(gi,j) − gi,j)xi+1yj+1 Xj=0 Xi=0 = p(x, y) ⋆ g(x, y), for some p(x, y) ∈ R[x, y; ρ, θ]. It is easy to see that p(x, y) is constant and p(x, y)g0,0 = 0. Since g(x, y) is a right divisor of (xl −λ1) ⋆ (ys −λ2) 2-D SKEW CONSTACYCLIC CODES 9 and λ1, λ2 are units, then g0,0 is a unit. Hence p(x, y) = 0 and so ρθ(gi,j) = gi,j for all 0 ≤ i ≤ l − k − 1 and 0 ≤ j ≤ s − t − 1. Therefore g(x, y) ∈ Rρθ[x, y; ρ, θ]. For the converse, assume that g(x, y) ∈ Rρθ[x, y; ρ, θ]. Thus xy ⋆ gi,j = ρθ(gi,j)xy = gi,jxy for all 0 ≤ i ≤ l − k − 1 and 0 ≤ j ≤ s − t − 1. Hence g(x, y)xy = xy ⋆ g(x, y) ∈ C, so we are done. (cid:3) Definition 3.8. For two l × s arrays c = (ci,j) and d = (di,j) we define: s−1 l−1 c ⊙ d := ci,jdi,j. Xj=0 Xi=0 We say that c, d are orthogonal if c ⊙ d = 0. The dual code of a linear code C of length ls is the set of all l × s arrays orthogonal to all codewords of C. The dual code of C is denoted by C ⊥. 1 , λ−1 2 )-constacyclic. In particular, if λ2 Theorem 3.9. Let hρi l, hθi s and λ1, λ2 be units in R such that λ1 ∈ Rρ, λ2 ∈ Rθ. Suppose that C is a code of lenght ls over R. Then C is 2-D skew (λ1, λ2)-constacyclic if and only if C ⊥ is 2-D skew (λ−1 2 = 1, then C is 2-D skew (λ1, λ2)-constacyclic if and only if C ⊥ is 2-D skew (λ1, λ2)- constacyclic. Proof. First of all, λ1 ∈ Rρ, λ2 ∈ Rθ imply that λ1 ∈ Rρ−1, λ2 ∈ Rθ−1. Let c = (ci,j) ∈ C and b = (bi,j) ∈ C ⊥ be two l × s arrays. Since C is a column λ1-constacyclic code, then 1 = λ2 ρl−1(λ1c1,0) ρl−1(λ1c2,0) ... ρl−1(λ1c1,1) ρl−1(λ1c2,1) ... ρl−1(λ1cl−1,0) ρl−1(λ1cl−1,1) ρl−1(c0,0) ρl−1(c0,1) ρl−1(λ1c1,s−1) ρl−1(λ1c2,s−1) . . . . . . . . . . . . ρl−1(λ1cl−1,s−1) . . . ρl−1(c0,s−1) ...   is in C and so it is orthogonal to b = (bi,j), i.e.,   l−1 s−1 0 = Xj=0 Xi=1 = λ1 l−1 Xi=1 Therefore s−1 Xj=0 s−1 Xj=0 Xj=0 s−1 ρl−1(λ1ci,j)bi−1,j + ρl−1(c0,j)bl−1,j ρl−1(ci,j)bi−1,j + ρl−1(c0,j)λ−1 1 bl−1,j! l−1 Xi=1 s−1 Xj=0 ρl−1(ci,j)bi−1,j + s−1 Xj=0 ρl−1(c0,j)λ−1 1 bl−1,j = 0. 10 Hence MOSTAFANASAB 0 = ρ(0) = l−1 Xi=1 s−1 Xj=0 ci,jρ(bi−1,j) + s−1 Xj=0 c0,jλ−1 1 ρ(bl−1,j), which shows that λ−1 1 ρ(bl−1,0) λ−1 ρ(b0,0) 1 ρ(bl−1,1) ρ(b0,1) ... ... 1 ρ(bl−1,s−1) ρ(b0,s−1) . . . λ−1 . . . . . . . . . ...   ρ(bl−2,0) ρ(bl−2,1) ρ(bl−2,s−1) is orthogonal to c = (ci,j) and so it belongs to C ⊥. Thus C ⊥ is column λ−1 1 -constacyclic. On the other hand, since C is a row λ2-constacyclic code, then θs−1(λ2c0,1) θs−1(λ2c1,1) ... θs−1(λ2cl−1,1) . . . . . . . . . . . . θs−1(λ2c0,s−1) θs−1(λ2c1,s−1) ... θs−1(c0,0) θs−1(c1,0) ... θs−1(λ2cl−1,s−1) θs−1(cl−1,0)   is in C and so it is orthogonal to b = (bi,j), i.e.,     θs−1(λ2ci,j)bi,j−1 + θs−1(ci,0)bi,s−1 θs−1(ci,j)bi,j−1 + θs−1(ci,0)λ−1 2 bi,s−1! l−1 Xi=0 Xi=0 l−1 s−1 l−1 0 = Xi=0 Xj=1 = λ2 s−1 Xj=1 s−1 l−1 Thus l−1 Xi=0 Xj=1 So we have θs−1(ci,j)bi,j−1 + Xi=0 θs−1(ci,0)λ−1 2 bi,s−1 = 0. l−1 Xi=0 0 = θ(0) = s−1 Xj=1 l−1 Xi=0 ci,jθ(bi,j−1) + l−1 Xi=0 ci,0λ−1 2 θ(bi,s−1), which shows that λ−1 2 θ(b0,s−1) λ−1 2 θ(b1,s−1) ... θ(b0,0) θ(b1,0) ... λ−1 2 θ(bl−1,s−1) θ(bl−1,0)   . . . . . . . . . . . . θ(b0,s−2) θ(b1,s−2) ... θ(bl−1,s−2)   2-D SKEW CONSTACYCLIC CODES 11 is orthogonal to c = (ci,j) and so it belongs to C ⊥. Thus, it follows that C ⊥ is row λ−1 2 )- constacyclic. The converse holds by the fact that (C ⊥)⊥ = C. 2 -constacyclic. Consequently C ⊥ is 2-D skew (λ−1 1 , λ−1 (cid:3) The ring R[x, y; ρ, θ] can be localized to the right at the multiplica- tive set S = {xiyj i, j ∈ N}. The existance of the localization R[x, y; ρ, θ]S −1 follows from [19, Theorem 2] since S verifies the fol- lowing two necessary and sufficient conditions: (1) For all xi1yj1 ∈ S and f (x, y) ∈ R[x, y; ρ, θ], there exists xi2yj2 ∈ S and g(x, y) ∈ R[x, y; ρ, θ] such that f (x, y)xi1yj1 = xi2yj2 ⋆ g(x, y). To prove this note to the multiplication rule xiyj ⋆ a = ρiθj(a)xiyj. (2) If for xi1yj1 ∈ S and f (x, y) ∈ R[x, y; ρ, θ] we have xi1yj1 ⋆ f (x, y) = 0, then there exists xi2yj2 ∈ S such that f (x, y)xi2yj2 = 0. But since xi2yj2 is never a zero divisor, f (x, y) must be zero. Now, we consider the ring R[x, y; ρ, θ]S −1 consisting of the elements t k x−iy−jai,j, where the coefficients are on the right and where the Pj=0 Pi=0 multiplication rule is given by ax−1y−1 = x−1y−1ρθ(a). Proposition 3.10. Let ψ : R[x, y; ρ, θ] → R[x, y; ρ, θ]S −1 be defined by ψ( t k Xj=0 Xi=0 ai,jxiyj) = t k Xj=0 Xi=0 x−iy−jai,j. Then ψ is a ring anti-isomorphism. Proof. Similar to [4, Theorem 4.4]. (cid:3) From now on, we assume that λ1, λ2 ∈ Rρ,θ, hρi l and hθi s. Theorem 3.11. Let a(x, y) = be in R[x, y; ρ, θ]. Suppose that λ2 ditions are equivalent: s−1 Pj=0 l−1 Pi=0 ai,jxiyj and b(x, y) = bi,jxiyj 1 = λ2 2 = 1. Then the following con- s−1 Pj=0 l−1 Pi=0 (1) The coefficient matrix of a(x, y) is orthogonal to the coefficient matrix of xiyj(cid:0)xl−1ys−1 ⋆ ψ(b(x, y))(cid:1) for all i ∈ {0, 1, . . . , l − 1} and all j ∈ {0, 1, . . . , s − 1}; 12 MOSTAFANASAB (2) The coefficient matrix of a(x, y) is orthogonal to bl−1,s−1 ρ(bl−2,s−1) ... θ(bl−1,s−2) ρθ(bl−2,s−2) ... ρl−1(b0,s−1) ρl−1θ(b0,s−2) A =  θs−1(bl−1,0) ρθs−1(bl−2,0) . . . . . . . . . . . . ρl−1θs−1(b0,0) ...   and all of its column skew λ1-constacyclic shifts and row skew λ2-constacyclic shifts; (3) a(x, y) ⋆ b(x, y) = 0 in R◦. Proof. (1)⇔(2) With a routine computation we can deduce that the coefficient matrix of xl−1ys−1 ⋆ ψ(b(x, y)) is equal to A, also the coef- ficient matrices of xiyj(cid:0)xl−1ys−1 ⋆ ψ(b(x, y))(cid:1) for 0 ≤ i ≤ l − 1 and 0 ≤ j ≤ s − 1 are precisely column skew λ1-constacyclic shifts and row skew λ2-constacyclic shifts of A. (2)⇔(3) Suppose that a(x, y) ⋆ b(x, y) = ci,jxiyj ∈ R◦. For s−1 Pt=0 l−1 Pk=0 k ∈ {0, 1, . . . , l − 1} and t ∈ {0, 1, . . . , s − 1} we have that ck,t = X j+v=t X i+u=k ai,jρiθj(buv) + X j+v=t X i+u=k+l 0≤j,v≤s−1 0≤i,u≤l−1 0≤j,v≤s−1 0≤i,u≤l−1 ai,j ρiθj(λ1buv) + X j+v=t+s X i+u=k 0≤j,v≤s−1 0≤i,u≤l−1 ai,jρiθj(λ2buv) + X j+v=t+s X i+u=k+l 0≤j,v≤s−1 0≤i,u≤l−1 ai,jρiθj (λ1λ2buv) = λ1λ2(cid:0) X j+v=t X i+u=k 0≤j,v≤s−1 0≤i,u≤l−1 ai,j ρk−uθt−v(λ1λ2buv) + X j+v=t X i+u=k+l 0≤j,v≤s−1 0≤i,u≤l−1 + X j+v=t+s X i+u=k 0≤j,v≤s−1 0≤i,u≤l−1 + X j+v=t+s X i+u=k+l 0≤j,v≤s−1 0≤i,u≤l−1 ai,jρk+l−uθt−v(λ2buv) ai,jρk−uθt+s−v(λ1buv) ai,jρk+l−uθt+s−v(buv)(cid:1) which is the ⊙-product of a0,0 a1,0 ... al−1,0 λ1λ2  . . . . . . . . . . . . a0,s−1 a1,s−1 ... al−1,s−1   in the matrix 2-D SKEW CONSTACYCLIC CODES 13 λ1λ2bk,t θ(λ1λ2bk,t−1) . . . ρ(λ1λ2bk−1,t) ρθ(λ1λ2bk−1,t−1) . . . θt(λ1λ2bk,0) θt+1(λ1bk,s−1) . . . θs−1(λ1bk,t+1) ρθt(λ1λ2bk−1,0) ρθt+1(λ1bk−1,s−1) . . . ρθs−1(λ1bk−1,t+1) ... ... ... ... ... ρk(λ1λ2b0,t) ρk+1(λ2bl−1,t) ρkθ(λ1λ2b0,t−1) . . . ρk+1θ(λ2bl−1,t−1) . . . ρkθt(λ1λ2b0,0) ρk+1θt(λ2bl−1,0) ρkθt+1(λ1b0,s−1) . . . ρk+1θt+1(bl−1,s−1) . . . ρkθs−1(λ1b0,t+1) ρk+1θs−1(bl−1,t+1) ... ... ... ... ... ρl−1(λ2bk+1,t) ρl−1θ(λ2bk+1,t−1) . . . ρl−1θt(λ2bk+1,0) ρl−1θt+1(bk+1,s−1) . . . ρl−1θs−1(bk+1,t+1)   Therefore a(x, y) ⋆ b(x, y) = 0 in R◦ if and only if ck,t = 0 for all k ∈ {0, 1, . . . , l − 1} and all t ∈ {0, 1, . . . , s − 1} which is true if and only if .   a0,0 a1,0 ... al−1,0   a0,s−1 a1,s−1 . . . . . . . . . . . . al−1,s−1 ...   is orthogonal to A and all of its column skew λ1-constacyclic shifts and row skew λ2-constacyclic shifts. (cid:3) Proposition 3.12. Let λ2 2 = 1, g(x, y) be a right divisor of (xl − λ1) ⋆ (ys − λ2) in R[x, y; ρ, θ] and h(x, y) := (xl−λ1)⋆(ys−λ2) with deg(h(x, y)) = (k, t). Suppose that C is a 2-D skew (λ1, λ2)-constacyclic code of length ls over R that is generated by g(x, y). Then the bivariate skew polynomial xkyt ⋆ψ(h(x, y)) is a right divisor of (xl −λ1)⋆(ys −λ2) and xkyt ⋆ ψ(h(x, y)) ∈ C ⊥. 1 = λ2 g(x,y) Proof. By the assumtions that λ1, λ2 ∈ Rρ,θ, hρi l and hθi s we can see that (cid:16)ψ(g(x, y)) ⋆ λ1λ2xl−kys−t(cid:17) ⋆ (cid:16)xkyt ⋆ ψ(h(x, y))(cid:17) = ψ(g(x, y)) ⋆ λ1λ2xlys ⋆ ψ(h(x, y)) = λ1λ2xlys ⋆ ψ(g(x, y)) ⋆ ψ(h(x, y)) = λ1λ2xlys ⋆ ψ(h(x, y) ⋆ g(x, y)) (since ψ is a ring anti-isomorphism) = λ1λ2xlys ⋆ ψ(cid:16)(xl − λ1) ⋆ (ys − λ2)(cid:17) = λ1λ2xlys ⋆ (x−ly−s − x−lλ2 − y−sλ1 + λ1λ2) = (xl − λ1) ⋆ (ys − λ2). Moreover, since g(x, y) ⋆ h(x, y) = 0 in R◦, then Theorem 3.11 implies that xkyt ⋆ ψ(h(x, y)) ∈ C ⊥. (cid:3) Now we can state the following open problem. Problem 3.13. In Proposition 3.12, is C ⊥ generated by xkyt⋆ψ(h(x, y))? 14 MOSTAFANASAB Theorem 3.14. Let g(x, y) be a monic right divisor of (xl − λ1) ⋆ (ys − λ2) in R[x, y; ρ, θ] and h(x, y) := (xl−λ1)⋆(ys −λ2) . Suppose that C is a 2- D skew (λ1, λ2)-constacyclic code of length ls over R that is generated by g(x, y). Then for f (x, y) ∈ R[x, y; ρ, θ], f (x, y) ∈ C if and only if f (x, y) ⋆ h(x, y) = 0 in R⋄ = R[x, y; ρ, θ]/h(xl − λ1) ⋆ (ys − λ2)il . g(x,y) Proof. Let f (x, y) ∈ C. Then f (x, y) = q(x, y) ⋆ g(x, y) for some q(x, y) ∈ R[x, y; ρ, θ]. So we have f (x, y) ⋆ h(x, y) = q(x, y) ⋆ g(x, y) ⋆ h(x, y) = q(x, y) ⋆ (xl − λ1) ⋆ (ys − λ2) = 0, in R⋄. Conversely, assume that f (x, y) ⋆ h(x, y) = 0 in R⋄. Then f (x, y) ⋆ h(x, y) = q(x, y) ⋆ (xl − λ1) ⋆ (ys − λ2) for some q(x, y) ∈ R[x, y; ρ, θ]. Therefore f (x, y) ⋆ h(x, y) = q(x, y) ⋆ g(x, y) ⋆ h(x, y). Now, since h(x, y) is monic, f (x, y) = q(x, y) ⋆ g(x, y) ∈ C. (cid:3) Acknowledgments The author would like to thank the referee for constructive comments which will help to improve the quality of the paper. References 1. M. C. V. Amarra and F. R. Nemenzo, On (1 − u)-cyclic codes over Fpk + uFpk , App. Math. Letters, 21 (2008), 1129 -- 1133. 2. A. Bonnecaze and P. Udaya, Cyclic codes and self-dual codes over F2 + uF2, IEEE Trans. Inf. Theory, 45 (1999), 1250 -- 1255. 3. D. Boucher, W. Geiselmann and F. Ulmer, Skew-cyclic codes, Appl. Algebra Eng., Commun. Comput, 18 (4) (2007), 379 -- 389. 4. D. Boucher, P. Sole and F. Ulmer, Skew constacyclic codes over Galois rings, Adv. Math. Commun, 2 (3) (2008), 273 -- 292. 5. D. Boucher and F. Ulmer, Codes as modules over skew polynomial rings, Lect. Notes Comp. Sci, 5921 (2009), 38 -- 55. 6. D. Boucher and F. Ulmer, Coding with skew polynomial rings, J. Symb. Comput, 44 (2009), 1644 -- 1656. 7. H. Q. Dinh, Constacyclic codes of length 2s over Galois extension rings of F2 + uF2, IEEE Trans. Inf. Theory, 55 (2009), 1730 -- 1740. 8. H. Q. Dinh, Negacyclic codes of length 2s over Galois rings, IEEE Trans. Inf. Theory, 51 (2005), 4252 -- 4262. 9. H. Q. Dinh, Constacyclic codes of length ps over Fpm + uFpm, J. Algebra, 324 (2010), 940 -- 950. 10. H. Q. Dinh and S. R. L´opez-Permouth, Cyclic and negacyclic codes over finite chain rings, IEEE Trans. Inf. Theory, 50 (2004), 1728 -- 1744. 2-D SKEW CONSTACYCLIC CODES 15 11. C. Guneri and F. Ozbudak, A relation between quasi-cyclic codes and 2-D cyclic codes, Finite Fields Appl, 18 (2012), 123 -- 132. 12. A. Hammons, P. V. Kumar, A. R. Calderbank, N. J. A. Slone and P. Sole, The Z4 linearity of Kerdock, Preparata, Goethals and related codes, IEEE Trans. Inf. Theory, 40 (4) (1994), 301 -- 319. 13. T. Ikai, H. Kosako and Y. Kojima, Two-dimensional cyclic codes, Electron. Commun. Jpn, 57A (1975), 27 -- 35. 14. H. Imai, A theory of two-dimensional cyclic codes, Inf. Control, 34 (1977), 1 -- 21. 15. S. Jitman, S. Ling and P. Udomkavanich, Skew constacyclic codes over finite chain rings, Adv. Math. Commun, 6(1) (2012), 39 -- 63. 16. N. H. McCoy, Annihilators in polynomial rings, Amer. Math. Monthly, 64 (1957), 28 -- 29. 17. F. J. MacWilliams and N. J. A. Sloane, The theory of error correcting codes, North Holland, 1977. 18. J. F. Qian, L. N. Zhang and S. X. Zhu, (1 + u)-cyclic and cyclic codes over the ring F2 + uF2, Appl. Math. Letters, 19 (2006), 820 -- 823. 19. P. Ribenboim., Sur la localisation des anneaux non commutatifs, S´eminaire Dubreil. Alg´ebre et th´eorie des nombres, tome 24. (1970 -- 1971). 20. K. Saints, and C. Heegard, Algebraic-geometric codes and multidimensional cyclic codes: a unified theory and algorithms for decoding using Grobner bases, IEEE Trans. Inf. Theory IT-41, 5 (1993), 1733 -- 1751. 21. S. Sakata, Decoding binary 2-D cyclic codes by the 2-D Berlekamp-Massey al- gorithm, IEEE Trans. Inf. Theory IT-21, 37 (1991), 1200 -- 1203. 22. S. Sakata, General theory of doubly periodic arrays over an arbitrary finite field and its applications, IEEE Trans. Inf. Theory IT-24, 6 (1978), 719 -- 730. 23. S. Sakata, On determining the independent point set for doubly periodic arrays and encoding two-dimensional cyclic codes and their duals, IEEE Trans. Inf. Theory IT-21, 5 (1981), 556 -- 565. 24. P. Udaya and A. Bonnecaze, Decoding of cyclic codes over F2 + uF2, IEEE Trans. Inf. Theory, 45 (1999), 2148 -- 2157. 25. L. Xiuli and L. Hongyan, 2-D skew-cyclic codes over Fq[x, y; ρ, θ], Finite Fields Appl, 25 (2014), 49 -- 63. Hojjat Mostafanasab Eski Silahtaraga Elektrik Santrali, Kazim Karabekir, Istanbul Bilgi University, Cad. No: 2/1334060, Eyup Istanbul, Turkey. Email: [email protected]
1601.03436	1	1601	2016-01-13T22:50:34	On Semiprime Goldie Modules	[ "math.RA" ]	For an $R$-module $M$, projective in $\sigma[M]$ and satisfying ascending chain condition (ACC) on left annihilators, we introduce the concept of Goldie module. We also use the concept of semiprime module defined by Raggi et. al. in \cite{S} to give necessary and sufficient conditions for an $R$-module $M$, to be a semiprime Goldie module. This theorem is a generalization of Goldie's theorem for semiprime left Goldie rings. Moreover, we prove that $M$ is a semiprime (prime) Goldie module if and only if the ring $S=End_R(M)$ is a semiprime (prime) right Goldie ring. Also, we study the case when $M$ is a duo module.	math.RA	math	On Semiprime Goldie Modules Jaime Castro P´erez∗ Escuela de Ingenier´ıa y Ciencias, Instituto Tecnol´ologico y de Estudios Superiores de Monterrey Calle del Puente 222, Tlalpan, 14380, M´exico D.F., M´exico. Mauricio Medina B´arcenas† Jos´e R´ıos Montes‡ Angel Zald´ıvar§ Instituto de Matem´aticas, Universidad Nacional Aut´onoma de M´exico Area de la Investigaci´on Cient´ıfica, Circuito Exterior, C.U., 04510, M´exico D.F., M´exico. September 24, 2018 Abstract For an R-module M , projective in σ[M ] and satisfying ascending chain condition (ACC) on left annihilators, we introduce the concept of Goldie module. We also use the concept of semiprime module defined by Raggi et. al. in [16] to give necessary and sufficient conditions for an R-module M , to be a semiprime Goldie module. This theorem is a generalization of Goldie's theorem for semiprime left Goldie rings. Moreover, we prove that M is a semiprime (prime) Goldie module if and only if the ring S = EndR(M ) is a semiprime (prime) right Goldie ring. Also, we study the case when M is a duo module. Keywords: Prime module, Semiprime module, Goldie module, Essentially compressible module, Duo module. 2010 Mathematics Subject Classification: 16D50, 16D80, 16P50, 16P70. Introduction Goldie's Theorem states that a ring R has a semisimple artinian classical left quotient ring if and only if R is a semiprime ring with finite uniform dimension and satisfies ACC on left annihilators. Wisbauer proves in ([20], Theorem 11.6) ∗[email protected], corresponding author †[email protected] ‡[email protected] §[email protected] 1 a version of Goldie's Theorem in terms of modules. For a retractable R-module M with S = EndR(M ) the following conditions are equivalent: 1. M is non M -singular with finite uniform dimension and S is semiprime, 2. M is non M -singular with finite uniform dimension and for every N ≤e M there exists a monomorphism M → N , 3. EndR(cM ) is semisimple left artinian and it is the classical left quotient ring of S, here cM denotes the M -injective hull of M . Also, in [8] the authors study when the endomorphism ring of a semiprojective module is a semiprime Goldie ring. In this paper we give another generalization of Goldie's Theorem. For this, we use the product of submodules of a module M defined in [3] to say when a module is a semiprime module. This product extends the product of left ideals of a ring R, so R is a semiprime module (over itself) if and only if R is a semiprime ring in the usual sense. In order to have a definition of Goldie Module such that it extends the clas- sical definition of left Goldie ring, we introduce what ascending chain condition on left annihilators means on a module. A left annihilator in M is a submodule of the form AX = Tf ∈X Ker(f ) for some X ⊆ EndR(M ). This definition with R = M is the usual concept of left annihilator. The main concept of this work is that an R-module M is a Goldie module if M satisfies ACC on left annihilators and has finite uniform dimension. We prove some characterizations of semiprime Goldie modules (Theorem 2.8, Theorem 2.22 and Corollary 2.23) which generalize the Goldie's Theorem and extends the Theorem 11.6 of [20] and corollary 2.7 of [8]. We organize this paper in three sections. Section 1 proves several results for semiprime modules. We also generalize Theorem 10.24 of [10] to semiprime artinian modules. In section 2 we introduce the concept of Goldie modules. We prove the main Theorem of this paper and a characterization of semiprime Goldie modules. We also obtain some examples of Goldie modules. We also prove that if M has finitely many minimal prime submodules P1,...,Pt in M such that M/Pi (1 ≤ i ≤ t) has finite uniform dimension, then M is Goldie module if and only if each M/Pi is Goldie module for (1 ≤ i ≤ t). We also give a description of the submodule Z(N ) with N ∈ σ[M ]. In the last section we apply the previous results to duo modules which extend results for commutative rings. In [13] the authors say that they do not know a duo module with a quotient not duo, in this section we show an example. Throughout this paper R will be an associative ring with unit and R-Mod will denote the category of unitary left R-modules. A submodule N of an R- module M is denoted by N ≤ M . If N is a proper submodule we write N < M . We use N ≤e M for an essential submodule. Let M and X be R-modules. X is said to be M -generated if there exists an epimorphism from a direct sum of copies of M onto X. Every R-module X has a largest M -generated submodule called the trace of M in X, defined by trM (X) = P{f (M )f : M → X}. The category σ[M ] is defined as the smallest full subcategory of R-Mod containing all R-modules X which are isomorphic to a submodule of an M -generated module. 2 A module N ∈ σ[M ] is called singular in σ[M ] or M -singular, if there is an exact sequence in σ[M ], 0 → K → L → N → 0 with K ≤e L. The class S of all M -singular modules in σ[M ] is closed under submodules, quotients and direct sums. Therefore, any L ∈ σ[M ] has a largest M -singular submodule Z(L) = X{f (N )N ∈ S and f ∈ HomR(N, L)} L is called non M -singular if Z(L) = 0. Let M be an R-module. In [2] the annihilator in M of a class C of modules is defined as AnnM (C) = TK∈Ω K, where Ω = {K ≤ M there exists W ∈ C and f ∈ HomR(M, W) with K = Ker(f)} Also in [2], the author defines a product in the following way: Let N ≤ M . For each module X, N · X = AnnM (C) where C is the class of modules W such that f (N ) = 0 for all f ∈ HomR(M, W ). For an R-module M and K, L submodules of M , in [3] the product KM L is defined by KM L = P{f (K)f ∈ HomR(M, L)}. Moreover, in [2] it is showed that if M is projective in σ[M ], and N ≤ M , then N · X = NM X for every module X. A nonzero R-module M is called monoform if for each submodule N of M and each morphism f : N → M , f is either zero or a monomorphism. M has enough monoforms if each nonzero submodule of M contains a monoform submodule. Let M -tors be the frame of all hereditary torsion theories on σ[M ]. For a family {Mα} of modules in σ[M ], let χ({Mα}) the greatest element of M -tors for which all Mα are torsion free. Let ξ({Mα}) be the least element of M -tors for which all Mα are torsion. ξ({Mα}) and χ({Mα}) are called the hereditary torsion theory generated by the family {Mα} and the hereditary torsion theory cogenerated by the same family. In particular, the greatest and least elements in M -tors are denoted by χ and ξ respectively. If τ ∈ M − tors, let Tτ , Fτ and tτ denote the torsion class, the torsion free class and the preradical associated to τ , respectively. For details about concepts and terminology concerning torsion theories in σ[M ], see [19] and [20]. 1 Semiprime Modules Definition 1.1. Let M ∈ R − M od and K, L submodules of M . Put KM L = P{f (K)f ∈ HomR(M, L)}. For the properties of this product see [4] Propo- sition 1.3. Definition 1.2. Let M ∈ R − M od. We say a fully invariant submodule N ≤ M is a prime submodule in M if for any fully invariant submodules K, L ≤ M such that KM L ≤ N , then K ≤ N or L ≤ N . We say M is a prime module if 0 is a prime submodule. 3 Proposition 1.3. Let M be projective in σ[M ] and P a fully invariant sub- module of M . The following conditions are equivalent: 1. P is prime in M . 2. For any submodules K, L of M containing P and such that KM L ≤ P , then K = P or L = P . Proof. 1 ⇒ 2 : By Proposition 1.11 of [4]. 2 ⇒ 1 : Suppose that K, L are submodules of M such that KM L ≤ P . We claim that KM (L + P ) ≤ P . Since KM L ≤ L ∩ P , by Proposition 5.5 of [2] KM (L/L ∩ P ) = 0 so KM (L + P /P ) = 0. Thus KM (L + P ) ≤ P . On the other hand, (K + P )M (L + P ) = KM (L + P ) + PM (L + P ) ≤ P because P is fully invariant in M . Then, by hypothesis K + P = P or L + P = P , hence K ≤ P or L ≤ P . Definition 1.4. We say a fully invariant submodule N ≤ M is a semiprime sub- module in M if for any fully invariant submodule K ≤ M such that KM K ≤ N , then K ≤ N . We said M is a semiprime module if 0 is a semiprime submodule. Lemma 1.5. Let M be projective in σ[M ] and N a fully invariant submodule of M . The following conditions are equivalent: 1. N is semiprime in M . 2. For any submodule K of M , KM K ≤ N implies K ≤ M . 3. For any submodule K ≤ M containing N such that KM K ≤ N , then K = N . Proof. 1 ⇒ 2 : Let K ≤ M such that KM K ≤ N . Consider the submodule KM M of M . This is the minimal fully invariant submodule of M which contains K and KM X = (KM M )M X for every module X. Hence by Proposition 1.3 of [4] we have that KM K = (KM M )M K ≤ ((KM M )M K)M M ) ≤ NM M Since N is fully invariant submodule of M then NM M = N and by Proposition 5.5 of [2] (KM M )M (KM M ) = ((KM M )M K)M M ) ≤ N . Since N is semiprime in M , KM M ≤ N . Hence K ≤ N . 2 ⇒ 1 : By definition. 1 ⇔ 3 : Similar to the proof of Proposition 1.3. In Remark 1.26 below, we give an example where the associativity of the product (·)M (·) is not true in general. Definition 1.6. Let M ∈ R − M od and N a fully invariant submodule of M . We define the powers of N as: 4 1. N 0 = 0 2. N 1 = N 3. N m = NM N m−1 Lemma 1.7. Let M be projective in σ[M ] and N semiprime in M . Let J be a fully invariant submodule of M such that J n ≤ N then J ≤ N . Proof. By induction on n. If n = 1 the result is clear. Suppose n > 1 and the Proposition is valid for n−1. We have that 2n−2 ≥ n then so J 2n−2 ≤ N (J n−1)2 = J n−1 M J n−1 ≤ N since N is semiprime J n−1 ≤ N then J ≤ N . Proposition 1.8. Let S := EndR(M ) and assume M generates all its submod- ules. If N is a fully invariant submodule of M such that HomR(M, N ) is a prime (semiprime) ideal of S, then N is prime (semiprime) in M . Proof. Let K and L be fully invariant submodules of M such that KM L ≤ N . Put I = HomR(M, L) and J = HomR(M, K). Let m ∈ M and P figi ∈ IJ. Since gi ∈ J and gi(m) ∈ K then P fi(gi(m)) ∈ KM L ≤ N . Hence IJ ≤ HomR(M, N ). Since HomR(M, N ) is prime (semiprime) in S, then I ≤ HomR(M, N ) or J ≤ HomR(M, N ). Hence trM (L) := Hom(M, L)M ≤ N or trM (K) ≤ N and since M generates all its submodules then L ≤ N or K ≤ N . Thus N is a prime (semiprime) submodule. Next definition aper in [9] Definition 1.9. A module M is retractable if HomR(M, N ) 6= 0 for all 0 6= N ≤ M Corollary 1.10. Let S := EndR(M ) with M retractable. (semiprime) ring then M is prime (semiprime). If S is a prime Proof. Let K and L be fully invariant submodules of M such that KM L = 0. Since HomR(M, 0) is a prime (semiprime) ideal of S then by the proof of 1.8, trM (K) = 0 o trM (L) = 0. Since M is retractable, K = 0 or L = 0. Hence 0 is prime (semiprime) in M . Thus M is prime (semiprime). Proposition 1.11. Let M be projective in σ[M ] and N a proper fully invariant submodule of M . The following conditions are equivalent: 1. N is semiprime in M . 2. If m ∈ M is such that RmM Rm ≤ N , then m ∈ N . 3. N is an intersection of prime submodules. 5 Proof. 1 ⇒ 2 : By Lemma 1.5. 2 ⇒ 3 : Since N is proper in M , let 0 6= m0 ∈ M \ N . Then Rm0M Rm0 (cid:2) N . Now, let 0 6= m1 ∈ Rm0M Rm0 but m1 /∈ N Then Rm1M Rm1 (cid:2) N and Rm1M Rm1 ≤ Rm0M Rm0. We obtain a sequence of non-zero elements of M , {m0, m1, ...} such that mi /∈ N for all i and Rmi+1M Rmi+1 ≤ RmiM Rmi. By Zorn's Lemma there exists a fully invariant submodule P of M with N ≤ P , maximal with the property that mi /∈ P for all i . We claim P is a prime submodule. Let K and L submodules of M containing P . Since P ≤ K and P ≤ L, then there exists mi and mj such that mi ∈ K and mj ∈ L. Suppose i ≤ j, then RmiM Rmi ≤ K and by construction mj ∈ RmiM Rmi and thus mj ∈ K. If we put k = max{i, j}, then mk ∈ K and mk ∈ L. Hence, RmkM Rmk ≤ KM L, and so KM L (cid:2) P . By Proposition 1.3, P is prime in M . 3 ⇒ 1 : It is clear. Proposition 1.12. Let 0 6= M be a semiprime module and projective in σ[M ]. Then M has minimal prime submodules in M . Proof. By the proof Proposition of 1.11, M has prime submodules. Let P ≤ M be a prime submodule. Consider Γ = {Q ≤ P Q is prime}. This family is not empty because P ∈ Γ. Let C = {Qi} be a descending chain in Γ. Let N, K ≤ M be fully invariant submodules of M such that NM K ≤ T C. Suppose that N (cid:2) T C. Then there exists Qj such that N (cid:2) Qj and N (cid:2) Ql for all K ≤ T C. Therefore T C ∈ Γ. By Zorn's Lemma Γ has minimal elements. Ql ≤ Qj. Therefore K ≤ Ql for all Ql ≤ Qj, and since C is a chain then Remark 1.13. Notice that if M is projective in σ[M ] and M has prime sub- modules in M , then M has minimal prime submodules. Corollary 1.14. Let 0 6= M be a semiprime module and projective in σ[M ]. Then 0 = \{P ≤ M P is a minimal prime in M }. Proof. Let x ∈ T{P ≤ M P is a minimal prime in M } and Q ≤ M be a prime submodule in M . By Proposition 1.12 there exists a minimal prime submodule P such that P ≤ Q then x ∈ Q and x is in the intersection of all primes in M . By Proposition 1.11, x = 0. Lemma 1.15. Let M ∈ R − M od and N a minimal submodule of M . Then N 2 = 0 or N is a direct summand of M . Proof. Suppose that NM N 6= 0. Then there exists f : M → N such that f (N ) 6= 0. Since 0 6= f (M ) ≤ N and N is a minimal submodule, f (M ) = N . On the other hand, Ker(f ) ∩ N ≤ N , since f (N ) 6= 0 then Ker(f ) ∩ N = 0. We have that (M/Ker(f )) ∼= N and since N is a minimal submodule Ker(f ), then is a maximal submodule of M . Thus Ker(f ) ⊕ N = M . Corollary 1.16. Let M be a retractable module. If N is a minimal submodule in a semiprime module M , then N is a direct summand. 6 Proof. Since M is semiprime, NM N 6= 0. Theorem 1.17. The following conditions are equivalent for a retractable R- module M : 1. M is semisimple and left artinian. 2. M is semiprime and left artinian. 3. M is semiprime and satisfies DCC on cyclic submodules and direct sum- mands. Proof. 1 ⇒ 2 : If M is semisimple then it is semiprime. 2 ⇒ 3 : Since M is left artinian, then it satisfies DCC on cyclic submodules and direct summands. 3 ⇒ 1 : Since M satisfies DCC on cyclic submodules, there exists K1 a minimal submodule of M . By Corollary 1.16, M = K1 ⊕ L1. Now there exists K2 a minimal submodule of L1 and L1 = K2 ⊕ L2. With this process we obtain a descending chain of direct summands, which by hypothesis it is finite L1 ⊇ L2 ⊇ L3 ⊇ ... ⊇ Lm. Since Lm is simple and M = K1⊕K2 ⊕...⊕Km⊕Lm, then M is semisimple. Now, if M is semisimple and satisfies DCC on direct summands then M is artinian. Definition 1.18. Let M ∈ R − M od and N ≤ M . We say N is an annihilator submodule if N = AnnM (K) for some 0 6= K ≤ M . Lemma 1.19. Let M be semiprime and projective in σ[M ]. Let N, L ≤ M . If LM N = 0, then NM L = 0 and L ∩ N = 0. Proof. Since LM N = 0, then 0 = NM (LM N )M L = (NM L)M (NM L). Hence NM L = 0 . Now, since L ∩ N ≤ L and L ∩ N ≤ N , then (L ∩ N )M (L ∩ N ) ≤ LM N = 0. Thus L ∩ N = 0 Corollary 1.20. Let M be semiprime and projective in σ[M ]. If N ≤ M , then NM AnnM (N ) = 0. Proposition 1.21. Let M be semiprime and projective in σ[M ] and N ≤ M . Then N is an annihilator submodule if and only if N = AnnM (AnnM (N )) 7 Proof. ⇒: By Lemma 1.20 N ≤ AnnM (AnnM (N )). There is K ≤ M such that N = AnnM (K), hence KM N = KM AnnM (K) = 0 and thus K ≤ AnnM (N ). Therefore, AnnM (AnnM (N )) ≤ AnnM (K) = N It follows that N = AnnM (AnnM (N )). ⇐: By definition of annihilator submodule. Proposition 1.22. Let M be semiprime and N ≤ M . Then, AnnM (N ) is the unique pseudocomplement fully invariant of N . Moreover, N L AnnM (N ) intersects all fully invariant submodules of M . Proof. Let L ≤ M be a fully invariant pseudocomplement of N in M . Then LM N ≤ L ∩ N = 0 Thus L ≤ AnnM (N ). Observe that (AnnM (N ) ∩ N )M (AnnM (N ) ∩ N ) ≤ (AnnM (N ) ∩ N )M N = 0 Since M is semiprime, AnnM (N ) ∩ N = 0. Thus L = AnnM (N ). Lemma 1.23. Let M be a semiprime module and N ≤ M . Let S be the set of all minimal prime submodules of M which do not contain N . Then AnnM (N ) = T{P P ∈ S}. Proof. Put K = T{P P ∈ S}. Any element in K ∩ N is in the intersection of all minimal prime submodules of M which is zero. Then K ∩ N = 0. Since K is fully invariant in M , KM N ≤ K ∩ N = 0. Thus, K ≤ AnnM (N ). Now, let P ∈ S. Since AnnM (N )M N = 0 ≤ P and N (cid:2) P , then AnnM (N ) ≤ K. Lemma 1.24. Let M be projective in σ[M ]. If M is semiprime then M is retractable. Proof. Let N ≤ M and suppose HomR(M, N ) = 0. Then AnnM (N ) = M . So MM N = 0 but NM N ⊆ MM N = 0. Since M is semprime then N = 0 by Lemma 1.5. Proposition 1.25. Let M be projective in σ[M ] and semiprime. The following conditions are equivalent for N ≤ M : 1. N is a maximal annihilator submodule. 2. N is an annihilator submodule and is a minimal prime submodule. 3. N is prime in M and N is an annihilator submodule. 8 Proof. 1 ⇒ 2 : Let K ≤ M such that N = AnnM (K). Let L, H ≤ M be fully invariant submodules of M such that LM H ≤ N . Assume H (cid:2) N . Then 0 6= HM K. Hence AnnM (K) ≤ AnnM (HM K), but since AnnM (K) is a maximal annihilator submodule, then AnnM (K) = AnnM (HM K). As M is projective in σ[M ], by Proposition 5.5 of [2], we have that LM (HM (HM K)) = (LM H)M (HM K) ≤ NM (HM K) = 0 Now, since HM (HM K) ≤ HM K, then AnnM (K) = AnnM (HM K) ≤ AnnM (HM (HM K)) Therefore AnnM (HM K) = AnnM (HM (HM K)). Thus L ≤ AnnM (K) = N . Now, let P ≤ M be a prime submodule of M such that P < N . We have that NM K = 0 ≤ P . So K ≤ P < N . Hence KM K = 0. Thus K = 0, but M is semiprime, a contradiction. It follows that N is a minimal prime submodule of M . 2 ⇒ 3 : By hypothesis. 3 ⇒ 1 : Suppose N < K with K an annihilator submodule. Then AnnM (K)M K = 0 ≤ N Since N is prime in M , then AnnM (K) ≤ N < K. By Proposition 1.22 AnnM (K)∩K = 0, hence AnnM (K) = 0. Since K is an annihilator submodule, by Proposition 1.21, K = AnnM (AnnM (K)) = AnnM (0) = M . Remark 1.26. Following the notation of Example 1.12 of [4] Let R = Z2 ⋊ (Z2 ⊕Z2). This ring has only one maximal ideal I and it has three simple ideals: J1, J2, J3, which are isomorphic. Then, the lattice of ideals of R has the form R • I • J1• ✂✂✂✂✂✂✂✂ ❁❁❁❁❁❁❁❁❁ J2• 0 • ❁❁❁❁❁❁❁❁ ✂✂✂✂✂✂✂✂✂ J3• Moreover, R is artinian and R-Mod has only one simple module up to iso- morphism. Let S be a simple module. By Theorem 2.13 of [15], the lattice of fully invariant submodules of E(S) has tree maximal submodules N , L and K, and it has the form 9 E(S) • ❄❄❄❄❄❄❄❄❄ ⑧⑧⑧⑧⑧⑧⑧⑧⑧ N • K • ⑧⑧⑧⑧⑧⑧⑧⑧⑧ ❄❄❄❄❄❄❄❄❄ L • S • 0 • Put M = E(S). Since K ∩ L = S and KM L ≤ K ∩ L, then KM L ≤ S. On the other hand consider the composition M π / M/N ∼= / S i / L f = i ◦ π where π is the natural projection and i is the inclusion. Then, f (K) = S and S ≤ KM L. Thus, KM L = S. Notice that KM L ≤ N but K (cid:2) N and L (cid:2) N . Hence N is not prime in M . Analogously, we prove that neither K nor L are prime in M . We also note that KM K = S. Moreover, π(K) = S, so KM S = S. In the same way LM S = S and NM S = S Let g : M → K be a non zero morphism. If Ker(g) ∩ S = 0 then g is a monomorphism, a contradiction. So Ker(g) ∩ S = S. Thus SM K = 0 and AnnM (K) = S. Analogously AnnM (L) = S = AnnM (N ) = AnnM (S). Since SM S ≤ SM K, SM S = 0. Thus M is not semiprime. Hence, S is a maximal annihilator submodule of M which is not prime because KM K = S. With this we can see that associativity is not true in general, because LM (KM S) = LM S = S and (LM K)M S = SM S = 0. Notice that, in this example HomR(M, H) 6= 0 for all H ∈ σ[M ] in particular M is retractable, but M is not projective in σ[M ]. Proposition 1.27. Let M be projective in σ[M ] and semiprime. For N ≤ M , if N = AnnM (U ) with U ≤ M a uniform submodule, then N is a maximal annihilator in M . Proof. Suppose that N < K with K an annihilator submodule in M . Since N = AnnM (U ) by Proposition 1.22, K ∩ U 6= 0. By hypothesis U is uniform and thus K ∩ U ≤e U . Then (K ∩ U ) ⊕ AnnM (U ) ≤e U ⊕ AnnM (U ) Now, notice that if L ≤F I M , by Proposition 1.22 (U ⊕ AnnM (U )) ∩ L 6= 0. So ((K ∩ U ) ⊕ AnnM (U )) ∩ L 6= 0. Therefore, K ∩ L 6= 0 and K intersects all fully invariant submodules of M . Since K ∩ AnnM (K) = 0 and AnnM (K) ≤F I M , then AnnM (K) = 0. Thus, K = AnnM (AnnM (K)) = AnnM (0) = M . 10 / / / Proposition 1.28. Let M be projective in σ[M ] and semiprime with finite uniform dimension. Then: 1. M has finitely many minimal prime submodules. 2. The number of annihilators submodules is finite. 3. M satisfies ACC on annihilators submodules. Proof. 1 : Let U1, .., Un be uniform submodules of M such that U1 ⊕ ... ⊕ Un ≤e M . By Proposition s 1.25 and 1.27, Pi := AnnM (Ui) is a minimal prime submodule of M for each i. By Proposition 1.22, (U1 ⊕ ... ⊕ Un) ∩ AnnM (U1 ⊕ ... ⊕ Un) = 0 and P1 ∩ ... ∩ Pn ≤ AnnM (U1 ⊕ ... ⊕ Un) = 0. Now, if P is a minimal prime submodule of M , then P1M P2M ...M Pn ≤ P1 ∩ ... ∩ Pn = 0 ≤ P Hence, there exists j such that Pj ≤ P , a contradiction. 2 : By Lemma 1.23. 3 : It is clear by 2. 2 Goldie Modules The following definition was taken from [17] Definition 2.1. Let M ∈ R − M od. M is essentially compressible if for every essential submodule N ≤e M there exists a monomorphism M → N . Definition 2.2. Let M ∈ R − M od. We call a left annihilator in M a submod- ule AX = \{Ker(f )f ∈ X} for some X ⊆ EndR(M ). Definition 2.3. We say M is a Goldie module if it satisfies ACC on left anni- hilators and has finite uniform dimension. Lemma 2.4. Suppose M is projective in σ[M ]. compressible, then AnnM (N ) is a semiprime submodule of M . If N ∈ σ[M ] is essentially Proof. Let L ≤ M be a fully invariant submodule of M such that LM L ≤ AnnM (N ). Put Γ = {K ≤ N LM K = 0} Then Γ 6= ∅ and by Zorn's Lemma there exists a maximal independent family {Ki}I in Γ. Notice that LI Ki ∈ Γ because Ki = M LM M LM Ki = 0 Let 0 6= A ≤ N be a submodule. Since (LM L)M A = 0 then LM A ∈ Γ. I I 11 independent family in Γ. If LM A = 0 then A ∈ Γ and A ∩ LI Ki 6= 0 because {Ki} is a maximal Now, if LM A 6= 0 we also have (LM A) ∩LI Ki 6= 0 and (LM A) ∩LI Ki ≤ By hypothesis there exists a monomorphism θ : N → LI Ki. Then A ∩ LI Ki. Thus LI Ki ≤e N . θ(LM N ) ≤ LM M Ki = 0 and hence LM N = 0. Thus L ≤ AnnM (N ). I Proposition 2.5. Let M be projective in σ[M ]. If N ∈ σ[M ] is an M -singular module, then Ker(f ) ≤e M for all f ∈ HomR(M, N ). Proof. Let f ∈ HomR(M, N ). Since N is M -singular, there exists an exact sequence 0 / K i / L π / N / 0 in σ[M ] with K ≤e L. Since M is projective in σ[M ], there exists f : M → L such that π f = f : M f f ~⑥⑥⑥⑥⑥⑥⑥⑥ L π / N / 0 As K ≤e L, then f −1(K) ≤e M . Then f ( f −1(K)) = π( f ( f −1(K))) ≤ π(K) = 0. Therefore, f −1(K) ≤ Ker(f ) and hence Ker(f ) ≤e M . Proposition 2.6. Let M be projective in σ[M ]. If M is essentially compressible then M is non M -singular. Proof. Suppose Z(M ) 6= 0. If Z(M ) ≤e M , then there exists a monomorphism θ : M → Z(M ), by Proposition 2.5 Kerθ ≤e M , a contradiction. Therefore Z(M ) has a pseudocomplement K in M and thus Z(M )⊕K ≤e M . Hence, there exists a monomorphism θ : M → Z(M ) ⊕ K. Let π : Z(M ) ⊕ K → Z(M ) be the canonical projection, then Ker(πθ) ≤e M and so Ker(πθ) = θ−1(Kerπ) = θ−1(K) ≤e M . But Z(M ) ∩ θ−1(K) = 0,contradiction. Thus Z(M ) = 0. Lemma 2.7. Let M ∈ R − M od with finite uniform dimension. Then, for every monomorphism f : M → M , Im(f ) ≤e M . Proof. Let f : M → M be a monomrfism. If the uniform dimension of M is n, (U dim(M ) = n) and there exists K ≤ M such that f (M ) ∩ K = 0, then U dim(f (M ) ⊕ K) = n + 1, a contradiction. Theorem 2.8. Let M be projective in σ[M ] with finite uniform dimension. The following conditions are equivalent: 12 / / / / ~ / / 1. M is semiprime and non M -singular 2. M is semiprime and satisfies ACC on annihilators 3. Let N ≤ M , then N ≤e M if and only if there exists a monomorphism f : M → N . Proof. 1 ⇒ 2 : Since M is non M -singular and has finite uniform dimension then, by Proposition 3.6 of [6] M satisfies ACC on annihilators. This proves 2. 2 ⇒ 3 : Let N ≤ M . Suppose that N ≤e M . Since M is semiprime with uniform dimension and satisfies ACC on annihilators, then M is essentially compressible by Proposition 3.13 of [6]. Now, if f : M → N is a monomorphism then N ≤e M by lemma 2.7. 3 ⇒ 1 : It follows from Lemma 2.4 and Proposition 2.6. Remark 2.9. Notice that Theorem 2.8 is a generalization of Goldie's Theorem. See [11] Theorem 11.13. In Proposition 3.13 of [6], M is a generator of σ[M ], but by Lemma 1.24 this hypothesis is not necessary. Corollary 2.10. Let M be projective in σ[M ] and semiprime. Then, M has finite uniform dimension and enough monoforms if and only if M is a Goldie module. Proof. ⇒: Since M is semiprime with finite uniform dimension and enough monoforms, then M is non M -singular by Proposition 3.8 of [6]. By Theorem 2.8, M is a Goldie module. ⇐: If M is a Goldie module, M has finite uniform dimension and by Theorem 2.8 M is non M -singular. Hence the uniform submodules of M are monoform. Since M has finite uniform dimension every submodule of M contains a uniform, hence every submodule contains a monoform. For the definition of M -Gabriel dimension see [4] section 4. Corollary 2.11. Let M be projective in σ[M ] with finite uniform dimension. If M is a semiprime module and has M -Gabriel dimension, then M is a Goldie module. Proof. Let N ≤ M . Since M has M -Gabriel dimension, by Lemma 4.2 of [4], N contains a cocritical submodule L. Then L is monoform. By Proposition 2.10 M is a Goldie module. Corollary 2.12. Let M be projective in σ[M ] and semiprime with Krull di- mension. Then M is a semiprime Goldie module. Proof. Since M has Krull dimension, M has finite uniform dimension and enough monoforms. By Proposition 2.10 M is a Goldie module. Proposition 2.13. Suppose that M is progenerator of σ[M ]. Let N ∈ σ[M ], then Z(N ) = X{f (M )f : M → N ker(f ) ≤e M }. 13 Proof. By definition of M -singular module, it is clear that P{f (M )f : M → N ker(f ) ≤e M } ≤ Z(N ). Now, let n ∈ Z(N ) and consider Rn ≤ Z(N ). Since Rn ∈ σ[M ] there exists a natural number t and an epimorphism ρ : M t → Rn. Suppose that (m1, .., mt) is such that ρ(m1, ..., mt) = n. If ji : M → M t are the inclusions (i = 1, ..., t), then by Proposition 2.5 Ker(ρ ◦ ji) ≤e M . Thus, n = Pt i=1 ρ ◦ ji(mi) ∈ P{f (M )f : M → N ker(f ) ≤e M }. Remark 2.14. Let M ∈ R − M od and consider τg ∈ M − tors, where τg = ξ({S ∈ σ[M ]S is M − singular}). If M ∈ Fτg , by [20] Proposition. 10.2, we have that χ(M ) = τg. Let tτg be the preradical associated to τg. Then tτg (N ) = X{S ≤ N S ∈ Tτg } = X{S ≤ N S is M − singular} = Z(N ). Proposition 2.15. Suppose M is progenerator of σ[M ]. If M is semiprime Goldie, then Z(N ) = X f (M ) where the sum is over the f : M → N such that there exists α ∈ EndR(M ) monomorphism with α(M ) ≤e M and f α = 0. Proof. Let N ∈ σ[M ]. By Proposition 2.13 Z(N ) = X{f (M )f : M → N ker(f ) ≤e M }. If f : M → N with Ker(f ) ≤e M , by Theorem 2.8 there exists a monomorphism α : M → Ker(f ). We have that f α = 0 and by Lemma 2.7 α(M ) ≤e (M ). Let f : M → N such that there exists α : M → M f α = 0 and α(M ) ≤e (M ). Then α(M ) ≤ Ker(f ). Therefore Ker(f ) ≤e (M ). Remark 2.16. Let R be a ring such that R-Mod has an infinite set of non- isomorphic simples modules. Consider M = LI Si, I an infinite set, such that Si is a simple module for all i ∈ I and with Si ≇ Sj if i 6= j. This module does not have finite uniform dimension and, in M -tors, τg = χ. Then, if N ∈ σ[M ] tτg (N ) = Z(N ) = X f (M ) where the sum is over the f : M → N such that there exists α ∈ EndR(M ) monomorphism with α(M ) ≤e M and f α = 0. This example shows that the converse of the last Proposition is not true in general. Following [1] Definition 2.17. A module M is weakly compressible if for any nonzero sub- module N of M , there exists f : M → N such that f ◦ f 6= 0. Remark 2.18. Notice that if M is weakly compressible then M is a semiprime module. The converse hold if M is projective in σ[M ] 14 Next definition was taken from [8] Definition 2.19. A module M is a semiprojective module if I = Hom(M, IM ) for any cyclic right ideal I of EndR(M ) For other characterizations see [19]. Proposition 2.20. Let M be projective in σ[M ] and retractable. Then, S := EndR(M ) is semiprime if and only if M is semiprime. Proof. ⇒: Corollary 1.10. ⇐: If M is semiprime, since M is projective in σ[M ] then M is weakly com- pressible and semiprojective. Then, by [[8]. Theorem 2.6 (b)] S is semiprime. Lemma 2.21. Let M be projective in σ[M ] and retractable. M is non M - singular if and only if HomR(M/N, M ) = 0 for all N ≤e M . Proof. ⇒: If N ≤e M then M/N is M -singular, then HomR(M/N, M ) = 0. ⇐: Suppose Z(M ) 6= 0. Since M is retractable there exists 0 6= f : M → Z(M ). By Proposition 2.5 Ker(f ) ≤e M , so there exists a non zero morphism form M/Ker(f ) → M . For a retractable R-module M , Theorem 11.6 of [20] gives necessary and sufficient conditions in order to T := EndR(cM ) being semisimple, left artinian, and being the classical left quotient ring of S = EndR(M ). Also, in ([8], Corol- lary 2.7) the authors give necessary and sufficient conditions for a semiprojective module M to S being a semiprime right Goldie ring. We give an extension of these results. Theorem 2.22. Let M be projective in σ[M ], S = EndR(M ) and T = EndR(cM ). The following conditions are equivalent: 1. M is a semiprime Goldie module. 2. T is semisimple right artinian and is the classical right quotient ring of S. 3. S is a semiprime right Goldie ring. 4. M is weakly compressible with finite uniform dimension, and for all N ≤e M , HomR(M/N, M ) = 0 . Proof. 1 ⇒ 2 : By Proposition 2.20, S is a semiprime ring. Since M is a Goldie module, then M is non M -singular with finite uniform dimension, hence by [20] Proposition 11.6, T is right semisimple and is the classical right quotient ring of S. 2 ⇒ 3 : By [11] Theorem 11.13, S is a semiprime right Goldie ring . 3 ⇒ 4 : By [8] Corollary 2.7. 4 ⇒ 1 : Since M is weakly compressible then M is semiprime. By Lemma 2.21 M is non M -singular. Thus, by Theorem 2.8 M is a Goldie module. 15 Corollary 2.23. Let M be projective in σ[M ], S = EndR(M ) and T = EndR(cM ). The following conditions are equivalent: 1. M is a prime Goldie module. 2. T is simple right artinian and is the classical right quotient ring of S. 3. S is a prime right Goldie ring. 4. Given nonzero submodules N , K of M there exists a morphism f : M → N such that K * Ker(f ). M has finite uniform dimension and for all N ≤e M , Hom(M/N, M ) = 0. Proof. 1 ⇒ 2 : By Proposition 2.22, S is a semiprime ring and T is right semisimple and the classical right quotient ring of S. Let 0 6= I ≤ T be an ideal. Since T is semisimple, there exits an ideal J ≤ T such that T = I ⊕ J. Put M1 = IcM and M2 = JcM . Then M1 and M2 are fully invariant submodules of cM and M1 ∩ M2 = 0 because I ∩ J = 0. Consider M1 ∩ M and M2 ∩ M . If f ∈ S, then there exists f ∈ T such that f = f M . Let x ∈ M1 ∩ M . Then f (x) = f (x) ∈ M1 ∩ M since M1 is a fully invariant submodule of cM . Thus M1 ∩ M is a fully invariant submodule of M . In the same way, M2 ∩ M is fully invariant in M . Since (M1∩M )∩(M2 ∩ M ) = 0, then (M1∩M )M (M2 ∩ M ) = 0. Hence M1 ∩ M = 0 or M2 ∩ M = 0 because M is prime. On the other hand, M ≤e cM and so M1 = 0 or M2 = 0. Since 0 6= I, then M2 = 0. Thus J = 0, and it follows that T is a simple ring. 2 ⇒ 3 : By [11] Corollary 11.16, S is a prime right Goldie ring. 3 ⇒ 4 : Let N , K be nonzero submodules of M , if K ⊆ Ker(f ) for all f : M → N then 0 = HomR(M, N )Hom(M, K) ≤ S. Then HomR(M, N ) = 0 or HomR(M, K) = 0. By retractability, N = 0 or K = 0, a contradiction. 4 ⇒ 1 It is clear. Remark 2.24. Suppose that M and N are R-modules such that σ[N ] ⊆ σ[M ]. If N is non M -singular, then N is non N -singular. This is because if there exists an exact sequence 0 → L → K → N → 0 in σ[N ] such that L ≤e N , then this sequence is in σ[M ] which implies that N is M -singular, a contradiction. Proposition 2.25. Let M be projective in σ[M ] and semiprime with finitely many minimal prime submodules P1, ..., Pt. Suppose every quotient M/Pi (1 ≤ i ≤ t) has finite uniform dimension. Then M is a Goldie module if and only if each M/Pi is a Goldie module. Proof. ⇒: Suppose M is a Goldie module and Pi is a minimal prime submodule of M . By hypothesis, each M/Pi has finite uniform dimension. Notice that by proposition 1.14 Pi ⊆ AnnM (P1 ∩ ... ∩ Pi−1 ∩ Pi+1 ∩ ... ∩ Pn) Since M has finite uniform dimension there exist a uniform submodule Ui of P1 ∩ ... ∩ Pi−1 ∩ Pi+1 ∩ ... ∩ Pn. So Pi ⊆ AnnM (Ui). By Propositions 1.25 and 16 X and 1.27, Pi = AnnM (Ui). Then, there exists a monomorphism M/Pi → Ui since Ui is non M -singular, then M/Pi is non M -singular. Thus M/Pi is non (M/Pi)-singular by Remark 2.24. Since M/Pi is a prime module, by Theorem 2.8 M/Pi is a Goldie module. ⇐: By Corollary 1.14 there exists a monomorphism M → Lt i=1 M/Pi. Since each M/Pi has finite uniform dimension then M has finite uniform dimension. Let 0 6= N be a submodule of M . Since there exists a monomorphism M → L M/Pi then there exists 1 ≤ i ≤ t and submodules 0 6= K ≤ M/Pi and 0 6= N ′ ≤ N such that K ∼= N ′. We have that M/Pi is a Goldie module, thus it has enough monoforms. Hence N ′ has a monoform submodule, that is M has enough monoforms, and so by Corollary 2.10 M is Goldie module. Remark 2.26. Notice that if M is a semiprime Goldie module then M has finitely many minimal prime submodules by Proposition 1.28. So in the proof ⇒: of Proposition 2.25 this hypothesis is not used. Definition 2.27. Let M ∈ R − M od and N ≤ M . We say N is a regular submodule if there exists a monomorphism M → N . Denote Reg(M ) := {N ≤ M N regular submodule} Remark 2.28. There exists modules with regular submodules which are nonessen- tial. For example, a pure infinite module, see [12]. Proposition 2.29. Let M be projective in σ[M ] and a semiprime Goldie mod- ule. Then, N is a regular submodule of M if and only if N is essential in M . Proof. Since M is Goldie, every regular submodule is essential by Lemma 2.7. Now, let N ≤e M . By Theorem 2.8, N is a regular submodule. If K ∈ σ[M ], we say that K is Reg(M )-injective if any morphism f : N → K with N ∈ Reg(M ) can be extended to a endomorphism of M . Corollary 2.30. Let M be projective in σ[M ] and a semiprime Goldie module. Let K ∈ σ[M ]. If K is Reg(M )-injective, then K is M -injective. 3 Duo Modules Following [13] Definition 3.1. Let M ∈ R − M od. M is a duo module if every submodule of M is fully invariant in M . Examples: 1. If RS is a simple module then, S is a duo module. 2. If RM = LI Si with Si simple and Si not isomorphic to Sj i 6= j then M is a duo module. 17 3. An R-module M is called a multiplication module if every N ≤ M is of the form IM = N for some ideal I of R. These modules are examples of duo modules. See [18] 4. Consider the example in Remark 1.26 that was taken from [4]. In that paper it is proved that M/K ∼= S ∼= M/L ∼= M/N , hence L, K and N are maximal submodules of M . It follows that K/S, L/S and N/S are maximal submodules of M/S. Moreover, since K ∩ L = S = K ∩ N = N ∩ L, then M/S = K/S ⊕ L/S. Thus K/S ∼= M/S L/S ∼= M/L ∼= S This implies that K/S is simple, and analogously L/S and N/S are simple. Let 0 6= T < M . Since S ≤e M , then S ≤ T . If T = S, then T is fully invariant. Suppose that T 6= S and T /∈ {K, L, N }. We have that S ≤ T ∩ K ≤ K. Moreover, since K/S is simple, then T ∩ K = S or T ∩ K = K. If T ∩ K = K then K ≤ T < M ; but K is maximal, then K = T , a contradiction. Thus, T ∩K = S. Analogously T ∩L = S = T ∩N . Let 0 6= x ∈ M . If annR(x) = 0, there exists a monomorphism R → M and thus E(R) = M , a contradiction, because E(R) ∼= M ⊕ M (see [4], Example 1.12) and M is a indecomposable injective module. Thus, annR(x) 6= 0 for all 0 6= x ∈ M . Let 0 6= x ∈ T . Since annR(x) 6= 0, then annR(x) ∈ {I, J1, J2, J3}. By Theorem 2.13 of [15] we have that: • If annR(x) = I then x ∈ S • If annR(x) = J1 then x ∈ K ∩ T = S • If annR(x) = J2 then x ∈ L ∩ T = S • If annR(x) = J3 then x ∈ N ∩ T = S Therefore T ≤ S, a contradiction. Thus, all submodules of M are fully invariant. Remark 3.2. In [13] the authors state that they did not know an example of a duo module M and a submodule N such that M/N is not a duo module. In this example, M is a duo module, but M/S ∼= S ⊕ S is not a duo module. Proposition 3.3. M is a duo module as R-module and it generates all its submodules if and only if M is a multiplication module as EndR(M )-module. Proof. ⇒: Let S = EndR(M ) and let N be a submodule of M . Since M is a duo module, N is fully invariant, thus HomR(M, N ) is an ideal of S. Since M generates all its submodules, then N = trM (N ) = HomR(M, N )M . Thus, M is a multiplication module as EndR(M )-module. ⇐: It is clear. 18 Proposition 3.4. Let M be projective in σ[M ]. Suppose that M is a semiprime and non M -singular duo module. Then, for every subset X ⊆ EndR(M ) we have that: Ker(f ) AnnM (AnnM (\ Ker(f ))) = \ X X Proof. Since M is a duo module, by Proposition 1.22, AnnM (TX Ker(f )) is the unique pseudocomplement of TX Ker(f ). Then Ker(f ) ≤e AnnM (AnnM (\ Ker(f ))). \ X X Since M is non M -singular, TX Ker(f ) has no essential extensions in M by Lemma 3.5 of [6]. Thus, we have the equality. Proposition 3.5. Let M projective in σ[M ]. Suppose that M is a semiprime and non M -singular duo module. The following conditions are equivalent: 1. M has finite uniform dimension. 2. M has a finite number of minimal prime submodules. 3. The number of annihilators in M is finite. 4. M satisfies the ACC on annihilators. 5. M satisfies the ACC on pseudocomplements. Proof. 1 ⇒ 2 ⇒ 3 : Are true by Proposition 1.28. 3 ⇒ 4 : By Proposition 3.4. 4 ⇒ 5 : By Proposition 1.22. 5 ⇒ 1 : By [11] Proposition 6.30. Proposition 3.6. Let M be projective in σ[M ]. Suppose M is a prime duo module with finite uniform dimension. Then, U dim(M ) = 1 Proof. Since M is prime, 0 is the unique minimal prime submodule of M . By Proposition 1.28, there exists a uniform submodule U of M such that 0 = AnnM (U ). By Proposition 1.22, U ≤e M . Thus, U dim(M ) = 1. Theorem 3.7. Let M be projective in σ[M ]. If M is a semiprime duo module, then the following conditions are equivalent: 1. M is a prime Goldie module. 2. cM is indecomposable and M is non M -singular. 3. M is uniform and non M -singular. 19 Proof. 1 ⇒ 2 : Since M is a prime module, by Proposition 3.6, U dim(M ) = 1 and then cM is indecomposable. Since M is a Goldie module, by Theorem 2.8 M is non M -singular. 2 ⇒ 3 : Let 0 6= K ≤ M . Then, there exists L ≤ M such that K ⊕ L ≤e M . Hence, bK ⊕bL = M , but since cM is indecomposable, then L = 0. Thus K ≤e M . 3 ⇒ 1 : Let K and 0 6= L be submodules of M such that KM L = 0. Then, K ≤ AnnM (L), and thus K ∩ L = 0 by Proposition 1.22. Since M is uniform, K = 0. Thus, M is prime and by Theorem 2.8 M is Goldie. References [1] O.D. Avraamova, "A generalized density theorem", Abelian Group and Modules, no. 8, Tomsk. Gos. Univ., Tomsk, 3-16 (1989). [2] J. Beachy, "M-Injective Modules and Prime M -Ideals", Communications in Algebra, 30:10, 4649-4676 (2002) [3] L. Bican, P. Jambor, T. Kepka, P. Nemec. "Prime and coprime modules", Fundamenta matematicae, 107:33-44 (1980). [4] J. Castro, J. R´ıos "Prime Submodules and Local Gabriel Correspondence in σ[M ] ", Communications in Algebra, 40:1, 213-232 (2012) [5] J. Castro, J. R´ıos, "FBN Modules", Communications in Algebra, 40:12, 4604-4616 (2012) [6] J. Castro, J. R´ıos, "Krull Dimension and Classical Krull Dimension of Modules", Communications in Algebra, 42:7, 3183-3204 (2014) [7] Goldie, "Semiprime rings with maximum conditions", Proc. London Math. Soc., 10, 201-220 (1960) [8] A. Haghany, M.R. Vedadi, "Study of Semi-projective Retractable Mod- ules", Algebra Colloquium, 14:3, 489-496 (2007) [9] S. M. Khuri, "Endomorphism rings and lattice isomorphisms", J. Algebra, 59 (2) 401-408 (1979). [10] T. Y. Lam, A First Course in Noncommutative Rings, Springer-Verlag, New York Inc. (2001) [11] T. Y. Lam, Lectures on Modules and Rings, Grad. Texts in Math., vol 139, Springer, New York, (1998) [12] S. H. Mohamed, B. J. Muller, Continuous and Discrete Modules, Lon- don Math. Soc. Lecture Note Series No. 147 (Cambridge University Press, 1990). [13] A. Ozcan, A. Harmanci, P. F. Smith, "Duo Modules ", Glasgow Math. J., 533-545 (2006). 20 [14] F. Raggi, J. R´ıos, H. Rinc´on, R. Fern´andez-Alonso, C. Signoret, "Prime and Irreducible Preradicals", J. Algebra Appl., Vol. 4, No. 4, 451-466. (2005) [15] F. Raggi, J. R´ıos, H. Rinc´on, R. Fern´andez-Alonso, "Basic Preradicals and Main Injective Modules ", J. Algebra Appl. 8(1):1-16 (2009) [16] F. Raggi, J. R´ıos, H. Rinc´on, R. Fern´andez-Alonso, "Semiprime preradicals ", Communications in Algebra, 37, no. 8, 2811-2822 (2009). [17] P.F. Smith, M.R. Vedadi, "Essentially Compressible Modules and Rings ", J. of Algebra, 304, 812-831 (2006). [18] A. A. Tuganbaev, "Multiplication modules over noncommutative rings ", Sb. Math, 194:1837-1864 (2003). [19] R. Wisbauer, Foundations of Module and Ring Theory., Reading: Gordon and Breach (1991). [20] R. Wisbauer, Modules and Algebras: Bimodule Structure and Group Ac- tions on Algebras, England: Addison Wesley Longman Limited (1996). 21
1501.02512	1	1501	2015-01-12T00:32:22	Piggyback dualities revisited	[ "math.RA" ]	In natural duality theory, the piggybacking technique is a valuable tool for constructing dualities. As originally devised by Davey and Werner, and extended by Davey and Priestley, it can be applied to finitely generated quasivarieties of algebras having term-reducts in a quasivariety for which a well-behaved natural duality is already available. This paper presents a comprehensive study of the method in a much wider setting: piggyback duality theorems are obtained for suitable prevarieties of structures. For the first time, and within this extended framework, piggybacking is used to derive theorems giving criteria for establishing strong dualities and two-for-one dualities. The general theorems specialise in particular to the familiar situation in which we piggyback on Priestley duality for distributive lattices or Hofmann--Mislove--Stralka duality for semilattices, and many well-known dualities are thereby subsumed. A selection of new dualities is also presented.	math.RA	math	Piggyback dualities revisited B. A. Davey, M. Haviar, and H. A. Priestley Dedicated to the memory of Ervin Fried and Jiri Sichler Abstract. In natural duality theory, the piggybacking technique is a valuable tool for constructing dualities. As originally devised by Davey and Werner, and extended by Davey and Priestley, it can be applied to finitely generated quasivarieties of algebras having term-reducts in a quasivariety for which a well-behaved natural duality is already available. This paper presents a comprehensive study of the method in a much wider setting: piggyback duality theorems are obtained for suitable prevarieties of structures. For the first time, and within this extended framework, piggybacking is used to derive theorems giving criteria for establishing strong dualities and two-for- one dualities. The general theorems specialise in particular to the familiar situation in which we piggyback on Priestley duality for distributive lattices or Hofmann -- Mislove -- Stralka duality for semilattices, and many well-known dualities are thereby subsumed. A selection of new dualities is also presented. 1. Introduction This paper gives a systematic, general treatment of the method of piggy- backing in the context of natural dualities for structures. The principal results are Theorems 3.3 and 3.5. These subsume and extend previous uses of the piggybacking technique. For the first time, piggybacking is used to derive strong dualities, albeit under stringent conditions; see Theorems 3.7 and 3.8. In broad terms, duality theory seeks to use one category, X, to reason about another, A, with the categories linked by a dual adjunction or, better, by a dual equivalence. The more tightly the two categories are linked, the more powerful this general strategy will be. Specifically, assume we have contravari- ant functors D : A → X and E : X → A and that there exist a unit and counit e and ε so that hD, E, e, εi is a dual adjunction. An aspect of duality theory that has proved particularly fruitful is that in which D and E are hom-functors, with D = A(−, M) and E = X(−, M∼ ), where M ∈ A and M∼ ∈ X are ob- jects with the same underlying set M . Within this very general categorical framework, duality theory as a tool for algebra has a special niche. The the- ory of natural dualities, in which M is taken to be an algebra, usually finite, Presented by . . . Received . . . ; accepted in final form . . . 2010 Mathematics Subject Classification: Primary: 08C20; Secondary: 06D50, 06A12. Key words and phrases: natural duality, piggyback duality, distributive lattice, semilat- tice, Ockham algebra. The second author acknowledges support from Slovak grants VEGA 1/0212/13 and APVV-0223-10. 2 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. has been developed to a high level, as was already evident in the 1998 text by Clark and Davey [4]. Subsequent advances have featured dualisability (as witnessed by the monograph [23]) and the theory of full dualities and strong dualities ([4, Chapter 3] and [13]). We shall strike out in a different direc- tion. As we have just indicated, the existing core theory of natural dualities focuses on dual representations for algebras, specifically algebras in a finitely generated quasivariety. But, while this restricts A to be drawn from a very important class of categories, it is possible to encompass structures more gen- eral than algebras and in certain circumstances to remove the requirement of finite generation. The study of natural dualities for finitely generated quasivarieties of struc- tures was initiated by Davey [6] (and Hofmann [20] had earlier considered the more general setting of finitary limit sketches). In the present paper we fill in more of the overall duality picture in the setting of structures. In a not unrelated development, the present authors considered situations in which the topology could be moved from one side of a dual adjunction to the other, thereby creating pairs of dualities in partnership [8]; our Theorem 3.8 can be seen as a piggyback-based topology-swapping theorem. Motivation for consid- eration of paired dualities came in part from investigation of canonical exten- sions for lattice-based algebras (see [8, 7]). More recently, we have shown how these ideas fit into the broader framework of free constructions which can be viewed as zero-dimensional Bohr compactifications of structures [9]. The authors of [4] took a deliberate decision to restrict their treatment to finitely generated quasivarieties (of algebras). It was already recognised in [15] that finite generation is not a necessary condition for a natural duality to exist but, 30 years on, little general theory has been developed and non-finitely generated examples remain tantalisingly scarce: abelian groups (Pontryagin [24]); Ockham algebras [19, 16, 17]; certain semilattice-based algebras [12, Section 8]. In this paper we operate within a framework in which we do not restrict to generating structures which are finite. As a consequence, topological conditions arise which are absent in the finitely generated setting. Our main results, as presented in Section 3, require the structures under consideration to be total, that is, to contain no partial operations. Nevertheless, the general setting of infinite generating structures is a new departure, and we shall allow for partial operations where there is no reason to exclude them. Most importantly, we must comment on the role of Davey and Werner's piggybacking technique [16, 17] and its subsequent evolution. The fundamen- tal idea is very simple. Consider a prevariety B = ISP(N) for which a well- behaved full duality is already available -- prototypical examples would be D, the variety of bounded distributive lattices, with Priestley duality, and S, the variety of unital meet semilattices, with Hofmann -- Mislove -- Stralka duality. Take an algebra M generating a prevariety A = ISP(M) for which a dual- ity is sought. Assume that M has a term-reduct M♭ in B. Seek to hitch a piggyback ride: use this known duality for B to build the required duality Vol. 00, XX Piggyback dualities revisited 3 for A, using a carrier map ω ∈ B(M♭, N) to link the categories involved. As Davey and Werner showed, this idea allowed the construction of economical dualities, in situations where the existence of a duality was not in question but where general theory (in particular the NU Duality Theorem, where applica- ble) supplied dual structures too unwieldy to be of practical use. They also showed how the method had the potential to establish dualisability for certain non-finitely generated varieties, as witnessed by their treatment of the variety of Ockham algebras. Later, Davey and Priestley [14] enlarged the scope of the method by allowing a set of carrier maps in cases where a single map ω does not provide tight enough linkage between A and B for Davey and Werner's single-carrier piggyback theorem to apply. More details on the general method can be found in Section 3 below. We want to highlight one aspect of our piggyback theorems, namely the conditions we supply which ensure that dualities are strong. This is a com- pletely new feature within the piggybacking framework, and its development is made possible by our consideration of co-dualities, in which, loosely, the roles of M and M∼ are swapped. This in turn hinges on the symmetry inherent in our allowing M to be a structure rather than an algebra; for an algebra M, a dualising alter ego M∼ (with topology disregarded) will rarely be an algebra. We draw attention to our Piggyback Strong Duality Theorem 3.7 in which we identify conditions under which single-carrier piggybacking yields a dou- ble best-of-all-possible-worlds scenario: we have simultaneously a duality and a co-duality, both strong. It was observed long ago, for certain well-known varieties of D -- based algebras, most notably De Morgan algebras and Stone algebras, that a natural duality (obtainable by single-carrier piggybacking) 'coincides' with Priestley duality. It is this phenomenon and its co-duality analogue that are witnessed by the specialisation of Theorem 3.7 to D-based algebras, Theorem 4.4. We draw attention, too, to an immediate consequence of Theorem 3.7, the Two-for-one Piggyback Strong Duality Theorem 3.8. Here the duality and co-duality are dualities in partnership, as in [8], with each ob- tained from the other simply by topology-swapping. Theorem 4.6 provides a nice application, in which moreover the generating algebra is infinite: we give a purely piggyback-based proof of Goldberg's duality for the variety of Ockham algebras, which is strong, and show that it is in partnership with a strong co-duality. The paper is organised in the following way. In Section 2 we set the scene for what follows. Here, and subsequently, we shall assume that the reader is familiar with the basic theory of natural dualities as presented in [4] and shall focus on aspects of the theory not to be found there, including some notions we introduce ab initio. Section 3 begins with a presentation of our Basic Assumptions -- the framework within which we establish our principal results. These results are then set out, prefaced by general comments on the piggybacking strategy. The proofs are postponed to Section 6 and modularised via a series of lemmas. That section begins with a summary of the overall 4 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. strategy and of the contribution made by the various lemmas. In Section 6, as in Section 2, we shall where possible allow partial operations locally. Sections 4 and 5 do not rely explicitly on the technical material in Section 6. They specialise the main results to the two settings -- D-based algebras and S- based algebras -- in which piggybacking has hitherto principally been employed. Here we have two objectives: to demonstrate how previous applications of piggybacking fit into a wider framework and to present new results, specifically on co-dualities and concerning prevarieties with infinite generators. We wish also to emphasise what the present paper does not cover. Our examples are confined to applications of our theorems to piggybacking over D and over S. Piggybacking over other amenable base categories is not explored here (though we note an overture in this direction made by Cabrer and Priest- ley [3], using piggybacking over the variety of distributive bilattices). There is also scope for a more comprehensive investigation of natural dualities in the non-finitely generated case, and associated examples: the theory here is as yet embryonic. We have addressed multi-carrier piggybacking for prevarieties but not full-blown multisorted dualities; however we would not foresee obstacles to extending our theory to the multisorted case. In conclusion, we assert that our achievements in this paper should be seen as meeting interim objectives rather than final ones. Further work is expected to reveal additional insights and to contribute new examples. 2. Setting the scene In this section we introduce the setting in which we shall work. A basic reference for natural duality theory for structures, rather than algebras, is Davey [6]. A discussion of the notions introduced below can be found there. We shall want to consider classes of structures of the form ISP(M), where M is a structure of a suitable kind and M is not necessarily finite. We refer to ISP(M) as the prevariety generated by M; only when M is finite is this preva- riety guaranteed to be definable by quasi-equations, and so to be a quasivariety according to the normal usage of this term. We regard the prevariety ISP(M) as a category by taking as morphisms all structure-preserving maps. The structures we shall consider take the form M = hM ; G, H, Ri. Here G and H are sets of, respectively, total and partial operations on M , of finite arities, and R is a set of finitary relations on M . We assume that the relations in R and the domains of the partial operations in H are non-empty. Any of G, H and R may be empty. We say that M is a total structure if H = ∅, that it is a total algebra if H = R = ∅ and that it is purely relational if G = H = ∅. Given a set H of partial operations on a set M , we define dom(H) := { dom(h) h ∈ H } and graph(H) := { graph(h) h ∈ H }. We note that in certain contexts (see [4, pp. 40 -- 41] for more details) it is convenient to replace the members of G ∪ H by their graphs. Vol. 00, XX Piggyback dualities revisited 5 Let M = hM ; G, H, Ri be a structure. Define A = ISP(M) to be the prevariety generated by M and let A ∈ A. We say that a subset X of A(A, M) separates the structure A if the following equivalent conditions hold: (1) the natural map η : A → M X, given by η(a)(x) := x(a), for all a ∈ A and all x ∈ X, is an embedding of A into MX ; (i) for all a, b ∈ A with a 6= b, there exists a morphism x ∈ X with (2) x(a) 6= x(b), and (ii) for every (n-ary) relation r in dom(H A) ∪ RA and all a1, . . . , an in A with (a1, . . . , an) /∈ rA, there exists a morphism x ∈ X with (x(a1), . . . , x(an)) /∈ rM. Note that A(A, M) separates the structure A as A ∈ ISP(M), and if M is a total algebra then a subset X of A(A, M) separates the structure A if and only if (2)(i) holds, that is, X separates the points of A. We shall also wish to consider structures-with-topology of the form M∼ = hM ; G, H, R, Ti. Here hM ; G, H, Ri will be assumed to be a structure of the same type as we considered above, and T is a compact Hausdorff topology on M -- a priori, no compatibility is assumed between the structure and the If M is finite then T is necessarily discrete. We define X := topology. IScP+(M∼ ) to be the class of all structures-with-topology X of the same type as M∼ for which X is isomorphic to a closed substructure of a non-zero power of M∼ . Note that the empty structure-with-topology ∅∅∅ belongs to X in case there are no nullary operations in the type of M∼ . Relations and total and partial operations are lifted pointwise from M∼ to any member X of X, and the domain of such a lifted map is indicated by the appropriate superscript. We regard X as a category by taking as morphisms all continuous structure- preserving maps. Given a structure-with-topology M∼ , let M′ be the structure obtained by removing the topology. When properties relating to structure are said to hold for M∼ we shall mean that they are true in M′. We now need to clarify what is meant by saying that a structure with topology is an alter ego of a structure on the same underlying set M . Here we must extend to the case that M is not necessarily finite the notion of com- patible structures introduced by Davey in [6]. Let M := hM ; G1, H1, R1i and M′ := hM ; G2, H2, R2i be structures on the set M . We say that M′ is com- patible with M if, for all n ∈ N, each n-ary relation in dom(H2) ∪ R2 forms a substructure of Mn and each operation in G2 ∪H2 is a homomorphism with re- spect to M. It is a symbol-pushing exercise to show that M′ is compatible with M if and only if M is compatible with M′. Given a compact Hausdorff topol- ogy T on M , we say that M = hM ; G1, H1, R1i and M∼ = hM ; G2, H2, R2, Ti are compatible if (a) M and M′ are compatible, and (b) each n-ary relation in dom(H1) ∪ R1 is topologically closed with respect to the topology T and each operation in G1 ∪H1 is continuous with respect to the topology T, that is, MT := hM ; G1, H1, R1, Ti is a topological structure. 6 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. A more compact, though less revealing, way to say that M and M∼ are compat- ible is to require that, for all n ∈ N, each n-ary relation in graph(G1 ∪ H1) ∪ R1 n. If M and M∼ are compatible, forms a topologically closed substructure of M∼ then we say that M∼ is an alter ego of M. At the finite level, because the topology on M is discrete, we can swap the topology to the other side, that is, hM ; G1, H1, R1, Ti is an alter ego of hM ; G2, H2, R2i provided hM ; G2, H2, R2, Ti is an alter ego of hM ; G1, H1, R1i. This symmetry fails in general if M is infinite. Given a structure M and an alter ego M∼ for M, we can set up a dual adjunction hD, E, e, εi between A := ISP(M) and X = IScP+(M∼ ). The claims below are taken from [6]. However, their verification is straightforward (in the case that M is a finite total algebra we refer to [4, Section 1.5] or to [23, Section 1.3] for the details). We define contravariant hom-functors D : A → X and E : X → A as follows: on objects D(A) = A(A, M), on morphisms D(f ) = − ◦ f ; and on objects on morphisms E(X) = X(X, M∼ ), E(ψ) = − ◦ ψ. The well-definedness of these functors is a consequence of our compatibility assumption. For each structure A ∈ A, the hom-set A(A, M) forms a closed A and so D(A) (the dual of A) is a member of X. Likewise, substructure of M∼ for each structure X ∈ X, the set X(X, M∼ ) forms a substructure E(X) of MX and so E(X) (the dual of X) is a member of A. For each A ∈ A and each X ∈ X, we define the evaluation maps eA: A → ED(A) and εX : X → DE(X) by eA(a)(x) := x(a), for all a ∈ A and all x ∈ A(A, M), and εX(x)(α) := α(x), for all x ∈ X and all α ∈ X(X, M∼ ). Then hD, E, e, εi is a dual adjunction be- tween A and X and e : idA → ED and ε : idX → DE are natural transforma- tions. Moreover, the construction of A and X via ISP and IScP+, respectively, ensures that the maps eA: A → ED(A) and εX : X → DE(X) are embeddings, for all A ∈ A and all X ∈ X. We note that here an embedding in A means 'isomorphism onto a substructure' while in X it means 'isomorphism onto a topologically closed substructure'. The following definitions mimic those given in [4] for the special case in which M is a finite total algebra. The alter ego M∼ yields a duality on A, or more briefly M∼ dualises M, if the map eA is an isomorphism, for all A ∈ A. If the natural map εX : X → DE(X) is an isomorphism for all X ∈ X, then we say that M∼ yields a co-duality on A, or M∼ co-dualises M; if the emphasis is on the structure with topology, we say that M yields a duality on X, or M dualises M∼ . The alter ego M∼ yields a full duality on A, or more briefly M∼ Vol. 00, XX Piggyback dualities revisited 7 fully dualises M, if it yields both a duality and a co-duality on A. In this case the functors D and E give a dual equivalence between the categories A and X. While the notions of dualising and fully dualising alter ego parallel exactly those given in [4], more care is needed when extending from algebras to struc- tures the concept of a strong duality as presented in [4, Chapter 3]. We recall that, in the restricted setting, proving that a duality is strong has been the primary tool for establishing that the duality is full. For structures in general, there are two competing definitions for a full duality yielded by an alter ego M∼ to be a strong duality: (1) every closed substructure of a non-zero power of M∼ is hom-closed, or equivalently is term-closed, and (2) M∼ is injective in the category X (with respect to embeddings). Fortunately, the two definitions coincide when M is a total structure, as it always will be in our theorems. Accordingly, we shall say that M∼ yields a strong duality on A, or that M∼ strongly dualises M, if (2) holds. (If partial operations are permitted in the type of M, then (2) is too strong, and (1) is the more appropriate definition. See the discussion in [13, Section 4.2].) So far, the definitions we have given are those applicable to duality theory for structures. They are not specific to dualities of piggyback type. Our principal theorems involve extensions to a much wider setting of conditions which underpin the piggybacking method for prevarieties of algebras. We now introduce the requisite definitions, allowing also for the topological versions we shall need in order to obtain both duality and co-duality theorems. Given a structure M we denote the set of total unary term functions of M by Clo1(M). (Note that a total unary term function of M may result from composing operations and partial operations in the type of M.) If M∼ is an alter ego of M, then the compatibility between M and M∼ guarantees that Clo1(M∼ ) ⊆ End(M) and Clo1(M) ⊆ End(M∼ ). The compatibility also guar- antees that the dual of the free structure in A on one generator is isomorphic to M∼ ; it follows that if M∼ dualises M, then Clo1(M) = End(M∼ ) -- see the discussion in [6, p. 13]. If M∼ fully dualises M, then the topologically closed M generated by Clo1(M∼ ) is End(M) (so Clo1(M∼ ) = End(M) substructure of M∼ if M is finite) -- the proof is a simple generalisation of that given for the finite case in [10, Proposition 4.3]; see also [11, Corollary 3.9] and [13, Lemma 4.3]. We next extend to structures the idea of a term-reduct of a total algebra. For further details see [13]. Let M = hM ; G, H, Ri be a structure. For n ∈ N, an n-ary relation r on M is conjunct-atomic definable from M if r = { (a1, . . . , an) ∈ M n M = m i=1 αi(a1, . . . , an) }, where v1, . . . , vn are distinct variables and each αi(v1, . . . , vn) is an atomic formula in the language hG, H, Ri involving variables from the set {v1, . . . , vn}. A structure M′ = hM ; G′, H ′, R′i is a structural reduct of M if each relation in dom(H ′) ∪ R′ is conjunct-atomic definable from M, each g ∈ G′ belongs to the 8 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. enriched partial clone of M and each h ∈ H ′ has an extension in the enriched partial clone of M; see [13, Lemma 2.5]. Now assume that M has a structural reduct M♭ in a prevariety B := ISP(N), where N = hN ; Gν , H ν, Rνi is some structure. Then every structure A in A = ISP(M) also has a structural reduct A♭ = hA; Gν , H ν, Rνi in B. We can define the structure A♭ in two equivalent ways: either syntactically via the conjunct-atomic formulas and terms that define M♭ from M or semantically via the embedding of A into a power of M. Either way, we have A♭ ∈ ISP(M♭). When we say that M∼ = hM ; G, H, R, Ti ♭ = hM ; Gν, H ν, Rν, Ti in IScP+(N∼), we mean that has a structural reduct M∼ ♭ belongs M′ := hM ; Gν , H ν, Rνi is a structural reduct of hM ; G, H, Ri and M∼ to IScP+(N∼) -- it is not sufficient to know that M′ belongs to ISP(N′), where N′ denotes N∼ minus its topology. If M∼ has a structural reduct in IScP+(N∼), then every member X of IScP+(M∼ ) has a structural reduct X♭ in IScP+(N∼). In fact, in the non-topological case, ♭ is a functor from ISP(M) to ISP(N), and in the topological case, ♭ is a functor from IScP+(M∼ ) to IScP+(N∼). Assume that the structure N∼ is an alter ego of the structure N and that N∼ is injective in Y := IScP+(N∼). (In fact, injectivity at the finite level would suf- fice.) The fact that N and N∼ are compatible guarantees that N is structurally equivalent to a total structure, and may without loss of generality be taken to be a total structure. This observation justifies an assumption we make in the next section. The notion of entailment will be important in our theory, as it is in the setting of quasivarieties of algebras [4, Section 2.4 and Chapter 8]. Assume that M∼ is an alter ego of a structure M and define A := ISP(M) and X := IScP+(M∼ ). Let r ⊆ M n, for some n ∈ N. We say that M∼ entails r if r forms a substructure of Mn and, for all A ∈ A, every X-morphism α : A(A, M) → M∼ preserves r. Likewise, we say that M entails r if r forms a topologically closed n and, for all X ∈ X, every A-morphism u : X(X, M∼ ) → M substructure of M∼ preserves r. In connection with entailment we shall encounter the diagonal ∆M := { (a, a) a ∈ M }, which forms a substructure of M2. Those familiar with traditional piggybacking will find the next definition unsurprising. Let M be a structure, let N be a set, and let r ⊆ N n, for some n ∈ N. For ω1, . . . , ωn : M → N , define (ω1, . . . , ωn)−1(r) := { (a1, . . . , an) ∈ M n (ω1(a1), . . . , ωn(an)) ∈ r }. Given Ω ⊆ N M , define maxM Ω−1(r) := { s ⊆ M n s ≤ Mn with s maximal in (ω1, . . . , ωn)−1(r) for some ω1, . . . , ωn ∈ Ω } (here, and subsequently, ≤ denotes 'is a substructure of'). Remark 2.1. As above, let ω be a map from M to N . We shall encounter in the statements of certain of our theorems various entailment conditions relating to relations contained in ker(ω) = (ω, ω)−1(∆N ). Since ∆M ⊆ ker(ω) Vol. 00, XX Piggyback dualities revisited 9 and ∆M forms a substructure of M2, we deduce that if r ⊆ ∆M and r forms a substructure of M2 that is maximal in ker(ω), then r = ∆M and M entails r trivially. Similar considerations apply when M is replaced by M∼ . It follows that, in all occurrences of entailment conditions involving binary relations maximal in ker(ω), we may without loss of generality restrict to relations not contained in ∆M . For a structure M∼ with a topology, we define maxM ∼ Ω−1(r) := maxM′ Ω−1(r), where M′ is M∼ with its topology removed. Lemma 2.2 guarantees that under Ω−1(∆N ) minimal assumptions the elements of sets maxM are closed substructures of Mn, so that it will make sense to demand that such relations are entailed by M. This fact is tacitly used in various theorem statements later on, and in those lemmas in Section 6 in which entailment by M arises. Ω−1(r) and of maxM ∼ ∼ Lemma 2.2 (Closed Maximal Relations Lemma). Let M = hM ; G, Ri be a total structure and assume that T is a topology on M such that each g ∈ G is continuous with respect to T. Let N = hN ; Ti be a topological space, and let r be a closed subset of Nn, for some n ∈ N. If ω1, . . . , ωn : M → N are continuous with respect to the topologies on M and N , then every relation s ⊆ M n that forms a substructure of Mn and is maximal in (ω1, . . . , ωn)−1(r) is topologically closed. Proof. Assume that ω1, . . . , ωn : M → N are continuous and let s be a subset of M n that forms a substructure of Mn and is maximal in (ω1, . . . , ωn)−1(r). Since r is closed, it follows that (ω1, . . . , ωn)−1(r) is also closed. Hence the closure s of s is a subset of (ω1, . . . , ωn)−1(r). Since s is a substructure of Mn, it is closed under each g in G. By continuity of g, the set s is also closed under g. Hence s forms a substructure of Mn. The maximality of s implies that s = s, whence s is topologically closed. (cid:3) The remainder of this section concerns notions which will not arise sub- sequently until we reach our co-duality theorems, and the results dependent on these. Because we are using the usual setting, in which the empty struc- ture is allowed in X = IScP+(M∼ ) (when the type of M∼ includes no nullary operations) but is not permitted in A = ISP(M), we need to take care with naming constants in the type of M∼ when considering co-dualities. Since it simplifies the discussion and is all we need here, we shall restrict ourselves to the situation where both M and M∼ are total structures. We say that a struc- ture A has named constants if the value of every constant unary term function of A is the value of a nullary term function. (Of course, to guarantee that A has named constants, it suffices to make the value of just one constant term function into the value of a nullary operation.) We require another definition. Given a total structure M = hM ; G, Ri and a ∈ M , we say that {a} forms a 10 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. complete one-element substructure of M if g(a, . . . , a) = a, for all g ∈ G and (a, . . . , a) ∈ r, for all r ∈ R. Define C1 to be the topologically closed substruc- ture of M∼ generated by the set of values of the constant unary term functions. If M∼ has constant unary term functions but no nullary operations, then the empty structure ∅∅∅ belongs to X and it is easy to see that E(∅∅∅) and E(C1) are isomorphic complete one-element structures, so it is impossible for M∼ to co-dualise M. (It also follows that DE(C1) is in a one-to-one correspondence with the complete one-element substructures of M.) Consequently, M∼ having named constants is a necessary condition for M∼ to co-dualise M. Indeed, this assumption guarantees that ε∅∅∅ : ∅∅∅ → DE(∅∅∅) is an isomorphism whenever ∅∅∅ belongs to X. The other conditions we give in our co-duality theorems are sufficient to prove that εX : X → DE(X) is an isomorphism, for all non-empty X in X. For a detailed discussion of named constants in the finite case, see [6, Lemma 6.1] and [4, Lemma 3.1.2]. In some of our results, we shall not need to postulate explicitly that M∼ has named constants: the following lemma ensures that this will follow from other assumptions we shall make. Lemma 2.3 (Named Constants Lemma). Let M and N be non-trivial to- tal structures (not necessarily of the same type), and let M∼ and N∼ be total structures that are alter egos of M and N, respectively. Define B := ISP(N) and Y := IScP+(N∼), denote the induced dual adjunction between B and Y by hH, K, k, κi, and assume that N∼ co-dualises N. Assume that M and M∼ have ♭ in Y, respectively. Then M∼ has named structural reducts M♭ in B and M∼ constants. Proof. Let a ∈ M and assume that M∼ has a constant unary term function ta with value a. Since M∼ is compatible with M, the map ta is an endomorphism of M and hence {a} forms a complete one-element substructure of M -- here we use the fact that each relation r in the type of M is non-empty. Hence {a} forms a complete one-element substructure of M♭. As M♭ is nontrivial and M♭ belongs to B = ISP(N), there exists ω ∈ B(M♭, N). As {a} forms a complete one-element substructure of M♭, it follows that ω(a) forms a com- plete one-element substructure of N. Let C1 denote the topologically closed substructure of N∼ generated by the set of values of the constant unary term functions of N∼. As we noted above, HK(C1) is in a one-to-one correspon- dence with the complete one-element substructures of N and so HK(C1) is non-empty. Since N∼ co-dualises N, the map εC1 : C1 → HK(C1) is an iso- morphism and consequently C1 is non-empty. But N∼ has named constants, again since N∼ co-dualises N, and so the type of N∼ includes a nullary opera- ♭ is a structural reduct of M∼ in Y, it follows that M∼ has a tion σ. Since M∼ nullary term function sa with value σM . Hence, M∼ has named constants. (In particular, ta(sa) is a nullary term function of M∼ with value a.) (cid:3) ∼ ♭ Vol. 00, XX Piggyback dualities revisited 11 3. The Piggyback Duality Theorems Now we set up the framework of Basic Assumptions, within which we shall develop our piggyback duality and co-duality theorems, and the associated notation. The benefits in terms of simplicity we gain by working uniformly with the Basic Assumptions outweigh the extra generality gained at the margins by tailoring each theorem to a minimal set of assumptions. For example, injectivity of N∼ in Y is explicitly needed in the Piggyback Duality Theorem but not in the Piggyback Co-duality Theorem, whereas injectivity of N in B is needed in the latter theorem but not the former. Thus the conditions laid out in the Basic Assumptions are a little stronger than we strictly need in every case. For this reason, in Section 6, where we give the proofs of our theorems, we shall include in each lemma only the conditions needed in the proof. Basic Assumptions 3.1. • M and N are non-trivial total structures (not necessarily of the same type), and M∼ and N∼ are total structures that are alter egos of M and N, respectively. • A := ISP(M) and X := IScP+(M∼ ) and the induced dual adjunction be- tween A and X is denoted by hD, E, e, εi. • B := ISP(N) and Y := IScP+(N∼) and the induced dual adjunction be- tween B and Y is denoted hH, K, k, κi. In addition, (i) N∼ yields a full duality on B; (ii) N∼ is injective in Y and N is injective in B. We shall add additional conditions as needed. In particular we shall, theorem by theorem, impose appropriate relationships between M and N or between M∼ and N∼. We adopt the following notation. Let Ω ⊆ N M and let E ⊆ M M . We write Ω ◦ E := { ω ◦ u ω ∈ Ω & u ∈ E } ⊆ N M . If Ω = {ω}, then we write simply ω ◦ E. Operating under our Basic Assumptions, we now briefly review the ideas behind the piggybacking method, as presented in [16, 17, 14]. We assume that the structure M, for which we wish to find a dualising alter ego, has a structural reduct M♭ in the category B, for which we already have a full duality. An alter ego M∼ will dualise M provided that, for each A ∈ A, we can find a one-to-one map α 7→ dα from ED(A) to KH(A♭) that commutes with the evaluations, that is, d− ◦ eA = kA♭ ; see Figure 1. We now come to the key idea underlying piggybacking: how we might construct the map d− such that d− ◦ eA = kA♭ , as in the commuting triangle diagram in Figure 1. We seek to use an element ω, or a set Ω of elements, in B(M♭, N) to link ED(A) to KH(A), for each A ∈ A. Following the usage in [14] we shall refer to such maps as carriers. Explicitly, given ω ∈ B(M♭, N) 12 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. eA A ED(A) kA♭ d− KH(A♭) Figure 1. The commuting triangle: the equation d− ◦ eA = kA♭ and A ∈ A, we may define a map ΦA ω : A(A, M) → B(A♭, N) by ΦA ω (x) := ω ◦ x, for all x ∈ A(A, M). In a best-case scenario there will exist a single carrier ω making ΦA ω surjective for each A ∈ A. In that case, given α : A(A, M) → M , we have a chance of "defining" dα : B(A♭, N) → N by dα(ω ◦ x) = ω(α(x)), for x ∈ A(A, M). The success of this construction then depends on our being able to select ω and the alter ego M∼ in such a way that d− is well defined and one-to-one. Historically, the piggybacking technique was devised in two stages. What we call single-carrier piggybacking was formulated by Davey and Werner [16, 17]. In [14], Davey and Priestley initiated the theory of multisorted natural dualities and thereby showed how to dispense with the restriction that there be a single carrier, in the following way. Suppose that we could find some subset Ω of B(M♭, N) such that, for each A, the family {ΦA ω }ω∈Ω is jointly surjective in the sense that the union of the images of the maps ΦA ω , for ω ∈ Ω, is the whole of B(A♭, N). Then we may still try to "define" dα as before, but now with ω, as well as x, varying. Presuming joint surjectivity, we would then need to choose Ω and the alter ego so that d− is well defined (and this is more of an issue than when there is a single ω) and one-to-one. We are finally ready to state our piggyback duality and co-duality theorems. We state each theorem first for the single-carrier case. We single this out for two reasons. Firstly, the conditions we give for single-carrier piggybacking to work are not merely specialisations of those for the general case. Secondly, our strong-duality results rely only on the single-carrier versions of the duality and co-duality theorems. So too do our results on semilattices in Section 5. We note that in Condition (3)(ii)(a) of Theorems 3.2 and Theorem 3.4 we could insert the restriction that r * ∆M ; recall Remark 2.1. Similar restrictions can be imposed in other theorems, likewise. Theorem 3.2 (Single-carrier Piggyback Duality Theorem). The setting for this theorem is provided by the Basic Assumptions. Assume that M has a structural reduct M♭ in B and let ω ∈ B(M♭, N). Assume that N∼ = hN ; Gν , Rν, Ti. Then M∼ dualises M provided (0) -- (3) below all hold. Vol. 00, XX Piggyback dualities revisited 13 (0) ω is continuous with respect to the topologies on M∼ and N∼. (1) ω ◦ Clo1(M∼ ) separates the structure M♭. (2) One of the following conditions holds: (i) N∼ is purely relational, or (ii) M∼ has a structural reduct M∼ (i) The structure M∼ entails every relation in maxM{ω}−1(r), for each ♭ in Y with ω ∈ Y(M∼ ♭, N∼). relation r ∈ Rν , and (3) (ii) either (a) M∼ entails each binary relation r on M which forms a substructure of M2 that is maximal in ker(ω), or (b) ω ◦ Clo1(M) separates the points of M . In the above theorem, Condition (3)(ii)(b) implies Condition (3)(ii)(a), but it is convenient to have (3)(ii)(b) recorded explicitly. (Indeed, it is an almost trivial exercise to show that if ω ◦ Clo1(M∼ ) separates the points of M , then each binary relation on M that forms a substructure of M2 and is contained in ker(ω) is contained in ∆M . A maximal such relation must therefore be ∆M , which is trivially entailed by any alter ego.) We now give the Piggyback Duality Theorem in its general form. It is worth drawing attention to the structure of the statement of the theorem. As occurs in Theorem 3.2 too, we have a list of conditions (0) -- (3), which together with the initial assumptions yield the conclusion. In Condition (2) there is a dichotomy: (2)(i) covers the case in which N∼ is purely relational, whereas (2)(ii) allows for operations (albeit at most unary) in N∼ but then imposes a rather stringent condition on how M∼ is related to N∼. An analogous dichotomy appears in Condition (2) in the Piggyback Co-duality Theorem. Theorem 3.3 (Piggyback Duality Theorem). The setting for this theorem is provided by the Basic Assumptions. Assume that M has a structural reduct M♭ in B and let Ω be a subset of B(M♭, N). Assume that N∼ = hN ; Gν , Rν, Ti. Then M∼ dualises M provided either Ω is a singleton set {ω} and the conditions for Theorem 3.2 are met or (0) -- (3) below all hold. (0) Ω is finite and each ω ∈ Ω is continuous with respect to the topologies on M∼ and N∼. (1) Ω ◦ Clo1(M∼ ) separates the structure M♭. (2) One of the following conditions holds: (i) N∼ is purely relational, or (ii) M∼ has a structural reduct M∼ and every operation in Gν is unary or nullary. ♭ in Y, each ω ∈ Ω belongs to Y(M∼ ♭, N∼), (3) (i) The structure M∼ entails every relation in maxM Ω−1(r), for each re- lation r ∈ Rν , and (ii) M∼ entails every relation in maxM Ω−1(∆N ). We now move on to consider co-dualities. We seek, mutatis mutandis, to mimic the strategy we described in our brief survey of traditional piggybacking. 14 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. If M∼ has a structural reduct M∼ X♭ in Y, and if ω ∈ Y(M∼ ♭ in Y, then each X ∈ X has a structural reduct ♭, N∼) and X ∈ X, we may define a map ω : X(X, M∼ ) → Y(X♭, N∼) ΨX by ΨX ω (α) := ω ◦ α, for all α ∈ X(X, M∼ ). The usual calculations show that Ψ− ω is a natural transformation between the set-valued hom-functors X(−, M∼ ) and Y((−)♭, N∼) out of X, and similarly Φ− ω , as defined earlier, is a natural transformation between A(−, M) and B((−)♭, N). Theorem 3.4 (Single-carrier Piggyback Co-duality Theorem). The setting for this theorem is provided by the Basic Assumptions with the additional assump- tion that M∼ has named constants. Assume that M∼ has a structural reduct ♭, N∼). Assume that N = hN ; Gν , Rνi. Then M∼ M∼ co-dualises M provided (1) -- (3) below all hold. ♭. (1) ω ◦ Clo1(M) separates the structure M∼ (2) One of the following conditions holds: ♭ in Y and let ω ∈ Y(M∼ (i) N is purely relational, or (ii) M has a structural reduct M♭ in B such that ω ∈ B(M♭, N). (3) The structures M = hM ; G1, R1i and M∼ = hM ; G2, R2, Ti satisfy the following conditions: (i) (a) the operations in G2 are continuous, and (b) the structure M entails every relation in maxM {ω}−1(r), for each ∼ r ∈ Rν . (ii) Either (a) M entails each binary relation r on M which forms a substructure of M∼ 2 that is maximal in ker(ω), or (b) ω ◦ Clo1(M∼ ) separates the points of M . Regarding Condition (3)(i)(a) we note that continuity of the operations in G2 is not a consequence of M∼ 's being an alter ego of M, as one might perhaps be tempted to assume. Theorem 3.5 (Piggyback Co-duality Theorem). The setting for this theorem is provided by the Basic Assumptions with the additional assumption that M∼ ♭ in Y and has named constants. Assume that M∼ has a structural reduct M∼ ♭, N∼). Assume that N = hN ; Gν , Rνi. Then M∼ let Ω be a subset of Y(M∼ co-dualises M provided either Ω is a singleton set {ω} and the conditions of Theorem 3.4 are met or (1) -- (3) below all hold. (1) Ω ◦ Clo1(M) separates the structure M∼ (2) One of the following conditions holds: ♭. (i) N is purely relational, or (ii) M has a structural reduct M♭ in B, each ω ∈ Ω belongs to B(M♭, N), and every operation in Gν is unary or nullary. (3) The structures M = hM ; G1, R1i and M∼ = hM ; G2, R2, Ti satisfy the following conditions: Vol. 00, XX Piggyback dualities revisited 15 (i) (a) the operations in G2 are continuous, and (b) the structure M entails every relation in maxM r ∈ Rν ; (ii) M entails every relation in maxM ∼ Ω−1(∆N ). Ω−1(r), for each ∼ The Piggyback Co-duality Theorem can be applied in two quite different ways. It can be used to prove that a duality given by the Piggyback Duality Theorem is in fact full. It is used in this way in the D-based Piggyback Strong Duality Theorem 4.4 and therefore also in its specialisation to Ockham algebras, Theorem 4.6. Alternatively, it can be used in a topology-swapping situation, where we take a duality that has been obtained via the Piggyback Duality Theorem and swap the topology from the 'relational' category to the 'algebraic' category. This leads to the D-based and S-based Piggyback Co- duality Theorems 4.2 and 5.2. In certain applications, in particular to D-based prevarieties, it may be convenient to make use of the following consequences of the piggyback duality and co-duality theorems, stated as a single corollary. In it, we do as we shall do subsequently and flag the duality and co-duality versions with (D) and (coD). The corollary follows from Lemma 6.3. Corollary 3.6. The setting is that of the Basic Assumptions. (D) Assume that M has a structural reduct M♭ in B and let Ω be a subset of B(M♭, N). Assume that N∼ = hN ; Gν , Rν, Ti. Then M∼ dualises M if Conditions (0), (1), (2) and (3)(i) of Theorem 3.2 (with Ω = {ω}) or of Theorem 3.3 hold and also (3)(ii)′ R contains a reflexive, antisymmetric binary relation ⊑. ♭ in Y and let Ω be a subset of Y(M∼ (coD) Assume that M∼ has named constants. Assume that M∼ has a structural ♭, N∼). Assume that N = reduct M∼ hN ; Gν, Rνi. Then M∼ co-dualises M if Conditions (1), (2) and (3)(i) in either Theorem 3.4 (with Ω = {ω}) or of Theorem 3.5 hold and also (3)(ii)′ R contains a reflexive, antisymmetric binary relation ⊑. When M has a plentiful supply of both unary term functions and endo- morphisms, we can combine the single-carrier theorems 3.2 and 3.4 to obtain various piggyback strong duality theorems -- we present three such under a common umbrella. Parts (I) and (II) of Theorem 3.7 are typically applied when we are piggybacking on Priestley duality for distributive lattices with or without bounds. Part (III) is typically applied when we are piggybacking on Hofmann -- Mislove -- Stralka duality for semilattices with or without bounds [21] (see also [15, 2.4 (p. 157)]). Note that, in the following two results, Lemma 2.3 guarantees that we do not need to add any assumptions about named constants. Theorem 3.7 (Piggyback Strong Duality Theorem). The setting for this the- orem is the Basic Assumptions. Assume that M and M∼ have structural reducts M♭ in B and M∼ ♭ in Y, respectively, and let ω ∈ B(M♭, N) ∩ Y(M∼ ♭, N∼). 16 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. Consider the following three sets of additional conditions. (I) (II) (III) (1) N is a total algebra, N∼ = hN ; Gν, Rν, Ti, and (i) ω ◦ Clo1(M∼ ) separates the points of M , (2) (ii) ω ◦ Clo1(M) separates the structure M∼ (iii) M∼ entails every relation in maxM{ω}−1(r), for each r ∈ Rν . ♭, and (1) N = hN ; Gν , Rνi, N∼ is a total algebra, and (2) (i) ω ◦ Clo1(M∼ ) separates the structure M♭, (ii) ω ◦ Clo1(M) separates the points of M , and (iii) if Rν 6= ∅, the operations in the type of M∼ are continuous and M∼ entails every relation in maxM{ω}−1(r), for each r ∈ Rν . (1) Both N and N∼ are total algebras. (2) (i) ω ◦ Clo1(M∼ ) separates the points of M , and (ii) ω ◦ Clo1(M) separates the points of M . Assume that (I), (II) or (III) applies. Then (a) M∼ fully dualises M, (b) M is injective in A and M∼ is injective in X, and consequently the duality is strong, and (c) for all A ∈ A and all X ∈ X, D(A)♭ ∼= H(A♭) and E(X)♭ ∼= K(X♭) via the maps ΦA ω and ΨX ω , respectively. The following result is an immediate consequence of the Piggyback Strong Duality Theorem. In its statement, we need to swap a topology between two to- tal structures with the same underlying set. Consequently, we need to suspend briefly the notation M∼ for an alter ego of a structure M. Given a structure M = hM ; G, H, Ri and a topology T on M , we define MT := hM ; G, H, R, Ti. We will also have occasion to use MT as an alternative notation for MT. Note that if MT 2 are topological structures with the same underlying set M , then MT 2 is an alter ego of M1 if and only if MT 1 is an alter ego of M2. 1 and MT 1 and NT 2 be total topological structures and assume that MT Theorem 3.8 (Two-for-one Piggyback Strong Duality Theorem). Let MT 1 , MT 2 , NT 2 and NT 2 are alter egos of M1 and N1, respectively. For i ∈ {1, 2}, define Ai := ISP(Mi), Bi := ISP(Ni), Xi := IScP+(MT i ). Assume that NT 1 injective in Y1 and N2 injective in B2, and that NT 2 injective in Y2 and N1 injective in B1. Assume that M1 and M2 have structural reducts M♭ 1)T in Y1 and (M♭ 2, N2). 2)T in Y2. Let ω be a continuous map in B1(M♭ 1 fully dualises N2 with NT 2 fully dualises N1 with NT i ) and Yi := IScP+(NT 1, N1) ∩ B2(M♭ 1 and M♭ 2 with (M♭ Consider the following two sets of additional conditions. (I) (1) N1 is a total algebra and N2 = hN ; Gν, Rνi; (i) ω ◦ Clo1(M2) separates the points of M , (2) (ii) ω ◦ Clo1(M1) separates the structure M♭ 2, and Vol. 00, XX Piggyback dualities revisited 17 (iii) both M2 and MT 2 entail every relation in maxM1{ω}−1(r), for each r ∈ Rν . (II) (1) Both N1 and N2 are total algebras; (2) (i) ω ◦ Clo1(M2) separates the points of M , and (ii) ω ◦ Clo1(M1) separates the points of M . Assume that (I) or (II) applies. Then (a) MT 2 fully dualises M1 with M1 injective in B1 and MT 2 injective in X2, (b) MT 1 fully dualises M2 with M2 injective in B2 and MT 1 injective in X1, and (c) the isomorphisms in the base categories given in Theorem 3.7(c) apply to both the duality between A1 and X2, with base categories B1 and Y2, and the duality between A2 and X1, with base categories B2 and Y1. In practice, when applying the Two-for-one Piggyback Strong Duality The- Indeed, if N1 = 2 i are finite compatible total structures with 2 strongly 1 i and N2 = hN ; Gν 2 finite and both N1 and N2 have named constants, then NT 2 strongly dualises N1. orem, it suffices to know that NT hN ; Gν Rν dualises N1 if and only if NT 1 , Rν 1 and Rν 2, Rν 1 strongly dualises N2, by [6, Theorem 6.9]. 4. Applications to distributive-lattice-based algebras In this section we consider piggybacking on Priestley duality, demonstrating in particular how the original piggyback duality theorems in [16, 17, 14] for algebras fit into the general framework for piggybacking developed in this paper. Our co-duality results, by contrast, are wholly new. We shall restrict attention to bounded distributive lattices. Analogous theorems are available when one bound is omitted from the type, or both bounds are omitted. As before, D will denote the variety of bounded distributive lattices. This is generated, as quasivariety, by D, the two-element algebra in D. We take N to be D and its alter ego N∼ to be the discretely topologised two-element chain D∼ = h{0, 1}; 6, Ti, so that Y := IScP+(D∼) is the category P of Priestley spaces. Here all the conditions demanded of N, N∼, B = ISP(N) and Y = IScP+(N∼) in the Basic Assumptions are satisfied. We say that a structure M is D-based if it has a structural reduct M♭ in D. The D-based Piggyback Duality Theorem 4.1, which is well known and has been used often in the case that M is a total algebra, follows immediately from Corollary 3.6(D). In the single carrier case Ω = {ω} (and similarly in Theorem 4.2), Condition (2) can be restricted to relations r * ∆M ; recall Remark 2.1. Theorem 4.1 (D-based Piggyback Duality Theorem). Let M be a D-based total structure with structural reduct M♭ in D. Then an alter ego M∼ of M dualises M provided that there is a finite subset Ω of D(M♭, D) such that 18 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. (0) each ω ∈ Ω is continuous with respect to the topologies on M∼ and D∼, (1) Ω ◦ Clo1(M∼ ) separates the points of M , and (2) M∼ entails every relation in maxM Ω−1(6). The theorem tells us that every finite D-based total structure M is dual- isable: choose Ω = D(M♭, D), then M∼ := hM ; maxM Ω−1(6), Ti dualises M. In fact, the NU Duality Theorem for total structures [6, Theorem 4.10] already ′ := hM ; S(M2), Ti, where S(M2) is the set tells us that M is dualised by M∼ of all non-empty substructures of M2. The advantage of the D-based Piggy- back Duality Theorem over the NU Duality Theorem is that in general the set maxM Ω−1(6) is much smaller than the set S(M2). A slightly perplexing consequence of the Piggyback Duality Theorem in the D-based case (and of the NU Duality Theorem) is that if M1 = hM ; G, R1i and M2 = hM ; G, R2i are finite D-based total structures with the same set G of fundamental operations, then there is a single alter ego, namely the alter ego M∼ := hM ; maxM Ω−1(6), Ti, that dualises both M1 and M2. We turn now to the co-duality theorem. Here we piggyback on the Bana- schewski duality between ordered sets and Boolean topological bounded dis- tributive lattices [1] (see also [8]). Let 2 = h{0, 1}; 6i be the two-element chain and 2∼ = h{0, 1}; ∨, ∧, 0, 1, Ti be the two-element bounded distributive lattice endowed with the discrete topology, so that Q := ISP(2) and DT := IScP+( 2∼) are, respectively, the cate- gory of non-empty ordered sets -- an extremely easy exercise -- and the category of non-trivial Boolean-topological bounded distributive lattices -- a non-trivial fact due to Numakura [22], see also [5, Example 8.2]. Banaschewski's duality theorem tells us that 2∼ fully dualises 2. Moreover, 2 is injective in Q and 2∼ is injective in DT. The following result is an immediate consequence of the Piggyback Co- duality Theorem 3.5. Theorem 4.2 (D-based Piggyback Co-duality Theorem). Let M = hM ; G, Ri be a total structure with named constants, let T be a Boolean topology on M and assume that MT := hM ; G, R, Ti has a structural reduct M♭ T in DT and that the operations in G are continuous. Let M′ be a total structure that is compatible with MT. Then M′ dualises MT provided that there is a subset Ω of DT(M♭ (1) Ω ◦ Clo1(M′) separates the points of M , and (2) M′ entails every relation in maxM Ω−1(6). T, 2∼) such that Corollary 4.3. Every Boolean-topological D-based total structure MT is dual- isable. Proof. Choose Ω = DT(M♭ MT by Theorem 4.2. T, 2∼). Then M′ := hM ; maxM Ω−1(6), Ti dualises (cid:3) Our two final theorems in this section are immediate corollaries of the cor- responding general results. Vol. 00, XX Piggyback dualities revisited 19 Theorem 4.4 (D-based Piggyback Strong Duality Theorem). Let M be a D-based total structure with structural reduct M♭ in D, let M∼ be an alter ego of M and define A := ISP(M) and X := IScP+(M∼ ). Assume there is an order ♭ := hM ; 6, Ti relation 6 that is conjunct-atomic definable from M∼ such that M∼ is a Priestley space, and there exists ω ∈ D(M♭, D) ∩ P(M∼ (1) M∼ entails each binary relation r on M which forms a substructure of M2 ♭, D∼) such that that is maximal in (ω, ω)−1(6), (2) ω ◦ Clo1(M∼ ) separates the points of M , and (3) ω ◦ Clo1(M) separates the structure M∼ ♭, that is, if a (cid:10) b in M∼ ♭, then there exists t ∈ Clo1(M) such that ω(t(a)) = 1 and ω(t(b)) = 0. Then the conclusions (a) -- (c) of Theorem 3.7 hold. Theorem 4.5 (D-based Two-for-one Piggyback Strong Duality Theorem). Let M1 and M2 be compatible total structures and let T be a Boolean topology on M such that both MT 2 are topological structures, and assume that 1 and MT (i) there are binary term functions ∨ and ∧ and nullary term functions 0 and 1)T := hM ; ∨, ∧, 0, 1, Ti is a Boolean topological 1 on M1 such that (M♭ bounded distributive lattice, and (ii) there is an order relation 6 that is conjunct-atomic definable from M2 such that (M♭ Assume that there exists a map ω ∈ D(M♭ 2)T := hM ; 6, Ti is a Priestley space. 1, D) ∩ P((M♭ 2)T, D∼) such that (1) ω ◦ Clo1(M2) separates the points of M , (2) ω ◦ Clo1(M1) separates the structure M♭ 2, that is, if a (cid:10) b in M♭ 2, then there exists t ∈ Clo1(M1) such that ω(t(a)) = 1 and ω(t(b)) = 0, and (3) M2 and MT 2 entail every binary relation r on M which forms a substruc- ture of M2 that is maximal in (ω, ω)−1(6). Then the conclusions (a) -- (c) of Theorem 3.8 hold. Leaving aside strongness of the co-duality involved, which has not been recognised before, Theorems 4.4 and 4.5 explain the observed behaviour of certain algebras, for example Stone algebras, double Stone algebras and De Morgan algebras, whose natural and Priestley duals 'coincide'. A discussion of this phenomenon of coincidence, in the context of arbitrary finitely generated quasivarieties of D-based algebras, is given by Cabrer and Priestley in [2, Section 2]; see in particular [2, Corollary 2.4]. They showed more generally how, within the ambit of D-based piggybacking, to pass from the natural dual space D(A) of an algebra A to H(A♭). The motivation in [2], in the context of a study of coproducts, was to harness simultaneously the categorical virtues of a natural duality and the pictorial nature of Priestley duality. This isolated illustration -- and we could have provided others -- indicates that the usefulness of piggybacking extends beyond the derivation of natural dualities, whether strong or not. 20 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. Applications to Ockham algebras. An algebra A = hA; ∨, ∧, ¬, 0, 1i is an Ockham algebra if A♭ := hA; ∨, ∧, 0, 1i belongs to D, and ¬ is a dual endomorphism of A♭, that is, ¬0 = 1, ¬1 = 0, and ¬ satisfies the De Morgan laws: ¬(a ∨ b) = ¬a ∧ ¬b and ¬(a ∧ b) = ¬a ∨ ¬b. The variety O of Ockham algebras has provided a valuable example for a num- ber of developments in duality theory. Below we shall show how Theorem 4.5 can be applied to O. Along the way, we recapture Davey and Werner's piggy- back duality for O, using essentially their argument; this duality was originally obtained, without piggybacking, by Goldberg [19]. We first recall some well-known facts. Now let γ : N0 → N0 be the suc- cessor function: γ(n) := n + 1 and let c denote Boolean complementation on {0, 1}. Define M1 := h{0, 1}N0 ∨, ∧, ¬, 0, 1i, where ∨ and ∧ are defined pointwise, 0 and 1 are the constant maps onto 0 and 1, respectively, and, for all a ∈ {0, 1}N0 we have ¬(a) := c ◦ a ◦ γ. Thus, ¬ is given by shift left and then negate; for example, ¬(0110010 . . . ) = (001101 . . . ). Here, and subsequently, we write elements of {0, 1}N0 as binary strings. Then M1 is an Ockham alge- bra. Moreover, M1 is subdirectly irreducible and O = ISP(M1) ([25]; see also [19]). The topology on {0, 1}N0 will be the product topology T coming from the discrete topology on {0, 1}. It is an easy exercise to see that the operations ∨, ∧ and ¬ on M1 are continuous with respect to T. Hence MT 1 is a topological algebra. We now set up a structure M2 = h{0, 1}N0; u, 4i that is compatible with M1. Let u : {0, 1}N0 → {0, 1}N0 be the left shift operator, given by u(a) := a ◦ γ. Thus, for example, u(0110010 . . . ) = (110010 . . . ). Then u ∈ End(M1) and u is clearly continuous. Define 4 to be the alternating order on {0, 1}N0, that is, for all a, b ∈ {0, 1}N0, a 4 b ⇐⇒ a(0) 6 b(0) & a(1) > b(1) & a(2) 6 b(2) & · · · . Since 4 forms a subalgebra of M2 1, it follows that M2 is compatible with M1. It is an elementary exercise to show that 4 is closed in the product topology on {0, 1}N0 × {0, 1}N0, and so MT 1)T := h{0, 1}N0; ∨, ∧, 0, 1, Ti is a Boolean-topological bounded distributive lattice and (M♭ 2)T := h{0, 1}N0; 4, Ti is a Priestley space, conditions (i) and (ii) of Theo- rem 4.5 are satisfied. 2 is a topological structure. Since (M♭ Define ω := π0 : {0, 1}N0 → {0, 1}. Clearly, π0 ∈ D(M♭ 1, D) ∩ P((M♭ The set π0 ◦ Clo1(M2) separates the points of M = {0, 1}N0: a, b ∈ {0, 1}N0 with a 6= b, then 2)T, D∼). indeed, let a 6= b =⇒ (∃n ∈ N0) a(n) 6= b(n) =⇒ (∃n ∈ N0) un(a)(0) = (a ◦ γn)(0) 6= (b ◦ γn)(0) = un(b)(0) =⇒ (∃n ∈ N0) (π0 ◦ un)(a) 6= (π0 ◦ un)(b). Vol. 00, XX Piggyback dualities revisited 21 As u is in Clo1(M2), so is un. Hence π0 ◦ Clo1(M2) separates the points of M , that is, Condition (1) of Theorem 4.5 holds. We now show that Condition (2) of Theorem 4.5 holds. We must prove that 2, then there π0 ◦ Clo1(M1) separates the structure M♭ exists t ∈ Clo1(M1) such that π0(t(a)) = 1 and π0(t(b)) = 0. We have 2, that is, if a 64 b in M♭ a 64 b in M♭ 2 ⇐⇒ (∃n ∈ N0)(a(n) = 1 & b(n) = 0, a(n) = 0 & b(n) = 1, n even n odd =⇒ (∃n ∈ N0) ¬n(a)(0) = 1 & ¬n(b)(0) = 0 =⇒ (∃n ∈ N0) π0(¬n(a)) = 1 & π0(¬n(b)) = 0 as required, with t(v) := ¬n(v). Finally, to establish Condition (3) of Theorem 4.5, we must find the bi- 1 that are maximal in nary relations r on M which form substructures of M2 (π0, π0)−1(6). We have (π0, π0)−1(6) = { (a, b) ∈ ({0, 1}N0)2 a(0) 6 b(0) }. Let r be a subalgebra of M2 1 with r ⊆ (π0, π0)−1(6). Then (a, b) ∈ r =⇒ (∀n ∈ N0) (¬n(a), ¬n(b)) ∈ r =⇒ (∀n ∈ N0) ¬n(a)(0) 6 ¬n(b)(0) =⇒ a(0) 6 b(0) & a(1) > b(1) & a(2) 6 b(2) & · · · ⇐⇒ a 4 b. 1 and 4 ⊆ (π0, π0)−1(6), and Thus r ⊆ 4. Since 4 forms a subalgebra of M2 r is a maximal such relation, it follows that r = 4 is the unique such rela- tion. (For related results, see [2, Section 3], which applies to finitely generated D-based quasivarieties and more particularly [14, Lemma 3.5 and 3.6], con- cerning Ockham algebra quasivarieties.) Since 4 is part of the type of M2, it is completely trivial that M2 and MT 2 entail 4, whence Condition (3) of Theorem 4.5 holds. We therefore have the following theorem. Theorem 4.6 (Two-for-One Strong Duality Theorem for Ockham algebras). Let M1 and MT 2 be as defined above. Then conclusions (a) -- (c) of Theorem 3.8 hold. In particular, MT 1 := h{0, 1}N0; ∨, ∧, ¬, 0, 1, Ti strongly dualises M2. 2 := h{0, 1}N0; u, 4, Ti strongly dualises M1 and MT Not surprisingly, our proof that MT 2 dualises the algebra M1 is essen- tially the same as the original Davey -- Werner piggyback-based proof. We note that Goldberg [19, Corollary 9] proved fullness via a direct calculation that X ∼= DE(X) for each X ∈ IScP+(MT 2 ). Earlier, Goldberg [18, Theorem 4.4] had proved that M1 is injective in O, whence the Injectivity Lemma (see [4, Lemma 3.2.10]) tells us that Goldberg's duality is strong. Our piggyback- based proof that the duality is strong is new. Conclusion (c) of Theorem 3.8 tells us, in particular, that the natural and Priestley duals of each Ockham 22 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. algebra A ∈ ISP(M1) and each Ockham space X ∈ IScP+(MT 2 ) coincide; more formally, D(A)♭ ∼= H(A♭) and E(X)♭ ∼= K(X♭). These isomorphisms were first proved by Goldberg [19, Theorem 8]. The fact that we can swap the topology and conclude that the topological algebra MT 1 strongly dualises the structure M2 is new. Remark 4.7. We have focused on the variety O rather than on its finitely generated subquasivarieties for several reasons. First of all, O provides a good example of the applicability of our machinery in the non-finitely generated set- ting. Secondly, natural dualities for subquasivarieties of O generated by finite subdirectly irreducible algebras (in particular those which are also varieties) have been very extensively studied, both as a tool for investigating Ockham algebras and perhaps more importantly as test case examples during the evo- lution of natural duality theory; see [4, Chapter 4] and [19, 16, 14, 8]. The most interesting in the present context are those Ockham varieties to which Theorems 4.4 and 4.5 apply and we thereby obtain new information. For these varieties both the duality, and the co-duality obtained by topology-swapping, are strong, and the forgetful functors on both sides give isomorphisms. Vari- eties coming under this umbrella include De Morgan algebras, Stone algebras and MS-algebras (and more generally any variety generated by a finite subdi- rectly irreducible Ockham algebra in one of the three infinite classes described in [14, Lemma 3.9]). Kleene algebras are not covered by Theorems 4.4 and 4.5. See the references cited above for further details of the varieties concerned and of their piggyback natural dualities. 5. Applications to semilattice-based algebras We concentrate in this section on the variety S of meet-semilattices with 1, but note (see [12] and Example 5.6) that simple modifications produce corre- sponding results for meet-semilattices, meet-semilattices with 0, and bounded meet-semilattices. When producing a piggyback duality theorem based on an underlying meet- semilattice structure, we are forced to have semilattice operations in the clones of M and M∼ ; denote these by ∧ and ⊓, respectively. Since M∼ is an alter ego of M, the operation ⊓ must be a homomorphism from M2 to M; in particular, we must have (a ∧ c) ⊓ (b ∧ d) = (a ⊓ b) ∧ (c ⊓ d), for all a, b, c, d ∈ M . But it follows easily from this that ⊓ = ∧; indeed, a ∧ b = (a ∧ b) ⊓ (b ∧ a) = (a ⊓ b) ∧ (b ⊓ a) = a ⊓ b. Thus we shall assume that in the clone of M there is a meet operation, ∧, that is continuous with respect to the topology on M∼ and is a homomorphism from M2 to M, and we shall assume that ∧ is part of the structure on M∼ . Let S be the two-element meet-semilattice with 1, let S∼ be S with the discrete topology added, and take N to be S and N∼ to be S∼. Then the Vol. 00, XX Piggyback dualities revisited 23 base categories S := ISP(S) and Y := IScP+(S∼) are, respectively, the cate- gory of meet-semilattices with 1 -- an extremely easy exercise -- and the cate- gory of Boolean-topological meet-semilattices with 1 -- see [15, 2.4 (SEP)] for a straightforward proof, and [5, Example 2.6 and Theorem 4.3] to see the result from a more general perspective. Moreover, the parts of the Basic Assumptions concerning the base categories are satisfied. We shall apply the piggyback duality and co-duality theorems in their single- carrier versions. Our first result generalises Davey, Jackson, Pitkethly and Talukder's Semilattice Piggyback Duality Theorem [12, Theorem 7.1] from semilattice-based algebras to semilattice-based total structures. Note that the semilattice operations in M♭ and M∼ Theorem 5.1 (S-based Piggyback Duality Theorem). Let M be a total struc- ture with a structural reduct M♭ in S, let M∼ be a structure with a structural ♭ in Y and assume that M∼ is an alter ego of M. Then M∼ dualises reduct M∼ ♭, S∼) such that M provided there exists ω ∈ Y(M∼ (1) ω ◦ Clo1(M∼ ) separates the points of M , and (2) M∼ entails each binary relation r on M which forms a substructure of M2 ♭ must agree. that is maximal in ker(ω). Proof. Since M♭ is a total algebra (in fact a meet-semilattice with 1), the assumptions guarantee that Conditions (0), (1), (2)(ii), (3)(i) and (3)(ii)(a) of the Single-carrier Piggyback Duality Theorem 3.2 hold (with (3)(i) holding vacuously). Hence M∼ dualises M. (cid:3) The S-based Piggyback Co-duality Theorem, which is new, follows in a similar way to its non-co counterpart. Note that Lemma 2.3 ensures that we do not need to mention named constants in this theorem. In fact, since the ♭ belongs to Y, base category Y has 1 as a nullary operation in its type and M∼ it follows that M∼ has a nullary operation and so has named constants. Theorem 5.2 (S-based Piggyback Co-duality Theorem). Let M be a total structure with a structural reduct M♭ in S, let M∼ be a structure with continuous ♭ in Y. Then an alter ego M∼ of M operations and with a structural reduct M∼ co-dualises M provided there exists ω ∈ Y(M∼ (1) ω ◦ Clo1(M) separates the points of M , and (2) M entails each binary relation r on M which forms a substructure of M∼ ♭, S∼) such that 2 that is maximal in ker(ω). Like the corresponding strong duality theorem for D-based total structures, our semilattice-based strong duality theorem is new. It is an immediate conse- quence of the Piggyback Strong Duality Theorem 3.7(III). By simply adding the assumption that M∼ is a topological structure, we could upgrade the theo- rem to the S-based Two-for-one Piggyback Strong Duality Theorem. Theorem 5.3 (S-based Piggyback Strong Duality Theorem). Let M be a total structure with a structural reduct M♭ in S, let M∼ be a total structure with a 24 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. ♭ in Y and assume that M∼ is an alter ego of M. Assume structural reduct M∼ that there exists ω ∈ Y(M∼ (1) ω ◦ Clo1(M∼ ) separates the points of M , and (2) ω ◦ Clo1(M) separates the points of M . Then conclusions (a) -- (c) of Theorem 3.7 hold. ♭, S∼) such that By applying this theorem when M∼ is simply M with an appropriate compact topology added, we obtain sufficient conditions for a semilattice-based total structure to be self-dualising. The result in the case in which M is a total algebra was proved in [12, Theorem 7.4]. Applications of the self-dualising version of the theorem, including examples in which M is infinite, may be found in [12, Section 8]. We now present a new example that is closely related to the infinite example studied in [12, Section 8]. Let V = h{0, 1}N0; ∧, u, 1i, where ∧ is defined pointwise, 1 is the constant map onto 1, and u is the left shift operator, that is, for all a ∈ {0, 1}N0 we have u(a) := a ◦ γ where γ : N0 → N0 is the successor function: γ(n) := n + 1. Let T be the product topology on {0, 1}N0. Theorem 5.4. Let W be topologically closed subalgebra of V. Then W is strongly self dualising, that is, W∼ := hW ; ∧, u, 1, Ti strongly dualises V, where T is the subspace topology. In particular, V itself and every finite subalgebra of V is self dualising. Proof. The fact that W be topologically closed subalgebra of V guarantees that W∼ is an alter ego of W. The calculations given for the Ockham algebra M1 in the proof of Theorem 4.6 show that π0 ◦ Clo1(W) separates the points of W . It follows at once from Theorem 5.3 that W∼ strongly dualises W. (cid:3) Example 5.5. Some interesting examples of topologically closed subalgebras of V are listed below -- see Figure 2. (1) CF := { a ∈ V a−1(1) is cofinite } forms a closed subalgebra of V. (2) For k ∈ N, define ak : N0 → {0, 1} by ak(ℓ) = 1 ⇐⇒ ℓ > k. Then, for all n > 3, the set Cn := {ak 1 6 k 6 n − 2} ∪ {0, 1} forms a subalgebra of V and C∞ := {ak k ∈ N} ∪ {0, 1} forms a closed subalgebra of V. (3) Fix n ∈ N and define b : N0 → {0, 1} by b(ℓ) = 1 ⇐⇒ ℓ ≡ 0 (mod n). For k ∈ Zn, define bk := uk(b). Then Mn := { bk k ∈ Zn } ∪ {0, 1} forms a subalgebra of V. (4) Define a := 01010101 . . . , b := 10101010 . . . and c := 10000 . . . . Then N5 := {0, a, b, c, 1} forms a subalgebra of V. Example 5.6. The algebra E := h{0, a1, 1}; ∧, ui, obtained by removing 1 from the type of C3, is (isomorphic to) the entropic closure semilattice studied by Davey, Jackson, Pitkethly and Talukder [12]. By applying the variant of the S-based Piggyback Strong Duality Theorem 5.3 obtained by piggybacking on the duality between semilattices and Boolean topological bounded semilattices, Vol. 00, XX Piggyback dualities revisited 25 1 a1 ... an−3 an−2 0 1 a1 a2 a3 ... 0 Cn C∞ 1 b0 · · · b1 bn−1 0 Mn a b c 1 0 N5 Figure 2. Some closed subalgebras of V we can see immediately that E∼ := h{0, a1, 1}; ∧, u, 0, 1, Ti strongly dualises E. This was proved directly in [12, Theorem 6.1]. 6. Proofs of the Piggyback Duality Theorems The proof of the Piggyback Duality Theorem 3.3 will be built from mod- ularised components set out in a series of lemmas. These components are combined to yield the proof of the theorem which is given after Lemma 6.10. The section concludes with the proof of the Strong Duality Theorem 3.7, based on the Injectivity Lemma 6.12. We shall begin with a summary of our strategy for proving the piggyback duality and co-duality theorems, in both their single carrier and general ver- sions. As noted already, we flag duality and co-duality results with the tags (D) and (coD). We now outline the roles of our key lemmas, in their (D) versions. For simplicity, assume for the purposes of this summary that we are working under the Basic Assumptions (not every lemma will require all the assumptions). Our objective is to demonstrate that, under suitable assumptions, we can construct the one-to-one map d− so that the diagram in Figure 1 commutes. Assume, pro tem, that we have identified a candidate alter ego M∼ for M with a structural reduct M♭ in B and also that Ω is a selected subset of B(M♭, N). (1) The Commuting Triangle Lemma 6.1 presupposes that the map d− is well defined, for each A ∈ A, and that, for each α ∈ ED(A), the map dα has domain B(A♭, N), that is, ΦA ω (A(A, M)) = B(A♭, N). [ω∈Ω 26 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. With these provisos, the lemma establishes that d− is one-to-one and that the diagram in Figure 3 below commutes. (2) The Existence Lemma 6.2 addresses the issue of well-definedness of d−. Here the choice of alter ego M∼ comes into play, and for the first time we make use of entailment. (3) The Relation Preservation Lemma 6.4 and the Operation Preservation Lemma 6.7 combine to give various conditions sufficient, in combination with the conditions of the Existence Lemma, to ensure that each map dα is a Y-morphism (once we know that the family {ΦA ω }ω∈Ω is jointly surjective, for all A ∈ A). The Operation Preservation Lemma is called on only when N∼ is not purely relational. Topological input is provided by the Continuity Lemma 6.5 and, for the Operation Preservation Lemma, also by the Morphism Lemma 6.6. We comment on both these ancillary lemmas shortly. (4) Our goal is then to give conditions under which joint surjectivity holds. The key steps towards this are provided by two important lemmas, the Substructure Lemma 6.8 and the Density Lemma 6.9. In the former we need to impose restrictions on the structure N; and in the latter we invoke the special properties of the duality for the base category included in the Basic Assumptions. (5) The Joint Surjectivity Lemma 6.10 shows that, under conditions encoun- ω }ω∈Ω are jointly surjec- tered already in (1) -- (4), the maps in the set {ΦA tive, as demanded in (1) and (3). We note that the multi-carrier case has certain features not present in the single-carrier case. In the Substructure Lemma, and in the Joint Surjectivity Lemma which makes use of it, compatibility issues arise when Ω > 1 which force us to assume that any operations in N∼ are unary or nullary. The (coD) strand of the theory parallels the (D) strand quite closely, the In almost every case our principal difference being in the role of topology. lemmas have separate (D) and (coD) parts. In the proofs we shall normally need to attend to topological issues in just one of these (of course when M and N are finite no topological considerations arise) and then to prove one of the two structure assertions, calling on symmetry to obtain the other. In such cases we shall simply specify which of (D) and (coD) we elect to prove, leaving it tacit that the unproved assertion also follows. Exceptions occur with the Continuity Lemma and the Morphism Lemma, to which we alluded in (3) above. The Continuity Lemma has no (coD) component and is required for the Piggyback Duality Theorem but not at all for the Piggyback Co-duality Theorem. The Morphism Lemma does not have separated (D) and (coD) claims; in it, the way in which a carrier map ω relates to B and Y simultaneously is crucial. We issue a reminder that Lemma 2.2 comes into play whenever we encounter entailment by M. This applies to Part (coD) in each of the Existence Lemma, the ⊑-Lemma discussed below, and the Relation Preservation Lemma. Vol. 00, XX Piggyback dualities revisited 27 The Existence Lemma involves entailment conditions linking the structures M and M∼ to the structures N and N∼ via the carrier maps in Ω. The ⊑- Lemma 6.3 shows that these conditions hold in particular if there exists a reflexive, antisymmetric relation ⊑ on N with suitable properties. It is used to obtain Corollary 3.6. The motivation here is to synthesise the behaviour of the order relation on D∼ in the D-based case. The conditions imposed in Theorems 3.3 and 3.5 guarantee that the appro- priate sets of carrier maps are jointly surjective. It is nonetheless of interest to ask whether this is an indispensable requirement if piggybacking is to be pos- sible. We are able to present the Extended Commuting Triangle Lemma 6.11 which does not demand joint surjectivity. We are now ready to carry out our indicated programme. Lemma 6.1 (Commuting Triangle Lemma, for the jointly surjective case). Let M and N be structures and let M∼ and N∼ be alter egos of M and N, respectively. Define A := ISP(M), B := ISP(N), X := IScP+(M∼ ) and Y := IScP+(N∼). (D) Assume that M has a structural reduct M♭ in B and let Ω be a sub- set of B(M♭, N). Let A ∈ A and assume that for every morphism α : A(A, M) → M∼ there is a map dα : [ω∈Ω ΦA ω (A(A, M)) → N such that dα ◦ ΦA ω = ω ◦ α, for all ω ∈ Ω (see Figure 3). A(A, M) ΦA ω ΦA ω (A(A, M)) [ω∈Ω α dα M ω N Figure 3. The equation dα ◦ ΦA ω = ω ◦ α (1) If Ω ◦ Clo1(M∼ ) separates the points of M , then the function α 7→ dα is one-to-one. (2) Assume that the joint image of the family {ΦA ω }ω∈Ω is B(A♭, N). Then deA(a) = kA♭(a), for all a ∈ A. (coD) Assume that M∼ has a structural reduct M∼ ♭ in Y and let Ω be a sub- ♭, N∼). Let X ∈ X and assume that for every morphism set of Y(M∼ u : X(X, M∼ ) → M there is a map du : [ω∈Ω ΨX ω (X(X, M∼ )) → N such that du ◦ ΨX ω = ω ◦ u, for all ω ∈ Ω. 28 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. (1) If Ω ◦ Clo1(M) separates the points of M , then the function u 7→ du is one-to-one. (2) Assume that the joint image of the family {ΨX ω }ω∈Ω is Y(X♭, N∼). Then dεX(x) = κX♭(x), for all x ∈ X. Proof. We prove the (D) assertions. Those for (coD) are obtained likewise. (D)(1) Let α, β : A(A, M) → M∼ be X-morphisms with α 6= β. We shall show that dα 6= dβ. Since α 6= β, there exists x ∈ A(A, M) with α(x) 6= β(x). Thus, since Ω ◦ Clo1(M∼ ) separates the points of M , there exists g ∈ Clo1(M∼ ) and ω ∈ Ω with ω(g(α(x))) 6= ω(g(β(x))). Since g is a unary term function of M∼ , both α and β preserve g, and g ◦ x ∈ A(A, M) as g is an endomorphism of M. Hence, since dα ◦ ΦA ω = ω ◦ α and dβ ◦ ΦA ω (g ◦ x)) = ω(α(g ◦ x)) = ω(α(gA(A,M)(x))) ω = ω ◦ β, we have dα(ω ◦ g ◦ x) = dα(ΦA = ω(g(α(x))) 6= ω(g(β(x))) = ω(β(gA(A,M)(x))) = ω(β(g ◦ x)) = dβ(ΦA = dβ(ω ◦ g ◦ x). ω (g ◦ x)) Thus, dα 6= dβ, whence the map α 7→ dα is one-to-one. (D)(2) Let z ∈ B(A♭, N). By the assumed joint surjectivity, there exist ω = ω ◦ eA(a), we have ω ∈ Ω and x ∈ A(A, M) with z = ΦA ω (x). As deA(a) ◦ ΦA deA(a)(z) = deA(a)(ΦA ω (x)) = ω(eA(a)(x)) = ω(x(a)) = kA♭(a)(ω ◦ x) = kA♭(a)(z). (cid:3) Lemma 6.2 (Existence Lemma). Let M and N be structures and let M∼ and N∼ be alter egos of M and N, respectively. Let A := ISP(M), B := ISP(N), X := IScP+(M∼ ) and Y := IScP+(N∼). (D) Assume that M is a total structure that has a structural reduct M♭ in B and let Ω be a subset of B(M♭, N). Let A ∈ A and let α : A(A, M) → M∼ be an X-morphism. Then, under either of the conditions listed below, there exists a map dα : [ω∈Ω ΦA ω (A(A, M)) → N such that dα ◦ ΦA (1) Ω = {ω} and either ω = ω ◦ α, for all ω ∈ Ω. (i) M∼ entails each binary relation r on M which forms a substruc- ture of M2 that is maximal in ker(ω), or (ii) ω ◦ Clo1(M) separates the points of M (in which case ΦA ω is one-to-one). (2) The structure M∼ entails each relation in maxM Ω−1(∆N ). (coD) Assume that M∼ = hM ; G, R, Ti is a total structure that has a structural ♭, N∼). Let X ∈ X and reduct M∼ let u : X(X, M∼ ) → M be an A-morphism. Then, under either of the ♭ in Y and let Ω be a subset of Y(M∼ Vol. 00, XX Piggyback dualities revisited 29 conditions listed below, there exists a map du : [ω∈Ω ΨX ω (X(X, M∼ )) → N such that du ◦ ΨX (1) Ω = {ω} and either ω = ω ◦ u, for all ω ∈ Ω. (i) the operations in G are continuous and M entails each binary 2 that is max- relation r on M which forms a substructure of M∼ imal in ker(ω), or (ii) ω ◦ Clo1(M∼ ) separates the points of M (in which case ΨX ω is one-to-one). (2) The operations in G are continuous and the structure M entails each relation in maxM ∼ Ω−1(∆N ). Proof. We shall prove the (coD) results. First note that the condition in (coD)(1)(i) for the single-carrier case is a special instance of condition (coD)(2) 2 that is maximal in since a binary relation on M forms a substructure of M∼ ker(ω) = (ω, ω)−1(∆N ) if and only if it belongs to maxM {ω}−1(∆N ). ∼ ω1(α) = ΨX Now assume (coD)(2) holds. For γ ∈ ΨX ω (X(X, M∼ )), we "define" du(γ) := ω(u(α)), where α is any element of X(X, M∼ ) for which ΨX ω (α) = γ. Of course, we must show that this choice is independent of the choice of ω and α. Assume that ω1, ω2 ∈ Ω. We must prove that ω1(u(α)) = ω2(u(β)) whenever α, β ∈ X(X, M∼ ) are such that ΨX ω2(β), that is, ω1 ◦ α = ω2 ◦ β. Since M∼ is a total structure, the image of the natural product morphism 2, that is, the relation (α ⊓ β)(X) = { (α(x), β(x)) x ∈ X }, is α ⊓ β : X → M∼ 2, and since ω1 ◦ α = ω2 ◦ β it follows at once that a closed substructure of M∼ 2 that contains (α ⊓ β)(X) ⊆ (ω1, ω2)−1(∆N ). Let s be a substructure of M∼ (α ⊓ β)(X) and is maximal in (ω1, ω2)−1(∆N ); thus s ∈ maxM Ω−1(∆N ), and consequently u preserves s, as M entails s, by assumption. We have (α, β) ∈ sX(X,M ∼ ), by construction, and hence (u(α), u(β)) ∈ s, as u preserves s. Since s ⊆ (ω1, ω2)−1(∆N ), it follows that ω1(u(α)) = ω2(u(β)), as required. ∼ It remains to consider the case when (coD)(1)(ii) holds. To show that there ω (X(X, M∼ )) → N satisfying du ◦ ΨX ω = ω ◦ u, it suffices ω ) ⊆ ker(ω ◦ u). In fact, we shall prove something much ω is one-to-one. Let ω ), so ω ◦ α = ω ◦ β, and choose x ∈ X. exists a map du : ΨX to prove that ker(ΨX stronger, namely that ker(ΨX α, β ∈ X(X, M∼ ) with (α, β) ∈ ker(ΨX Then, for all t ∈ Clo1(M∼ ), we have ∼ ), that is, ΨX ω ) = ∆X(X,M ω(t(α(x))) = ω(α(t(x))) as α preserves t = ω(β(t(x))) as ω ◦ α = ω ◦ β = ω(t(β(x))) as β preserves t. Since ω ◦ Clo1(M∼ ) separates the points of M , it follows that α(x) = β(x) and so α = β, as x ∈ X was chosen arbitrarily. Hence ker(ΨX (cid:3) ω ) = ∆X(X,M ∼ ). 30 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. We shall now present the lemma which gives sufficient conditions for (D)(2) and (coD)(2) in the Existence Lemma to hold. This result will immediately yield Corollary 3.6 once the piggyback duality and co-duality theorems have been established. Note that in the (coD) part of Corollary 3.6, we assume that ⊑ is part of the structure on N. As N∼ is an alter ego of N, the relation ⊑ is therefore topologically closed. Lemma 6.3 (⊑-Lemma). Let M and N be structures and let M∼ and N∼ be alter egos of M and N respectively. Let B := ISP(N) and Y := IScP+(N∼). (D) Assume that M has a structural reduct M♭ in B and let Ω be a subset of B(M♭, N). Assume that ⊑ is a reflexive, antisymmetric binary relation on N such that M∼ entails each relation in maxM Ω−1(⊑). Then M∼ entails each relation in maxM Ω−1(∆N ). (coD) Let M∼ = hM ; G, R, Ti with the operations in G continuous. Assume that ♭, N∼). M∼ has a structural reduct M∼ Assume that ⊑ is a topologically closed reflexive, antisymmetric binary Ω−1(⊑). Then relation on N such that M entails each relation in maxM M entails each relation in maxM ♭ in Y and let Ω be a subset of Y(M∼ Ω−1(∆N ). ∼ ∼ Proof. Topology plays no overt role in the proof here, so it would suffice to prove either statement, but to align with the proof of the Existence Lemma we establish the (coD) statement. Under the assumptions made there we can Ω−1(⊒). We shall show that also assert that M entails each relation in maxM M entails every relation in maxM Ω−1(∆N ). ∼ ∼ ∼ Let s ∈ maxM Ω−1(∆N ). Thus s is a subuniverse of M2 and there exist ω1, ω2 ∈ Ω with s maximal in (ω1, ω2)−1(∆N ). As ⊑ is reflexive, ∆N is contained in ⊑ and so s ⊆ (ω1, ω2)−1(⊑) and s ⊆ (ω1, ω2)−1(⊒). Let t1 and t2 be subuniverses of M2 with s ⊆ t1 and s ⊆ t2 such that t1 and t2 are maximal in (ω1, ω2)−1(⊑) and (ω1, ω2)−1(⊒), respectively; so t1 ∈ maxM Ω−1(⊑) and Ω−1(⊒). Let (a, b) ∈ t1 ∩ t2. Then ω1(a) ⊑ ω2(b) because t1 ⊆ t2 ∈ maxM (ω1, ω2)−1(⊑), and ω1(a) ⊒ ω2(b) because t2 ⊆ (ω1, ω2)−1(⊒). Hence ω1(a) = ω2(b), as ⊑ is antisymmetric, and so t1 ∩ t2 ⊆ (ω1, ω2)−1(∆N ). As s ⊆ t1 ∩ t2 and t1 ∩ t2 is a subuniverse, the maximality of s in (ω1, ω2)−1(∆N ) guarantees that s = t1 ∩ t2. Since M entails t1 and t2, by assumption, it follows that M entails s. (cid:3) ∼ ∼ Lemma 6.4 (Relation Preservation Lemma). Let M and N be structures and let M∼ and N∼ be alter egos of M and N, respectively. Define A := ISP(M), B := ISP(N), X := IScP+(M∼ ) and Y := IScP+(N∼). (D) Assume that M is a total structure that has a structural reduct M♭ in B and let Ω be a subset of B(M♭, N). Let A ∈ A and let α : A(A, M) → M∼ be an X-morphism and assume that there is a map dα : [ω∈Ω ΦA ω (A(A, M)) → N Vol. 00, XX Piggyback dualities revisited 31 ω = ω ◦ α, for all ω ∈ Ω. Let r be a subuniverse of Nn, such that dα ◦ ΦA for some n ∈ N, and extend r pointwise to the joint image of the family of maps {ΦA ω }ω∈Ω. Then dα preserves r provided the structure M∼ entails each relation in maxM Ω−1(r). (coD) Assume that M∼ is a total structure with continuous operations that has a ♭, N∼). Let X ∈ X structural reduct M∼ and let u : X(X, M∼ ) → M be an A-morphism and assume that there is a map ♭ in Y and let Ω be a subset of Y(M∼ du : [ω∈Ω ΨX ω (X(X, M∼ )) → N such that du ◦ ΨX ω = ω ◦ u, for all ω ∈ Ω. Let r be a topologically closed subuniverse of Nn, for some n ∈ N, and extend r pointwise to the joint image of the family of maps {ΨX ω }ω∈Ω. Then du preserves r provided the structure M entails each relation in maxM Ω−1(r). ∼ Let z1, . . . , zn ∈ Sω∈Ω ΦA Proof. We prove (D). The proof of (coD) is a simple modification. Let r be a subuniverse of Nn and assume that M∼ entails every relation in maxM Ω−1(r). ω (A(A, M)) with (z1, . . . , zn) ∈ rB(A♭,N). Thus there exist ωi ∈ Ω and xi ∈ A(A, M) with Φωi(xi) = zi, that is ωi ◦ xi = zi, for all i ∈ {1, . . . , n}. Since dα(zi) := ωi(α(xi)), we must prove that (ω1(α(x1)), . . . , ωn(α(xn))) ∈ r. As M is a total structure, the image of the natural product morphism x1 ⊓ · · · ⊓ xn : A → Mn, that is, the relation (x1 ⊓ · · · ⊓ xn)(N ) = { (x1(a), . . . , xn(a)) a ∈ A }, is a substructure of Mn. Since (ω1 ◦ x1, . . . , ωn ◦ xn) ∈ rB(A♭,N), we have (ω1(x1(a)), . . . , ωn(xn(a))) ∈ r, for all a ∈ A, by the definition of rB(A♭,N). It follows at once that (x1 ⊓· · ·⊓xn)(N ) ⊆ (ω1, . . . , ωn)−1(r). Let s be a substruc- ture of Mn that contains (x1⊓· · ·⊓xn)(N ) and is maximal in (ω1, . . . , ωn)−1(r); Ω−1(r). Then (x1, . . . , xn) ∈ sA(A,M) as (x1 ⊓· · ·⊓xn)(N ) ⊆ s, thus s ∈ maxM and hence (α(x1), . . . , α(xn)) ∈ s, as α preserves s, by assumption. Since s ⊆ (ω1, . . . , ωn)−1(r), it follows that (ω1(α(x1)), . . . , ωn(α(xn))) ∈ r. (cid:3) ∼ The Continuity Lemma concerns continuity of the maps dα. There is no (coD) assertion to be made about the maps du. Lemma 6.5 (Continuity Lemma). Let M and N be structures and let M∼ and N∼ be alter egos of M and N, respectively. Define A := ISP(M) and B := ISP(N). Assume that M has a structural reduct M♭ in B. (1) Let ω ∈ B(M♭, N). Then ΦA ω : A(A, M) → B(A♭, N) is continuous for all A ∈ A if and only if ω is continuous with respect to the topologies on M∼ and N∼. (2) Let Ω be a finite subset of B(M♭, N) with each ω ∈ Ω continuous. Let A ∈ A, let α : A(A, M) → M be a continuous map and let dα : [ω∈Ω ΦA ω (A(A, M)) → N 32 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. be a map such that dα ◦ ΦA ω = ω ◦ α, for all ω ∈ Ω. Then dα is continuous. Proof. (1) Assume that ω is continuous and let A ∈ A. Given sets S and T , and s ∈ S and U ⊆ T , define (s; U ) := { x ∈ T S x(s) ∈ U }. Thus a sub-basic open set in B(A♭, N) is of the form (a; U ) ∩ B(A♭, N), for some a in A and U open in N . We have ω(cid:1)−1 (cid:0)ΦA ((a; U ) ∩ B(A♭, N)) = {x ∈ A(A, M) ΦA ω (x) ∈ (a; U ) } = {x ∈ A(A, M) ω(x(a)) ∈ U } = {x ∈ A(A, M) x(a) ∈ ω−1(U ) } = (a; ω−1(U )) ∩ A(A, M), which is open in A(A, M) as ω is continuous. Hence ΦA ω is continuous. Now assume that ΦA ω is continuous for all A ∈ A. In particular, ΦF1 ω is continuous, where F1 denotes the one-generated free algebra in A. Hence graph(ΦF1 1, N). Let the free v1 : N F1 → N be the generator of F1 be v1 and let πM projections. Then ω ) is closed in the product space A(F1, M) × B(F♭ v1 : M F1 → M and πN graph(ω) = { (a, ω(a)) a ∈ A } = { (x(v1), ω(x(v1)) x ∈ A(F1, M) } = (πM v1 × πN v1 )(graph(ΦF1 ω )). As the topologies on M and N are compact, the projection πM v1 is a closed map and consequently graph(ω) is closed in M × N . Since graph(ω) is closed and the topologies on M and N are compact and Hausdorff, it follows that ω is continuous. v1 × πN (2) By (1), ΦA ω is continuous and so is a closed map, for all ω ∈ Ω. Let U be a closed subset of D. A simple calculation shows that d−1 α (U ) = [ω∈Ω ΦA ω (α−1(ω−1(U ))), which is closed in S{ ΦA ω is continuous, each ΦA continuous. ω (A(A, M)) ω ∈ Ω } since α is continuous, each ω is a closed map, and Ω is finite. Hence, dα is (cid:3) Up to this point our lemmas have not reflected the dichotomy in Condi- tions (2) in our duality and co-duality theorems. Now we address this. The Morphism Lemma 6.6 and the Operation Preservation Lemma 6.7 below will be needed to handle two scenarios. (D) In Theorem 3.3, alternative (ii) under Assumption (2) when N∼ is not purely relational, and (coD) In Theorem 3.5, alternative (ii) under Assumption (2) when N is not purely relational, and likewise when the corresponding conditions in the single-carrier versions of the theorems hold. Vol. 00, XX Piggyback dualities revisited 33 We have already noted that the Morphism Lemma does not have separate (D) and (coD) claims. In it we require a carrier map ω which acts both as a B-morphism and as a Y-morphism. In a (D) assertion, ω is usually a B- ω : A(A, M) → B(A♭, N) morphism, as this ensures that the associated map ΦA is well defined. But in order to assert that ΦA ω is a Y-morphism we shall need ω to preserve the operations and relations not of N but of N∼. Recall that, if M and M∼ have structural reducts M♭ in B and M∼ ♭ in Y, respectively, then we have corresponding forgetful functors ♭ : A → B and ♭ : X → Y. Lemma 6.6 (Morphism Lemma). Let M and N be structures and let M∼ and N∼ be alter egos of M and N, respectively. Define A := ISP(M), B := ISP(N), X := IScP+(M∼ ) and Y := IScP+(N∼). Assume that M and M∼ have structural ♭, N∼). reducts M♭ in B and M∼ ω : A(A, M)♭ → B(A♭, N) is a Let A ∈ A and X ∈ X. Then the map ΦA Y-morphism and the map ΨX ♭ in Y, respectively. Let ω ∈ B(M♭, N) ∩ Y(M∼ ω : X(X, M∼ )♭ → Y(X♭, N∼) is a B-morphism. ω is a Y-morphism. We note that ΦA Proof. By symmetry, it is sufficient to prove the stronger statement, namely that ΦA ω is continuous, by Lemma 6.5. Since a map preserves an operation or partial operation g if and only if it preserves graph(g), to prove that ΦA ω is a Y-morphism it suffices to prove that ΦA ω preserves every relation preserved by ω. Let r be an n-ary relation symbol (not necessarily in the type of M∼ , nor of N∼), let rM and rN be interpretations of r on M and N , respectively, and assume that ω preserves r. Since rM and rN are extended pointwise to A(A, M) and B(A♭, N), respectively, we have, for all x1, . . . , xn ∈ A(A, M), (x1, . . . , xn) ∈ rA(A,M) =⇒ (∀a ∈ A) (x1(a), . . . , xn(a)) ∈ rM =⇒ (∀a ∈ A) (ω(x1(a)), . . . , ω(xn(a))) ∈ rN =⇒ (ω ◦ x1, . . . , ω ◦ xn) ∈ rB(A♭,N) =⇒ (ΦA ω (x1), . . . , ΦA ω (xn)) ∈ rB(A♭,N). Hence ΦA ω preserves r. (cid:3) ♭, N∼). Lemma 6.7 (Operation Preservation Lemma). Let M and N be structures and let M∼ and N∼ be alter egos of M and N, respectively. Define A := ISP(M), B := ISP(N), X := IScP+(M∼ ) and Y := IScP+(N∼). Assume that M and ♭ in Y, respectively, and let Ω ⊆ M∼ have structural reducts M♭ in B and M∼ B(M♭, N) ∩ Y(M∼ (D) Assume that N∼ = hN ; Gν , Rν, Ti is a total structure, let A ∈ A and let α : A(A, M) → M∼ be an X-morphism. Assume that the family of maps ω : A(A, M) → B(A♭, N), for ω ∈ Ω, is jointly surjective and that ΦA dα : B(A♭, N∼) → N is a map such that dα ◦ ΦA ω = ω ◦ α, for all ω ∈ Ω. If Ω = 1 or every operation in Gν is unary or nullary, then dα preserves the operations in Gν . 34 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. (coD) Assume that N = hN ; Gν , Rνi is a total structure, let X ∈ X and let u : X(X, M∼ ) → M be an A-morphism. Assume that the family of maps ω : X(X, M∼ ) → Y(X♭, N∼), for ω ∈ Ω, is jointly surjective and that ΨX du : Y(X♭, N∼) → N is a map such that du ◦ ΨX ω = ω ◦ u, for all ω ∈ Ω. If Ω = 1 or every operation in Gν is unary or nullary, then du preserves the operations in Gν . ♭ is a structural reduct of M∼ , the map Proof. We shall prove (D). As M∼ ♭ is a Y-morphism and so preserves the operations in Gν. α : A(A, M)♭ → M∼ Under the given assumptions Lemma 6.6 applies, and tells us that the maps ω are Y-morphisms, and so they also preserve the operations in Gν . Fix ΦA ω ∈ Ω. As N∼ is a total structure, Yω := ΦA ω (A(A, M)) is a substructure of B(A♭, N). Since dα ◦ ΦA ω , α and ω preserve the operations in Gν , it follows easily that dα↾Yω : Yω → N preserves the opera- tions in Gν -- use the fact that if A, B and C are algebras and u : A → B and w : A → C are homomorphisms with u surjective, then any map v : B → C satisfying v ◦ u = w is necessarily a homomorphism. ω = ω ◦ α and the maps ΦA The joint-surjectivity assumption tells us that B(A♭, N) = Sω∈Ω Yω. If Ω = {ω}, then B(A♭, N) = Yω, and hence dα = dα↾Yω preserves the operations in Gν. If every operation in Gν is unary or nullary, then dα preserves the operations in Gν since its restriction to each of the subuniverses Yω does. (cid:3) We now work towards sufficient conditions for joint surjectivity. We prepare the way by establishing two lemmas concerning substructures. They are very different in the assumptions required to arrive at the desired conclusions. Lemma 6.8 (Substructure Lemma). Let M and N be structures and let M∼ and N∼ be alter egos of M and N, respectively. Define A := ISP(M), B := ISP(N), X := IScP+(M∼ ) and Y := IScP+(N∼). (D) Assume that M has a structural reduct M♭ in B and let Ω be a finite subset of B(M♭, N). Assume that one of the following conditions holds: (i) N∼ = hN ; Rν, Ti is purely relational and each ω ∈ Ω is continuous; (ii) N∼ = hN ; Gν , Rν, Ti is a total structure, M∼ has a structural reduct ♭, N∼), and either Ω = 1 or ♭ in Y, each ω ∈ Ω belongs to Y(M∼ M∼ every operation in Gν is unary or nullary. Then, for all A ∈ A, the joint image of the family {ΦA substructure of N∼ A, and therefore of B(A♭, N). ω }ω∈Ω is a closed (coD) Assume that M∼ has a structural reduct M∼ ♭ in Y and let Ω be a subset of Y(M∼ ♭, N∼) and assume that one of the following conditions holds: (i) N = hN ; Rνi is purely relational; (ii) N = hN ; Gν , Rνi is a total structure, M has a structural reduct M♭ in B, each ω ∈ Ω belongs to B(M♭, N), and either Ω = 1 or every operation in Gν is unary or nullary. Then, for all X ∈ X, the joint image of the family {ΨX structure of NX , and therefore of Y(X♭, N∼). ω }ω∈Ω is a sub- Vol. 00, XX Piggyback dualities revisited 35 Proof. Consider (D). Define Z to be the joint image of the family {ΦA ω }ω∈Ω. If assumption (i) holds then it is trivial that Z is a substructure of B(A♭, N) as Gν = ∅. By Lemma 6.5, each ΦA ω is continuous since ω is. Thus Z is topologically closed as Ω is finite. Now assume that (ii) holds. By Lemma 6.6, each ΦA ω is a Y-morphism. Hence, as N∼ is a total structure, the image of each ω is a topologically closed substructure of B(A♭, N). If Ω = {ω}, then Z is ΦA ω and so is a closed substructure of B(A♭, N). If every simply the image of ΦA operation in Gν is unary or nullary, then Z is a substructure since, in this case, a union of substructures is again a substructure. The set Z is topologically closed since Ω is finite and each ΦA ω is continuous. Hence (D) follows. The proof of (coD) is almost identical, minus the topological arguments. (cid:3) The proof of the next lemma, in which we need to assume that N∼ fully dualises N, is similar to the proof in the case that M is a total algebra given by Davey and Priestley [14, Proposition 1.11]; we give the details for completeness. Lemma 6.9 (Density Lemma). Let N be a structure, let N∼ be a fully dualising alter ego of N, define B = ISP(N) and Y := IScP+(N∼) and let hH, K, k, κi be the associated dual equivalence between B and Y. Let B ∈ B and Y ∈ Y. (D) Assume that N∼ is injective in Y. If Z is a subset of B(B, N) that separates the structure B, then the topologically closed substructure of B(B, N) ≤ N∼ B generated by Z is B(B, N) itself. (coD) Assume that N is injective in B. If C is a subset of Y(Y, N∼) that sepa- rates the structure Y, then the substructure of Y(Y, N∼) ≤ NY generated by C is Y(Y, N∼) itself. Proof. We shall prove (D). The proof of (coD) is a simple modification obtained by replacing the closed substructure of H(B) generated by Z by the substruc- ture of K(Y) generated by C. Assume that Z is a subset of H(B) = B(B, N) that separates the structure B. Define Y to be the closed substructure of H(B) generated by Z. Let µ : Y → H(B) be the inclusion map and consider K(µ) : KH(B) → K(Y). We claim that K(µ) is an embedding. Assume that N = hN ; Gν, H ν, Rνi. Let r be an n-ary relation in dom(H ν) ∪ Rν ∪ {∆N } and let α1, . . . , αn ∈ KH(B) with (α1, . . . , αn) /∈ rKH(B). Since N∼ dualises N, there exist b1, . . . , bn ∈ B with (b1, . . . , bn) /∈ rB and αi = kB(bi), for all i ∈ {1, . . . , n}. Since Z separates the structure B, there exists z ∈ Z ⊆ Y with (z(b1), . . . , z(bn)) /∈ rN. Hence (α1(z), . . . , αn(z)) = (kB(b1)(z), . . . , kB(bn)(z)) = (z(b1), . . . , z(bn)) /∈ rN. Thus (α1↾Y , . . . , αn↾Y ) /∈ rK(Y), and consequently K(µ) is an embedding. As µ is an embedding and N∼ is injective in Y, the map K(µ) is surjective. Thus K(µ) is an isomorphism. As N∼ fully dualises N, it follows that µ is an isomorphism, whence Y = H(B). (cid:3) In Lemma 6.10 we shall require Ω◦End(M) to separate the structure M♭ and ♭. In the earlier lemmas, by contrast, Ω ◦ End(M∼ ) to separate the structure M∼ 36 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. we required that the possibly smaller sets Ω◦Clo1(M∼ ) and Ω◦Clo1(M) separate the points of M . Lemma 6.10 (Joint Surjectivity Lemma). Let M and N be structures and let M∼ and N∼ be alter egos of M and N, respectively. Define A := ISP(M), B := ISP(N), X := IScP+(M∼ ) and Y := IScP+(N∼). (D) Assume that M has a structural reduct M♭ in B and let Ω be a subset of B(M♭, N). (1) (i) If the family {ΦM ω }ω∈Ω is jointly surjective, then Ω ◦ End(M) separates the structure M♭. (ii) Assume that Ω◦ End(M) separates the structure M♭. Then, for ω }ω∈Ω separates each A ∈ A, the joint image of the family {ΦA the structure A♭. (2) Assume that Ω is finite, N∼ fully dualises N with N∼ injective in Y, and that one of the following conditions holds: (i) N∼ = hN ; Rν, Ti is purely relational and each ω ∈ Ω is continu- ous; (ii) N∼ = hN ; Gν , Rν, Ti is a total structure, M∼ has a structural ♭, N∼), and either reduct M∼ Ω = 1 or every operation in Gν is unary or nullary. ♭ in Y, each ω ∈ Ω belongs to Y(M∼ Assume that Ω ◦ End M separates the structure M♭. Then, for each A ∈ A, the family {ΦA ω }ω∈Ω is jointly surjective. (coD) Assume that M∼ has a structural reduct M∼ ♭ in Y and let Ω be a subset ♭, N∼). of Y(M∼ (1) (i) If the family {Ψ M ∼ ω }ω∈Ω is jointly surjective, then Ω ◦ End(M∼ ) separates the structure M∼ ♭. (ii) Assume that Ω◦ End(M∼ ) separates the structure M∼ each X ∈ X, the joint image of the family {ΨX the structure X♭. ♭. Then, for ω }ω∈Ω separates (2) Assume that N∼ fully dualises N with N is injective in B. Assume ♭, and that one that the set Ω ◦ End(M∼ ) separates the structure M∼ of the following conditions holds: (i) N = hN ; Rνi is purely relational; (ii) N = hN ; Gν, Rνi is a total structure, M has a structural reduct M♭ in B, each ω ∈ Ω belongs to B(M♭, N), and either Ω = 1 or every operation in Gν is unary or nullary. Assume that Ω ◦ End(M∼ ) separates the structure M∼ each X ∈ X, the family {ΨX ω }ω∈Ω is jointly surjective. ♭. Then, for Proof. We prove (D). The proof of (coD) is almost identical, but the finiteness of Ω is not required since there is no topology involved. (D)(1)(i) Assume that the family {ΦM ω }ω∈Ω is jointly surjective. Assume that N = hN ; Gν , H ν, Rνi and let r ∈ dom(H ν )∪Rν ∪{∆N } be n-ary. Suppose that a1, . . . , an ∈ M with (a1, . . . , an) /∈ rM♭ . As M♭ ∈ ISP(N), there exists a Vol. 00, XX Piggyback dualities revisited 37 morphism z ∈ B(M♭, N) with (z(a1), . . . , z(an)) /∈ rN. By assumption, there exists ω ∈ Ω and x ∈ A(M, M) = End(M) with ω ◦ x = ΦA ω (x) = z. Hence Ω ◦ End(M) separates the structure M♭. (D)(1)(ii) Assume that Ω ◦ End(M) separates the structure M♭ and as- sume that N = hN ; Gν, H ν, Rνi. Let A ∈ A. Let r be an n-ary relation in dom(H ν) ∪ Rν ∪ {∆N } and let a1, . . . , an ∈ A with (a1, . . . , an) /∈ rA♭ . Since A♭ ∈ ISP(M♭), there exists x ∈ B(A♭, M♭) with (x(a1), . . . , x(an)) /∈ rM♭ . As Ω ◦ End(M) separates the structure M♭, there exists ω ∈ Ω and u ∈ End(M) satisfying (ω(u(x(a1))), . . . , ω(u(x(an)))) /∈ rN. As u ∈ End(M), it follows that u ◦ x ∈ A(A, M) and hence (ΦA ω (u ◦ x)(an)) /∈ rN. ω }ω∈Ω separates the structure A♭. Therefore the joint image of the family {ΦA (cid:3) (D)(2) follows from (D)(1) and Lemmas 6.8 and 6.9. ω (u ◦ x)(a1), . . . , ΦA We have now assembled all the components for the proof of the Piggyback Duality Theorem 3.3. That of the Piggyback Co-Duality Theorem 3.5 is a simple modification. Proof of Theorem 3.3. Assume that the Basic Assumptions are in force and that Conditions (0) -- (3) in the statement of the theorem hold. For ease of reference we list these again here. Recall that N∼ = hN ; Gν, Rν, Ti. (0) Ω is finite and each ω ∈ Ω is continuous with respect to the topologies on M∼ and N∼. (1) Ω ◦ Clo1(M∼ ) separates the structure M♭. (2) One of the following conditions holds: (i) N∼ is purely relational, or (ii) M∼ has a structural reduct M∼ and every operation in Gν is unary or nullary. ♭ in Y, each ω ∈ Ω belongs to Y(M∼ ♭, N∼), (3) (i) The structure M∼ entails every relation in maxM Ω−1(r), for each re- lation r ∈ Rν, and (ii) M∼ entails every relation in maxM Ω−1(∆N ). We now give the proof. Let A ∈ A and let α : A(A, M) → M∼ be an X-morphism, that is, α ∈ ED(A). We must prove that there exists a ∈ A with eA(a) = α. Since each ω ∈ Ω is continuous, Conditions (1) and (2) ensure that the family {ΦA ω }ω∈Ω is jointly surjective by the Joint Surjectiv- ity Lemma 6.10(D)(2). Condition (3)(ii) now ensures the existence of a map dα : B(A♭, N) → N satisfying dα ◦ ΦA ω = ω ◦ α, for all ω ∈ Ω, by the Exis- tence Lemma 6.2(D). Conditions (3)(i) and (0) allow us to invoke the Relation Preservation Lemma 6.4(D) and the Continuity Lemma 6.5(2) to conclude that dα preserves the relations in Rν and is continuous. If (2)(i) holds we deduce immediately that dα is a Y-morphism. To obtain the same conclusion when (2)(ii) holds we must confirm also that dα preserves the operations in Gν . For this we can invoke the Operation Preservation Lemma 6.7(D). Now we have a map d− : ED(A) → KH(A♭), which, by the Commuting Triangle Lemma 6.1(2), satisfies deA(a) = kA♭ (a), for all a ∈ A. Since N∼ 38 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. dualises N, the map kA♭ : A♭ → KH(A♭) is surjective. So there exists a ∈ A with kA♭ (a) = dα. Hence deA(a) = kA♭ (a) = dα. Condition (1) guarantees that d− : ED(A) → KH(A♭) is one-to-one, by Lemma 6.1(1). Hence eA(a) = α. We have now established the version of the Piggyback Duality Theorem which is not specialised to the single-carrier case. For the proof for the single carrier case, that is, the proof of Theorem 3.2, we note that essentially the same arguments as above apply when the conditions are stated in the simplified form which is possible when Ω = 1. (cid:3) We imposed sufficient conditions above to guarantee joint surjectivity. The following lemma indicates that we may, in suitable circumstances, be able to circumvent the need for joint surjectivity. This opens the way to the possibility of pushing through the piggyback constructions in the absence of joint surjec- tivity, subject to specified conditions being met -- see [16, 3.8] for an application of the Extended Commuting Triangle Lemma to finite pseudocomplemented semilattices. Since the examples considered in this paper are covered by the results stated in Section 3, we do not pursue this topic further in this paper. Lemma 6.11 (Extended Commuting Triangle Lemma). Let M and N be structures and let M∼ and N∼ be alter egos of M and N, respectively. Define A := ISP(M), B := ISP(N), X := IScP+(M∼ ) and Y := IScP+(N∼). (D) Assume that M has a structural reduct M♭ in B and let Ω be a sub- set of B(M♭, N). Let A ∈ A and assume that for every morphism ω (A(A, M)) → N such α : A(A, M) → M∼ there is a map dα : Sω∈Ω ΦA that dα ◦ ΦA (1) If Ω ◦ Clo1(M∼ ) separates the points of M , then the function α 7→ dα ω = ω ◦ α, for all ω ∈ Ω (as in Figure 3). is one-to-one. (2) Assume that (i) the closed substructure of B(A♭, N) generated by the joint im- age of the family {ΦA ω }ω∈Ω is B(A♭, N), and (coD) Assume that M∼ has a structural reduct M∼ (ii) dα extends to a Y-morphism bdα : B(A♭, N) → N∼. Then bdeA(a) = kA♭(a), for all a ∈ A. ♭ in Y and let Ω be a sub- ♭, N∼). Let X ∈ X and assume that for every morphism ω (X(X, M∼ )) → N such u : X(X, M∼ ) → M there is a map du : Sω∈Ω ΨX that du ◦ ΨX (1) If Ω ◦ Clo1(M) separates the points of M , then the function u 7→ du ω = ω ◦ u, for all ω ∈ Ω. set of Y(M∼ is one-to-one. (2) Assume that (i) the substructure of Y(X♭, N∼) generated by the joint image of the family {ΨX ω }ω∈Ω is Y(X♭, N∼), and (ii) du extends to a B-morphism bdu : Y(X♭, N∼) → N. Then bdεX(x) = κX♭(x), for all x ∈ X. Vol. 00, XX Piggyback dualities revisited 39 Proof. The differences between the Commuting Triangle Lemma 6.1 and this lemma lie in the weakened assumptions in (D)(2) and (coD)(2). We consider kA♭ (a) are Y-morphisms from B(A♭, N) to N∼, it suffices to prove that they (D)(2) only. We must prove that bdeA(a) = kA♭(a), for all a ∈ A. As bdeA(a) and agree on the generating set Z :=Sω∈Ω ΦA ω (A(A, M)). As bdeA(a) extends deA(a), this is immediate from the proof given for Lemma 6.1(D)(2). (cid:3) We now turn to strong dualities. Our final lemma shows that when we impose sufficient conditions for M∼ to fully dualise M, the resulting duality will in fact be strong. Lemma 6.12 (Injectivity Lemma). Let M and N be structures and let M∼ and N∼ be alter egos of M and N, respectively. Let A := ISP(M), B := ISP(N), X := IScP+(M∼ ) and Y := IScP+(N∼). (D) Assume that M has a structural reduct M♭ in B, let ω ∈ B(M♭, N) and ω : A(A, M) → B(A♭, N) is a bijection for all A ∈ A. If assume that ΦA N is injective in B, then M is injective in A. (coD) Assume that M∼ has a structural reduct M∼ ♭, N∼) and ω : X(X, M∼ ) → Y(X♭, N∼) is a bijection for all X ∈ X. If ♭ in Y, let ω ∈ Y(M∼ assume that ΨX N∼ is injective in Y, then M∼ is injective in X. S → M∼ such that u = v◦ϕ. As ϕ♭ : X♭ → (M∼ S be an embedding in X, for Proof. We shall prove only (coD). Let ϕ : X → M∼ some non-empty set S, and let u : X → M∼ be an X-morphism. We must find an ♭)S is X-morphism v : M∼ a Y-embedding and ω ◦ u♭ : X♭ → N∼ is a Y-morphism, the injectivity of N∼ in Y ♭)S → N∼ with ω◦u = γ◦ϕ. As guarantees the existence of a Y-morphism γ : (M∼ ♭)S, N∼) is surjective, there exists an X-morphism Ψ v : M∼ S → M∼ with ω ◦ v = γ. Since u, v ◦ ϕ ∈ X(X, M∼ ), we can write S, M∼ ) → Y((M∼ S)♭ = (M∼ ω : X(M∼ S M ∼ ΨX ω (u) = ω ◦ u = γ ◦ ϕ = ω ◦ (v ◦ ϕ) = ΨX ω (v ◦ ϕ). As ΨX ω is one-to-one, it follows that u = v ◦ ϕ, as required. (cid:3) Now we present the proof of the Piggyback Strong Duality Theorem 3.7. Proof of Theorem 3.7. (a) Since (III) is a special case of both (I) and (II), it suffices to consider the assumptions in (I) and (II). Both of these sets of assumptions guarantee that Conditions (0), (1), (2)(ii) and (3)(ii)(b) of the Single-carrier Piggyback Duality Theorem 3.2 hold, and that Conditions (1), (2)(ii) and (3)(ii)(b) of the Single-carrier Piggyback Co-duality Theorem 3.4 hold. Hence M∼ fully dualises M. Part (b) will follow immediately from the Injectivity Lemma 6.12 once we have proved Part (c). We shall prove (c) under the assumption that (I) holds. By the Morphism Lemma 6.6, ΨX ω is a Y-morphism, and by the Existence Lemma 6.2(coD)(1)(ii) and the Joint Surjectivity Lemma 6.10(coD)(2) (in 40 B. A. Davey, M. Haviar, and H. A. Priestley Algebra univers. which (ii) is satisfied), ΨX bra, it follows that ΨX Y(X♭, N∼) = K(X♭). ω is one-to-one and onto. Since N is a total alge- ω is an isomorphism between X(X, M∼ )♭ = E(X)♭ and We can use the duality and the fact that E(X)♭ ∼= K(X♭) to prove that D(A)♭ ∼= H(A♭) as follows: since E(X)♭ ∼= K(X♭), for all X ∈ X, we have E(D(A))♭ ∼= K(D(A)♭), and so D(A)♭ ∼= H(K(D(A)♭)) ∼= H(E(D(A))♭) ∼= H(A♭). ω : A(A, M)♭ → B(A♭, N) is an Alternatively, we can prove directly that ΦA isomorphism in Y as follows. By the Morphism Lemma 6.6, ΦA ω is a Y- homomorphism, and by the Existence Lemma 6.2(D)(1)(ii) and the Joint Sur- jectivity Lemma 6.10(D)(2) (in which (ii) is satisfied), ΦA ω is one-to-one and onto. It remains to show that, for each (n-ary) relation r ∈ Rν, (x1, . . . , xn) /∈ r in A(A, M)♭ =⇒ (ΦA ω (xn)) /∈ r in Y(A♭, N). Let x1, . . . , xn ∈ A(A, M) with (x1, . . . , xn) /∈ r in A(A, M)♭. Thus there ♭. As ω ◦ Clo1(M) separates the exists a ∈ A with (x1(a), . . . , xn(a)) /∈ r in M∼ structure M∼ in N∼. Since x1, . . . , xn preserve t it follows that ♭, there exists t ∈ Clo1(M) with(cid:0)ω(t(x1(a))), . . . , ω(t(xn(a)))(cid:1) /∈ r (cid:0)ΦA ω (xn)(t(a))(cid:1) =(cid:0)(ω ◦ x1)(t(a)), . . . , (ω ◦ xn)(t(a))(cid:1) =(cid:0)ω(x1(t(a))), . . . , ω(xn(t(a)))(cid:1) =(cid:0)ω(t(x1(a))), . . . , ω(t(xn(a)))(cid:1) /∈ r, ω (x1)(t(a)), . . . , ΦA ω (x1), . . . , ΦA and so (ΦA ω (x1), . . . , ΦA ω (xn)) /∈ r in Y(A♭, N). Hence (c) holds. (cid:3) Acknowledgement. The authors would like to thank Leonardo Cabrer for his careful reading of several versions of this paper. His well-targetted com- ments pinpointed some initial obscurities and have also helped us make the end product much more user-friendly. References [1] Banaschewski, B.: Remarks on dual adjointness. In: Nordwestdeutsches Kategorienseminar, Tagung, Bremen, 1976. Math.-Arbeitspapiere 7, Teil A: Math. Forschungspapiere, pp. 3 -- 10. Univ. Bremen, Bremen (1976) [2] Cabrer, L.M., Priestley, H.A.: Coproducts of distributive lattice-based algebras. Algebra Universalis (in press). Available at arxiv:1308.4650 [3] Cabrer, L.M., Priestley, H.A.: Natural dualities for varieties of distributive bilattice expansions (2014, preprint) [4] Clark, D.M., Davey, B.A.: Natural Dualities for the Working Algebraist. Cambridge University Press, Cambridge (1998) [5] Clark, D.M., Davey, B.A., Freese, R.S., Jackson, M.: Standard topological algebras: syntactic and principal congruences and profiniteness. Algebra Universalis 52, 343 -- 376 (2004) [6] Davey, B.A.: Natural dualities for structures. Acta Univ. M. Belii Ser. Math. 13, 3 -- 28 (2006). Available at http://actamath.savbb.sk/pdf/acta1301.pdf [7] Davey, B.A., Gouveia, M.J., Haviar, M., Priestley, H.A.: Natural extensions and profinite completions of algebras. Algebra Universalis 66, 205 -- 241 (2011) [8] Davey, B.A., Haviar, M., Priestley, H.A.: Natural dualities in partnership. Appl. Categ. Structures 20, 583 -- 602 (2012) Vol. 00, XX Piggyback dualities revisited 41 [9] Davey, B.A., Haviar, M., Priestley, H.A.: Zero-dimensional Bohr compactifications of algebras and structures (in preparation) [10] Davey, B.A., Haviar, M., Willard, R.: Full does not imply strong, does it? Algebra Universalis 54, 1 -- 22 (2005) [11] Davey, B.A., Haviar, M., Willard, R.: Structural entailment. Algebra Universalis 54, 397 -- 416 (2005) [12] Davey, B.A., Jackson, M., Pitkethly, J.G., Talukder, M.R.: Natural dualities for semilattice-based algebras. Algebra Universalis 57, 463 -- 490 (2007) [13] Davey, B.A., Pitkethly, J.G., Willard, R.: The lattice of alter egos. Internat. J. Algebra Comput. 22 (2012). DOI:10.1142/S021819671100673X [14] Davey, B.A., Priestley, H.A.: Generalized piggyback dualities and applications to Ockham algebras. Houston J. Math. 13, 151 -- 197 (1987) [15] Davey, B.A., Werner, H.: Dualities and equivalences for varieties of algebras. In: Huhn, A.P., Schmidt, E.T. (eds.) Contributions to Lattice Theory (Szeged, 1980). Coll. Math. Soc. J´anos Bolyai, vol. 33, pp. 101 -- 275. North-Holland, Amsterdam (1983) [16] Davey, B.A., Werner, H.: Piggyback-Dualitaten. Bull. Austral. Math. Soc. 32, 1 -- 32 (1985) (German) [17] Davey, B.A., Werner, H.: Piggyback dualities. In: Szab´o, L., Szendrei, ´A (eds.) Lectures in Universal Algebra (Szeged, 1983). Colloq. Math. Soc. J´anos Bolyai, vol. 43, pp. 61 -- 83. North-Holland, Amsterdam (1986) [18] Goldberg, M.S.: Distributive Ockham algebras: free algebras and injectivity. Bull. Austral. Math. Soc. 24, 161 -- 203 (1981) [19] Goldberg, M.S.: Topological duality for distributive Ockham algebras. Studia Logica 42, 23 -- 31 (1983) [20] Hofmann, D.: A generalization of the duality compactness theorem. J. Pure Appl. Algebra 171, 205 -- 217 (2002) [21] Hofmann, K.H., Mislove, M., Stralka, A.: The Pontryagin duality of compact O-dimensional semilattices and its applications. Lecture Notes in Mathematics, vol. 396. Springer (1974) [22] Numakura, K.: Theorems on compact totally disconnected semigroups and lattices. Proc. Amer. Math. Soc. 8, 623 -- 626 (1957) [23] Pitkethly, J.G., Davey, B.A.: Dualisability: Unary Algebras and Beyond. Advances in Mathematics, vol. 9. Springer (2005) [24] Pontryagin, L.S.: The theory of topological commutative groups. Ann. Math. 35, 361 -- 388 (1934) [25] Urquhart, A.: Lattices with a dual homomorphic operation. Studia Logica 38, 201 -- 209 (1979) B. A. Davey Department of Mathematics and Statistics, La Trobe University, Victoria 3086, Australia e-mail : [email protected] M. Haviar Faculty of Natural Sciences, Matej Bel University, Tajovsk´eho 40, 974 01 Bansk´a Bystrica, Slovak Republic e-mail : [email protected] H. A. Priestley Mathematical Institute, University of Oxford, Radcliffe Observatory Quarter, Oxford OX2 6GG, United Kingdom e-mail : [email protected]
1808.02199	1	1808	2018-08-06T10:37:44	Classification of 7-dimensional subalgebras of 8-dimensional Clifford algebra	[ "math.RA" ]	All 7-dimensional subalgebras of the 8-dimensional Clifford algebra over the field C of complex numbers are found. Canonical bases are used throughout the determination. It is found that the 8-dimensional Clifford algebra over C has exactly eight 7-dimensional subalgebras and each of these eight is over the complex numbers. It follows that the 8-dimensional Clifford algebra over C has no 7-dimensional subalgebra over the real numbers. Another consequence is that the eight 7-dimensional subalgebras of the 8-dimentional Clifford over C are in fact 7-dimensional subalgebras of all greater dimensional Clifford algebras over C.	math.RA	math	Classification of 7-dimensional subalgebras of the 8-dimensional Clifford algebra 1 Author: Dr. Uladzimir Shtukar, Associate Professor, North Carolina Central University, USA. e-mail: [email protected] Post address: 1906 Raj Drive, Durham, NC 27703. U. Shtukar 15A21, 17B30 Abstract. All 7-dimensional subalgebras of the 8-dimensional Clifford algebra over the field C of complex numbers are found. Canonical bases are used throughout the determination. It is found that the 8-dimensional Clifford algebra over C has exactly eight 7-dimensional subalgebras and each of these eight is over the complex numbers. It follows that the 8-dimensional Clifford algebra over C has no 7-dimensional subalgebra over the real numbers. Another consequence is that the eight 7- dimensional subalgebras of the 8-dimentional Clifford over C are in fact 7-dimensional subalgebras of all 2n-dimensional Clifford algebras over C for 𝑛 ≥ 3. Key words: Clifford algebra; subalgebras; canonical bases. Let 𝑉𝑛 be the n−dimensional vector space with its standard basis 𝑒1, 𝑒2, … , 𝑒𝑛. The Clifford algebra over 𝑉𝑛 is the associative algebra whose generators are 𝑒𝐴 = 𝑒𝑖1𝑒𝑖2 … 𝑒𝑖𝑘 for each possible index 𝐴 = (𝑖1𝑖2 … 𝑖𝑘) that is a subset of {1,2, …, n} with 𝑖1 < 𝑖2 < ⋯ < 𝑖𝑘. It follows that (as a vector space) the dimension of the Clifford algebra over 𝑉𝑛 is 2𝑛. The multiplicative identity element of the Clifford algebra is 𝑒∅. It is also denoted by 𝑒0 and 1; and, it has the following property 𝑒0𝑒𝑖 = 𝑒𝑖𝑒0 = 𝑒𝑖, 𝑒0𝑒0 = 1. The vectors 𝑒1, 𝑒2, … , 𝑒𝑛 are included in the generators of the Clifford algebra (via index sets A with k = 1) and the products of these particular vectors satisfy the following conditions 𝑒𝑖𝑒𝑗 = −𝑒𝑗𝑒𝑖, (𝑖 ≠ 𝑗), and 𝑒𝑖 2 = −1, where 𝑖, 𝑗 = 1,2, … , 𝑛. As mentioned, the multiplication operation of the Clifford algebra is associative, so (𝑎𝑏)𝑐 = 𝑎(𝑏𝑐) for any elements 𝑎, 𝑏, 𝑐 of the Clifford algebra. An arbitrary element of the Clifford algebra over 𝑉𝑛 is , using all indices 𝐴 = (𝑖1𝑖2 … 𝑖𝑘). We denote the Clifford algebra over 𝑉𝑛 by 𝑔(n,C), when the coefficients 𝑎𝐴 are complex numbers and by g(n,R), when the coefficients 𝑎𝐴 are real numbers. For more information about Clifford algebras, see, for example, [1]. Classification of subalgebras is a fundamental way to gain understanding of any algebra. What can we say about subalgebras of the Clifford algebra 𝑔(n,C)? First, it is easy to see that the Clifford algebra 𝑔(n,C) is a 2𝑛 −dimensional subalgebra of the 2𝑛+𝑘 −dimensional Clifford algebra 𝑔(n+k,C) with 𝑘 ≥ 0. This leads to the more general and equally obvious statement. Lemma. If 𝒉 is a subalgebra of 𝟐𝒏 −dimensional Clifford algebra 𝒈(n,C), then 𝒉 is also a subalgebra of 𝟐𝒏+𝒌 −dimensional Clifford algebra 𝒈(n+k,C), where 𝒌 ≥ 𝟎. A lot of properties of Clifford algebras have been established, and some concern subalgebras. For example, equal rank subalgebras are discussed by E. Meinrenken in his book [2]. An answer to the following question would be real progress in the analysis of Clifford algebras: Does the Clifford algebra 𝑔(n,C) have 𝑘 −dimensional subalgebras for 2𝑛−1 < 𝑘 < 2𝑛? This article answers this AAAaae question for the case 𝑘 =7, 𝑛 =3 by finding all 7-dimensional subalgebras of the 8 −dimensional Clifford algebra 𝑔(3,C). 2 For simplicity, the following notation will be used for the generators of 𝑔(3,C): 𝑒0 = 1, 𝑒1, 𝑒2, 𝑒3, 𝑖 = 𝑒1𝑒2, 𝑗 = 𝑒1𝑒3, 𝑘 = 𝑒2𝑒3, 𝑧 = 𝑒1𝑒2𝑒3. Products of the generators of 𝑔(3,C) are recorded in the following table: . The following statement contains the total description of all 7-dimensional subalgebras of the 8- dimensional Clifford algebra 𝑔(3,C). Theorem. The 8-dimensional Clifford algebra 𝒈(3,C) has the following four 1-parameter sets of 7-dimensional subalgebras over the field of complex numbers 𝑪 ℎ1 = 𝑆𝑝𝑎𝑛{1, 𝑒1 + 𝑘, 𝑒2 + 𝑎𝑘, 𝑒3 + √−1 − 𝑎2𝑘, 𝑖 − √−1 − 𝑎2𝑘, 𝑗 + 𝑎𝑘, 𝑧}, 𝑎 ∈ 𝐶; ℎ2 = 𝑆𝑝𝑎𝑛{1, 𝑒1 + 𝑘, 𝑒2 + 𝑎𝑘, 𝑒3 − √−1 − 𝑎2𝑘, 𝑖 + √−1 − 𝑎2𝑘, 𝑗 + 𝑎𝑘, 𝑧}, 𝑎 ∈ 𝐶; ℎ3 = 𝑆𝑝𝑎𝑛{1, 𝑒1 − 𝑘, 𝑒2 + 𝑎𝑘, 𝑒3 + √−1 − 𝑎2𝑘, 𝑖 + √−1 − 𝑎2𝑘, 𝑗 − 𝑎𝑘, 𝑧}, 𝑎 ∈ 𝐶; ℎ4 = 𝑆𝑝𝑎𝑛{1, 𝑒1 − 𝑘, 𝑒2 + 𝑎𝑘, 𝑒3 − √−1 − 𝑎2𝑘, 𝑖 − √−1 − 𝑎2𝑘, 𝑗 − 𝑎𝑘, 𝑧}, 𝑎 ∈ 𝐶; and four isolated 7-dimensional subalgebras over the field of complex numbers 𝑪 ℎ5 = 𝑆𝑝𝑎𝑛{1, 𝑒1, 𝑒2 + 𝑗, 𝑒3 + √−1𝑗, 𝑖 + √−1𝑗, 𝑘, 𝑧}, ℎ6 = 𝑆𝑝𝑎𝑛{1, 𝑒1, 𝑒2 + 𝑗, 𝑒3 − √−1𝑗, 𝑖 − √−1𝑗, 𝑘, 𝑧}, ℎ7 = 𝑆𝑝𝑎𝑛{1, 𝑒1, 𝑒2 − 𝑗, 𝑒3 + √−1𝑗, 𝑖 − √−1𝑗, 𝑘, 𝑧}, ℎ8 = 𝑆𝑝𝑎𝑛{1, 𝑒1, 𝑒2 − 𝑗, 𝑒3 − √−1𝑗, 𝑖 + √−1𝑗, 𝑘, 𝑧}. The Clifford algebra 𝒈(3,C) has no 7-dimensional subalgebras over the field of real numbers. Proof. Since 7-dimensional subalgebras of 𝑔(3,C) are 7-dimensional subspaces of 𝑔(3,C), we first identify the 7-dimensional subspaces of 𝑔(3,C) and then we identify which of these subspaces are subalgebras. To find 7-dimensional subspaces of the 8-dimensional 𝑔(3,C), we use the canonical bases for (𝑛 − 1) −dimensional subspaces of 𝑛 −dimensional vector space that are detailed in [3]. Our case is 𝑛 = 8, and the corresponding canonical bases for this case are 1231231123221333122133123213211111111111eeeijkzeeeijkzeeijeezkeeikezejeejkzeeiiieezkjejjezekiekkzeejiezzkjieee3 (1) 𝑎1 = 1 + 𝑎18𝑧, 𝑎2 = 𝑒1 + 𝑎28𝑧, 𝑎3 = 𝑒2 + 𝑎38𝑧, 𝑎4 = 𝑒3 + 𝑎48𝑧, 𝑎5 = 𝑖 + 𝑎58𝑧, 𝑎6 = 𝑗 + 𝑎68𝑧, 𝑎7 = 𝑘 + 𝑎78𝑧; (2) 𝑎1 = 1 + 𝑎17𝑘, 𝑎2 = 𝑒1 + 𝑎27𝑘, 𝑎3 = 𝑒2 + 𝑎37𝑘, 𝑎4 = 𝑒3 + 𝑎47𝑘, 𝑎5 = 𝑖 + 𝑎57𝑘, 𝑎6 = 𝑗 + 𝑎67𝑘, 𝑎7 = 𝑧; (3) 𝑎1 = 1 + 𝑎16𝑗, 𝑎2 = 𝑒1 + 𝑎26𝑗, 𝑎3 = 𝑒2 + 𝑎36𝑗, 𝑎4 = 𝑒3 + 𝑎46𝑗, 𝑎5 = 𝑖 + 𝑎56𝑗, 𝑎6 = 𝑘, 𝑎7 = 𝑧; (𝟒) 𝑎1 = 1 + 𝑎15𝑖, 𝑎2 = 𝑒1 + 𝑎25𝑖, 𝑎3 = 𝑒2 + 𝑎35𝑖, 𝑎4 = 𝑒3 + 𝑎45𝑖, 𝑎5 = 𝑗, 𝑎6 = 𝑘, 𝑎7 = 𝑧; (𝟓) 𝑎1 = 1 + 𝑎14𝑒3, 𝑎2 = 𝑒1 + 𝑎24𝑒3, 𝑎3 = 𝑒2 + 𝑎34𝑒3, 𝑎4 = 𝑖, 𝑎5 = 𝑗, 𝑎6 = 𝑘, 𝑎7 = 𝑧; (𝟔) 𝑎1 = 1 + 𝑎13𝑒2, 𝑎2 = 𝑒1 + 𝑎23𝑒2, 𝑎3 = 𝑒3, 𝑎4 = 𝑖, 𝑎5 = 𝑗, 𝑎6 = 𝑘, 𝑎7 = 𝑧; (𝟕) 𝑎1 = 1 + 𝑎12𝑒1, 𝑎2 = 𝑒2, 𝑎3 = 𝑒3, 𝑎4 = 𝑖, 𝑎5 = 𝑗, 𝑎6 = 𝑘, 𝑎7 = 𝑧; (𝟖) 𝑎1 = 𝑒1, 𝑎2 = 𝑒2, 𝑎3 = 𝑒3, 𝑎4 = 𝑖, 𝑎5 = 𝑗, 𝑎6 = 𝑘, 𝑎7 = 𝑧. Every 7-dimensional subalgebra of 𝑔(3,C) is a subspace of 𝑔(3,C) and hence is associated with exactly one of these 8 canonical bases for some choice of the parameters 𝑎𝑖𝑗 . To determine which subspaces of 𝑔(3,C) are subalgebras, we use the well-known fact, that a subspace ℎ of 𝑔(3,C) is a subalgebra of 𝑔(3,C) if and only if 𝑥𝑦 ∈ ℎ for any two elements 𝑥 ∈ ℎ and 𝑦 ∈ ℎ. So, for each of the 8 canonical bases we find all products 𝑎𝑖𝑎𝑗 for 𝑖, 𝑗 = 1, 2, 3, 4, 5, 6, 7, and then determine whether or not all products 𝑎𝑖𝑎𝑗 are in ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7}. Specifically, for each of the 8 canonical bases we find all products 𝑎𝑖𝑎𝑗 for 𝑖, 𝑗 = 1, 2, 3, 4, 5, 6, 7, and then determine whether or not 𝑥1, 𝑥2, 𝑥3, 𝑥4, 𝑥5, 𝑥6, 𝑥7 exist in C so that 𝑎𝑖𝑎𝑗 = 𝑥1𝑎1 + 𝑥2𝑎2 + 𝑥3𝑎3 + 𝑥4𝑎4 + 𝑥5𝑎5 + 𝑥6𝑎6 + 𝑥7𝑎7 . The solutions to this equation are found by expressing each 𝑎𝑖 in terms of generators of 𝑔(3,C) and then equating the coefficients from each side of the equation for corresponding generators. This technique either a) establishes conditions on the parameters 𝑎𝑖𝑎𝑗 for a solution for the 𝑥𝑖 to exist or b) establishes that there is no solution. In case a) a subalgebra of 𝑔(3,C) is associated with the canonical basis. In case b) no subalgebra of 𝑔(3,C) is associated with the canonical basis. Basis (1). Products 𝑎2𝑎7, 𝑎7𝑎2, 𝑎3𝑎4, 𝑎4𝑎3, 𝑎3𝑎5, 𝑎5𝑎3 play crucial roles for this basis. Start with 𝑎3𝑎4, and 𝑎4𝑎3. Using the table of products above, we have 𝑎3𝑎4 = (𝑒2 + 𝑎38𝑧)(𝑒3 + 𝑎48𝑧) = 𝑘 + 𝑎38𝑎48 + 𝑎48𝑗 − 𝑎38𝑖. To check whether or not this is in ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7}, we look for 𝑥1, 𝑥2, 𝑥3, 𝑥4, 𝑥5, 𝑥6, 𝑥7 so that 𝑘 + 𝑎38𝑎48 + 𝑎48𝑗 − 𝑎38𝑖 = 𝑥1𝑎1 + 𝑥2𝑎2 + 𝑥3𝑎3 + 𝑥4𝑎4 + 𝑥5𝑎5 + 𝑥6𝑎6 + 𝑥7𝑎7. By expressing each 𝑎𝑖 in terms of generators of 𝑔(3,C) and equating the coefficients from each side of the equation for corresponding generators, we find a solution exists ( 𝑥1 = 𝑎38𝑎48, 𝑥2 = 0, 𝑥3 = 0, 𝑥4 = 0, 𝑥5 = −𝑎38, 𝑥6 = 𝑎48, 𝑥7 = 1 ), if 𝑎18𝑎38𝑎48 − 𝑎38𝑎58 + 𝑎48𝑎68 + 𝑎78 = 0. Similarly, the product 𝑎4𝑎3 requires the following condition for it to be in h: 𝑎18𝑎38𝑎48 − 𝑎38𝑎58 + 𝑎48𝑎68 − 𝑎78 = 0. From the last two equations, we must have 𝑎78 = 0 for both 𝑎3𝑎4, and 𝑎4𝑎3 to be in h. Next, we have 𝑎3𝑎5 = (𝑒2 + 𝑎38𝑧)(𝑖 + 𝑎58𝑧) = 𝑒1 + 𝑎38𝑎58 − 𝑎38𝑒3 + 𝑎58𝑗. Using the same procedure that we used above, we find 𝑎3𝑎5 is in h, if 𝑎18𝑎38𝑎58 + 𝑎28 − 𝑎38𝑎48 + 𝑎58𝑎68 = 0. Similarly, the product 𝑎5𝑎3 requires the condition 𝑎28 − 𝑎38𝑎48 + 𝑎58𝑎68 = 0 for it to be in ℎ. From the last two equations, we must have 𝑎28 = 0 for both 𝑎3𝑎5 and 𝑎5𝑎3 to be in h. Now compute 𝑎2𝑎7 = (𝑒1 + 𝑎28𝑧)(𝑘 + 𝑎78𝑧) = 𝑧 + 𝑎28𝑎78 − 𝑎78𝑘 − 𝑎28𝑒1. Using the same 2 = 1. Product 𝑎7𝑎2 requires the same procedure, we find 𝑎2𝑎7 is in h, if 𝑎18𝑎28𝑎78 − 𝑎28 condition to be in h. If our previous requirements 𝑎78 = 0 and 𝑎28 = 0 are substituted into 2 − 𝑎78 4 𝑎18𝑎28𝑎78 − 𝑎28 for all 𝑎𝑖𝑎𝑗 to be in h. So, Basis (1) doesn't generate any subalgebra in algebra 𝑔(3, 𝐶). 2 = 1, the contradiction 0 = 1 is produced. This means that it is impossible 2 − 𝑎78 Basis (2). This basis consists of vectors 𝑎1 = 1 + 𝑎17𝑘, 𝑎2 = 𝑒1 + 𝑎27𝑘, 𝑎3 = 𝑒2 + 𝑎37𝑘, 𝑎4 = 𝑒3 + 𝑎47𝑘, 𝑎5 = 𝑖 + 𝑎57𝑘, 𝑎6 = 𝑗 + 𝑎67𝑘, 𝑎7 = 𝑧. Find all products 𝑎𝑖𝑎𝑗 where 𝑖, 𝑗 = 1, 2, 3, 4, 5, 6, 7, and find conditions so that all products 𝑎𝑖𝑎𝑗 are in ℎ where ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7}. Using the table of products, we have 𝑎1𝑎7 = (1 + 𝑎17𝑘)𝑧 = 𝑧 − 𝑎17𝑒1. To determine if this is in h, set it equal to 𝑥1𝑎1 + 𝑥2𝑎2 + 𝑥3𝑎3 + 𝑥4𝑎4 + 𝑥5𝑎5 + 𝑥6𝑎6 + 𝑥7𝑎7, express each 𝑎𝑖 in terms of the generators of 𝑔(3, C), and solve for the 𝑥𝑖. This procedure yields a solution ( 𝑥1 = 𝑥3 = 𝑥4 = 𝑥5 = 𝑥6 = 0, 𝑥2 = −𝑎17 , 𝑥7 = 1 ) , if the condition 𝑎17𝑎27 = 0 is met. For the product 𝑎2𝑎7 we have 𝑎2𝑎7 = (𝑒1 + 𝑎27𝑘)𝑧 = −𝑘 − 𝑎27𝑒1. By following the same technique, this is in ℎ, if 𝑎27 condition (𝑎17𝑎27 = 0) is equivalent to 𝑎17 = 0. Next, the product 𝑎3𝑎7 = (𝑒2 + 𝑎37𝑘)𝑧 = 𝑗 − 𝑎37𝑒1 is in ℎ, if −𝑎27𝑎37 + 𝑎67 = 0. The product 𝑎4𝑎7 = (𝑒3 + 𝑎47𝑘)𝑧 = −𝑖 − 𝑎47𝑒1 is in ℎ, if 𝑎27𝑎47 + 𝑎57 = 0. The product 𝑎5𝑎7 = (𝑖 + 𝑎57𝑘)𝑧 = −𝑒3 − 𝑎57𝑒1 is in ℎ, if 𝑎27𝑎57 + 𝑎47 = 0. This is not a new condition. Since 𝑎27 condition for 𝑎4𝑎7 to be in h. The product 𝑎6𝑎7 = (𝑗 + 𝑎67𝑘)𝑧 = 𝑒2 − 𝑎67𝑒1 is in ℎ, if −𝑎27𝑎67 + 2 = 1, it is equivalent to −𝑎27𝑎37 + 𝑎67 = 0 which 𝑎37 = 0. This condition is also not new. Since 𝑎27 is the condition for 𝑎3𝑎7 to be in h. The following initial list summarizes the conditions that the parameters 𝑎17, 𝑎27, 𝑎37, 𝑎47, 𝑎57, 𝑎67 must satisfy to insure that all products 𝑎𝑖𝑎𝑗 considered thus far will be in h = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7} : 2 = 1, it is equivalent to 𝑎27𝑎47 + 𝑎57 = 0 which is the 2 = 1. Under this condition, the previous 𝑎17 = 0, 𝑎27 2 = 1, 𝑎57 = −𝑎27𝑎47, and 𝑎67 = 𝑎27𝑎37. Next note that 𝑎1𝑎1 = (1 + 𝑎17𝑘)(1 + 𝑎17𝑘) = 1 − 𝑎17 𝑎17 = 0 which is one of the conditions in our initial list. Similarly, we have 𝑎2𝑎2 ∈ ℎ, 𝑎3𝑎3 ∈ ℎ, 𝑎4𝑎4 ∈ ℎ, 𝑎5𝑎5 ∈ ℎ, 𝑎6𝑎6 ∈ ℎ, 𝑎7𝑎7 ∈ ℎ. Also, since 𝑎17 = 0, we have 𝑎1 = 1 and, therefore, 𝑎1𝑎2 = 𝑎2𝑎1 = 𝑎2, 𝑎1𝑎3 = 𝑎3𝑎1 = 𝑎3, 𝑎1𝑎4 = 𝑎4𝑎1 = 𝑎4, 𝑎1𝑎5 = 𝑎5𝑎1 = 𝑎5, 𝑎1𝑎6 = 𝑎6𝑎1 = 𝑎6. So, all of these products are in h without requiring any additional conditions for parameters 𝑎17, 𝑎27, 𝑎37, 𝑎47, 𝑎57, 𝑎67. 2 + 2𝑎17𝑘. It turns out that this is in h, if Additionally, the products 𝑎7𝑎1, 𝑎7𝑎2, 𝑎7𝑎3, 𝑎7𝑎4, 𝑎7𝑎5, 𝑎7𝑎6 are in h without requiring any additional conditions for parameters 𝑎17, 𝑎27, 𝑎37, 𝑎47, 𝑎57, 𝑎67. These products are equal to 𝑎1𝑎7, 𝑎2𝑎7, 𝑎3𝑎7, 𝑎4𝑎7, 𝑎5𝑎7, 𝑎6𝑎7 respectively and all of these products were used in constructing the initial list of conditions on the parameters. Now consider the remaining products of basic elements for Basis (2). For each, we will use the technique employed above to find any condition that must be met for the product to belong to ℎ. The product 𝑎2𝑎3 = (𝑒1 + 𝑎27𝑘)(𝑒2 + 𝑎37𝑘) = 𝑖 + 𝑎37𝑧 + 𝑎27𝑒3 − 𝑎27𝑎37 is in ℎ, if 𝑎27𝑎47 + 𝑎57 = 0. The same condition is required for 𝑎3𝑎2 to be in ℎ. This condition is in the initial list. Next, 𝑎2𝑎4 = (𝑒1 + 𝑎27𝑘)(𝑒3 + 𝑎47𝑘) = 𝑗 + 𝑎47𝑧 − 𝑎27𝑒2 − 𝑎27𝑎47 is in ℎ, if −𝑎27𝑎37 + 𝑎67 = 0. The same condition is required for 𝑎4𝑎2 to be in ℎ. This condition is in the initial list. Next, 𝑎2𝑎5 = (𝑒1 + 𝑎27𝑘)(𝑖 + 𝑎57𝑘) = −𝑒2 + 𝑎57𝑧 + 𝑎27𝑗 − 𝑎27𝑎57 is in ℎ, if − 𝑎37 + 𝑎27𝑎67 = 0. The same condition is required for 𝑎5𝑎2 to be in ℎ. Since 𝑎27 to 𝑎67 = 𝑎27𝑎37 which is in the initial list. 2 = 1, this condition is equivalent 5 Next, 𝑎2𝑎6 = (𝑒1 + 𝑎27𝑘)(𝑗 + 𝑎67𝑘) = −𝑒3 + 𝑎67𝑧 − 𝑎27𝑖 − 𝑎27𝑎67 is in ℎ, if − 𝑎47 − 𝑎27𝑎57 = 2 = 1, this condition is equivalent to 0. The same condition is required for 𝑎6𝑎2 to be in h. Since 𝑎27 𝑎57 = −𝑎27𝑎47 which is in the initial list. Next, 𝑎3𝑎4 = (𝑒2 + 𝑎37𝑘)(𝑒3 + 𝑎47𝑘) = 𝑘 − 𝑎47𝑒3 − 𝑎37𝑒2 − 𝑎37𝑎47 is in h, if −𝑎37𝑎37 − 𝑎47𝑎47 = 1. The same condition is required for 𝑎4𝑎3 to be in h. This is a new condition. Next, 𝑎3𝑎5 = (𝑒2 + 𝑎37𝑘)(𝑖 + 𝑎57𝑘) = 𝑒1 − 𝑎57𝑒3 + 𝑎37𝑗 − 𝑎37𝑎57 is in h, if 𝑎27 − 𝑎47𝑎57 + 𝑎37𝑎67 = 0. The same condition is required for 𝑎5𝑎3 to be in h. This is a new condition. Next, 𝑎3𝑎6 = (𝑒2 + 𝑎37𝑘)(𝑗 + 𝑎67𝑘) = −𝑧 − 𝑎67𝑒3 − 𝑎37𝑖 − 𝑎37𝑎67 is in ℎ, if − 𝑎47𝑎67 − 𝑎37𝑎57 = 0. The same condition is required for 𝑎6𝑎3 to be in h. This is a new condition. Next, 𝑎4𝑎5 = (𝑒3 + 𝑎47𝑘)(𝑖 + 𝑎57𝑘) = 𝑧 − 𝑎57𝑒2 + 𝑎47𝑗 − 𝑎47𝑎57 is in ℎ, if 𝑎37𝑎57 + 𝑎47𝑎67 = 0. The same condition is required for 𝑎5𝑎4 to be in h. This condition is the same as the new condition found for 𝑎3𝑎6. Next, 𝑎4𝑎6 = (𝑒3 + 𝑎47𝑘)(𝑗 + 𝑎67𝑘) = 𝑒1 + 𝑎67𝑒2 − 𝑎47𝑖 − 𝑎47𝑎67 is in ℎ, if 𝑎27 + 𝑎37𝑎67 − 𝑎47𝑎57 = 0. The same condition is required for 𝑎6𝑎4 to be in h. This condition is the same as the new condition found for 𝑎3𝑎5. Next, 𝑎5𝑎6 = (𝑖 + 𝑎57𝑘)(𝑗 + 𝑎67𝑘) = 𝑘 − 𝑎67𝑗 − 𝑎57𝑖 − 𝑎57𝑎67 is in h, if 𝑎57𝑎57 + 𝑎67𝑎67 = −1. The same condition is required for 𝑎6𝑎5 to be in h. This is equivalent to the new condition −𝑎37𝑎37 − 𝑎47𝑎47 = 1 found for 𝑎3𝑎4. To see this, note that the three conditions 𝑎57 = −𝑎27𝑎47, 𝑎67 = 𝑎27𝑎37 and 𝑎27 2 = 1 are in the initial list and together imply 𝑎57 2 and 𝑎67 2 = 𝑎47 2 = 𝑎37 2 . The following list summarizes the conditions that the parameters 𝑎17, 𝑎27, 𝑎37, 𝑎47, 𝑎57, 𝑎67 must satisfy to insure that all products 𝑎𝑖𝑎𝑗 considered thus far will be in ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7} for 𝑖, 𝑗 = 1, 2, 3, 4, 5, 6, 7: 𝑎17 = 0, 𝑎27 2 = 1, 𝑎57 = −𝑎27𝑎47, 𝑎67 = 𝑎27𝑎37, −𝑎37𝑎37 − 𝑎47𝑎47 = 1, −𝑎47𝑎67 − 𝑎37𝑎57 = 0, 𝑎27 + 𝑎37𝑎67 − 𝑎47𝑎57 = 0. 2 + 𝑎47 2 = −1. The last equation has no solution If 𝑎27 = 1, then 𝑎57 = −𝑎47, 𝑎67 = 𝑎37, and 𝑎37 in the field of real numbers but it has solutions in the field of complex numbers. For convenience, denote 𝑎37 = 𝑎. Then the parameter values 𝑎17 = 0, 𝑎27 = 1, 𝑎37 = 𝑎, 𝑎47 = ±√−1 − 𝑎2, 𝑎57 = ∓√−1 − 𝑎2, and 𝑎67 = 𝑎 clearly satisfy the first five of the seven conditions that are listed above as sufficient to put all products 𝑎𝑖𝑎𝑗 in ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7}. It is easy to check that these parameter values also satisfy the last two conditions. Therefore, the following 1-parameter sets of subalgebras in algebra 𝑔(3, 𝐶) are found: ℎ1 = 𝑆𝑝𝑎𝑛{1, 𝑒1 + 𝑘, 𝑒2 + 𝑎𝑘, 𝑒3 + √−1 − 𝑎2𝑘, 𝑖 − √−1 − 𝑎2𝑘, 𝑗 + 𝑎𝑘, 𝑧}, 𝑎 ∈ 𝐶; ℎ2 = 𝑆𝑝𝑎𝑛{1, 𝑒1 + 𝑘, 𝑒2 + 𝑎𝑘, 𝑒3 − √−1 − 𝑎2𝑘, 𝑖 + √−1 − 𝑎2𝑘, 𝑗 + 𝑎𝑘, 𝑧}, 𝑎 ∈ 𝐶. 2 = −1. The last equation has no solution If 𝑎27 = −1, then 𝑎57 = 𝑎47, 𝑎67 = −𝑎37, 𝑎37 in the field of real numbers but it has solutions in the field of complex numbers. For convenience, denote 𝑎37 = 𝑎. Then the parameter values 𝑎17 = 0, 𝑎27 = −1, 𝑎37 = 𝑎, 𝑎47 = ±√−1 − 𝑎2, 𝑎57 = ±√−1 − 𝑎2 and 𝑎67 = −𝑎 clearly satisfy the first five of the seven conditions that are listed above as sufficient to put all products 𝑎𝑖𝑎𝑗 in ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7}. It is easy to check 2 + 𝑎47 that these parameter values also satisfy the last two conditions. Therefore, the following 1-parameter sets of subalgebras in algebra 𝑔(3, 𝐶) are found: 6 ℎ3 = 𝑆𝑝𝑎𝑛{1, 𝑒1 − 𝑘, 𝑒2 + 𝑎𝑘, 𝑒3 + √−1 − 𝑎2𝑘, 𝑖 + √−1 − 𝑎2𝑘, 𝑗 − 𝑎𝑘, 𝑧}, 𝑎 ∈ 𝐶; ℎ4 = 𝑆𝑝𝑎𝑛{1, 𝑒1 − 𝑘, 𝑒2 + 𝑎𝑘, 𝑒3 − √−1 − 𝑎2𝑘, 𝑖 − √−1 − 𝑎2𝑘, 𝑗 − 𝑎𝑘, 𝑧}, 𝑎 ∈ 𝐶. Basis (3). This basis consists of vectors 𝑎1 = 1 + 𝑎16𝑗, 𝑎2 = 𝑒1 + 𝑎26𝑗, 𝑎3 = 𝑒2 + 𝑎36𝑗, 𝑎4 = 𝑒3 + 𝑎46𝑗, 𝑎5 = 𝑖 + 𝑎56𝑗, 𝑎6 = 𝑘, 𝑎7 = 𝑧. Consider subspace ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7}. Find all products 𝑎𝑖𝑎𝑗 where 𝑖, 𝑗 = 1, 2, 3, 4, 5, 6, 7, and find conditions so that all products 𝑎𝑖𝑎𝑗 are in ℎ. Using the table of products, we have 𝑎1𝑎7 = (1 + 𝑎16𝑗)𝑧 = 𝑧 + 𝑎16𝑒2. This turns out to be in h, if 𝑎16𝑎36 = 0. For the product 𝑎2𝑎7 we have 𝑎2𝑎7 = (𝑒1 + 𝑎26𝑗)𝑧 = −𝑘 + 𝑎26𝑒2. This is in h, if 𝑎26𝑎36 = 0. Next, compute the product 𝑎3𝑎7 = (𝑒2 + 𝑎36𝑗)𝑧 = 𝑗 + 𝑎36𝑒2. 2 = 1. Under this new condition, the two previous conditions (𝑎16𝑎36 = 0 and This is in h, if 𝑎36 𝑎26𝑎36 = 0) are equivalent to requiring 𝑎16 = 0 and 𝑎26 = 0. Now compute product 𝑎4𝑎7 = (𝑒3 + 𝑎46𝑗)𝑧 = −𝑖 + 𝑎46𝑒2. This is in h, if 𝑎36𝑎46 − 𝑎56 = 0. Next, compute 𝑎5𝑎7 = (𝑖 + 𝑎56𝑗)𝑧 = −𝑒3 + 𝑎56𝑒2. This is in ℎ, if 𝑎36𝑎56 − 𝑎46 = 0. This condition is not new. Since 2 = 1, it is equivalent to 𝑎36𝑎46 − 𝑎56 = 0 which is the condition for 𝑎4𝑎7 to be in h. Compute 𝑎36 product 𝑎6𝑎7 = 𝑘𝑧 = −𝑒1. This is in h, if 𝑎26 = 0. This is also a condition we have already found. The following initial list summarizes the conditions that the parameters 𝑎16, 𝑎26, 𝑎36, 𝑎46, 𝑎56 must satisfy to insure that all products 𝑎𝑖𝑎𝑗 considered thus far will be in ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7} : 𝑎16 = 0, 𝑎26 = 0, 𝑎36 2 = 1, and 𝑎56 = 𝑎36𝑎46. Next note that 𝑎1𝑎1 = (1 + 𝑎16𝑗)(1 + 𝑎16𝑗) = 1 − 𝑎16 𝑎16 = 0 which is one of the conditions listed above. Similarly, we have 𝑎2𝑎2 ∈ ℎ, 𝑎3𝑎3 ∈ ℎ, 𝑎4𝑎4 ∈ ℎ, 𝑎5𝑎5 ∈ ℎ, 𝑎6𝑎6 ∈ ℎ, 𝑎7𝑎7 ∈ ℎ. Also, since 𝑎16 = 0, we have 𝑎1 = 1 and, therefore, 𝑎1𝑎2 = 𝑎2𝑎1 = 𝑎2, 𝑎1𝑎3 = 𝑎3𝑎1 = 𝑎3, 𝑎1𝑎4 = 𝑎4𝑎1 = 𝑎4, 𝑎1𝑎5 = 𝑎5𝑎1 = 𝑎5, 𝑎1𝑎6 = 𝑎6𝑎1 = 𝑎6. So, all of these products are in h without requiring any additional conditions for parameters 𝑎16, 𝑎26, 𝑎36, 𝑎46, 𝑎56. 2 + 2𝑎16𝑗. It turns out that this is in h, if Additionally, the products 𝑎7𝑎1, 𝑎7𝑎2, 𝑎7𝑎3, 𝑎7𝑎4, 𝑎7𝑎5, 𝑎7𝑎6 are in h without requiring any additional conditions for parameters 𝑎16, 𝑎26, 𝑎36, 𝑎46, 𝑎56. These products are equal to 𝑎1𝑎7, 𝑎2𝑎7, 𝑎3𝑎7, 𝑎4𝑎7, 𝑎5𝑎7, 𝑎6𝑎7 respectively and all of these products were used in constructing the initial list of conditions on the parameters. Now consider the remaining products of basic elements for Basis (3). For each, we will use the technique employed above to find any condition that must be met for the product to belong to ℎ. The product 𝑎2𝑎3 = (𝑒1 + 𝑎26𝑗)(𝑒2 + 𝑎36𝑗) = 𝑖 − 𝑎36𝑒3 − 𝑎26𝑧 − 𝑎26𝑎36 is in ℎ, if −𝑎36𝑎46 + 𝑎56 = 0. The same condition is required for 𝑎3𝑎2 to be in h. This condition is in the initial list above. Next, 𝑎2𝑎4 = (𝑒1 + 𝑎26𝑗)(𝑒3 + 𝑎46𝑗) = 𝑗 − 𝑎46𝑒3 − 𝑎26𝑒1 − 𝑎26𝑎46 is in h, if −𝑎26𝑎26 − 𝑎46𝑎46 = 1. The same condition is required for 𝑎4𝑎2 to be in h. Since 𝑎26 = 0, this is equivalent to 𝑎46𝑎46 = −1. This is a new condition. Next, 𝑎2𝑎5 = (𝑒1 + 𝑎26𝑗)(𝑖 + 𝑎56𝑗) = −𝑒2 − 𝑎56𝑒3 − 𝑎26𝑘 − 𝑎26𝑎56 is in h, if −𝑎36 − 𝑎46𝑎56 = 0. The same condition is required for 𝑎5𝑎2 to be in h. Since 𝑎46𝑎46 = −1 (the new condition for 𝑎2𝑎4), this condition is equivalent to 𝑎56 = 𝑎36𝑎46 which is in the initial list. Next, 𝑎2𝑎6 = (𝑒1 + 𝑎26𝑗)𝑘 = 𝑧 + 𝑎26𝑖 is in h, if 𝑎26𝑎56 = 0. The same condition is required for 𝑎6𝑎2 to be in h. This condition is a consequence of 𝑎26 = 0 which is in the initial list. Next, 𝑎3𝑎4 = (𝑒2 + 𝑎36𝑗)(𝑒3 + 𝑎46𝑗) = 𝑘 − 𝑎46𝑧 − 𝑎36𝑒1 − 𝑎36𝑎46 is in ℎ, if 𝑎26𝑎36 = 0. The same condition is required for 𝑎4𝑎3 to be in h. This condition is a consequence of 𝑎26 = 0 which is in the initial list. 7 Next, 𝑎3𝑎5 = (𝑒2 + 𝑎36𝑗)(𝑖 + 𝑎56𝑗) = 𝑒1 − 𝑎56𝑧 − 𝑎36𝑘 − 𝑎36𝑎56 is in h, if 𝑎26 = 0. Product 𝑎5𝑎3 requires the same condition to be in h. This condition is in the initial list. Next, 𝑎3𝑎6 = (𝑒2 + 𝑎36𝑗)𝑘 = −𝑒3 + 𝑎36𝑖 is in h, if −𝑎46 + 𝑎36𝑎56 = 0. Product 𝑎6𝑎3 requires 2 = 1, this condition is equivalent to 𝑎56 = 𝑎36𝑎46 which is the same condition to be in h. Since 𝑎36 in the initial list. Next, 𝑎4𝑎5 = (𝑒3 + 𝑎46𝑗)(𝑖 + 𝑎56𝑗) = 𝑧 + 𝑎56𝑒1 − 𝑎46𝑘 − 𝑎46𝑎56 is in h, if 𝑎26𝑎56 = 0. Similarly, product 𝑎5𝑎4 requires the same condition to be in h. This condition is a consequence of 𝑎26 = 0 which is in the initial list. Next, 𝑎4𝑎6 = (𝑒3 + 𝑎46𝑗)𝑘 = 𝑒2 + 𝑎46𝑖 is in h, if 𝑎36 + 𝑎46𝑎56 = 0. Product 𝑎6𝑎4 requires the same condition to be in h. Since 𝑎46𝑎46 = −1(the new condition for 𝑎2𝑎4), this condition is equivalent to 𝑎56 = 𝑎36𝑎46 which is in the initial list. Next, 𝑎5𝑎6 = (𝑖 + 𝑎56𝑗)𝑘 = −𝑗 + 𝑎56𝑖 is in h, if 𝑎56𝑎56 = −1. Product 𝑎6𝑎5 requires the same condition to be in h. This is a new condition. The following list summarizes the conditions that the parameters 𝑎16, 𝑎26, 𝑎36, 𝑎46, 𝑎56 must satisfy to insure that all products 𝑎𝑖𝑎𝑗 considered thus far will be in h = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7} for 𝑖, 𝑗 = 1, 2, 3, 4, 5, 6, 7: 𝑎16 = 0, 𝑎26 = 0, 𝑎36 2 = 1, 𝑎56 = 𝑎36𝑎46, 𝑎46𝑎46 = −1, 𝑎56𝑎56 = −1. If 𝑎36 = 1, then the parameter values 𝑎16 = 0, 𝑎26 = 0, 𝑎36 = 1, 𝑎46 = ±√−1, and 𝑎56 = 𝑎46 = ±√−1 clearly satisfy the six conditions that are listed above as sufficient to put all products 𝑎𝑖𝑎𝑗 in ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7}. If 𝑎36 = −1, then the parameter values 𝑎16 = 0, 𝑎26 = 0, 𝑎36 = −1, 𝑎46 = ±√−1, and 𝑎56 = −𝑎46 = ∓√−1 clearly satisfy the six conditions that are listed above as sufficient to put all products 𝑎𝑖𝑎𝑗 in ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7}. Therefore, the system has four solutions, and four subalgebras of algebra 𝑔(3, 𝐶) are found: ℎ5 = 𝑆𝑝𝑎𝑛{1, 𝑒1, 𝑒2 + 𝑗, 𝑒3 + √−1𝑗, 𝑖 + √−1𝑗, 𝑘, 𝑧}, ℎ6 = 𝑆𝑝𝑎𝑛{1, 𝑒1, 𝑒2 + 𝑗, 𝑒3 − √−1𝑗, 𝑖 − √−1𝑗, 𝑘, 𝑧}, ℎ7 = 𝑆𝑝𝑎𝑛{1, 𝑒1, 𝑒2 − 𝑗, 𝑒3 + √−1𝑗, 𝑖 − √−1𝑗, 𝑘, 𝑧}, ℎ8 = 𝑆𝑝𝑎𝑛{1, 𝑒1, 𝑒2 − 𝑗, 𝑒3 − √−1𝑗, 𝑖 + √−1𝑗, 𝑘, 𝑧}. Basis (4). This basis consists of vectors 𝑎1 = 1 + 𝑎15𝑖, 𝑎2 = 𝑒1 + 𝑎25𝑖, 𝑎3 = 𝑒2 + 𝑎35𝑖, 𝑎4 = 𝑒3 + 𝑎45𝑖, 𝑎5 = 𝑗, 𝑎6 = 𝑘, 𝑎7 = 𝑧. Consider the subspace ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7} and the specific products 𝑎5𝑎6 and 𝑎6𝑎5. We have 𝑎5𝑎6 = 𝑗𝑘 = 𝑖, and 𝑎6𝑎5 = 𝑘𝑗 = −𝑖. But, it is easy to check that vectors 𝑖 and −𝑖 don't belong to the subspace ℎ. Therefore, ℎ is not a subalgebra of algebra 𝑔(3, 𝐶). Basis (5). This basis consists of vectors 𝑎1 = 1 + 𝑎14𝑒3, 𝑎2 = 𝑒1 + 𝑎24𝑒3, 𝑎3 = 𝑒2 + 𝑎34𝑒3, 𝑎4 = 𝑖, 𝑎5 = 𝑗, 𝑎6 = 𝑘, 𝑎7 = 𝑧. Consider the subspace ℎ = 𝑆𝑝𝑎𝑛{𝑎1, 𝑎2, 𝑎3, 𝑎4, 𝑎5, 𝑎6, 𝑎7} and the specific products 𝑎4𝑎7 and 𝑎7𝑎4. We have 𝑎4𝑎7 = 𝑖𝑧 = −𝑒3, and 𝑎7𝑎4 = 𝑧𝑖 = −𝑒3. But, it is easy to check that vector 𝑒3 doesn't belong to the subspace ℎ. Therefore, ℎ is not a subalgebra of algebra 𝑔(3, 𝐶). Bases (6), (7), (8). These bases don't generate any subalgebra of algebra 𝑔(3, 𝐶). These three cases are very similar to the cases of Bases (4) and (5); so, the details are omitted. 8 The proof is finished. Remark. According to Lemma, we can say that the eight 7-dimensional subalgebras of 𝑔(3,C) found here are 7-dimensional subalgebras of each 2𝑛 −dimensional Clifford algebra 𝑔(n,C) for 𝑛 ≥ 3. This article begins the classification of small dimensional subalgebras for Clifford algebras. The results are obtained by using canonical bases. Author thanks Professor Russell Gosnell for his useful advise. References [1] J. Chisholm, A. Common. "Clifford Algebras and Their Applications in Mathematical Physics". Reidel, Dordrecht, 1986. [2] E. Meinrenken. "Clifford Algebras and Lie Theory". Springer, Berlin -- Heidelberg, 2013. [3] U. Shtukar. "Classification of Canonical Bases for (𝑛 − 1) −Dimensional Subspaces of 𝑛 −Dimensional Vector Space". Journal of Generalized Lie Theory and Applications. Vol. 10, Issue 1, 2016, 241-245.
1502.00764	2	1502	2015-07-22T13:53:37	The geometry of polynomial identities	[ "math.RA" ]	We discuss some geometric invariants of polynomial identities of algebras deduced from Kemer's theory and deduce some quantitative information on codimension and co--length	math.RA	math	The geometry of Polynomial Identities Claudio Procesi July 23, 2015 Contents 1 Introduction 2 Growth and the theory of Kemer 2.0.1 Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Three fundamental theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1 Alternating polynomials . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.2 Use of Schur–Weyl duality . . . . . . . . . . . . . . . . . . . . . . . . 2.1.3 The Kemer index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Fundamental algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 The role of traces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 T –ideals of finite Kemer index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.3 Generic elements 3 The canonical filtration 3.3 Representation varieties 3.1 Rationality and a canonical filtration . . . . . . . . . . . . . . . . . . . . . . 3.2 A close look at the filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Polynomial maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Kemer polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 The varieties Wn1,...,nu 3.3.2 The support of Kemer polynomials . . . . . . . . . . . . . . . . . . . 3.3.3 Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.4 The dimension of relatively free algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Partition functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 The Theory of Dahmen–Micchelli . . . . . . . . . . . . . . . . . . . . 3.4.3 U invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.4 Cocharacters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Invariants of several copies of V . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.1 Colength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Cocharacters 3.5 3 5 5 6 6 7 8 10 12 14 14 16 16 19 19 20 22 22 24 25 27 28 28 30 31 34 35 36 1 2 4 Model algebras 4.1 The canonical model of fundamental algebras . . . . . . . . . . . . . . . . . 4.1.1 Description of F ¯A,s(X) . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Some complements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Generalized identities . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Moduli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 A canonical structure of Cayley–Hamilton algebra . . . . . . . . . . 38 38 40 41 41 42 44 3 1 Introduction In the last 50 years, or more, several papers have dealt with connections between non– commutative algebra and algebraic geometry. Sometimes the term non–commutative algebraic geometry has been used, which I find mis- leading since there are several quite disjoint instances of these connections. Loosely the only unifying ideas are some information on various sorts of spectral theory of operators. In this paper we want to examine a particular theory, that of algebras with polynomial identities, and show how a sequence of varieties of semi–simple representations, see 3.3.1, appears naturally associated to a PI theory, see Theorems 3.1 and 3.13. These varieties are among a series of discrete, geometric and combinatorial invariants of PI theories, together with these one should also consider some moduli spaces, which parametrise certain coherent sheaves on these representation varieties. At the moment these objects seem to be hard to treat although some hints will be given in the last part of the paper. In this paper algebra will always mean associative algebra over some field F . We will also assume that F is of characteristic 0, otherwise too many difficult and sometimes unsolved problems arise, when convenient we shall also assume that F is algebraically closed. Let us recall the main definitions and refer the reader to Rowen’s book for a comprehensive discussion of the basic material, [40] or V. Drensky, [15] and V. Drensky–E. Formanek [16]. We start from the free associative algebra F hXi, in some finite or countable set of variables xi, with basis the words in these variables, and whose elements we think of as non commutative polynomials. Given any algebra A we have for every map, X → A, xi 7→ ai an associated evaluation of polynomials f (x1, . . . , xm) 7→ f (a1, . . . , am) ∈ A and then Definition 1.1. • A polynomial f (x1, . . . , xm) ∈ F hXi is a polynomial identity of A, short PI, if f (a1, . . . , am) = 0 for all evaluations in A. • A PI algebra is an algebra A which satisfies a non–zero polynomial identity. • Two algebras A, B are PI–equivalent if they satisfy the same polynomial identities. The set Id(A) of PI’s of A in the variables X is an ideal of F hXi closed under all possible endomorphisms of F hXi, that is substitutions of the variables xi with polynomials. Such an ideal I is called a T –ideal, the quotient F hXi/I satisfies all the identities in I. In fact F hXi/Id(A) is a free algebra in the category of all algebras which satisfy the PI’s of A. When X is countable and I is a T –ideal, then I is the ideal of all polynomial identities of F hXi/I. One usually says that F hXi/I is a relatively free algebra. Remark 1.2. There is a small annoying technical point, sometimes it is necessary to work with algebras without 1 (in the sense that either there is truly not 1 or we do not consider it as part of the axioms). In this case the free algebra has to be taken without 1. This implies 4 a different notion of T –ideals, since in the case of algebras with 1, a T –ideal is also stable under specialising a variable to 1, while this is not allowed in the other case. We leave to the reader to understand in which setting we are working. As usual polynomials have a dual aspect of symbolic expressions or functions, the same happens in PI theory where F hXi/I can be identified to an algebra of polynomial maps AX → A which commute with the action of the automorphism group of A. Its algebraic combinatorial nature is quite complex and usually impossible to describe in detail even for very special algebras A. In general the algebra of all polynomial maps commuting with the action of the automorphism group of A is strictly larger. A major example is when A = Mn(F ) is the algebra of n × n matrices over F where one quickly finds a connection with classical invariant theory. Even this basic example cannot be described in full except in the trivial case n = 1 and the non–trivial case n = 2 cf. [32]. A major difficulty in the Theory is the fact that relatively free algebras F hXi/I, even when X is finite, almost never satisfy the Noetherian condition, cf. Amitsur [6], that is the fact that ideals (right, left or bilateral) have a finite set of generators. This is not just a technical point but reflects some deep combinatorics appearing. This is in part overcome by the solution of the Specht problem by Kemer, Theorem 2.6. Section §2 is devoted to a quick overview of the required Kemer theory. In particular we shall stress the role of fundamental algebras, see Definition 2.19 and Theorem 2.23, which are the building pieces of PI equivalence classes of finite dimensional algebras. In §3 we start to draw some consequences of Kemer’s theory. The first goal of this paper is to show, Theorem 3.1, that, when X is finite, for a given relatively free algebra F hXi/I there is a canonical finite filtration by T –ideals Ki such that the quotients Ki/Ki−1 have natural structures of finitely generated modules over special finitely generated commutative algebras Ti. In fact each piece Ki/Ki−1 is a two sided ideal in some special trace algebras which are finitely generated modules over these commutative algebras. These algebras Ti are coordinate rings of certain representation varieties, §3.3.1 and the word geometry appearing in the title refers to the geometric description of the algebraic varieties supporting the various modules Ki/Ki−1, see Theorem 3.13. As next step we show that these varieties are natural quotient varieties parametrising semi– simple representations. As an important consequence in Corollary 3.14 we show that: if two fundamental algebras are PI equivalent then they have the same semisimple part. Then from some explicit information on these varieties we shall deduce a number of corollaries computing the dimension of the relatively free algebras, Proposition 3.18 and the growth of the cocharacter sequence, see Definition 2.3, using some general facts of invariant theory, Corollary 3.29. In the final section §4 we shall indicate a method to classify finite dimensional algebras up to PI–equivalence §4.1. 5 2 Growth and the theory of Kemer In this section we recall basic results of PI theory which motivate our research and are needed for the material of this paper 2.0.1 Growth Given n, we let Vn denote the space of multilinear polynomials of degree n in the vari- ables x1, . . . , xn. The space Vn has as basis the monomials xσ(1) . . . xσ(n) as σ runs over all permutations of 1, 2, . . . , n, so dim Vn = n!. Identities can always be multilinearized, hence the subset Id(A) ∩ Vn plays a special role and, in characteristic zero, the ideal Id(A) is completely determined by the sequence of multilinear identities {Id(A) ∩ Vn}n≥1. In order to study dim (Id(A) ∩ Vn) we introduce the quotient space Vn/(Id(A) ∩ Vn) and its dimension cn(A) := dim (cid:18) Vn Id(A) ∩ Vn(cid:19) . The integer cn(A) is the n-th codimension of A. Clearly cn(A) determines dim (Id(A) ∩ Vn) since dim Vn is known. The study of growth for PI algebra A is mostly the study of the rate of growth of the sequence cn(A) of its codimensions, as n goes to infinity. For a full survey we refer to [37]. We have the following basic property proved by Regev. Theorem 2.1. [36] cn(A) is always exponentially bounded. We have then the integrality theorem of Giambruno–Zaicev. Theorem 2.2. [20] Let A be a PI algebra over a filed F with char(F ) = 0, then the limit lim n→∞ exists and is an integer, called exponent. cn(A)1/n ∈ N The space Vn/(Id(A)∩Vn) is a representation of the symmetric group Sn acting by permuting variables and Definition 2.3. The Sn character of that space, χSn(Vn/(Id(A) ∩ Vn)) is denoted χn(A) = χSn (cid:18) Vn Id(A) ∩ Vn(cid:19) , and is called the n-th cocharacter of A. Since cn(A) = deg χn(A), cocharacters are refinement of codimensions, and are important tool in their study. By a theorem of Amitsur–Regev and of Kemer, χn(A) is supported on some (k, ℓ) hook. Shirshov’s Height Theorem, [43], then implies that the multiplicities of the irreducible characters, in the cocharacter sequence, are polynomially bounded. One of the goals of this paper is to discuss, see §3.3.4 and §3.4, some further informations one can gather on these numbers using geometric methods. 6 2.1 Three fundamental theorems We need to review the results and some of the techniques of the Theory of Kemer, presented in the monograph [25], see also [3] or the forthcoming book [2]. A fundamental Theorem of Kemer states that Theorem 2.4. [Kemer] If X is a finite set, a non–zero T –ideal I of F hXi is the ideal of polynomial identities in the variables X for a finite dimensional algebra A. In other words any finitely generated PI algebra is PI equivalent to a finite dimensional algebra. This is in fact the first part of a more general statement, let us consider the Grassmann algebra, thought of as super–algebra, in countably many odd generators G := V[e1, e2, . . .] decomposed as G = G0 ⊕ G1 into its even and odd part. Theorem 2.5. [Kemer] Every PI algebra R is PI equivalent to the Grassmann envelope G0 ⊗ A0 ⊕ G1 ⊗ A1 of a finite dimensional super–algebra A = A0 ⊕ A1. The algebra A is of course not unique, nevertheless some normalisations in the choice of A can be made and the purpose of this paper is to show that there is a deep geometric structure of the algebra F hXi/I which reflects the structure of these normalised algebras. A major motivation of Kemer was to solve the Specht problem, that is to prove Theorem 2.6. [Kemer] All T –ideals are finitely generated as T –ideals. This implies that, when working with T –ideals, we can use the standard method of Noethe- rian induction that is, every non empty set of T –ideals contains maximal elements. 2.1.1 Alternating polynomials Kemer’s theory is based on the existence of some special alternating polynomials which are not identities, that is do not belong to a given T –ideal Γ. So let us start by reviewing this basic formalism. Let us fix some positive integers µ, t, s, we want to construct multilinear polynomials in some variables X and possibly other variables Y . We want to have N = µt + s(t + 1) variables X decomposed as µ disjoint subsets called small layers X1, . . . , Xµ, with t elements each and s big layers Z1, . . . , Zs, with t + 1 elements each, we consider polynomials alternating in each layer. Such a space of polynomials is obtained, by operations of substitution of variables, from a finite dimensional space Mµ,t,s(X, W ) constructed as follows. Take N + 1 variables w1, w2, . . . , wN +1 and consider the space spanned by the N! monomials w1xσ(1)w2xσ(2) . . . wN xσ(N )wN +1, σ ∈ SN +1. In this space the subgroup G := Sµ t+1 acts permuting the monomials and thus we have a subspace Mµ,t,s(X, W ) of dimension N!/t!µ(t + 1)!s with basis all possible polynomials alternating in these layers. t × Ss 7 Since in general we work with algebras without 1 we also need to add all polynomials obtained from these by specialising some of the variables wi to 1. We obtain thus a space MX = Mµ,t,s(X, W ) of polynomials in X, and some of the W , alternating in the layers of X, so that, if we take any polynomial f (X, Y ) which is multilinear and alternating in the layers of X, and depends on other variables Y , this polynomial is obtained as linear combination of elements of MX after substitution of the variables wi with polynomials in the variables Y . In particular this shows how from a space Mµ,t,s(X, W ) one may deduce, by variable substi- tutions, larger spaces Mµ′,t′,s′(X, W ), µ′ ≥ µ, t′ ≥ t, s′ ≥ s. Particular importance have the Capelli polynomials, introduced by Razmyslov, [34] Cm(x1, x2, . . . , xm; w1, w2, . . . , wm+1) := Xσ∈Sm ǫσw1xσ(1)w2xσ(2) . . . wmxσ(m)wm+1. (1) In fact this polynomial plays a role analogous to that of the classical Capelli identity, which is instead an identity of differential operators. Since one often needs to analyse algebras without 1, it is useful to introduce the Capelli list Cm of all polynomials deduced from Cm(x1, x2, . . . , xm; w1, w2, . . . , wm+1) by specialising one or more of the variables wi to 1. One of the facts of the theory, consequence of Theorem 2.4, is Proposition 2.7. A PI algebra is PI equivalent to a finite dimensional algebra if and only if it satisfies some Capelli identity.1 In this paper we want to concentrate on finite dimensional algebras so we shall from now on assume that some Capelli identity is satisfied. In this way we do not need to introduce super–algebras nor apply Theorem 2.5. Nevertheless most results could be extended to super–algebras in a more or less straightforward way as will be presented in the forthcoming book [2]. 2.1.2 Use of Schur–Weyl duality Recall that, given a vector space V with dim V = k, on each tensor power V ⊗d act the general linear group GL(V ) and the symmetric group Sd which span two algebras each the centralizer of the other. In characteristic 0 both algebras are semi–simple and thus we have the Schur–Weyl duality decomposition in isotypic components, indexed by partitions λ ⊢ d of height ≤ dim V = k: V ⊗d = ⊕λ⊢d, ht(λ)≤kSλ(V ) ⊗ Mλ. (2) The Mλ are the irreducible representations of the symmetric group Sd, constructed from the theory of Young symmetrizers, with character χλ. The modules Sλ(V ), are irreducible representations of GL(V ) which in fact can be thought of as polynomial functors on vector spaces called the Schur functors (cf. [33]). 1for algebras without 1 we mean that it also satisfies all the identities of the Capelli list. 8 We consider the free algebra F hXi as the tensor algebra T (V ) over an infinite dimensional vector space V , with basis the variables X := {x1, x2, . . .} and take a T ideal I. Remark 2.8. If we want to stress the basis X of V we also write Sλ(V ) = Sλ(X). Since a T –ideal is stable under variable substitutions it is in particular stable under the action of linear group of V , that is GL(V ) := ∪mGL(m, F ) and we can decompose T (V )/I into irreducible representations of this group deduced From Formula (2). If we assume that I contains a Capelli identity Cm+1 (or a Capelli list) we have a restriction, deduced from these Capelli identities, on the height of the partitions appearing in T (V )/I: T (V )/I := ⊕d ⊕λ⊢d, ht(λ)≤m nλSλ(V ). (3) One can then apply the theory of highest weight vectors, this theory belongs to the Theory of Lie and algebraic groups. In our case the notion of weight is just the multidegree in the variables xi. An element of weight the sequences k1, k2, . . . is a non commutative polynomial homogeneous of degree ki in each xi. Weights are usually equipped with the dominance order which in Lie theory arises from the theory of roots, in our case simply k1, k2, . . . is greater than h1, h2, . . . in dominance order if k1 ≥ h1, k1 + k2 ≥ h1 + h2, . . . , k1 + . . . + ki ≥ h1 + . . . + hi, . . .. Consider in GL(V ) the unipotent group U of those (strictly upper triangular) linear trans- formations of type xi 7→ xi +Pj<i ai,jxi. One know that in the space Sλ(V ) the subspace Sλ(V )U of U invariants is 1–dimensional and generated by an element of weight λ that is homogeneous in each xi of degree hi where hi is the length of the ith row of λ, in fact this is a highest weight vector using the dominance order of weights. Thus if the height of λ is ≤ m this element depends only upon the first m variables. On the other hand since Sλ(V ) is an irreducible representation of GL(V ) it is generated by this highest weight vector. One deduces from Theorem 2.4 Theorem 2.9. A T –ideal Γ ⊂ F hXi, in countably many variables X is the ideal of identities of a finite dimensional algebra if and only if it contains a Capelli list. Remark 2.10. This allows us, in order to study the multiplicities nλ to restrict the number of variables, hence V to any chosen finite number with the constraint to be ≥ m. 2.1.3 The Kemer index A main tool in the Theory of Kemer is given by introducing a pair of non negative integers β(Γ), γ(Γ), called the Kemer index of Γ, invariants of a T –ideal Γ which contains some Capelli identities. These numbers give a first measure of which of the spaces Mµ,t,s(X, W ) are not entirely contained in Γ, (or not polynomial identities of some given algebra). 9 Definition 2.11. For every T –ideal Γ, which contains some Capelli identities, we let β(Γ) to be the greatest integer t such that for every µ ∈ N there exists a µ-fold t-alternating (in the µ layers Xi with t elements) polynomial not in Γ: f (X1, . . . , Xµ, Y ) /∈ Γ. We then let γ(Γ) to be the maximum s ∈ N for which there exists, for all µ, a polynomial f (X1, . . . , Xµ, Z1, . . . , Zs, Y ) /∈ Γ, alternating in µ small layers Xi with β(Γ) elements and in s big layers Zj with β(Γ) + 1 elements. The pair (β(Γ), γ(Γ)) is the Kemer index of Γ denoted IndΓ. For an algebra A we define the Kemer index of A, denoted by Ind(A), to be the Kemer index of the ideal of polynomial identities of A. We order the Kemer indices lexicographically and then observe that Remark 2.12. If Γ1 ⊂ Γ2 are two T –ideals we have If Γ = ∩k i=1Γi is an intersection of T –ideals then IndΓ1 ≥ IndΓ2. IndΓ = max IndΓi If A = ⊕k i=1Ai is a direct sum of algebras Ind(A) = max Ind(Ai). Remark 2.13. By definition, denoting s := γ(Γ), there is a minimum µ0 = µ0(Γ) such that there is no polynomial f (X1, . . . , Xµ, Z1, . . . , Zs+1, Y ) /∈ Γ, alternating in µ ≥ µ0 layers Xi with β(Γ) elements and in s + 1 layers Zj with β(Γ) + 1 elements. Definition 2.14. A polynomial f (X1, . . . , Xµ, Z1, . . . , Zs, Y ) /∈ Γ alternating in µ layers Xi with β(Γ) elements and in s = γ(Γ) layers Zj with β(Γ) + 1 elements with µ > µ0 + 1 will be called a µ–Kemer polynomial. A Kemer polynomial is by definition a µ–Kemer polynomial for some µ > µ0 + 1. Remark 2.15. A Kemer polynomial f (X1, . . . , Xµ, Z1, . . . , Zs, Y ), which is also linear in a variable w, not in the layers Zj, has the following property. Fix one of the layers Xi which does not contain the variable w, add to it the variable w and alternate these t + 1 variables. If w is also in no layer Xj, we produce a polynomial alternating in µ − 1 ≥ µ0 layers with t elements and in s + 1 layers with t + 1 elements, otherwise if w ∈ Xj, i 6= j we produce a polynomial alternating in µ − 2 ≥ µ0 layers with t elements and in s + 1 layers with t + 1 elements. By the definition of µ0 and of Kemer polynomial this is always an element of Γ, so if Γ = Id(A) a polynomial identity of A. 10 Example 2.16. 1) Conversely, using Capelli polynomials, (1), the product (taken in an ordered way) If A = Mn(F ), then every (n2 + 1)-alternating polynomial is in Id(A). Cµ,n2(X1, . . . , Xµ, Y ) := µ Yi=1 Cn2(xi,1, . . . , xi,n2, yi,1, . . . , yi,n2, yi,n2+1), (4) Xi := {xi,1, . . . , xi,n2} evaluated in suitable eij’s can take any value ehk for every µ so, in particular, it is not an identity. Hence β(Mn(F )) = n2, γ(Mn(F )) = 0. 2) On the opposite if A is a finite dimensional nilpotent algebra and As 6= 0, As+1 = 0 we see that its Kemer index is (0, s). In a subtle, and not completely understood way, all other cases are a mixture of these two special cases. Remark 2.17. We could take a slightly different point of view and define as µ–Kemer polynomials for a T ideal with Kemer index t, s to be the elements of Mµ,t,s(X, W ) which are not in Γ, then deduce some larger T -ideal by evaluations of these polynomials. 2.2 Fundamental algebras We need to recall, without proofs, some steps of Kemer’s theory and fix some notations. Let A be a finite dimensional algebra over a field F of characteristic 0, J := rad A be its Jacobson radical and let A := A/J, a semi–simple algebra. By Wedderburn’s principal Theorem, cf. [1], we can decompose A = A ⊕ J = R1 ⊕ · · · ⊕ Rq ⊕ J, Ri simple (5) where, if we assume F algebraically closed, for every i the simple algebra Ri is isomorphic to Mni(F ) for some ni. Due to Example 2.16 2), we will assume from now on that ¯A 6= 0 and call it the semisimple part of A. Definition 2.18. We set tA := dim F A/J = dim F A, and sA + 1 be the nilpotency index of J, that is J sA 6= 0, J sA+1 = 0. We call the pair tA, sA the t, s–index of A. It is easily seen that the t, s index of A is greater or equal than the Kemer index Ind(A), so it is important to understand when these two indices coincide. With the previous notations consider the quotient map π : A → A/J = ⊕q 1 ≤ j ≤ q, i=1Ri and let for R(i) := ⊕q i6=j, j=1Rj and Ai := π−1R(i) ⊂ A. (6) We have for all i that Ai satisfies all polynomial identities of A and tAi < tA. 11 We need to construct a further algebra A0 which satisfies all polynomial identities of A and tA0 = tA but sA0 < sA. This algebra is constructed as the free product ¯A ⋆ F hXi modulo the ideal generated by all polynomial identities of A, for a sufficiently large X (in fact we shall see that s variables suffice, Lemma 4.3), and finally modulo the ideal of elements of degree ≥ s in the variables X. By definition tA0 = tA, sA0 = sA − 1. We have by construction Id(A) ⊂ ∩q then A is PI equivalent to ⊕q than the t, s–index of A. This suggests i=1Id(Ai) ∩ Id(A0) i=1Ai ⊕ A0. All the t, s–indices of these algebras are strictly less i=1Id(Ai) ∩ Id(A0), and if Id(A) = ∩q Definition 2.19. We say that A is fundamental if Id(A) ( ∩q i=0Id(Ai). (7) A polynomial f ∈ ∩q i=0Id(Ai) \ Id(A) will be called fundamental. Example 2.20. An algebra of block triangular matrices is fundamental. By induction one has that Proposition 2.21. Every finite dimensional algebra A is PI equivalent to a finite direct sum of fundamental algebras. Let us now see the properties of a fundamental multilinear polynomial f . By definition we have an evaluation in A which is different from 0. We may assume that each variable has been substituted to a semisimple resp. radical element, we call such an evaluation restricted. Let us call, by abuse of notation, semisimple resp. radical the corresponding variables. Lemma 2.22. Take a non–zero restricted evaluation, in A, of a multilinear fundamental polynomial f . 1. We then have that there is at least one semisimple variable evaluated in each Ri, for all i = 1, . . . , q (Property of being full). 2. We have exactly sA radical substitutions (Property K). The next result is usually presented in the literature divided in two parts, called the first and second Kemer Lemma. Theorem 2.23. A finite dimensional algebra A is fundamental if and only if its Kemer index equals the t, s index. In this case, given any fundamental polynomial f , we have µ–Kemer polynomials in the T –ideal hf i generated by f , for every µ. Remark 2.24. Conversely there is a µ such that all µ–Kemer polynomials are fundamental. 12 We then introduce a definition Definition 2.25. A T –ideal I is called primary if it is the ideal of identities of a fundamental algebra. A T –ideal I is irreducible if it is not the intersection I = J1 ∩ J2 of two T –ideals J1, J2 properly containing I. We then see Proposition 2.26. Every irreducible T –ideal containing a Capelli list is primary and every T –ideal containing a Capelli identity is a finite intersection of primary T –ideals. Proof. By Noetherian induction every T –ideal is a finite intersection or irreducible T –ideals, otherwise there is a maximum one which has not this property and we quickly have a contradiction. If a T –ideal I contains a Capelli list it is the T –ideal of PI’s of a finite dimensional algebra, which by Proposition 2.21 is PI equivalent to a direct sum of fundamental algebras. Hence I is the intersection of the ideals of polynomial identities of these algebras, since it is irreducible it must coincide with the ideal of polynomial identities of one summand, a fundamental algebra. As in the Theory of primary decomposition we may define an irredundant decomposition I = J1 ∩ . . . ∩ Jk of a T –ideal I in primary ideals. In a similar way every finite dimensional algebra B is PI equivalent to a direct sum A = ⊕iAi of fundamental algebras which is irredundant in the sense that A is not PI equivalent to any proper algebra ⊕i6=jAi. We call this an irredundant sum of fundamental algebras. It is NOT true that the T –ideal of PI of a fundamental algebra is irreducible. The following example has been suggested to me by Belov. Consider the two fundamental algebras A1,2, A2,1 of block upper triangular 3 × 3 matrices stabilizing a partial flag formed by a subspace of dimension 1 or 2 respectively. They have both a semisimple part isomorphic to M2(F ) ⊕ F . By a Theorem of Giambruno–Zaicev [23] the two T –ideals of PI are respectively I2I1 and I1I2 where Ik denotes the ideal of identities of k × k matrices. By a Theorem of Bergman–Lewin [10] these two ideals are different. Now in their direct sum A1,2 ⊕ A2,1 consider the algebra L which on the diagonal has equal entries in the 2 two by two and in the two one by one blocks. It is easy to see that L is PI equivalent to A1,2 ⊕ A2,1 and that it is fundamental. Its T –ideal is not irreducible but it is I2I1 ∩ I1I2. 2.2.1 The role of traces Let A be a fundamental algebra over a field L, with index t, s and f a µ–Kemer polynomial for µ sufficiently large so that f is also fundamental. It follows from Lemma 2.22 that the evaluation in the small layers factors through the radical. 13 So let us fix one of the small layers of the variables, say x1, . . . , xt, and denote for simplicity by Y the remaining variables. Thus, having fixed some evaluation, which we denote by ¯Y , of the variables Y outside the chosen small layer of variables, we deduce from f a multilinear alternating map from A/J to A in t variables (still denoted by f ), which must then have the form: f (x1, x2, . . . , xt, ¯Y ) = det(¯x1, . . . , ¯xt)u( ¯Y ), where the ¯xi are the classes in ¯A of the evaluations of the variables xi and we have chosen a trivialization of Vt ¯A. Since f is fundamental and there are s radical evaluations we also have u( ¯Y ) ∈ J s. As a consequence we deduce the important identity f (zx1, zx2, . . . , zxt, Y ) = det(¯z)f (x1, x2, . . . , xt, Y ) (8) where det(¯z) means the determinant of left multiplication of ¯z on A/J. When we polaryze this identity we have on the right hand side the characteristic polynomial and in particular the identity as functions on A: t Xk=1 f (x1, . . . , xk−1, zxk, xk+1, . . . , xt, Y ) = tr(¯z)f (x1, . . . , xt, Y ), (9) where tr(¯z) is again the trace of left multiplication by ¯z in A/J. We now observe that: the left hand side of this formula is still alternating in all the layers in which f is alternating and not an identity, since for generic z we have tr(¯z) 6= 0, thus it is again a Kemer polynomial. We can then repeat the argument and it follows that Lemma 2.27. The function resulting by multiplying a Kemer polynomial, evaluated in A, by a product tr(¯z1)tr(¯z2) . . . tr(¯zk), where the zi are new variables, is the evaluation in A of a new Kemer polynomial (involving also the variables zi). This discussion easily extends to a direct sum A = ⊕q radical Ji) over some field L all with the same Kemer index (t, s) getting a formula i=1Ai of fundamental algebras (with q Xi=1 ti(¯z)fi(x1, . . . , xt, Y ) = t Xk=1 f (x1, . . . , xk−1, zxk, xk+1, . . . , xt, Y ), (10) where fi(x1, . . . , xt, Y ) is the projection to Ai of the evaluation f (x1, . . . , xt, Y ), while ti(¯z) is the trace of left multiplication by the class of z on Ai. In this case if µ is sufficiently large and f is a µ–Kemer polynomial for one of the algebras Ai then it is either a PI or a µ–Kemer polynomial for each of the Aj. It is then convenient to think of A as a module over the direct sum of q copies of L. Then we can write t(¯z) := (t1(¯z), . . . , tq(¯z)) ∈ Lq, and we have: t(¯z)f (x1, . . . , xt, Y ) = t Xk=1 f (x1, . . . , xk−1, zxk, xk+1, . . . , xt, Y ). Notice furthermore that if µ is sufficiently large and f is a generic µ–Kemer polynomial it is not a PI for any of the algebras Ai. 14 2.2.2 T –ideals of finite Kemer index. We have remarked that a T –ideal in the free algebra in countably many variables is of finite Kemer index if and only if it contains some Capelli list. The following fact could also be obtained from the theory of Zubrilin, [45]. Proposition 2.28. If Γ is a T –ideal with finite Kemer index t, s, there is a ν ∈ N such that, every ν–Kemer polynomial f (X1, . . . , Xν, Z1, . . . , Zs, W ) ∈ R(X) has the property that, for two extra variables y, z /∈ X one has, modulo Γ ∀i = 1, . . . , ν : f (zX1, . . . , Xν, Z1, . . . , Zs, W ) mod. Γ= f (X1, . . . , zXi, . . . , Xν, Z1, . . . , Zs, W ); f (zyX1, X2, . . . , Xν, Z1, . . . , Zs, W ) mod. Γ= f (yzX1, X2, . . . , Xν, Z1, . . . , Zs, W ). (11) Proof. By Theorem 2.9, Proposition 2.21 we know that Γ is the ideal of identities of a finite direct sum of fundamental algebras A = ⊕m i=1Ai. Also by Remark 2.12 the Kemer index of Γ is the maximum of the Kemer indices of the fundamental algebras Ai. Thus, we may assume that in the list Ai the first k algebras have this maximum Kemer index and decompose A = B ⊕ C where B = ⊕k i=1Ai. It follows that there is some ν ∈ N such that any ν–Kemer polynomial for Γ is a ν–Kemer polynomial for B while it is a polynomial identity for C. Then Formula (11) is valid if and only if it is valid modulo Γ′ := Id(B) that is as functions on B and this is insured by Formula (8). 2.2.3 Generic elements Let A be a finite dimensional algebra over an algebraically closed field F , and dim F A = n, fix a basis a1, . . . , an of A. Given m ∈ N (or m = ∞) consider L, the rational function field F (Λ) where Λ = {λi,j, i = 1, . . . , n; j = 1, . . . , m} are nm indeterminates. i=1Ai is a direct sum of finite dimensional algebras (usually we shall assume to be We can construct m generic elements ξj := Pn i=1 λi,jai for A, and in A ⊗ L we construct the algebra FA(m) = F hξ1, . . . , ξmi generated by these generic elements. This is clearly isomorphic to the relatively free algebra quotient of the free algebra modulo the identities of A in m variables. If A = ⊕k fundamental) let J = ⊕Ji its radical and A/J = ⊕k We may choose a basis a1, . . . , an of A over F union of bases of the summands Ai, we may also choose for each summand Ai decomposed as ¯Ai ⊕ Ji the basis to be formed of a basis of Ji and one of ¯Ai. Then when we construct m generic elements ξj := Pn i=1 λi,jai for A, by the choice of the basis each is the sum ξj = Pk i=1 ξj,i + ηj,i where the ξj,i + ηj,i are generic for Aj. The ξj,i are generic for ¯Aj, while ηj,i are generic for the radical Jj, and all involve disjoint variables. For each j = 1, . . . , k we also have the relatively free algebra FAj (m) = F hξ1,j, . . . , ξm,ji and an injection ¯Ai = Ai/Ji. i=1 FA(m) ⊂ ⊕k j=1FAj (m) ⊂ A ⊗F L = ⊕k j=1Aj ⊗F L. 15 Notice that the radical of A ⊗F L is J ⊗F L and modulo the radical this is ¯A ⊗F L which is some direct sum of matrix algebras ⊕iMni(L). The projection p : A → ¯A = ⊕k generated by generic elements ¯Ai of coordinates p1, p2, . . . , pk, induces a map of algebras i=1 p : FA → ⊕iF ¯Ai ⊂ ⊕i ¯Ai ⊗F L, p : ξi 7→ (ξi,1, . . . , ξi,k). Remark 2.29. The Kernel of p is a nilpotent ideal while the image is isomorphic to the domain FMn(F ) of generic n × n matrices where n is the maximum of the degrees ni (where ¯A = ⊕iMni(F )). We then set a ∈ FA, t(a) := (t1(a), . . . , tk(a)) ∈ L⊕k (12) by setting ti(a) to be the trace of left multiplication of the image of a, under pi, in the summand ¯Ai ⊗F L. We let TA(m) := F [t(a)]a∈FA ⊂ L⊕k (13) be the (commutative) algebra generated over F by all the elements t(a), a ∈ FA(m). From now on we assume m fixed and drop the symbol (m) write simply FA(m) = FA, TA(m) = TA. Theorem 2.30. TA is a finitely generated F algebra and TAFA is a finitely generated module over TA. 1 an2 Proof. The proof uses a basic tool of PI theory, the Shirshov basis, that is the existence of a finite number N of monomials ai in the generators ξj such that every monomial in the 2 . . . anN variables ξj, is a linear combination with coefficients in F of products of powers an1 [40]. We first claim that every element a ∈ FA satisfies a monic polynomial with coefficients in TA, in fact the coefficients are polynomials in t(aj) for j ≤ max(dim ¯Ai). For this let ni := dim ¯Ai. The projection of a in ¯Ai ⊗F L satisfies Hni(x) where we take for Hni(x) the Cayley–Hamilton polynomial induced by left multiplication on ¯Ai ⊗F L. This is a universal expression in x and the elements ti(aj), j ≤ ni where ti(aj) is the trace of the the left action on ¯Ai ⊗F L of the projection of aj. Thus if we use the formal Cayley Hamilton polynomial for ni ×ni matrices, but using as trace of aj the k–tuple t(aj) = (t1(aj), . . . , t1(aj)) for all i we see that, if ¯a denotes the image of a i=1 Hni(¯a)¯a = 0 in ⊕i ¯Ai ⊗F L we have Hni(¯a)¯a ∈ ⊕i ¯Ai ⊗F L has 0 in the ith component, so Qk in p(TAFA) ⊂ ⊕i ¯Ai ⊗F L. Now every element of the Kernel of p is nilpotent of some fixed degree s and finally we deduce that N k We have multiplied by a since we do not assume that the algebra has a 1. Hni(a)a)s = 0, in TAFA. ( Yi=1 (14) 16 We take a Shirshov basis for FA then, since we know that t(t(a)b) = t(a)t(b), it follows that TA is generated by the traces t(M) where M is a monomial in the Shirshov basis with i=1 Hni(x)x)s and FATA is spanned over TA by this finite exponents less that the degree of (Qk number of monomials. Example 2.31. A basic example is given by A = Mt(F ) the algebra of matrices. In this case the algebra of generic elements is known as the generic matrices. The commutative algebra TA(Y ) (will be denoted by Tt(Y )) equals the algebra of invariants of m := Y matrices under conjugation and the algebra TAFA is the algebra of equivariant maps (under conjugation) between m–tuples of matrices to matrices. Assume now that A = ⊕m i=1Ai is a direct sum of fundamental algebras. Decompose A = B⊕C where we may assume that B = ⊕k i=1Ai is the sum of the Ai with maximal Kemer index, the same Kemer index as A and C the remaining algebras. For µ sufficiently large a µ– Kemer polynomial for A is a polynomial identity on C and either a Kemer polynomial or a polynomial identity for Ai, i = 1, . . . , k. So in this case we call Kemer polynomial for A, one with this property. By formula (9), as extended to direct sums, a Kemer polynomial evaluated in FA ⊂ A ⊗F L = ⊕iAi ⊗F L satisfies formula (9) where tr(¯z) is the element of TA of Formula (12). In fact, since such polynomials vanish on C, the formulas factor through FB, TB which are quotients of FA, TB. We can then interpret Lemma 2.27 as Corollary 2.32. The ideal KA of FA spanned by evaluations of Kemer polynomials, is a TA submodule and thus a common ideal in FA and FATA. In fact under the quotient map FA → FB the ideal KA maps isomorphically to the corre- sponding ideal KB, in other words the action of FATA on KA factors through FBTB. The importance of this corollary is in the fact that the non–commutative object KA is, by Theorem 2.30, a finitely generated module over a finitely generated commutative algebra, so we can apply to it all the methods of commutative algebra. This is the goal of the next sections. At this point the ideal KA depends on A and not only on the T –ideal but as we shall see one can also remove this dependence and define an intrinsic object which plays the same role. 3 The canonical filtration 3.1 Rationality and a canonical filtration We want to draw some interesting consequences from the theory developed. Let R(X) = R := F hXi/I be a relatively free algebra in a finite number k of variables X. We have seen that I is the T –ideal of identities in k variables of a finite dimensional algebra A = ⊕iAi direct sum of fundamental algebras, we may also choose this irredundant. 17 Choosing such an A we write R = FA(X) and identify FA(X) to the corresponding algebra of generic elements of A. We can decompose this direct sum in two parts A = B ⊕ C where B is the direct sum of the Ai with the same Kemer index as A while B the sum of those of strictly lower Kemer index. We have Id(A) = Id(B) ∩ Id(C) and so an embedding FA(X) ⊂ FB(X) ⊕ FC(X) of the corresponding relatively free algebras. Let us drop X for simplicity. We can apply to B Theorem 2.30, and embed FB ⊂ FBTB which is a finitely generated module over the finitely generated commutative algebra TB, (both graded). Let K0 ⊂ R be the T ideal generated by the Kemer polynomials of B for sufficiently large µ. Since these polynomials are PI of C it follows that under the embedding FA(X) ⊂ FB(X) ⊕ FC(X) the ideal K0 maps isomorphically to the corresponding ideal in FB which, by Corollary 2.32, is a finite module over TB. Since K0 ⊂ R is a T ideal the algebra R1 := R/K0 is also a relatively free algebra, now with strictly lower Kemer index. So we can repeat the construction and let K1 ⊂ R be the T ideal with K0 ⊂ K1 and K1/K0 the T ideal in R1 generated by the corresponding Kemer polynomials. If we iterate the construction, since at each step the Kemer index strictly decreases, we must stop after a finite number of steps. Theorem 3.1. 1. We have a filtration 0 ⊂ K0 ⊂ K1 ⊂ . . . ⊂ Ku = R of T ideals, such that Ki+1/Ki is the T ideal, in Ri := R/Ki, generated by the corresponding Kemer polynomials for µ suitably large. 2. The Kemer index of Ri is strictly smaller than that of Ri−1. 3. Each algebra Ri has a quotient ¯Ri which can be embedded ¯Ri ⊂ ¯RiT ¯Ri in a finitely gen- erated module over a finitely generated commutative algebra T ¯Ri and such that Ki+1/Ki is mapped injectively to an ideal of ¯RiT ¯Ri. 4. In particular Ki+1/Ki has a structure of a finitely generated module over the finitely generated commutative algebra T ¯Ri. Corollary 3.2 (Belov [8]). If R := F hXi/I is a relatively free algebra in a finite number of variables X, its Hilbert series HR(t) := ∞ Xk=0 dim (Ri) ti is a rational function of the form p(t) j=1(1 − thj ) QN , hj ∈ N, p(t) ∈ Z[t]. (15) 18 Proof. Clearly HR(t) = Pu−1 i=0 HKi+1/Ki(t). We know that Ki+1/Ki is a finitely generated module over a finitely generated graded algebra TRi. If TRi is generated by some elements a1, . . . , am of degrees hi by commutative algebra one has that HKi+1/Ki(t) = , pi(t) ∈ Z[t]. pi(t) j=1(1 − thj ) Qm Summing this finite number of rational functions we have the result. In Theorem 3.19 and Corollary 3.21 we will apply a deeper geometric analysis in order to compute from the Hilbert series the dimension of the relatively free algebras. One can generalise these results by considering a relatively free algebra in k variables X as multi-graded by the degrees of the variables X, and then we write its generating series of the multi-grading HR(x) = Xh1,...,hk dim (Rh1...hk)xh1 1 . . . xhk k (16) this is of course the graded character of the induced action of the torus of diagonal matrices (which is a standard choice for a maximal torus of the general linear group GL(k)) acting linearly on the space of variables. Thus the series of Formula (16) should be interpreted in terms of the representation Theory of GL(k). In each degree d the homogeneous part Rd of the algebra R is some quotient of the representation V ⊗d, dim V = k. By Schur–Weyl duality discussed in §2.1.2 and by Formula (3) we have Rd = ⊕λ⊢dmλSλ(V ), mλ ≤ χλ(1). That is mλ is ≤ the dimension of the corresponding irreducible representation of the sym- metric group Sd and in fact, by Remark 2.10: Proposition 3.3. If the number of variables is larger than m where R satisfies the Capelli list Cm we have that mλ equals the multiplicity of χλ in the cocharacter of A of Definition 2.3. The character of Sλ(V ) is the corresponding Schur function Sλ(x1, . . . , xk) (a symmetric function) so that finally we have HR(x) = Xd Xλ⊢d mλSλ(x1, . . . , xk). (17) We have seen that the numbers mλ are the multiplicities of the cocharacters, so it will also be interesting to write directly a generating function for these multiplicities. ¯HR(t1, . . . , tk) = Xd Xλ⊢d mλtλ. (18) This has to be interpreted using the theory of highest weights. Recall that a highest weight vector is an element of a given representation of G = GL(k) which is invariant under U and it is a weight vector under the maximal torus of diagonal matrices (x1, . . . , xk). The weight is then a dominant weight sum of of the fundamental weights ωi := Qi is the highest weight of Vi F k. If mi is the number of columns of length i of the partition λ, i=1(Qi then the corresponding dominant weight is Pi miωi that is the character Qk j=1 xj)mi, we identify partitions with dominant weights and thus write 19 j=1 xj which λ = Xi miωi, ti := tωi, tλ = Yi tmi i . A highest weight vector vλ of weight λ generates an irreducible representation Sλ(F k) and Sλ(F k)U is 1–dimensional spanned by vλ. In the next paragraphs, using the Theory of highest weight vectors we shall show that also these two functions (of Formulas (17) and (18)) are rational and of special type and connected to the Theory of partition functions. Finally in Theorem 3.37 we will apply this theory to give a precise quantitative result on the growth of the colength of R. 3.2 A close look at the filtration Our next goal is to show that the commutative rings TRi, which we have deduced from some fundamental algebras can be derived formally only from properties of the T –ideals, for this we need to recall the theory of polynomial maps. 3.2.1 Polynomial maps The notion of polynomial map is quite general and we refer to Norbert Roby, [38] and [39]. A polynomial map, homogeneous of degree t, f : M → N between two vector spaces factors through the map m 7→ m⊗t to the symmetric tensors St(M) := (M ⊗t)St and a linear map St(M) → N. If A is an algebra both A⊗t and St(A) = (A⊗t)St are algebras and we have the following general fact. Definition 3.4. A polynomial map, homogeneous of degree t, between two algebras A, B, is said to be multiplicative if f (ab) = f (a)f (b), ∀a, b ∈ A. Proposition 3.5. [Roby [39]] Given a multiplicative polynomial map, homogeneous of degree t, between two algebras A, B, then the induced map St(A) → B is an algebra homomorphism. We can apply this theory to A equal to a free algebra F hY i, a positive integer t and the subalgebra of symmetric tensors St(F hY i) = (F hY i⊗t)St. We can treat the map z 7→ z⊗t as a universal multiplicative polynomial map, homogeneous of degree t. A general theorem of Ziplies [44] interpreting the Second Fundamental theorem of matrix invariants of Procesi and Razmyslov, that is the Procesi-Razmyslov theory of trace identities [31], [34, 35], states the following: 20 Theorem 3.6 (Ziplies). The maximal abelian quotient of St(F hY i) is isomorphic to the algebra Tt(Y ) of invariants of t × t matrices in the variables Y . (Vaccarino) The previous isomorphism is induced by the explicit multiplicative map det : F hY i → Tt(Y ), det : f (y1, . . . , ym) 7→ det f (ξ1, . . . , ξm), where the ξi are generic t × t matrices. The second part is due to Vaccarino, [42].2 3.2.2 Kemer polynomials The proof of Corollary 2.32 is based on the fact that a T ideal Γ can be presented as ideal of identities of a finite dimensional algebra A direct sum of fundamental algebras. We now want to show that this structure of the T ideal of Kemer polynomials is independent of A and thus gives some information on the possible algebras A having Γ as ideal of identities. Denote by R(Y ) and R(Y ∪ X) the relatively free algebras in the variety associated to Γ in these corresponding variables. Let us first use an auxiliary algebra A and let KA ⊂ R(Y ∪X) be the space of Kemer polynomials previously defined starting from A. Changing the algebra A to some A′ may change the space KA but only for the µ–Kemer polynomials up to some µ1, we shall see soon how to free ourselves from this irrelevant constraint. Let Kν be the space of ν–Kemer polynomials in the relatively free algebra F hXi/Γ in which the small layers are taken from the variables in X (and may depend also on the variables Y ). By Proposition 2.28 there is an intrinsic ν ∈ N such that if f (X1, . . . , Xν, W ) ∈ Kν we interpret Formula (11) as follows. If z ∈ F hY i we have a linear map which we shall denote by z given by (choosing one of small layers): z : f (x1, x2, . . . , xt, X, W ) 7→ f (zx1, zx2, . . . , zxt, X, W ) mod. Γ. (19) The operators z do not depend on the small layer chosen and commute and the map z 7→ z, is a multiplicative polynomial map homogeneous of degree t from the free algebra F hY i to a commutative algebra of linear operators. Therefore the Theorem of Zieplies–Vaccarino 3.6, tells us that Formula (19) defines a module structure on Kν by the algebra Tt(Y ) of invariants of t × t matrices in the variables Y . This module structure does not depend on A and thus is independent of the embedding of the relatively free algebra in A ⊗ L. On the other hand, choose A = ⊕k algebras, and A = B ⊕ C with B = ⊕u index (t, s) equal to the Kemer index of Γ. If ν is sufficiently large we know that Kν vanishes on C and in the previous embedding z coincides with the multiplication by (det(¯zi)) ∈ Lu. When we polaryze from z ∈ R(Y ) we see by Formula (9) that multiplication by tr(z) := (tr(¯z1), . . . , tr(¯zu)) maps Kν into Kν and i Ai so that Γ = Id(A), a direct sum of fundamental i Ai the direct sum of the Ai with maximal Kemer 2At the moment the Theorem is proved only in characteristic 0. 21 by definition lies in TA = TA(Y ) (which now depends on A) and in fact by definition TA is generated by these elements. We claim that Lemma 3.7. The action of TA on Kν is faithful so that the algebra TA(Y ) is the quotient of Tt(Y ) modulo the ideal annihilator of the module Kν. This ideal is independent on ν for ν large. Proof. Now we have constructed TA from Formula (13) as contained in the direct sum of the algebras TAi, i = 1, . . . u. By definition TA depends only from the semisimple part of A so it is the coordinate ring of a variety which depends only upon the algebra ⊕iMni(F ) that is from the numbers ni. So in the end the action of Tt(Y ), invariants of t × t matrices in the variables Y , on K factors through the map π : Tt(Y ) → TA(Y ) ⊂ ⊕u i=1TAi(Y ). (20) We claim that the composition of π with any projection to a summand TAi(Y ) is surjec- tive. In fact the first ring is generated by the traces of the monomials evaluated in all t—dimensional representations while the second is generated by the same traces but only those representations which factor through the left action of ¯Ai. We shall see in §3.3.1 the nature of this subvariety. Let Kν,i be the image of Kν in FAiTAi(Y ). We have that each Kν,i is torsion free over TAi, which is a domain, since Kν,i ⊂ J s Since Kν ⊂ ⊕u TA acts faithfully on Kν so that we have defined the homomorphism of Formula (20). i=1Kν,i and for each i the restriction to Kν,i is non–zero, we finally have that i ⊗ L is contained in a vector space over the field L. We want now to free ourselves from the auxiliary variables X and evaluate in all possible ways the elements of Kν in the relatively free algebra R(Y ) of A in the variables Y , obtaining thus a T ideal KR in R(Y ). We see that Theorem 3.8. The module action of Tt(Y ) on Kν induces a unique module action on KR compatible with substitutions of variables in Y . This action factors through a faithful action of its image ¯TA(Y ). in TA(Y ). Proof. If we work inside the non intrinsic algebra generated by R(Y ) and TA we have that KR is a TA submodule and this module structure is by definition compatible with substitutions of variables in Y . On the other hand since the elements of KR can be obtained by specializing the elements of K there is a unique way in which the module action of Tt(Y ) on K can induce a module action on KR compatible with substitutions of variables in Y . It is a faithful ¯TA(Y ) action since TA acts faithfully on KRL. Notice that KR ⊂ B ⊗ L so all the computations are just for the algebra B = ⊕iAi and from now on we shall just assume A = B and C = 0. 22 3.3 Representation varieties Our next task is to describe the algebraic varieties of which the rings TA are coordinate rings. 3.3.1 The varieties Wn1,...,nu For a given t and m consider the space Mt(F )m of m–tuples of t × t matrices. We think of this as the set of t–dimensional representations of the free algebra F hXi in m variables x1, . . . , xm. On this space acts by simultaneous conjugation the projective linear group P GL(t, F ) so that its orbits are the isomorphism classes of such representations. It is well known, [7], that the closed orbits correspond to semi–simple representations, so by geometric invariant theory the quotient variety Vt(m) := Mt(F )m//P GL(t) parametrizes equivalence classes of semisimple representations of dimension t. As soon as m ≥ 2, its generic points correspond to irreducible representations, which give closed free orbits, so the variety has dimension (m − 1)t2 + 1. The coordinate ring of this variety is the ring of P GL(t, F ) invariants, which we shall denote by Tt,m or as before Tt(Y ), if we denote by Y the m matrix variables. In characteristic 0 the algebra Tt(Y ) is generated by the traces of the monomials in the matrix variables, while in all characteristic by work of Donkin [14], we need all coefficients of characteristic polynomials of monomials, which can be taken to be primitive. Given non negative integers hi, ni with Pi hini = t we may consider, inside the variety Mt(F )m//P GL(t), the subvariety Wh1,...,hu;n1,...,nu, of semisimple representations which can be obtained as direct sum ⊕ihiNi, from semisimple representations Ni of dimension ni for each i = 1, . . . , u. Of course generically each Ni is irreducible. It will be interesting for us the special case hi = ni which we denote by Wn1,...,nu. Wh1,...,hu;n1,...,nu is the natural image of the product Qu i=1 Vni(m), where Vni(m) is the variety of semisimple representations of m–tuples of ni×ni matrices, under the map j : N1, . . . , Nu 7→ ⊕ihiNi. We see in fact that this map j can be considered as a restriction. Inside the algebra of t × t matrices, consider the subalgebra of block diagonal matrices ⊕u i=1hiMni(F ), where the block Mni(F ) appears embedded into hi equal blocks, which is isomorphic of course by definition to Mn1,...,nu := ⊕u i=1Mni(F ) we call jh1,...,jn this inclusion isomorphism. When we restrict the invariants Tt(Y ) to this subalgebra we see that when z is some polyno- i=1 hitri(z) where tri(z) is in the algebra Tni(Y ). mial in Y the restriction of the function tr(z) to this subalgebra equals Pu This means that Tt(Y ) maps to the G = Qu ring of Qu We have shown that, as Qu under the group G = Qu i=1 Vni(m) which is Tn1(Y ) ⊗ . . . ⊗ Tnu(Y ). i=1 P GL(ni, F ) invariants, that is the coordinate i=1 Vni(m) is the quotient of m copies of the space ⊕u i=1Mni(F ) i=1 P GL(ni, F ) acting by conjugation, we have the commutative diagram, where the two maps πG, πP GL(t,F ) are quotients under the two groups (⊕u i=1Mni(F ))m jh1,...,jn −−−−−→ Mt(F )m 23 (21) i=1 Vni(m) −−−→ Wh1,...,hu;n1,...,nu j πP GL(t,F )y πGy Qu Every representation of the form ⊕ihiNi with Ni of dimension ni can be conjugated into ⊕u i=1hiMni(F ), but usually this can be done in several different ways, thus we have Remark 3.9. The map j : Qu i=1 Vni(m) → Wn1,...,nu is surjective but it is almost never an isomorphism. It is isomorphism only when u = 1. We claim that the two varieties have the same dimension. For this it is enough to show that the generic fiber is finite. i=1 Vni(m) and in Wn1,...,nu we have that all the The generic fiber is obtained when in Qu summands Ni are irreducible and not just semisimple. In this case if we have several indices i with the same ni then in the expression ⊕iniNi we may permute the indices of the irreducible representations of the same dimension so they come from different ways of arranging them in the factors Vni(m). Thus if we have certain multiplicities h1, . . . , hk of the different indices ni we see that the generic fiber is formed by Qi hi! points. In fact even if there are no multiplicities, so the map j is birational, the same argument may show that in non generic fibers we may perform some of these permutations and so the map is not usually bijective. Example 3.10. t = 2 = 1 + 1, the variety V1(m) is just affine space with coordinate ring the polynomial ring F [x1, . . . , xm] so V1(m) × V1(m) is 2m dimensional affine space with coordinate ring the polynomial ring F [x1, . . . , xm, y1, . . . , ym]. For the other ring we take monomials in the elements (xi, yi) which should be thought of as diagonal 2 × 2 matrices. Such a monomial is the pair formed by a monomial in the xi and the same in the yi. Its trace is the sum over these two monomials, a symmetric function in the exchange τ between xi, yi. The map is generically 2 to 1, the image of coordinate rings is the ring of τ invariants. In fact we have an even stronger statement, let Ai denote the coordinate ring of the variety i=1 Vni(m). We claim Vni(m) so A1 ⊗ A2 ⊗ . . . ⊗ Au is the coordinate ring of the variety Qu that Lemma 3.11. A1 ⊗ A2 ⊗ . . . ⊗ Au is integral over TA(m), spanned by monomials of degree bounded by some number independent of m. Proof. Consider a polynomial f (t) = tm − a1tm−1 + a2tm−2 + (−1)mam 24 with roots x1, . . . , xm so that the elements ai are the elementary symmetric functions in the xi, we may even take the xi as indeterminates. Given an integer k ∈ N and a set S ⊂ {1, . . . , m} with h elements set S := Xi∈S Consider next the polynomial of degree N := (cid:0)m h(cid:1) YS⊂{1,...,m}, S=h (−1)ibitN −i := Xi=1 tN + N X k xk i (t − X k S). The coefficients bi of this polynomial are clearly symmetric functions in the variables xi so they are expressible as polynomials in the elements ai, in fact bi is a polynomial of degree ki in the variables xi so it is a polynomial of this weight when we give to ai the weight i. Each Ai is generated by the traces tr(Mi) of the monomials M acting on the ith summand. Thus the element tr(Mi) is a sum of ni eigenvalues out of the d eigenvalues of the monomial M we deduce a universal polynomial of degree (cid:0) d polynomials in the elements tr(M k), k = 1, . . . , d. ni(cid:1) satisfied by tr(Mi) with coefficients 3.3.2 The support of Kemer polynomials We now apply the previous discussion to Kemer polynomials, first let us take a fundamental algebra D with semisimple part ⊕q i . We have constructed the map from Tt(Y ) to TD. i=1Mni(F ), t = Pi n2 Lemma 3.12. The algebra TD is the coordinate ring of the irreducible subvariety Wn1,...,nu of Mt(F )m//P GL(t). Proof. By definition TD is the algebra generated by the traces of m–tuples of elements of ⊕q i acting on itself by left multiplication so the Lemma follows from the previous discussion as summarized by Formula (21). i=1Mni(F ), t = Pi n2 Let A = ⊕u (t, s). To each Ai we have associated the irreducible subvariety of Mt(F )m//P GL(t, F ): i=1Ai be a direct sum of fundamental algebras all with the same Kemer index Wi := Wn1,...,nqi , ¯Ai = ⊕qi j=1Mnj (F ), qi Xj=1 n2 j = t. We have thus the rather interesting fact. Theorem 3.13. The image of the algebra Tt(Y ) acting on the space of ν–Kemer polynomials, for large ν, KR is the coordinate ring of a, possibly reducible, subvariety of Mt(F )m//P GL(t) union of the subvarieties Wi, i = 1, . . . , u. 25 In other words the Tt(Y )–module KR of ν–Kemer polynomials, for large ν is supported on this subvariety. Notice that Tt(Y ) is an object intrinsecally defined and then also Si Wi is intrinsic being the reflect the structure of the semisimple parts support of KR, the subvarieties Wi = Wn1,...,nqi of the fundamental algebras Ai which may appear as summands of maximal Kemer index of an algebra A having as identities the given T –ideal. There are some subtle points in this construction, first of all by KR we mean the T –ideal generated by Kν for ν large, in the sense that this variety stabilizes for ν large. It is possible that some variety Wi is contained in another Wj, this gives an embedded component which may not be visible just by the structure of the module KR but depends on the embedding KR ⊂ ⊕iKi which in turn depends on the particular choice of A so that the ideal Γ = IdA = ∩u i=1IdAi. This appears as some primary decomposition and it is worth of further investigation. A specific element of KR vanishes on one of the varieties Wi if and only if it is a polynomial identity for the corresponding summand Ai. Corollary 3.14. If R is the relatively free algebra associated to a fundamental algebra A with semisimple part ¯A = ⊕q j=1Mnj (F ), then the module of Kemer polynomial is supported on the irreducible variety Wn1,...,nq. In particular if two fundamental algebras are PI equivalent then they have the same semisim- ple part. In general if we have two equivalent PI algebras A = ⊕iAi, B = ⊕jBj each a direct sum of fundamental algebras, we see that the semisimple components which give maximal varieties of representations are uniquely determined. This answers at least partially the question on how intrinsic are the constructions associ- ated to a particular choice of an algebra A having a chosen T –ideal as ideal of polynomial identities. 3.3.3 Dimension For an associative algebra R with 1, over a field F , Gel’fand–Kirillov [18] have defined a dimension as follows. Let V ⊂ R be a finite dimensional subspace with 1 ∈ V . Let Vn denote the span of all products of n elements of V and set dV (n) := dim Vn. Definition 3.15. [Gel’fand–Kirillov–Dimension (GK–Dimension)] Dim R := sup lim sup V n→∞ log dV (n)/ log n. If R is generated by V then Dim R = lim sup n→∞ log dV (n)/ log n. 26 In general a finitely generated non–commutative algebra may have infinite dimension or a dimension which is not an integer, [11]. A special case is when R is graded and its Hilbert series HR(t) := P∞ rational function of type of Formula (15) (this is a rather strong constraint on R). Since k=0 dim F Rktk is a 1 1 − th = ∞ tih Xi=0 one has Pr j=0 ajtj QN j=1(1 − thj ) = r Xj=0 ajtj N ∞ ( Yj=1 Xi=0 tihj ). j=1 ijhj, ij ∈ N. j=1(1 − thj )−1 is the Hilbert series of the polynomial algebra in generators k=0 cktk we see that ck is a non negative integer which counts in how many ways the integer k can be written in the form The function QN x1, . . . , xN with xi of degree hi. Write QN k = PN Such a function ck is classically known as a partition function, it coincides on the positive integers with a quasi–polynomial of degree N − 1. Quasi–polynomial means in this case that, when we restrict ck on each coset of Z modulo the least common multiple of the hj, on the positive integers in this coset this function coincides with a polynomial of degree N − 1. There is an extensive literature on such functions (cf. [13]). j=1(P∞ i=0 tihj ) = P∞ If we develop the rational function of Formula (15) in power series P∞ k=0 dktk we still have that after a finite number of steps the function dk coincides with a quasi–polynomial D(k), but its degree may be strictly lower than N − 1. One has Theorem 3.16. The dimension of R is the order of the pole of HR(t) at t = 1 and equals n + 1 where n is the degree of the quasi polynomial D(k). If R is a commutative algebra, finitely generated over a field F , has a finite dimension which can be defined in several ways, either as Krull dimension, that is the length of a maximal chain of prime ideals or as Gelfand–Kirillov dimension, or finally when R is a domain as the trascendence degree of the field of fractions of R over F . For a finitely generated commutative graded algebra R, the Hilbert series is a rational function of type of Formula (15), the dimension equals the dimension of its associated affine variety V (R). For a module M we have the notion of support of M, that is the set of point p ∈ V (R) where M is not zero, that is M ⊗R Rp 6= 0 where Rp is the local ring at p. Then the dimension of M equals the dimension of its support. The support is computed as follows Remark 3.17. It is well known that a finitely generated module M over a commutative Noetherian ring R has a finite filtration 0 ⊂ M1 ⊂ M2 . . . ⊂ Mk = M such that Mi+1/Mi = R/Pi where Pi is a prime ideal. If R, M are graded and R finitely generated, the Mi can be taken graded and the Hilbert series is the sum of the Hibert series of the R/Pi, the dimension is the maximum of the dimensions of the R/Pi. 27 In the language of algebraic geometry R defines the algebraic variety V (R) and the Pi some subvarieties with coordinate rings R/Pi, one sees by induction that M is supported on the union of these subvarieties and thus its dimension is the dimension of its support. If M is torsion free over a domain R then it contains a free module Rk ⊂ M so that M/Rk is supported in a proper subvariety, hence it has lower dimension and the dimension of M equals that of R. We will apply this to the non commutative relatively free algebras. 3.3.4 The dimension of relatively free algebras We take as definition of dimension for a relatively free algebra, the one given by its Hilbert series, which measures growth and one can see equals the Gelfand–Kirillov dimension. It is almost never equal to the Krull dimension which instead equals the dimension of R/J where J is the nilpotent radical and it is well known that R/I is a ring of generic matrices since the only semiprime T –ideals are the T –ideals of identities of matrices. Let us first analyze a fundamental algebra D, with semi–simple part ⊕q Take the relatively free algebra in m variables FD(m) := F hξ1, . . . , ξmi for D. We have an inclusion of FD(m) ⊂ FD(m)TD(m) and also an inclusion of the ideal KD ⊂ FD(m) generated by Kemer polynomials. Since both KD and FD(m)TD(m) are finitely generated torsion free modules over TD(m) it follows that the dimension of the 4 Hilbert series of KD, FD(m), FD(m)TD(m), TD(m) are all equal to the Gel’fand Kirillov dimension of the algebras FD(m), FD(m)TD(m), TD(m). i is also the We have computed the dimension dim TD(m) = (m − 1)t + q here t = Pq first Kemer index. i=1Mni(F ). i=1 n2 Proposition 3.18. 1. The GK dimension of the relatively free algebra in m variables for a fundamental algebra D is (m − 1)t + q where t is the first Kemer index. 2. The GK dimension of the relatively free algebra in m variables for a direct sum ⊕iDi of fundamental algebras is the maximum of the GK dimension of the relatively free algebras in m variables for the fundamental algebras Di. 3. If Γ is a T –ideal containg a Capelli list and Γ = Ti Γi with the T –ideals Γi irreducible then the dimension of F hXi/Γ is the maximum of the dimensions of F hXi/Γi, each one of these being the relatively free algebra of a fundamental algebra. Proof. The first part we have just proved, as for the second remark that the relatively free algebra in m variables for a direct sum ⊕iDi is contained in the direct sum of the relatively free algebras in m variables for the summands Di, so its dimension is smaller or equal than the maximum of the dimensions of the algebras relative to the summands. On the other hand each relatively free algebra in m variables for the summands Di is also a quotient of the relatively free algebras in m variables for the direct sum so the claim follows.The third part follows from the second and the fact that an irreducible T –ideal containing a Capelli list is the ideal of PI’s of a fundamental algebra (2.26). 28 We can also approach in a more intrinsic form using the filtration. By Theorem 3.13 and the remark following the dimension is the maximum of the dimensions of the modules Ki+1/Ki each one is supported in a union of varieties Wn1,...,nk but on each of these subvarieties the module is torsion free so it has the same dimension as its support, we deduce: Theorem 3.19. The dimension of R is the maximum dimension of the varieties Wn1,...,nk. Now the dimension of Wn1,...,nk is mt − Pq i − 1) = (m − 1)t + q, so we see that as m grows the maximum is obtained from the factors Ri for which the first Kemer index equals the Kemer index t of R and among these the one with maximum q. i=1(n2 Definition 3.20. The integer q is an invariant of the T –ideal, called q invariant. Corollary 3.21. The dimension of R(m) for m large is (m − 1)t + q where t is the first Kemer index and q is the q invariant. Observe that when R is the ring of generic n × n matrices, then t = n2 and q = 1, the formula is valid for all m ≥ 2. By work of Giambruno Zaicev a similar statement, that the formula is valid for all m ≥ 2, is true for block triangular matrices with q equal the number of blocks and t the dimension of the semisimple part. 3.4 Cocharacters We want now to extend work of Berele [9] and Belov [8] in which they show how cocharacter multiplicities are described by partition functions. This requires some standard preliminaries. 3.4.1 Partition functions The notion of partition function can be discussed for a sequence of integral vectors S := {a1, . . . , am}, ai ∈ Zp for which there is a linear function f with f (ai) > 0, ∀i. 3 Then one defines the partition function on Zp b ∈ Zp, PS(b) = #{t1, . . . , tm ∈ N m Xi=1 tiai = b}. Of course PS(b) = 0 unless b ∈ C(S) := {Pi xiai xi ∈ R+} the positive cone generated by the elements ai. The assumption f (ai) > 0, ∀i means that C(S) is pointed, that is it does not contain any line (only half lines). It is customary to express the partition function by a generating function, this is a series in p variables x1, . . . , xp. When b = (b1, . . . , bp) ∈ Zp we set xb := Qp i=1 xbi i and then setting PS = Xb∈Zp PS(b)xb 3This restriction is essential otherwise the value of the partition function is ∞ 29 . one sees that PS = 1 i=1(1 − xai) Qm This formal series is also truly convergent on some region of space. We can interpret this in terms of graded algebras (or geometrically as torus embedding). Let RS = F [y1, . . . , ym] be the polynomial algebra in m variables yi to which we give a Zp grading by assigning to yi the degree ai. For every graded vector space V = ⊕a∈ZpVa, for which dim (Va) < ∞, ∀a, we can define its graded Hilbert series HV = Xa∈Zp dim (Va)xa. One has to be careful when manipulating such series since in general a product of two such series makes no sense, so if we have two graded vector spaces V, W with the previous restriction on graded dimensions, in general V ⊗ W does not satisfy this restriction. The product makes sense if we restrict to series supported in a given pointed cone, in this case we have that HV ⊗W = HV HW . In general let us consider a finitely generated graded RS module M then we have the Lemma 3.22. The partition function of S coincides with the Hilbert series of RS. The Hilbert series of M has the form HM = p(x) i=1(1 − xai) Qm , p(x) ∈ Z[x±1 1 , . . . , x±1 p ]. (22) That is p(x) is a finite linear combination with integer coefficients of Laurent monomials. If S ⊂ Np and M is graded in Np then the elements p(x) are polynomials. Definition 3.23. In [9], Berele calls a rational function of the type of Formula (22) with p(x) a polynomial a nice rational function. If we set all the variables xi equal to t in a nice rational function H we have a nice rational function of t, the order of the pole at t = 1 of this nice rational function will be called the dimension of H. This dimension gives an information on the growth of the coefficients of the generating function H(t) := P∞ k=0 cktk = p(t) Q(1−thi ) . In fact: Proposition 3.24. The function ck for k sufficiently large is a quasi–polynomial, that is a polynomial on each coset modulo the least common multiple m of the hi. If ck ≥ 0 for k >> 0 then the dimension which equals the order of the pole of this function at t = 1 equals the (maximum) degree of these polynomials plus 1. Proof. Let m be the least common multiple of the integers hi so that m = hiki, =⇒ (1 − tm) = (1 − thi)( ki−1 Xj=0 thij). 30 It then follows that H(t) can be written in the form P (t) Since for i > 0 we have (1−tm)d with P (t) some polynomial. tm (1 − tm)i = 1 − (1 − tm) (1 − tm)i = 1 (1 − tm)i − 1 (1 − tm)i−1 , it then follows that H(t) has an expansion in partial fractions H(t) = d Xi=1 pi(t) (1 − tm)i + q(t) with all the pi(t) polynomials of degree < m in t and q(t) a polynomial. Then remark that the generating function of tj (1 − tm)i = ∞ Xk=0 (cid:18)i + k − 1 i − 1 (cid:19)tmk+j, 0 ≤ j < m gives a polynomial on the coset Zm + j with positive values of degree i − 1. It follows that after a finite number of steps (given by the polynomial q(t)) the fucntion ck is a polynomial on each coset, and of degree d − 1 on some cosets where the coefficients of pd(t) are different from 0. If ck is definitely positive it follows that all the coefficients of pd(t) are non negative, in particular p(1) 6= 0 and the order of the pole at 1 of H(t) is clearly d. 3.4.2 The Theory of Dahmen–Micchelli We want to recall quickly some features of the theory of partition functions. Let S be as before a list of integral vectors. We want to decompose the cone C(S) into big cells and define its singular and regular points. The big cells are defined as the connected components of the open set of C(S) of regular points, which is obtained when removing from C(S) the singular points which are defined as all linear combinations of some subset of S which does not span Rp. One finally needs the notion of quasi–polynomial on Zp. This is a function f on Zp for which there exists a subgroup Λ ⊂ Zp of finite index, so that f , restricted to each coset of Λ in Zp coincides with some polynomial. In the theory of partition functions plays a major role a finite dimensional space of quasi– polynomials introduced by Dahmen–Micchelli. Let us recall their Theory. • Given a list of integral vectors S := {a1, . . . , am}, ai ∈ Zp spanning Rp we call a subset Y of S a cocircuit if it is minimal with the property that S \ Y does not span Rp. The set of all cocircuits of S will be denoted by E(S). • To S is associated a remarkable convex polytope the Zonotope Z(S) := {Xi tiai 0 ≤ ti ≤ 1}. 31 • The faces of this zonotope come in opposite pairs corresponding to the cocircuits of S. Moreover Z(S) can be paved by parallelepipeds associated to all the bases of Rp which can be extracted from S. Each of these parallelepipeds has volume a positive integer, the absolute value of the determinant of the corresponding basis elements. • For a list of integral vectors Y we then define the difference operator ∇Y := Yy∈Y ∇y, ∇yf (x) = f (x) − f (x − y). • The space DM(S) of Dahmen–Micchelli is the space of integral valued functions on Zp solutions of the system of difference equations ∇Y f = 0, Y ∈ E(S). • The space DM(S) is a space of quasi–polynomials of degree m − p, it is a free abelian group of dimension δ(S) the weighted number of bases extracted from S, or the volume of Z(S). • In fact if we take a generic shift a − Z(S), a ∈ Rp we have that a − Z(S) ∩ Zp has exactly δ(S) elements and the restriction of DM(S) to a−Z(S)∩Zp is an isomorphism with the space of integral valued functions on a − Z(S) ∩ Zp. The main Theorem on partition functions for which we refer to [13] is, assuming that S spans Rp. Theorem 3.25. PS is supported on the intersection of the lattice Zp with the cone C(S). Given a big cell c of C(S) and a point a ∈ c very close to 0 one has that a − Z(S) ∩ Zp intersects C(S) only in 0. PS coincides on each big cell c with the quasi polynomial in the space DM(S) which is 1 at 0 and equals 0 in the other points of a − Z(S) ∩ Zp. Finally the partition function may be interpreted as counting the number of integral points in the variable convex compact polytope V (b) := {(t1, . . . , tm) ti ∈ R+, Pi tiai = b}. Thus the partition function is asymptotic to the volume of this variable polytope. This volume function, denoted by TS(b) is a spline that is a piecewise polynomial function on C(S), polynomial, in each big cell, of degree m − p and again there is a remarkable theory behind these functions, cf. [13]. 3.4.3 U invariants In order to apply the previous theory to cocharacters we need to recall some basic facts on highest weight vectors and U invariants, where U is the subgroup of the linear group of strictly upper triangular matrices with 1 on the diagonal. Recall that we have seen, Proposition 3.4, that the multiplicity mλ of a cocharacter equals the multiplicity of the highest weight vectors of weight λ in the relatively free algebra in k 32 variables, if we are considering an algebra of dimension k or more generally if it satisfies the Capelli list Ck+1(Remark 2.10). By Theorem (3.1) the Hilbert series of R is the sum of the contributions of the finitely many factors Ki+1/Ki which are all modules over finitely generated algebras. Moreover these are all stable under the action of the linear group G = GL(k) so that we have some decomposition into irreducible representations Ki+1/Ki = ⊕λSλ(F k)⊕mi,λ. (23) These multiplicities mi,λ can be computed by taking the vector space of U invariants, (Ki+1/Ki)U = ⊕λ(Sλ(F k)U )⊕mi,λ. Since dim Sλ(F k)U = 1 is generated by a single vector of weight λ, we have that the graded Hilbert series of (Ki+1/Ki)U , which is Nk graded by the coordinates mi of the dominant weights, is the generating function of the multiplicities mi,λ. We thus have that (Ki+1/Ki)U is a graded module over a graded polynomial algebra in finitely many variables ω1, . . . , ωk and the number mi,λ is the dimension of its component of degree λ = Pj njωj. So our aim is to prove that (Ki+1/Ki)U is finitely generated as graded module over the polynomial algebra in the variables ω1, . . . , ωk. Then we can apply lemma 3.22 and Theorem 3.25 which properly interpreted give the desired result. This actually is a standard fact on reductive groups and let us explain its proof. Let G be a reductive group in characteristic 0. Its coordinate algebra F (G) decomposes, under left and right action by G, as F (G) = ⊕Vi ⊗ V ∗ i , where Vi runs over all irreducible rational representations of G. If we fix a Borel subgroup B with unipotent radical U the algebra F (G)U = ⊕Vi ⊗ (V ∗ i )U (U acting on the right) is a finitely generated algebra over which G acts on the left. In fact it is generated by the irreducible representations associated to the fundamental weights. This algebra is the coordinate ring of an affine variety G/U which contains as open orbit G/U. For every subgroup H of G consider the invariants under the right action F (G)H. Given a representation V of G, the group G then acts diagonally on V ⊗ F (G)H and: Lemma 3.26. For every representation V of G we have the equality V H = (V ⊗ F (G)H)G. (24) In fact this is given by restricting to (V ⊗ F (G)H)G the explicit map π : V ⊗ F (G) → V, π : v ⊗ f (g) 7→ vf (1). If V is an algebra with G action as automorphisms this identification is an isomorphism of algebras. 33 Proof. The space of polynomial maps from G to V is clearly V ⊗F (G) where v⊗φ corresponds to the map g 7→ φ(g)v. We act on such maps with G × G, the right action is f h(g) := f (gh) while for the left action we use hf (g) := hf (h−1g). These two actions commute, and the left action on maps is the tensor product of the action on V and the left action on F (G), that is if f = v ⊗ φ we have hf = hv ⊗h φ. . A map f is G equivariant, where on G we use the left action if f (hg) = hf (g), this means that f is invariant under the left action on maps. For such a map we have f (g) = f (g1) = gf (1), conversely given any vector v ∈ V the map g 7→ gv is G equivariant so we have a canonical identification j : (V ⊗ F (G))G ∼= V, j(f ) := f (1), ⇐⇒ j : v ⊗ φ 7→ φ(1)v. Moreover the map j is G equivariant if on (V ⊗ F (G))G we use the right G action, since for an equivariant map f we have f h(g) := f (gh) (induced from right action), which maps to f h(1) = f (h) = hf (1). by some subgroup H it means that for the corresponding map f : G → V we have that f (gh) = f (g), ∀h ∈ H, thus f (h) = f (1) and, if f is G equivariant we see that this is Under this identification, if a = Pj vj ⊗ φj(g) ∈ V ⊗ F (G)H is invariant under right action equivalent to the fact that Pj vj ⊗ φj(1) ∈ V H. As for the second statement it is enough to observe that π is a homomorphism of algebras. This allows us to replace for G modules, the invariant theory for U, which is not ruled by Hilbert’s theory since U is not reductive, with the one of the reductive group G, and obtain in this way the desired statements on finite generation. In fact a simple argument as in Hilbert theory shows the following. Let M be a module finitely generated over a finitely generated commutative algebra A. Assume that a linearly reductive group G acts on M, and on A by automorphisms in a compatible way, that is g(am) = g(a)g(m), a ∈ A, m ∈ M then Lemma 3.27. The space of invariants M G is finitely generated as a module over the finitely generated algebra AG. Proof. Consider the A submodule AM G of M. m1, . . . , mk ∈ M G. It is finitely generated by some elements G equivariant so it commutes with the projection to the invariants (in A called the Reynolds Thus if u ∈ M G we have u = Pi aimi, ai ∈ A. Now the map Ak → M given by Pi aimi is operator R), so u = Pi R(ai)mi, R(ai) ∈ AG hence M G is generated over AG by the elements mi. At this point we only have to apply the theory of graded modules over graded algebras, here the grading is by the semigroup of dominant weights (which one can identify to Nk) and use the fact that the algebra F (G)U is finitely generated by elements which have as weight the fundamental weights. Thus if V is a finitely generated module over the finitely generated algebra A with G action, we have that V ⊗ F (G)U is a finitely generated module over the finitely generated algebra A ⊗ F (G)U with diagonal G action. We deduce 34 Theorem 3.28. V U is a finitely generated module over the finitely generated algebra AU . Corollary 3.29. If V and A are as before, and V = ⊕λmλSλ(F k), the generating function Pλ mλtλ is a rational function, in the variables ti := tωi with denominator a product of 1 − tµi = 1 − tPi miωi = 1 −Qi tmi The µi = Pi miωi, mi ∈ N are the dominant weights of some finite set of irreducible representations generating A as algebra. . i Proof. If V = ⊕λmλSλ(F k) we have V U = ⊕λmλSλ(F k)U , and Sλ(F k)U is 1–dimensional generated by a vector of weight λ so Pλ mλtλ is the Hilbert series of the graded module V U (graded by dominant weights). Then compute the generating function of V U using its identification with (A ⊗ F (G)U )G and then apply Lemma 3.22 and Lemma 3.27, we only need to remark that the torus T acts on (A ⊗ F (G)U )G by acting on F (G) on the right and under the identification the weight is preserved. In the algebras which we are studying and the various graded objects associated we have the action of Gl(k) which is in fact a polynomial action, that is no inverse of the determinant appears. The action of the torus of diagonal matrices a1, . . . , ak determines the weight decomposition so the multi–grading and the grading. The fundamental weight ti := ωi = a1a2 . . . ai has degree i so the ordinary Hilbert series of n is given by substituting ti := tωi 7→ ti so that tPi miωi 7→ tPi mii. V U , Pn dim V U Notice that in degree n the dimension of V U components in which Vn decomposes. n equals the length or number of irreducible Corollary 3.30. The length of Vn is a nice rational function with numerator a polynomial in Z[t] and denominator a product as factors of type 1 − thi. 3.4.4 Cocharacters We want to apply the previous Theory to cocharacters. Let A be a PI algebra satisfying a Capelli identity (or rather a Capelli list) Cm. By Kemer’s theorem this is in fact PI equivalent to a finite dimensional algebra. ters χλ associated to partitions with height < m. We have then that the cocharacter χk = Pλ⊢k ht(λ)<m nk,λχλ is a sum of irreducible charac- Consider the generating function Pk(Pλ⊢k ht(λ)<m nλtλ). We write tλ = Qm−1 i=1 tni Theorem 3.31. The generating function Pλ nλtλ of the multiplicities of the cocharacters tn, n := (n1, . . . , nm−1) where ni equals the number of columns of λ of length i. is a nice rational function (3.23) that is it has the form := i Hco = , p(t) ∈ Z[t1, . . . , tm−1]. (25) p(t) Qm i=1(1 − tni) The generating function of the colengths is also a nice rational function ([9]). 35 Proof. From what we have seen nλ is the multiplicity of the irreducible representation Sλ(X) in the relatively free algebra FA(X) associated to A. We know that this multiplicity equals the multiplicity of Sλ(X)U (Remark 2.8 and 2.10) which equals Sλ({x1, . . . , xm−1})U . Now we have for the free algebra in m − 1 variables the canonical filtration, where each Ki/Ki−1 is a GL(k) module, so it has a generating series of cocharacters H i co. Furthermore Ki/Ki−1 satisfies the hypotheses of Theorem 3.28 with V = Ki/Ki−1 and A = Ti. So for each Ki/Ki−1 we may apply corollary 3.29, giving a contribution to Formula (25) of the same type. For the colength we apply Corollary 3.30. co and Hco = Pi H i Remark 3.32. In principle the rational function describing the generating series of the cocharacters contains all the information on the codimension. The series of codimensions is obtained from the series of cocharacters by a formal linear substitutions of a monomial ta by χa(1)ta. It may be worth of further investigation the properties of this linear map on the space of power series which are expressed by rational functions. 3.5 Invariants of several copies of V In the next section we shall deduce some precise estimates on the dimension (cf. Definition 3.23) of the rational functions expressing cocharaters and colength for a fundamental algebra. We need some general facts first. Let us ask the following question, let V be a vector space of some dimension k and G a semisimple group acting on V . The invariants of m copies V m under the action of G are also a representation of GL(m, F ), in fact from Cauchy’s Formula we have S[(V ∗)m] = ⊕λSλ(V ∗) ⊗ Sλ(F m) =⇒ S[(V ∗)m]G = ⊕λSλ(V ∗)G ⊗ Sλ(F m). If we are interested in understanding the multiplicity with which a given representation Sλ(F m) appears we know that it equals dim Sλ(V )G provided that m is larger than the height of λ but it may appear only if the height of λ is ≤ dim V . Thus this multiplicity stabilizes for m ≥ dim V . On the other hand if U is the unipotent group of SL(k, F ) of strictly upper triangular matrices we have S[(V ∗)k]G×U = ⊕λSλ(V ∗)G ⊗ Sλ(F k)U , dim Sλ(F k)U = 1. Thus the generating function of these multiplicities is the generating function of S[(V ∗)m]G×U . In particular for the growth we need to compute the dimension of the algebra S[(V ∗)m]G×U . This we compute as follows, from the previous section we have that S[(V ∗)k]G×U = (S[(V ∗)k] ⊗ F (SL(k))U )G×SL(k). (26) Theorem 3.33. If G acts faithfully on V and G × SL(k) acts freely on a non empty open set of the variety V k × SL(k)/U the dimension of S[(V ∗)k]G×U is k2 + k 2 − dim (G). (27) 36 Proof. From Formula (26) this dimension is the dimension of the quotient variety of G × SL(k) acting on the variety V k × W , where W is the variety of coordinate ring F (SL(k))U which contains SL(k)/U as dense open set so its dimension is (k2 − 1) − k2−k The group SL(k, F ) is semisimple and simply connected, so by a Theorem of Popov (cf. [29] Corollary of Proposition 1) the coordinate ring of SL(k, F ) is factorial, then, since U is a connected group, also the ring F (SL(k))U is factorial hence the variety V k × W is factorial. For a semisimple group H acting on an irreducible affine variety X which is also factorial and with the generic orbit equal to H or just with finite stabilizer one knows, by another Theorem of Popov, cf. [28], that the generic orbit is closed so equals the generic fiber and, by Hilbert’s theory, the quotient variety has dimension dim W − dim H. These hypotheses are satisfied and we have in our case dim W = k2 + k2+k−2 H = G × SL(k) we have dim H = dim G + k2 − 1. The formula follows. 2 = k2+k−2 and since . 2 2 From this Formula we see that only for somewhat small G we may have this strong condition. It is then useful the following criterion. Proposition 3.34. If G acts faithfully on a space V of dimension k and the generic stabilizer of the action of G on V is a torus, G × SL(k) acts freely on a non empty open set of V k × SL(k)/U and G × U acts freely on a non empty open set of V k. Proof. In fact let us look at the stabilizer of ((v1, . . . , vk), U) it is the subgroup of G × U stabilizing (v1, . . . , vk). Then, for a generic choice of (v1, . . . , vk) this is the stabilizer of a generic orbit of G × U on V k so it is enough to show that this is trivial. The action of (g, u) on a vector (v1, . . . , vk) is the action of u on (gv1, . . . , gvk). Moreover u = 1 + Λ is some triangular matrix with λj,i 6= 0, =⇒ j < i so finally. (g, u)(v1, . . . , vk) = (w1, . . . , wk), wi = gvi +Xj<i λj,igvj. (28) Hence if (g, u) stabilizes the vector (v1, . . . , vk) we must have vi = wi ∀i. In particular v1 = gv1, v2 = gv2 + λv1. So since v1 is generic g is a semisimple element being in a torus. Decompose V = V g ⊕ Vg the invariants and a stable complement, write each vi = ai + bi, ai ∈ V g, bi ∈ Vg. Consider thus a2 + b2 = v2 = gv2 + λv1 = a2 + gb2 + λa1, this implies that b2 − gb2 = λa1 ∈ Vg ∩ V g = {0} implies b2 − gb2 = 0 but this implies b2 ∈ V g hence b2 = 0. Since b2 is generic in Vg this implies Vg = 0 or g = 1. Thus form Formula (28) we deduce Pj<i λj,ivj = 0, ∀i. Then since (v1, . . . , vk) are generic they are linearly independent and this implies λj,i = 0 and also u = 1. 3.5.1 Colength We want to investigate now the colength of R = ⊕∞ n=0Rn where R is the relatively free algebra, in k variables, of some finite dimensional algebra. By definition the colength is the function ℓ(Rn) of n which measures the number of irreducible representations of GL(k, F ) 37 If k is larger than the degree m of a Capelli list i=1Tni(m) be the ring of invariants of m copies of a semisimple algebra A = decomposing the part Rn of degree n. satisfied by A we also know that the colength stabilizes. Let SA = ⊗q ⊕q i=1Mni(F ) under its automorphism group G = Qi P SL(ni). A comes from Formula (26) for V = A = ⊕q The algebra SU Qi P SL(ni) which acts faithfully on ¯A and it is semisimple. Moreover the generic element of A is a list of matrices each with distinct eigenvalues so that the stabilizer is a product of maximal tori. We thus have verified all the properties of Proposition 3.34 we can thus apply Theorem 3.33 and then the formula for the dimension of SU A is given by Formula (27) where i=1Mni(F ) under the group G = k = dim A = t = Pi n2 i while dim G = Pq We finally get for the quotient the dimension i=1(n2 i − 1) = t − q. Proposition 3.35. If A = ⊕q of Formula (26) equals t2−t 2 + q. i=1Mni(F ), G = Qi P SL(ni) the dimension of the algebra SU A Proof. Here we apply Formula (27) with k = t and dim G = t − q. We now want to apply this to FA the relatively free algebra of a fundamental algebra A and the trace ring TA. Proposition 3.36. If FA is the relatively free algebra of a fundamental algebra A with ¯A = ⊕q i=1Mni(F ) the dimension of the rational function of colengths is t2 − t 2 + q, t = q Xi=1 n2 i . Proof. By lemma 3.11 the ring of invariants S ¯A is a finite (torsion free) module over the trace ring TA so, by Theorem 3.28, SU A and hence the dimension of T U A . The proposed dimension is the dimension of the colength function of SA hence also of TAFA, which is a finite torsion free module over TA. But also FA contains an ideal KA which is an ideal for TAFA hence it has the same dimension. ¯A a finite (torsion free) module over T U A equals that of SU function Hℓ(R) := P∞ For a general relatively free algebra R, using the standard filtration we have that the colength of R is the sum of the colengths of the factors Ki+1/Ki and we have seen that the generating n=0 ℓ(Rn)tn is a nice rational function with denominator a product of factors 1 − thi. So, for large n the colength ℓ(Rn) is a quasi–polynomial of some degree d − 1 where d is the order of the pole of Hℓ(R) at t = 1. Each one of these factors Ki+1/Ki is a finite module over some finitely generated algebra T ¯Ri ⊂ ⊕jTAj , where ¯Ri quotient of Ri is the relatively free algebra associated to an algebra A = ⊕jAj, direct sum of fundamental algebras Aj all with the same Kemer index, the one of Ri. T ¯Ri ⊂ ⊕jTAj is the coordinate ring of some union of the varieties Wj associated to the fundamental algebras Aj. By Theorem 3.1 Moreover Ki+1/Ki ⊂ ⊕j(Ki+1/Ki)j, and each (Ki+1/Ki)j is torsion free over the corresponding TAj . We claim that the number d is also 38 the maximum of the order of the pole of the Hilbert series of the colength of TAj hence the dimension of T U Aj . Let us summarize these results for R(m) a relatively free in m variables satisfying some Capelli list Ck+1. Denote by ¯Ri the quotients of the standard filtration Theorem 3.37. The generating function of the colength of R(m) is a nice rational function which stabilizes for m ≥ k. When m ≥ k this rational function has dimension max( t2 index of ¯Ri while qi is its q invariant (Definition 3.20). i −ti 2 + qi) where ti is the first Kemer Proof. The only thing which requires some proof is the dimension. This computation de- pends on the following fact the dimension of a nice rational function associated to a gener- ating sequence Pi citi, ci > 0 is the order of the pole at t = 1. Hence one has easily, from to generating sequences Pi citi, ci > 0 equals the maximum of these dimensions. Proposition 3.24, that the dimension of the sum of several nice rational functions associated In our case one has to compute the maximum arising from the colength in the standard filtration and finally the argument is, using the previous discussion for fundamental algebras, like the argument of Theorem 3.19. 4 Model algebras 4.1 The canonical model of fundamental algebras i=1 n2 i . i=1Mni(F ) so that β(A) = t = Pq We want to discuss now the problem of choosing special fundamental algebras in the PI equivalence classes. In corollary 3.14 we have seen that two PI–equivalent fundamental algebras have isomorphic semi–simple parts. Thus it is natural to study fundamental algebras A with a given fixed semi–simple part ¯A = ⊕q From Lemma 2.22 it follows that the second Kemer index γ(A), which for a fundamental algebra coincides with the maximum s for which J s 6= 0, must be ≥ q − 1. The case γ(A) = q − 1 is attained by upper triangular matrices with semi–simple part ¯A. So our present goal is to analyse fundamental algebras with given ¯A and Kemer index t = dim ( ¯A), s ≥ q − 1. We use now Definition 2.25 and Proposition 2.26. Consider a T –ideal I of identities of a fundamental algebra A = ¯A ⊕ J, we have seen in Corollary 3.14 that the semisimple part ¯A is determined by I. We want to construct a canonical fundamental algebra having semisimple part ¯A and I as T –ideal I of identities. This algebra is constructed as in §2.2. First we construct a universal object. Take the free product ¯A⋆F hXi, for X = {x1, . . . , xm}. We assume m ≥ q where q is the number of simple blocks of ¯A. We now take this free product modulo the ideal of elements of degree ≥ s + 1 in the variables X, where s is some fixed integer, call F ¯A,s(X) the resulting algebra. F ¯A,s(X) is a finite dimensional algebra with semisimple part ¯A and Jacobson radical J of nilpotency s + 1 generated by the xi. It satisfies a universal property among such algebras. 39 Remark 4.1. Given a finite dimensional algebra A, with semisimple part ¯A and Jacobson radical J of nilpotency s+1 any map X → J, extends to a unique homomorphism of F ¯A,s(X) to A which is the identity on ¯A. Given a finite dimensional algebra A with given t, s index we define As(X) to be F ¯A,s(X) modulo the ideal generated by all polynomial identities of A. The Jacobson radical Js of As(X) (and also of F ¯A,s(X)) is the ideal generated by the xi and its semisimple part is ¯A. By construction and Remark 4.1, given any list of elements a1, . . . , am ∈ J there is a mor- phism π : As(X) → A which is the identity on ¯A and maps xi 7→ ai. For m sufficiently large this morphism may be chosen so that the radical Js maps surjectively to the radical J so also the map of As(X) to A is surjective, hence As(X) satisfies the same PI as A. Definition 4.2. We now define F ¯A,s and As to be the algebras F ¯A,s(X) and As(X) where X is formed by s variables. Lemma 4.3. As satisfies the same identities as A. Proof. By construction all identities of A are satisfied by As, so we need to prove the converse and we may take a multilinear polynomial F which is not a PI of A. Then there is a substitution of f in A, which we may assume to be restricted, which is non zero. In this substitution at most s variables are in the radical the others are in some matrix units of ¯A. We now take the same substitution for matrix units in ¯A and the remaining variables, which we may call y1, . . . , yk, k ≤ s we substitute in x1, . . . , xk ∈ As. By the universal property the evaluation of f in A factors through this evaluation in As which therefore is different from 0. Lemma 4.4. Let B, A = B/I be two finite dimensional algebras, J the radical of B and I ⊂ J so A, B have the same semisimple part ¯A. Assume that A, B have the same nilpotency index and A is fundamental, then B is funda- mental and with the same Kemer index as A. Proof. The assumption implies that A, B have the same t, s index, then the statement follows from Theorem 2.23 since by by hypothesis the Kemer index of A equals the t, s index, on the other hand clearly the Kemer index of B, which is less or equal than its t, s index, cannot be less than the Kemer index of A. Proposition 4.5. F ¯A,s and As are fundamental algebras with Kemer index t = dim ( ¯A), s. Proof. This follows from Theorem 2.23 since by construction its Kemer index equals the t, s index. Definition 4.6. We call As the canonical model of A. 40 Definition 4.7. The algebra F ¯A,s is the universal fundamental algebra for ¯A, s. Remark that only if m is sufficiently large we have that F ¯A,s(X) is fundamental, since we need the existence of some fundamental algebra quotient of F ¯A,s(X). We claim that we have the exact condition m ≥ q − 1. For m = q −1 we may take as fundamental algebra an algebra R of upper triangular matrices with ¯A as semisimple part it is easily seen that such an algebra is generated by q −1 elements over ¯A. In fact the algebra R can be described as the direct sum ⊕i≤j hom(Vi, Vj) where dim Vi = ni. ¯A = ⊕i hom(Vi, Vi) and hom(Vi, Vj) is an irreducible module under hom(Vi, Vi) ⊕ hom(Vj, Vj) for this we may take the elements Ei,i+1 a non zero matrix in the corresponding block hom(Vi, Vi+1). Remark also that by construction the nilpotent subalgebra of F ¯A,s(X) generated by the xi is a relatively free nilpotent algebra. Question Is the T –ideal of identities of F ¯A,s irreducible, and how is it described? 4.1.1 Description of F ¯A,s(X) Let V be the vector space with basis the elements xi. In degree h the algebra F ¯A,s(X) = F ¯A,s(V ) can be described as follows, consider Definition 4.8. Let M be the monoid in two generators a, b with relation a2 = a. Elements of this monoid correspond to words in a, b in which aa never appears as sub-word. When we multiply two such words, if we have the factor aa appearing we reduce it by the rule to a. Now to such a word w we associate a tensor product Tw of ¯A whenever we have an a and V when we have a b w = abbab 7→ Tw = ¯A ⊗ V ⊗ V ⊗ ¯A ⊗ V. The multiplication of two such words w1w2 according to the previous rule, induces a multi- plication Tw1Tw2 ⊂ Tw1w2. We take the corresponding tensor product and if we get a factor ¯A ⊗ ¯A we replace it by ¯A by multiplication. Then we see that F ¯A,s(X) is the direct sum ⊕Tw where w runs over all the words of previous type with at most s appearances of b. It is a graded algebra over this monoid, truncated at degree s in b. As representation of GL(m, F ) × G = GL(V ) × G, F ¯A,s(V ) in degree h, is a direct sum of ci,j spaces each isomorphic to ¯A⊗i ⊗ V ⊗h where ci,j is the number of words in M with i times a and j times b. The summands correspond to the type of elements in the free product which are monomials in j elements of V and i elements of ¯A. Corollary 4.9. Having fixed s the GL(V ) invariant subspaces of F ¯A,s(V ) all intersect F ¯A,s(Vs) where Vs is the subspace of dimension s spanned by the first s variables. 41 Proof. This is a property of each V ⊗h, h ≤ s. A GL(V ) invariant subspace is generated by its highest weight vectors which in V ⊗h depend on the first h elements of a chosen basis of V (the variables). 4.2 Some complements 4.2.1 Generalized identities The algebra F ¯A,s(X) should be thought of as a free algebra in a suitable category, so we give the following Definition 4.10. Given an algebra R, an R–algebra is any associative algebra S with a bimodule action of R on S satisfying r(s1s2) = (rs1)s2, (s1s2)r = s1(s2r), (s1r)s2 = s1(rs2), ∀r ∈ R, s1, s2 ∈ S. (29) Given an R–algebra S we can give to R ⊕ S a structure of algebra by setting (r1, s1)(r2, s2) = (r1s1, r1s2 + s1r2 + s1s2) the axioms (29) are the ones necessary and sufficient to have that R ⊕ S is associative. The canonical model has the following universal property, consider a nilpotent algebra R satisfying the PI’s of A, with Rs+1 = 0 and equipped with a ¯A algebra structure, according to Definition 4.10. Then any map of X → R extends to a map As → ¯A ⊕ R equal to j on ¯A. In particular F ¯A,s(X) and As behave as relatively free ¯A algebras. The endomorphisms of F ¯A,s(X) resp. As which are the identity on ¯A correspond to arbitrary substitutions of the variables xi with elements of the radical. This gives rise to the notion of T –ideal in F ¯A,s(X) or ideal of generalised identities. Recall that, in an algebra R a verbal ideal is an ideal generated by the evaluations in R of a T –ideal. In other words a verbal ideal is an ideal I of R minimal with respect to the property that R/I satisfies some given set of polynomial identities. The verbal ideals of F ¯A,s defining the algebras As are clearly T –ideals, but not all T –ideals are of this type as even the simplest examples show (cf. Example 4.11). In particular the action of the linear group GL(m, F ) on the vector space with basis the elements xi and also the automorphism group G of ¯A extend to give a group GL(m, F ) × G of automorphisms of F ¯A,s(X) and As. Thus the kernel of the quotient map F ¯A,s(X) → As is stable under the group GL(m, F ) × G of automorphisms. In fact the possible ideals of F ¯A,s(X) appearing in this way are all verbal ideals evaluations on F ¯A,s(X) of a T ideal Γ which contains the PI’s of ¯A. Of course, since ¯A = ⊕q i=1Mni(F ) this condition is that Γ contains the PI’s of n × n matrices where n = max ni. 42 Not all such verbal ideals can occur but only the ones for which the Kemer index is t, s. There is the further condition that do not decrease the nilpotency order s + 1, this must be included if we fix m. If we let m increase then it will be automatically satisfied. In fact the previous analysis confirms the choice of m = s given in 4.2 and proved in 4.3. This may also be interpreted as fixing a Capelli identity satisfied by the algebras under consideration. Example 4.11. ¯A = F, s = 1 then if e is the unit of F the algebra F ¯A,s is 5 dimensional with basis e, x, ex, xe, exe One can see that there are 4 verbal ideals contained in the radical, for the identities [x, y], [x, y]z, z[x, y], z[x, y]w. There are 5, T –ideals. The radical is a T –ideal but not verbal, the corresponding verbal ideal is ex, xe, exe. As for verbal ideals one may construct them as follows. Take a multilinear polynomial f (y1, . . . , yh) which is also a polynomial identity of ¯A, that is of n × n matrices where n = max ni. We may consider all possible evaluations of this polynomial in which some of the variables yj are substituted with matrix units in ¯A and the remaining variables are left unchanged but free to be evaluated in the radical. Then we may leave out of the semisimple evaluation at most s variables which we can evaluate in the elements x1, . . . , xs. In this way we have constructed a finite list of elements in F ¯A,s and the ideal they generated is the verbal ideal associated to f (y1, . . . , yh). 4.3 Moduli We have seen that the canonical fundamental algebras with a given ¯A and s are the quotients of the finite dimensional algebra F ¯A,s modulo verbal ideals. Thus it is natural to ask if these quotients arise in algebraic families. Proposition 4.12. Let A be a finite dimensional algebra, the set of ideals, respectively T – ideals is closed in the Grassmann variety of subspaces of codimension h. Proof. Let I be a subspace, the condition to be an ideal is that it should be stable under multiplication, left and right, by elements of A, instead the condition of being a T –ideal is to be stable also under endomorphisms of A. For any linear operator ρ : A → A the set of subspaces stable under ρ is closed in the Grassmann variety hence the claim. 43 The condition that I is verbal is that it coincides with the ideal generated by the elements f (a1, . . . , ah) for some set of multilinear polynomials. Since A is finite dimensional if I is generated by the evaluations of some set f of multilinear polynomials it is also generated by the evaluations of finitely many of them. So every verbal ideal is in some class Vk of verbal ideals which can be generated by some multilinear polynomials f (y1, . . . , yh) with h ≤ k. In particular thus given some sequence of possible codimensions d := {di}, i = 1, . . . , k we may consider the set Vh,d of verbal ideals of A of codimension h generated by T ideals of those given codimensions. Proposition 4.13. Let A be a finite dimensional algebra, the set of verbal ideals Vh,d is locally closed in the Grassmann variety of subspaces of codimension h. Proof. Let Mk be the space of multilinear polynomials in k variables and Gh(A) the Grass- mann variety of codimension h subspaces of A. In Mk × Gh(A) consider the subset Mk,h,A of these pairs f, U such that all evaluations of f in A lie in the subspace U, clearly Mk,h,A is closed, the projection map πk : Mk,h,A → Gh(A) has the property that the fibre of U is the linear subspace of Mk of the polynomials f such that all evaluations of f in A lie in the subspace U. Since the dimension of a fibre is semicontinuous we have a stratification of Gh(A) by locally closed sets where the dimension is fixed. In particular we have a locally closed subset Gh,d(A) where the dimension of the fibre of πi equals di for all i ≤ k. This then intersected with the closed subset of subspaces which are ideals defines again a locally closed set Xd, on this variety we have two vector bundles, the tautological bundle Ud which at some point U has as fibre the ideal U, and the bundle W := ⊕iWi whose fibre is the direct sum of the fibres Wi = (Wi)U at some ideal U, formed by the direct sum of the spaces of multilinear polynomials of degree i ≤ k which evaluated in A take values which lie in U. In this set we have to identify the subset of the ideals generated by the evaluations of the corresponding polynomials and we claim that this condition is open. This condition is given by imposing that a certain linear map has a maximal rank. The map is the following for a multilinear polynomial f in k variables the span of its image is the image of the map f : A⊗k → A. For a space W of multilinear polynomials we have a map Wk ⊗ A⊗k → A, globally we have have a map of vector bundles ⊕iWi ⊗ Ai → Ud. For the ideal generated by these polynomials we have to add the polynomials xf y with two extra variables x, y which will induce a map ⊕iWi ⊗ Ai+2 → Ud. The verbal ideals is the open set of Xd where this map of vector bundles is surjective. The condition to be surjective is then clearly open in U. Finally when we apply this to fundamental algebras we need the further restriction to take those verbal ideals of F ¯A,s which do not contain the part of degree s, this is again an open condition. 44 4.3.1 A canonical structure of Cayley–Hamilton algebra We finally make a further connection with a construction from invariant theory. We give to F ¯A,s a canonical structure of Cayley–Hamilton algebra as follows. Given a ∈ F ¯A,s we set t(a) := str(¯a) where ¯a ∈ ¯A = ⊕q sum of the traces of the components which are matrices. We claim that with this definition of trace the elements satisfy the (N +1)s Cayley–Hamilton i=1Mni(F ) and we take as trace the identity where N = Pi ni or even the Ns Cayley–Hamilton identity if the algebra has an identity. This follows from Formula (14). We then have universal embeddings F ¯A,s ⊂ M(N +1)s(U), As ⊂ M(N +1)s(UA) with U, UA = U/I commutative rings over which acts the projective linear group G := P GL((N + 1)s) and we have canonical isomorphisms and a commutative diagram F ¯A,s ∼−−−→ M(N +1)s(U)G As πy πy ¯A πy πy ∼−−−→ M(N +1)s(UA)G (30) ∼−−−→ M(N +1)s(U ¯A)G with the vertical maps surjective. We can make a further reduction, by special properties of U ¯A so that ¯A embeds in matrices in a standard way. This depends upon the fact that all embeddings of ¯A into M(N +1)s(F ) which are compatible with the given trace are conjugate, this should mean that U ¯A is the coordinate ring of a homogeneous space and then there is a standard reduction to the subgroup. We cannot expect to reduce the embedding as into the same matrix algebra over F . References [1] A. A. Albert, Structure of Algebras. American Mathematical Society Colloquium Pub- lications, vol. 24. American Mathematical Society, New York, 1939. xi+210 pp. [2] E. Aljadeff, A. Giambruno, C. Procesi, A. Regev, Rings with polynomial identities and finite dimensional representations of algebras Book in preparation [3] E. Aljadeff, A. Kanel-Belov,Y. Karasik, Kemer’s Th. for affine PI algebras over a field of characteristic zero, preprint. [4] S. A. Amitsur, The identities of PI-rings, Proc. Amer. Math. Soc. 4 (1953), 27-34. [5] S. A. Amitsur, The T-ideals of the free ring, J. London Math. Soc. 30 (1955), 470-475. 45 [6] S. A. Amitsur, A noncommutative Hilbert basis Theorem and subrings of matrices, Trans, A.M.S. Volume 149, May 1970 [7] M. Artin, On Azumaya algebras and finite-dimensional representations of rings, J. Algebra 11 (1969), 532–563. [8] A.Ya. Belov, Rationality of Hilbert series with respect to free algebras, Russian Math. Surveys 52 (1997) 394–395. [9] A. Berele, Applications of Belov’s theorem to the cocharacter sequence of p.i. algebras. J. Algebra 298 (2006), no. 1, 208–214. [10] Bergman, George M.; Lewin, Jacques The semigroup of ideals of a fir is (usually) free. J. London Math. Soc. (2) 11 (1975), no. 1, 2131. [11] W. Borho, H. Kraft, Uber die Gelfand–Kirillov-Dimension. Math. Ann. 220 (1976), no. 1, 124. [12] Jean-F. Boutot, , Singularit´es rationnelles et quotients par les groupes r´eductifs. Invent. Math. 88 (1987), no. 1, 65–68. [13] C. De Concini, C. Procesi, Topics in Hyperplane Arrangements, Polytopes and Box- Splines Series: Universitext, 2010, XXII, 381p. [14] S. Donkin, Invariants of several matrices, Inventiones Mathematicae, v. 110, 1992, n. 2, pp. 389–401 [15] V. Drensky, Free algebras and PI-algebras, Graduate Course in Algebra, Springer-Verlag, Singapore (2000). [16] V. Drensky, E. Formanek, Polynomial identity rings, Advanced Courses in Mathematics. CRM Barcelona. Birkhuser Verlag, Basel, 2004. [17] D. Eisenbud, Commutative Algebra: with a View Toward Algebraic Geometry (Springer Graduate Texts in Mathematics)– March 1, 1999 [18] I. M. Gel’fand, A. A. Kirillov, On fields connected with the enveloping algebras of Lie algebras. (Russian) Dokl. Akad. Nauk SSSR 167 1966 503505. [19] A. Giambruno and M. Zaicev, On codimension growth of finitely generated associative algebras, Adv. Math. 140 (1998), 145–155. [20] A. Giambruno and M. Zaicev, Exponential codimension growth of P.I. algebras: an exact estimate, Adv. Math. 142 (1999), 221-243. [21] A. Giambruno and M. Zaicev, Involution codimensions of finite dimensional algebras and exponential growth, J. Algebra 222 (1999),471–484. 46 [22] A. Giambruno and M. Zaicev, A characterization of algebras with polynomial growth of the codimensions, Proc. Amer. Math. Soc. 129 (2001), 59–67. [23] A. Giambruno and M. Zaicev, Minimal varieties of exponential growth, Adv. Math. 174, (2003), 310–323. [24] M. Hochster and J. L. Roberts, Rings of invariants of reductive groups acting on regular rings are Cohen-Macaulay. Advances in Math. 13 (1974), 115–175. [25] A. R. Kemer, Ideals of identities of associative algebras, Amer. Math. Soc. Translations of monographs 87 (1991). [26] G. R. Kempf, Some quotient varieties have rational singularities. Michigan Math. J. 24 (1977), no. 3, 347–352. [27] H. Kraft, Geometrische Methoden in der Invariantentheorie. F. Vieweg, 1985 - Math- ematics - 308 pages. [28] V. L. Popov, Criteria for the stability of the action of a semisimple group on a factorial manifold. Izv. Akad. Nauk SSSR Ser. Mat. 34 1970 523531. [29] V. L. Popov, Picard groups of homogeneous spaces of linear algebraic groups and one–dimensional homogeneous vector bundles, Math. USSR Izvestija Ser. Mat. Tom 38 (1974), No. 2 Vol. 8 [30] V. L. Popov. Contraction of the actions of reductive algebraic groups. Math. USSR- Sb. 58 (1987), 311–335. [31] C. Procesi, The invariant theory of n × n matrices, Adv. Math. 19 (1976), 300-381. [32] C. Procesi, Computing with 2 × 2 matrices, J. Algebra 87 (1984), 432-359. [33] C. Procesi, Lie Groups. An approach through invariants and representations, pp. xxiv+596, Springer, Universitext, 2007 . [34] Yu. P. Razmyslov, Trace identities of full matrix algebras over a field of characteristic zero, Math. USSR Izv. 8 (1974), 724-760. [35] Yu. P. Razmyslov, Identities of Algebras and Their Representations (Russian), Moscow 1989. Translation: Translations of Math. Monographs 138 AMS, Providence, R.I., 1994. [36] A. Regev, Existence of Identities in A ⊗ B, Israel J. Math. 11 (1972), 131-152. [37] A. Regev, Growth for Algebras Satisfying Polynomial Identities, preprint, 2014. [38] N. Roby, Lois polynomes et lois formelles en th´eorie des modules, Annales Scientifiques de l’´Ecole Normale Sup´erieure. Troisi`eme S´erie, v. 80, 1963, pp.213–348. 47 [39] N. Roby, Lois polynomes multiplicatives universelles, Comptes Rendus Hebdomadaires des S´eances de l’Acad´emie des Sciences. S´eries A et B, v. 290, 1980, n. 19, pp. A869– A871. [40] L. H. Rowen, Polynomial Identities in Ring Theory, Acad. Press (1980). [41] L. H. Rowen, Ring Theory, Vol I, II, Acad. Press (1988). [42] F. Vaccarino, Generalized symmetric functions and invariants of matrices, Mathematis- che Zeitschrift, v. 260, 2008, n. 3, pp. 509–526, [43] A. I. Shirshov, On rings with identity relations (Russian), Mat. Sb. 43 (1957), 277-283. [44] D. Ziplies, Abelianizing the divided powers algebra of an algebra, Journal of Algebra, v. 122, 1989, n. 2, pp. 261–274, [45] K. A. Zubrilin, On the largest nilpotent ideal in algebras that satisfy Capelli identities. (Russian) Mat. Sb. 188 (1997), no. 8, 93–102; translation in Sb. Math. 188 (1997), no. 8, 12031211
1810.12422	2	1810	2019-09-19T17:46:09	On the number of clonoids	[ "math.RA" ]	A clonoid is a set of finitary functions from a set $A$ to a set $B$ that is closed under taking minors. Hence clonoids are generalizations of clones. By a classical result of Post, there are only countably many clones on a 2-element set. In contrast to that, we present continuum many clonoids for $A = B = \{0,1\}$. More generally, for any finite set $A$ and any $2$-element algebra $\mathbf{B}$, we give the cardinality of the set of clonoids from $A$ to $\mathbf{B}$ that are closed under the operations of $\mathbf{B}$.	math.RA	math	ON THE NUMBER OF CLONOIDS ATHENA SPARKS Abstract. A clonoid is a set of finitary functions from a set A to a set B that is closed under taking minors. Hence clonoids are generalizations of clones. By a classical result of Post, there are only countably many clones on a 2-element set. In contrast to that, we present continuum many clonoids for A = B = {0, 1}. More generally, for any finite set A and any 2-element algebra B, we give the cardinality of the set of clonoids from A to B that are closed under the operations of B. Further, for any finite set A and finite idempotent algebra B without a cube term (with A, B ≥ 2) there are continuum many clonoids from A to B that are closed under the operations of B; if B has a cube term there are countably many such clonoids. 1. Introduction A clone on a set D is a set of finitary operations on D that contains all projections and is closed under composition of functions (see [6, page 97] for the definition). In particular, clones are closed under the usual manipulations of permuting variables, identifying variables, and introducing dummy variables in functions. For subsets A, B of D, the restriction of a clone on D to the functions from powers of A into B is not a clone anymore. However, this restriction is still closed under the variable manipulations mentioned above. More precisely, such a set of functions is closed under minors. For k ∈ N, let [k] := {1, . . . , k}. Definition 1.1. Let A, B be sets, k ∈ N, and f : Ak → B. For ℓ ∈ N and σ : [k] → [ℓ], the function f σ : Aℓ → B, (x1, . . . , xℓ) 7→ f (xσ(1), . . . , xσ(k)) is a minor of f . Date: September 20, 2019. 1991 Mathematics Subject Classification. Primary: 08A40; Secondary 06E30. Key words and phrases. clones, polymorphisms, Boolean functions, minors. This material is based upon work supported by the National Science Foundation under Grant No. DMS 1500254. 1 2 ATHENA SPARKS Sets of functions that are closed under minors have been investigated by Pippenger in [8]. He developed a Galois theory for them and sets of pairs of relations that generalizes the classical Galois theory for clones. These sets reappeared recently when Brakensiek and Guruswami clas- sified Promise Constraint Satisfaction Problems (PCSP) on Boolean, symmetric, self-dual relational structures via polymorphisms between relational structures A and B of the same type in [3]. Independently, these sets were used by Aichinger and Mayr to investigate equational theories of algebras in [1]. Following the notion introduced in that last paper we define: Definition 1.2. [1, Definition 4.1] Let A be a set and B = (B, F ) an and k ∈ N, we let Ck := C ∩ BAk . algebra. For a subset C ofSn∈N BAn We call C a clonoid with source set A and target algebra B if (1) C is closed under taking minors, and (2) for all k ∈ N, Ck is a subalgebra of BAk . The set of all clonoids with source A and target algebra B is denoted CA,B. Note that every subset C of Sn∈N BAn that is closed under taking minors is a clonoid with target algebra the set (B, ∅). Further, every clone C on a set A is a clonoid with source set A and target algebra (A, C). It is a well known result of Post that there are only countably many clones on a two element set [6, Theorem 3.1.1]. Janov and Mucnik showed that there are continuum many clones on any finite set with three or more elements [6, Theorem 8.1.3]. In light of these results, one may ask whether the number of clonoids for fixed source A and target B depends on the size of A and B. We will show that there are already continuum many clonoids with source and target of size 2 (see Corollary 1.6). Sn∈N BAn We introduce some more notation that will be needed in the following sections. Let A be a set and B = (B, F ) an algebra. For a set F ⊆ , the clonoid with source set A and target algebra B generated by the functions in F is denoted hF iB. If F = ∅, then we simply write hF i. Let P and Q be a pair of m-ary relations on A and B respectively. A function f : Ak → B is a polymorphism of (P, Q) if f applied component-wise to any k-tuple of elements of P is an element of Q. For a set of pairs of relations R := {(Pi, Qi) : i ∈ I} on A and B, the set of functions that are polymorphisms of all pairs of relations in R is denoted Pol(R). If R contains only a single pair of relations (P, Q), we write Pol(P, Q) instead. Note that if f ∈ Pol(R), then any minor of f is in Pol(R). A clonoid C with source set A and target algebra B ON THE NUMBER OF CLONOIDS 3 is finitely related if there exists a finite set of pairs of finitary relations R := {(Pi, Qi) : 1 ≤ i ≤ n} on A and B such that C = Pol(R). It can be easily shown that any finitely related clonoid is the polymorphism clonoid of a single relation. Let X be the k × (2k − 1) matrix with columns {x, y}k \ {x} where x = (x, x, . . . , x). A k-cube term of B is a (2k − 1)-ary term c in the operations of B such that c(X) = x for all x, y in B where c is applied to every row of X. A near-unanimity (NU) term of B is an k-ary (k ≥ 3) term f in the operations of B which satisfies f (y, x, x, . . . , x, x) = f (x, y, x, . . . , x, x) = · · · = f (x, x, x, . . . , x, y) = x for all x, y ∈ B. A Mal'cev term of B is a ternary term f in the operations of B which satisfies f (y, y, x) = f (x, y, y) = x for all x, y ∈ B. Clearly, if B has an NU-term or a Mal'cev term, then B has a cube term. The following main result of this paper gives more precise informa- tion about the cardinality of clonoids with target algebras of size 2. Theorem 1.3. Let CA,B denote the set of all clonoids with finite source A (A > 1) and target algebra B of size 2. Then (1) CA,B is finite iff B has an NU-term; (2) CA,B is countably infinite iff B has a Mal'cev term but no ma- jority term; (3) CA,B has size continuum iff B has neither an NU-term nor a Mal'cev term. Moreover, in cases (1) and (2) all clonoids in CA,B are finitely related. Following the case distinction of the theorem, we consider the size of CA,B for an arbitrary finite B with an NU-term in Section 2, for B with a cube term in Section 3, and for B without cube term in Section 4. In Section 5 we combine the results from these sections to prove Theorem 1.3. The backward direction of (1) and the forward direction of (3) hold for arbitrary finite algebras B of size at least 2; our proofs of the others require that B is Boolean. It is unknown if the forward direction of (1) holds for arbitrary finite algebras B of size at least 2, however, we know that the backward direction of (3) does not. An example of a target algebra B that has neither an NU-term nor a Mal'cev term where CA,B is countably infinite is given in Example 1.5. This example also shows that (2) does not hold for arbitrary target 4 ATHENA SPARKS algebras. It is not known if there exists B with a Mal'cev term but no NU-term where CA,B is finite. The following theorem addresses the size of CA,B for a finite idempo- tent algebra B. Theorem 1.4. Let A be a finite set and B a finite idempotent algebra with A, B > 1. Then CA,B has size continuum iff B has no cube term. This follows immediately from Theorem 3.1 and Lemma 4.2. It is unknown if this holds for arbitrary finite target algebras of size at least two. Given these results, the following example gives a target algebra B that has neither an NU-term nor a Mal'cev term where CA,B is count- ably infinite. Example 1.5. Let A be a finite set and B1 and B2 be algebras of size 2 and type (2,3). The binary operation t is interpreted in B1 as the projection onto to first coordinate and in B2 as the projection onto the second coordinate. The ternary operation s in B1 is the Mal'cev operation x − y + z (mod 2) and s is the ternary majority operation in B2. Because of t, we see that B1 and B2 are independent [5, Lemma 2.1]; that is, the term operations of B1 × B2 are exactly the functions of the form (B1 × B2)k → B1 × B2 ((x1, y1), . . . , (xk, yk)) 7→ (g(x1, . . . , xk), h(y1, . . . , yk)) for k ∈ B and g, h arbitrary term functions of B1, B2, respectively. In particular, B1 × B2 is an idempotent algebra that has a 3-cube term but neither an NU-term nor a Mal'cev term. By Theorem 1.3 (2), there are countably infinitely many clonoids with target algebra B1. Each clonoid C in CA,B1 can be identified with a clonoid C in CA,B1×B2 where C := { f : Ak → (B1 × B2), (x1, . . . , xk) 7→ (f (x1, . . . , xk), 0) f ∈ Ck}. Hence there are infinitely many clonoids with source A and target alge- bra B1 × B2. Therefore, by Theorem 1.4, the number of clonoids with source A and target B is countably infinite. Pippenger showed that there are continuum many clonoids where the target algebra is the set {0, 1} with no operations [8, Proposition 3.4 and following discussion]. Theorem 1.3 (3) gives a alternate proof to this result. Since each clonoid with a target {0, 1} is also a clonoid with target {0, . . . , n} for any n ≥ 1, we immediately have the following: ON THE NUMBER OF CLONOIDS 5 Corollary 1.6. For all m, n ≥ 1, there are continuum many clonoids with source {0, . . . , m} and target {0, . . . , n}. 2. NU-terms In this section, we will show that there are only finitely many clonoids with a finite source A and algebra B with an NU-term. In particular, we show that each such clonoid is the polymorphism clonoid of a single and let ΠAn pair of relations on A and B. We identify AAn be the set of all An-ary projections on A. with AAn A Theorem 2.1. Let A be a finite set of size greater than 1 and B a finite algebra with n-ary NU-term (n ≥ 3). Let C be a clonoid with source A and target B. Then C = Pol(ΠAn−1 , CAn−1). Hence there are only finitely many such clonoids with source A and target B. A Proof. Let f : Ak → B. We claim that f ∈ C iff all An−1-ary minors of f are in C. (2.1) This is equivalent to C = Pol(ΠAn−1 A , CAn−1). The forward direction of (2.1) is immediate from the definition of clonoids. For the reverse direction, note that the k-ary functions in C form a subalgebra Ck of BAk . By the Baker-Pixley Theorem [2], Ck is uniquely determined by its projections onto the subsets of Ak with n − 1 or fewer elements. More precisely, (2.2) f ∈ Ck iff ∀I ⊆ Ak with I ≤ n − 1, ∃g ∈ Ck so that f I = gI. Let Z be a matrix with n − 1 rows whose columns are the An−1 tuples of An−1 in some order. For fixed x1, . . . , xn−1 ∈ Ak, let X denote the matrix with rows x1, . . . , xn−1 and k columns. Let σ : [k] → [An−1] such that the i-th column of X is equal to the σ(i)-th column of Z. With functions acting on the rows of the corresponding matrices, we then have (2.3) f (X) = f σ(Z). With (2.2) and (2.3) it follows that f ∈ Ck. Thus (2.1) and the theorem are proved. (cid:3) 3. Cube term In this section, we will show that all clonoids with a finite source and a target algebra with a cube term, in particular, with a Mal'cev term, are finitely related. We will also construct infinitely many clonoids for a fixed algebra of size 2 with a Mal'cev term. 6 ATHENA SPARKS Theorem 3.1. Let A be a finite set and B a finite algebra with cube term. Then each clonoid C with source A and target B is finitely related. Hence there are at most countably many such clonoids. Proof. Let Inv(C) denote the set of all relational pairs on A and B preserved by C. Let {(Pi, Qi) : i ∈ N} be an enumeration of Inv(C). Then C = Pol(Inv(C)) by the Galois Connection given in [8]. Define Cj := Pol({(Pi, Qi) : i ≤ j}). Then C1 ⊇ C2 ⊇ · · · is a descending chain and (3.1) Cj = C. \j∈N By Theorem 5.3 in [1], CA,B satisfies the DCC. Hence there exists m ∈ N such that Cm = Cn for all n ≥ m. By (3.1), C = Cm and C is finitely related. (cid:3) Next we show that there actually are infinitely many clonoids with target any Mal'cev algebra of size 2 without an NU-term. By Post's classification of Boolean clones, the clone of each such algebra is con- tained in the clone of ({0, 1}, +, 0, 1), where 0, 1 are the unary constant functions. For algebras B and B′, if the clone of B′ is contained in the clone of B, then CA,B ⊆ CA,B′ for any set A. So it suffices to show the following: Lemma 3.2. There exists infinitely many clonoids with source A of size at least two and target algebra B = ({0, 1}, +, 0, 1). Proof. Let 0, 1 ∈ A and for k ∈ N define ek : Ak → {0, 1}, x 7→(1 if x = (1, . . . , 1), 0 else. We will show that he1iB ⊂ he2iB ⊂ . . . is an infinite ascending chain of clonoids with target B. The idea for this example was used by Bulatov in [4] to construct countably many expansions of (Z4, +). It is enough to show that k−1 Xi=1 (3.2) ek 6= aieσi i for any ai ∈ {0, 1} and σi : [i] → [k]. For any i < k and σi : [i] → [k], let the support of eσi i (x) = 1}. Note that the support of eσi eσi i be {x ∈ {0, 1}k : i has even size for any i < k ON THE NUMBER OF CLONOIDS 7 and σi : [i] → [k]. Hence Pk−1 i=1 aieσi ai, σi. Since the support of ek is odd, (3.2) follows immediately. i has support of even size for all (cid:3) 4. Without cube term In this section, we will show that there are continuum many clonoids with a finite source and finite idempotent target algebra without a cube term. Additionally, we will show there are continuum many clonoids with a finite source and Boolean target algebra without a cube term, or equivalently without an NU-term or Mal'cev term. Let A = {0, 1, . . . , d} and B = {0, 1, . . . , e} for d, e ≥ 1. Define the following n-ary relations on A and B, respectively, for all n ∈ N: Pn := {(1, 0, . . . , 0), (0, 1, 0, . . . , 0), . . . , (0, . . . , 0, 1)} ⊆ An, Qn := {0, 1}n \ {(1, . . . , 1)} ⊆ Bn. For U ⊆ N, let RU := {(Pn, Qn) : n ∈ U}. Note that 0 preserves RU for any U ⊆ N. Define the following k-ary functions for all k ∈ N: if x ∈ Pk, otherwise. fk : Ak → {0, 1}, x 7→(cid:26) 1 0 For U ⊆ N, let FU := {fk : k ∈ U}. We show some connections between these functions and relations that we need later. Lemma 4.1. (1) Let k, n ∈ N. Then fk preserves (Pn, Qn) iff k 6= n. (2) hFU i ⊆ Pol(RU ) for each U ⊆ N where U is the complement of U. Proof. For (1), we see that fk does not preserve (Pk, Qk) since 1 0 0 · · · · · · 1 ... 0 0 0 ... ... 0 1 ∈ ∈ · · · ∈ · · · Pk Pk Pk · · · fk−→ 1 fk−→ 1 ... ... fk−→ 1 ∈ Qk. Next assume n 6= k and x1, . . . , xk ∈ Pn. Let M be the n × k matrix where the jth column is xj for 1 ≤ j ≤ k. If n < k, then at least one row of M must have at least two entries equal to 1. Thus at least one entry of fk(M), the n-tuple obtained by applying fk to the rows of M, 6 8 ATHENA SPARKS is 0. Hence fk(M) is in Qn. If n > k, then at least one row of M is all zeros. So at least one entry of fk(M) is 0 and fk(M) is in Qn. This concludes the proof of (1). Item (2) is immediate from (1). (cid:3) Lemma 4.2. Let A be a finite set, B a finite idempotent algebra without a cube term, and A, B > 1. Then the number of clonoids from A to B is continuum. Proof. By [7, Theorem 2.1] B must have cube term blocker. That is, there exists a nonempty proper subset V of B such that Tn := Bn \ (B \ V )n is a subuniverse of B for all n. Without loss of generality, assume 0 ∈ V and 1 ∈ B \ V . Thus Qn ⊆ Tn. The statement is immediate from the following claim: (4.1) hFU iB ∩ FN = FU for each U ⊆ N. The inclusion ⊇ is clear. To prove the converse, let U ⊆ N \ {1} and n ∈ N such that fn ∈ hFU iB. Then fn = ϕ(f σ1 km ) for some m-ary ϕ in the clone of B, k1, . . . , km ∈ U, and maps σi : [ki] → [n] for 1 ≤ i ≤ m. If n = ki for some i, then fn ∈ FU . k1 , . . . , f σm Assume toward a contradiction that n 6= ki for any i. Let a1, . . . , an enumerate Pn. By Lemma 4.1, for each 1 ≤ i ≤ m. Therefore we have f σi ki (a1, . . . , an) =: bi ∈ Qn fn(a1, . . . , an) = ϕ(f σ1 km )(a1, . . . , an) k1 , . . . , f σm = ϕ(b1, . . . , bm) ∈ Tn since b1, . . . , bm ∈ Tn and ϕ preserves Tn. However fn(a1, . . . , an) = (1, . . . , 1) 6∈ Tn. This contradiction completes the proof of (4.1). (cid:3) Theorem 1.4 follows immediately from Theorem 3.1 and Lemma 4.2. Next we show that Lemma 4.2 generalizes to nonidempotent Boolean algebras. By Post's classification of Boolean clones, each clone on {0, 1} with- out an NU-term or a Mal'cev term is contained in a nonidempotent clone generated by one of the following sets of operations: (1) {∧, 0, 1} or {∨, 0, 1}, (2) {¬, 0}, (3) {→} or {6→}. ON THE NUMBER OF CLONOIDS 9 Thus there are 3 cases up to duality. We will show that for each case there are continuum many clonoids with source A and corresponding target algebra B by variations of the proof of Lemma 4.2. From this it follows that for algebras with smaller clone of term operations (e.g., the set ({0, 1}, ∅)), there are continuum many clonoids as well. Note that the maximal clones without a cube term on sets of size at least 3 are not explicitly know. Hence we do not know whether Lemma 4.2 generalizes to arbitrary nonidempotent algebras. We begin proving the 3 cases with the case where B = ({0, 1}, ∧, 0, 1). Lemma 4.3. The number of clonoids with finite source A and target algebra B = ({0, 1}, ∧, 0, 1) is continuum. Proof. Let B = ({0, 1}, ∧, 0, 1) and B′ = ({0, 1}, ∧). Note that the clone of B is the clone of B′ with the addition of the constant maps 0, 1. Hence for any subset U ⊆ N, we have hFU iB = hFU iB′ ∪ {0, 1}. By Lemma 4.2 there are continuum many Boolean clonoids of the form hFU iB′. (cid:3) Now we prove the case where B = ({0, 1}, ¬, 0). Lemma 4.4. The number of clonoids with finite source A and target algebra B = ({0, 1}, ¬, 0) is continuum. Proof. As in the proof of Lemma 4.2, the statement is immediate from the following claim: (4.2) hFU iB ∩ FN = FU for each U ⊆ N \ {1}. The inclusion ⊇ is clear. To prove the converse, let U ⊆ N \ {1} and ℓ ∈ N such that fℓ ∈ hFU iB. Then fℓ = f σ k ) for some k ∈ U and map σ : [k] → [ℓ]. In the former case, Lemma 4.1 yields ℓ = k and further ℓ ∈ U. To see that the latter cannot occur, let m ∈ N, m 6= ℓ, and let a = (1, 0, . . . , 0) ∈ Pm. We have k or fℓ = ¬(f σ fℓ(a, . . . , a) = ¬(f σ k )(a, . . . , a ) = ¬fk(a, . . . , a ) ℓ {z } {z } {z }m 6∈ Qm. k ) since k ∈ U, so k > 1 = (1, . . . , 1 Thus fℓ does not preserve (Pm, Qm). This contradicts Lemma 4.1 and completes the proof of (4.2). (cid:3) 10 ATHENA SPARKS The final case, where B = ({0, 1}, 6→), is given in the following lemma. Lemma 4.5. The number of clonoids with finite source A and target algebra B = ({0, 1}, 6→) is continuum. Proof. First we show hFU iB ⊆ Pol(RU ) for each U ⊆ N. (4.3) By Lemma 4.1, hFU i ⊆ Pol(RU ). Assume g, h ∈ hFU iB of arity k preserve (Pn, Qn) and let a1, . . . , ak ∈ Pn. Let d := g 6→ h. Then we have d(a1, . . . , ak) = g(a1, . . . , ak) ∧ (¬h(a1, . . . , ak)). Since g preserves (Pn, Qn), there must be at least one zero entry in g(a1, . . . , ak). Thus d(a1, . . . , ak) ∈ Qn. Hence (4.3) is proved. Let U, V ⊆ N such that U 6= V . We claim that (4.4) hFU iB 6= hFV iB. Without loss of generality, assume there exists n ∈ U \ V . From (4.3), we have that hFV iB preserves (Pn, Qn). Since n ∈ U, we have fn ∈ FU and thus hFU iB does not preserve (Pn, Qn) by Lemma 4.1. Therefore hFU iB 6= hFV iB. (cid:3) 5. Proof of Main Theorem In this section, we combine the results from the previous sections to give a proof of Theorem 1.3. Proof of Theorem 1.3. The reverse direction of (1) follows immediately from Theorem 2.1. To prove the reverse direction of (2), assume B has a Mal'cev term but no majority term. Then by Theorem 3.1, CA,B is at most count- ably infinite. Since B has no majority term, by Post's classification, the clone of B is contained in the clone of B′ := ({0, 1}, +, 0, 1). In Lemma 3.2 we show that there are infinitely many clonoids in CA,B′. Since CA,B′ ⊆ CA,B, there are countably many clonoids in CA,B. Now assume B has neither an NU-term nor a Mal'cev term. As mentioned in the beginning of Section 4, it follows from Lemmas 4.3, 4.4, and 4.5 and their duals that there are continuum many clonoids with target B. This proves the reverse direction of (3). The forward directions of (1), (2), and (3) follow because the cases (cid:3) are mutually exclusive and cover all possibilities. ON THE NUMBER OF CLONOIDS 11 Acknowledgments The author thanks Peter Mayr for discussions on the material in this paper and the anonymous referee for their comments and asking a question that led to Theorem 1.4. References [1] Erhard Aichinger and Peter Mayr. Finitely generated equational classes. J. Pure Appl. Algebra, 220(8):2816 -- 2827, 2016. [2] Kirby A. Baker and Alden F. Pixley. Polynomial interpolation and the Chinese remainder theorem for algebraic systems. Math. Z., 143(2):165 -- 174, 1975. [3] Joshua Brakensiek and Venkatesan Guruswami. Promise constraint satisfaction structure theory and a symmetric Boolean dichotomy. In Proceedings of the Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 1782 -- 1801. SIAM, Philadelphia, PA, 2018. [4] Andrei Bulatov. On the number of finite Mal'tsev algebras. In Contributions to general algebra, 13 (Velk´e Karlovice, 1999/Dresden, 2000), pages 41 -- 54. Heyn, Klagenfurt, 2001. [5] Alfred L. Foster. The identities of -- and unique subdirect factorization within -- classes of universal algebras. Math. Z., 62:171 -- 188, 1955. [6] Dietlinde Lau. Function algebras on finite sets. Springer Monographs in Mathe- matics. Springer-Verlag, Berlin, 2006. A basic course on many-valued logic and clone theory. [7] Petar Markovi´c, Mikl´os Mar´oti, and Ralph McKenzie. Finitely related clones and algebras with cube terms. Order, 29(2):345 -- 359, 2012. [8] Nicholas Pippenger. Galois theory for minors of finite functions. Discrete Math., 254(1-3):405 -- 419, 2002. Department of Mathematics, University of Colorado Boulder, USA E-mail address: [email protected]
1107.5922	1	1107	2011-07-29T10:28:49	Singular equivalences induced by homological epimorphisms	[ "math.RA", "math.RT" ]	We prove that a certain homological epimorphism between two algebras induces a triangle equivalence between their singularity categories. Applying the result to a construction of matrix algebras, we describe the singularity categories of some non-Gorenstein algebras.	math.RA	math	SINGULAR EQUIVALENCES INDUCED BY HOMOLOGICAL EPIMORPHISMS XIAO-WU CHEN Abstract. We prove that a certain homological epimorphism between two algebras induces a triangle equivalence between their singularity categories. Applying the result to a construction of matrix algebras, we describe the sin- gularity categories of some non-Gorenstein algebras. 1. Introduction Let A be a finite dimensional algebra over a field k. Denote by A-mod the category of finitely generated left A-modules, and by Db(A-mod) the bounded derived category. Following [17], the singularity category Dsg(A) of A is the Verdier quotient triangulated category of Db(A-mod) with respect to the full subcategory formed by perfect complexes; see also [4, 13, 11, 20, 2, 14] and [6]. The singularity category measures the homological singularity of an algebra: the algebra A has finite global dimension if and only if its singularity category Dsg(A) is trivial. Meanwhile, the singularity category captures the stable homological features of an algebra ([4]). A fundamental result of Buchweitz and Happel states that for a Gorenstein alge- bra A, the singularity category Dsg(A) is triangle equivalent to the stable category of (maximal) Cohen-Macaulay A-modules ([4] and [11]). This result specializes to Rickard's result ([20]) on self-injective algebras. For non-Gorenstein algebras, not much is known about their singularity categories ([5, 7]). The following concepts might be useful in the study of singularity categories. Two algebras A and B are said to be singularly equivalent provided that there is a triangle equivalence between Dsg(A) and Dsg(B). Such an equivalence is called a singular equivalence; compare [18]. In this case, if A is non-Gorenstein and B is Gorenstein, then Buchweitz-Happel's theorem applies to give a description of Dsg(A) in terms of Cohen-Macaulay modules over B. We observe that a derived equivalence of two algebras, that is, a triangle equivalence between their bounded derived categories, induces naturally a singular equivalence. The converse is not true in general. Let A be an algebra and let J ⊆ A be a two-sided ideal. Following [19], we call J a homological ideal provided that the canonical map A → A/J is a homological epimorphism ([9]), meaning that the naturally induced functor Db(A/J-mod) → Db(A-mod) is fully faithful. The main observation we make is as follows. Date: March 12, 2018. 1991 Mathematics Subject Classification. 18E30, 13E10, 16E50. Key words and phrases. singularity category, triangle equivalence, homological epimorphism, non-Gorenstein algebra. The author is supported by Special Foundation of President of The Chinese Academy of Sci- ences (No.1731112304061) and National Natural Science Foundation of China (No.10971206). E-mail: [email protected]. 1 2 XIAO-WU CHEN Theorem. Let A be a finite dimensional k-algebra and let J ⊆ A be a homological ideal which has finite projective dimension as an A-A-bimodule. Then there is a singular equivalence between A and A/J . The paper is structured as follows. In Section 2, we recall some ingredients and then prove Theorem. In Section 3, we apply Theorem to a construction of matrix algebras, and then describe the singularity categories of some non-Gorenstein al- gebras. In particular, we give two examples, which extend in different manners an example considered by Happel in [11]. 2. Proof of Theorem We will present the proof of Theorem in this section. Before that, we recall from [22] and [12] some known results on triangulated categories and derived categories. Let T be a triangulated category. We will denote its translation functor by [1]. For a triangulated subcategory N , we denote by T /N the Verdier quotient triangulated category. The quotient functor q : T → T /N has the property that q(X) ≃ 0 if and only if X is a direct summand of an object in N . In particular, if N is a thick subcategory, that is, it is closed under direct summands, we have that Ker q = N . Here, for a triangle functor F , Ker F denotes the (thick) triangulated subcategory consisting of objects on which F vanishes. The following result is well known. Lemma 2.1. Let F : T → T ′ be a triangle functor which allows a fully faithful right adjoint G. Then F induces uniquely a triangle equivalence T /Ker F ≃ T ′. Proof. The existence of the induced functor follows from the universal property of the quotient functor. The result is a triangulated version of [8, Proposition I. 1.3]. For details, see [3, Propositions 1.5 and 1.6]. (cid:3) Let F : T → T ′ be a triangle functor. Assume that N ⊆ T and N ′ ⊆ T ′ are triangulated subcategories satisfying F N ⊆ N ′. Then there is a uniquely induced triangle functor ¯F : T /N → T ′/N ′. Lemma 2.2. ([17, Lemma 1.2]) Let F : T → T ′ be a triangle functor which has a right adjoint G. Assume that N ⊆ T and N ′ ⊆ T ′ are triangulated subcategories satisfying that F N ⊆ N ′ and GN ′ ⊆ N . Then the induced functor ¯F : T /N → T ′/N ′ has a right adjoint ¯G. Moreover, if G is fully faithful, so is ¯G. Proof. The unit and counit of (F, G) induce uniquely two natural transformations IdT /N → ¯G ¯F and ¯F ¯G → IdT ′/N ′ , which are the corresponding unit and counit of the adjoint pair ( ¯F , ¯G); consult [16, Chapter IV, Section 1, Theorem 2(v)]. Note that the fully-faithfulness of G is equivalent to that the counit of (F, G) is an isomorphism. It follows that the counit of ( ¯F , ¯G) is also an isomorphism, which is equivalent to the fully-faithfulness of ¯G; consult [16, Chapter IV, Section 3, Theorem 1]. (cid:3) Let k be a field and let A be a finite dimensional k-algebra. Recall that A-mod is the category of finite dimensional left A-modules. We write AA for the regular left A-module. Denote by D(A-mod) (resp. Db(A-mod)) the (resp. bounded) derived category of A-mod. We identify A-mod as the full subcategory of Db(A-mod) consisting of stalk complex concentrated at degree zero; see [12, Proposition I. 4.3]. A complex of A-modules is usually denoted by X • = (X n, dn)n∈Z, where X n are A-modules and the differentials dn : X n → X n+1 are homomorphisms of modules satisfying dn+1 ◦ dn = 0. Recall that a complex in Db(A-mod) is perfect provided that it is isomorphic to a bounded complex consisting of projective modules. The full subcategory consisting of perfect complexes is denoted by perf(A). Recall from SINGULAR EQUIVALENCES INDUCED BY HOMOLOGICAL EPIMORPHISMS 3 [4, Lemma 1.2.1] that a complex X • in Db(A-mod) is perfect if and only if there is a natural number n0 such that for each A-module M , HomDb(A-mod)(X •, M [n]) = 0 for all n ≥ n0. It follows that perf(A) is a thick subcategory of Db(A-mod). Indeed, it is the smallest thick subcategory of Db(A-mod) containing AA. Let π : A → B be a homomorphism of algebras. The functor of restricting of scalars π∗ : B-mod → A-mod is exact, and it extends to a triangle functor Db(B-mod) → Db(A-mod), which will still be denoted by π∗. Following [9], we call the homomorphism π a homological epimorphism provided that π∗ : Db(B-mod) → Db(A-mod) is fully faithful. By [9, Theorem 4.1(1)] this is equivalent to that A B is an isomorphism in D(Ae-mod). Here, π ⊗L Ae = A ⊗k Aop is the enveloping algebra of A, and we identify Ae-mod as the category of A-A-bimodules. A B : B ≃ A ⊗L A B → B ⊗L Lemma 2.3. ([19, Proposition 2.2(a)]) Let J ⊆ A be an ideal and let π : A → A/J be the canonical projection. Then π is a homological epimorphism if and only if J 2 = J and TorA i (J, A/J) = 0 for all i ≥ 1. In the situation of the lemma, the ideal J is called a homological ideal in [19]. As a special case, we call an ideal J a hereditary ideal provided that J 2 = J and J is a projective A-A-bimodule; compare [19, Lemma 3.4]. A A/J, we get a triangle J ⊗L Proof. The natural exact sequence 0 → J → A → A/J → 0 of A-A-bimodules induces a triangle J → A → A/J → J[1] in Db(Ae-mod). Applying the functor − ⊗L A A/J[1]. Then π is a homological epimorphism, or equivalently π ⊗A A/J is an isomorphism if and only if J ⊗L A A/J = 0; see [10, Lemma I.1.7]. This is equivalent to that TorA 0 (J, A/J) ≃ J⊗AA/J ≃ J/J 2. (cid:3) i (J, A/J) = 0 for all i ≥ 0. We note that TorA A A/J → A/J → A/J ⊗L A A/J → J ⊗L Now we are in the position to prove Theorem. Recall that for an algebra A, its singularity category Dsg(A) = Db(A-mod)/perf(A). Moreover, a complex X • becomes zero in Dsg(A) if and only if it is perfect. Here, we use the fact that perf(A) ⊆ Db(A-mod) is a thick subcategory. Proof of Theorem. Write B = A/J. Since J, as an A-A-bimodule, has finite projective dimension, so it has finite projective dimension both as a left and right A- module. Consider the natural exact sequence 0 → J → A → B → 0. It follows that B, both as a left and right A-module, has finite projective dimension. Moreover, for a complex X • in Db(A-mod), J ⊗L Indeed, take a bounded projective resolution P • → J as an Ae-module. Then J ⊗L A X • ≃ P • ⊗A X •. This is a perfect complex, since each left A-module P i ⊗A X j is projective. A X • is perfect. Denote by π : A → B be the canonical projection. By the assumption, the functor π∗ : Db(B-mod) → Db(A-mod) is fully faithful. Since π∗(B) is perfect, the functor π∗ sends perfect complexes to perfect complexes. Then it induces a triangle functor ¯π∗ : Dsg(B) → Dsg(A). We will show that ¯π∗ is an equivalence. The functor π∗ : Db(B-mod) → Db(A-mod) has a left adjoint F = B ⊗L A − : Db(A-mod) → Db(B-mod). Here we use the fact that the right A-module B has finite projective dimension. Since F sends perfect complexes to perfect com- plexes, we have the induced triangle functor ¯F : Dsg(A) → Dsg(B). By Lemma 2.2 we have the adjoint pair ( ¯F , ¯π∗); moreover, the functor ¯π∗ is fully faithful. By Lemma 2.1 there is a triangle equivalence Dsg(A)/Ker ¯F ≃ Dsg(B). It remains to show that the essential kernel Ker ¯F is trivial. For this, assume that a complex X • lies in Ker ¯F . This means that the complex F (X •) in Db(B-mod) is perfect. Since π∗ preserves perfect complexes, it follows that π∗F (X •) is also perfect. The natural exact sequence 0 → J → A → B → 0 induces a triangle 4 XIAO-WU CHEN A X • → X • → π∗F (X •) → J ⊗L A X • is J ⊗L perfect. It follows that X • is perfect, since perf(A) ⊆ Db(A-mod) is a triangulated subcategory. The proves that X • is zero in Dsg(A). (cid:3) A X •[1] in Db(A-mod). Recall that J ⊗L The following special case of Theorem is of interest. Corollary 2.4. Let A be a finite dimensional algebra and J ⊆ A a hereditary ideal. Then we have a triangle equivalence Dsg(A) ≃ Dsg(A/J). Proof. It suffices to observe by Lemma 2.3 that J is a homological ideal. (cid:3) 3. Examples In this section, we will describe a construction of matrix algebras to illustrate Corollary 2.4. In particular, the singularity categories of sone non-Gorenstein alge- bras are studied. The following construction is similar to [15, Section 4]. Let A be a finite di- mensional algebra over a field k. Let AM and NA be a left and right A-module, respectively. Then M ⊗k N becomes an A-A-bimodule. Consider an A-A-bimodule monomorphism φ : M ⊗k N → A such that Im φ vanishes both on M and N . Here, we observe that Im φ ⊆ A is an ideal. The matrix Γ = (cid:18)A M N k (cid:19) becomes an associative algebra via the following multiplication n λ(cid:19) (cid:18)a′ m′ (cid:18)a m n′ λ′(cid:19) = (cid:18)aa′ + φ(m ⊗ n) am′ + λ′m na′ + λn′ λλ′ (cid:19) . Proposition 3.1. Keep the notation and assumption as above. Then there is a triangle equivalence Dsg(Γ) ≃ Dsg(A/Im φ). Proof. Set J = ΓeΓ with e = (cid:18)0 1(cid:19). Observe that Γ/J = A/Im φ. The ideal J is hereditary: J 2 = J is clear, while the natural map Γe ⊗k eΓ → J is an isomorphism of Γ-Γ-bimodules and then J is a projective Γ-Γ-bimodule. The isomorphism uses that φ is mono. Then we apply Corollary 2.4. (cid:3) 0 0 Remark 3.2. The above construction contains the one-point extension and co- extension of algebras, where M or N is zero. Hence Proposition 3.1 contains the results in [7, Section 4]. We will illustrate Proposition 3.1 by three examples. Two of these examples extend an example considered by Happel in [11]. In particular, based on results in [7], we obtain descriptions of the singularity categories of some non-Gorenstein algebras. Recall from [11] that an algebra A is Gorenstein provided that both as a left and right module, the regular module A has finite injective dimension. It follows from [4, Theorem 4.4] and [11, Theorem 4.6] that in the Gorenstein case, the singularity category Dsg(A) is Hom-finite. This means that all Hom spaces in Dsg(A) are finite dimensional over k. For algebras given by quivers and relations, we refer to [1, Chapter III]. Example 3.3. Let Γ be the k-algebra given by the following quiver Q with relations {δx, βx, xγ, xα, βγ, δα, βα, δγ, αβ − γδ}. We write the concatenation of paths from the left to the right. x ·1 α β * ∗ δ γ ·2 We have in Γ that 1 = e1 + e∗ + e2, where the e's are the primitive idempotents corresponding to the vertices. Set Γ′ = Γ/Γe1Γ. It is an algebra with radical square * j j * * j j SINGULAR EQUIVALENCES INDUCED BY HOMOLOGICAL EPIMORPHISMS 5 zero, whose quiver is obtained from Q by removing the vertex 1 and the adjacent arrows. We identify Γ with (cid:18) A kα k (cid:19), where the k in the southeast corner is identified kβ with e1Γe1, and A = (1 − e1)Γ(1 − e1). The corresponding Im φ equals kαβ, and we have A/Im φ = Γ′; consult the proof of Proposition 3.1. Then Proposition 3.1 yields a triangle equivalence Dsg(Γ) ≃ Dsg(Γ′). The triangulated category Dsg(Γ′) is completely described in [7]; also see [21]; in particular, it is not Hom-finite. More precisely, it is equivalent to the category of finitely generated projective modules on a von Neumann regular algebra. The algebra Γ′, or rather its Koszul dual, is related to the noncommutative space of Penrose tiling via the work of Smith; see [21, Theorem 7.2 and Example]. We point out that the algebra Γ is non-Gorenstein, since Dsg(Γ) is not Hom-finite. Example 3.4. Let Γ be the k-algebra given by the following quiver Q with relations {x2 2, x1α1, x2α1, β2α1, β2α1, x1α2, x2α2, β1α2, β2α2, α1β1 − x1x2, α2β2 − x2x1}. 1, x2 ·1 α1 β1 x1 * ∗ x2 β2 α2 ·2 We claim that there is a triangle equivalence Dsg(Γ) ≃ Dsg(khx1, x2i/(x1, x2)2). Here, khx1, x2i is the free algebra with two variables. We point out that the triangulated category Dsg(khx1, x2i/(x1, x2)2) is described completely in [7, Example 3.11]. Similar as in the example above, this algebra Γ is non-Gorenstein. To see the claim, we observe that the quiver Q has two loops and two 2-cycles. The proof is done by "removing the 2-cycles". We have a natural isomorphism kβ1 Γ = (cid:18) A kα1 k (cid:19), where k = e1Γe1 and A = (1 − e1)Γ(1 − e1). We observe that Proposition 3.1 applies with the corresponding Im φ = kα1β1. Set A/Im φ = Γ′. So Dsg(Γ) ≃ Dsg(Γ′). The quiver of Γ′ is obtained from Q by removing the vertex 1 and the adjacent arrows, while its relations are obtained from the ones of Γ by k (cid:19) with k = e2Γ′e2 and A′ = e∗Γ′e∗. Then Proposition 3.1 applies and we get the equivalence Dsg(Γ′) ≃ Dsg(khx1, x2i/(x1, x2)2). replacing α1β1 − x1x2 with x1x2. Similarly, Γ′ = (cid:18) A′ kα2 kβ2 This example generalizes directly to a quiver with n loops and n 2-cycles with similar relations. The corresponding statement for the case n = 1 is implicitly contained in [11, 2.3 and 4.8]. The last example is a Gorenstein algebra. Example 3.5. Let r ≥ 2. Consider the following quiver Q consisting of three 2-cycles and a central 3-cycle Z3. We identify γ3 with γ0, and denote by pi the path in the central cycle starting at vertex i of length 3. ·1′ α1 β1 ·1 / ·3 β3 ) α3 ) ·3′ γ1 γ2 ·2′ α2 ( β2 ( ·2 _??? γ3??? * j j * * Z Z j j H H i i / _ h h 6 XIAO-WU CHEN Let Γ be the k-algebra given by the quiver Q with relations {βiαi, γiαi, βiγi−1, βiαi− pr i = 1, 2, 3}. We point out that in Γ all paths in the central cycle of length i strictly larger than 3r + 1 vanish. Set A = kZ3/(γ1, γ2, γ3)3r, where kZ3 is the path algebra of the central 3-cycle Z3. The algebra A is self-injective and Nakayama ([1, p.111]). Denote by A-mod the stable category of A-modules; it is naturally a triangulated category; see [10, Theorem I.2.6]. We claim that there is a triangle equivalence Dsg(Γ) ≃ A-mod. For the claim, we observe an isomorphism A = Γ/Γ(e1′ + e2′ + e3′)Γ. We argue as in Example 3.4 by removing the three 2-cycles and applying Proposition 3.1 repeatedly. Then we get a triangle equivalence Dsg(Γ) ≃ Dsg(A). Finally, by [20, Theorem 2.1] we have a triangle equivalence Dsg(A) ≃ A-mod. Then we are done. We observe that the algebra Γ is Gorenstein with self injective dimension two. Moreover, this example generalizes directly to a quiver with n 2-cycles and a central n-cycle with similar relations. The case where n = 1 and r = 2 coincides with the example considered in [11, 2.3 and 4.8]. Acknowledgements. The results of this paper answer partially a question, which was raised by Professor Changchang Xi during a conference held in Jinan, June 2011. The author thanks Huanhuan Li for helpful discussion. References [1] M. Auslander, I. Reiten, and S.O. Smalø, Representation Theory of Artin Algebras, Cambridge Studies in Adv. Math. 36, Cambridge Univ. Press, Cambridge, 1995. [2] A. Beligiannis, The homological theory of contravariantly finite subcategories: Auslander- Buchweitz contexts, Gorenstein categories and (co-)stabilization, Comm. Algebra 28(10) (2000), 4547–4596. [3] A.I. Bondal and M.M. Kapranov, Representable functors, Serre functors, and reconstruc- tions, Izv. Akad. Nauk SSSR Ser. Mat. 53 (1989), 1183–1205. [4] R.O. Buchweitz, Maximal Cohen-Macaulay Modules and Tate Cohomology over Goren- stein Rings, Unpublished Manuscript, 1987. [5] X.W. Chen, Singularity categories, Schur functors and triangular matrix rings, Algebr. Represent. Theor. 12 (2009), 181–191. [6] X.W. Chen, Relative singularity categories and Gorenstein-projective modules, Math. Nathr. 284 no.2-3 (2011), 199–212. [7] X.W. Chen, The singularity category of an algebra with radical square zero, arXiv:1104.4006v1. [8] P. Gabriel and M. Zisman, Calculus of Fractions and Homotopy Theory, Springer-Verlag New York, Inc., New York, 1967. [9] W. Geigle and H. Lenzing, Perpendicular categories with applications to representations and sheaves, J. Algebra 144 (1991), no. 2, 273–343. [10] D. Happel, Triangulated Categories in the Representation Theory of Finite Dimensional Algebras, London Math. Soc., Lecture Notes Ser. 119, Cambridge Univ. Press, Cambridge, 1988. [11] D. Happel, On Gorenstein algebras, In: Progress in Math. 95, Birkhauser Verlag, Basel, 1991, 389–404. [12] R. Hartshorne, Duality and Residue, Lecture Notes in Math. 20, Springer, Berlin, 1966. [13] B. Keller and D. Vossieck, Sous les cat´egories d´eriv´ees, C.R. Acad. Sci. Paris, t. 305 S´erie I (1987) 225–228. [14] H. Krause, The stable derived category of a noetherian scheme, Compositio Math. 141 (2005), 1128–1162. [15] S. Koenig and H. Nagase, Hochschild cohomology and stratifying ideals, J. Pure Appl. Algebra 213 (2009), 886–891. [16] S. Mac Lane, Categories for the Working Mathematicians, Grad. Text in Math. 5, Springer, 1998. [17] D. Orlov, Triangulated categories of singularities and D-branes in Landau-Ginzburg mod- els, Trudy Steklov Math. Institute 204 (2004), 240–262. SINGULAR EQUIVALENCES INDUCED BY HOMOLOGICAL EPIMORPHISMS 7 [18] D. Orlov, Triangulated categories of singularities and equivalences between Landau- Ginzburg models, Matem. Sbornik 197 (2006), 1827–1840. [19] J.A. de la Pena and C.C. Xi, Hochschild cohomology of algebras with homological ideals, Tsukuba J. Math. 30 (1) (2006), 61–79. [20] J. Rickard, Derived categories and stable equivalence, J. Pure Appl. Algebra 61 (1989), 303–317. [21] S.P. Smith, Equivalence of categories involving graded modules over path algebras of quivers, arXiv:1107.3511. [22] J.L. Verdier, Categories derive´ees, in SGA 4 1/2, Lecture Notes in Math. 569, Springer, Berlin, 1977. Xiao-Wu Chen, School of Mathematical Sciences, University of Science and Technology of China, Hefei 230026, Anhui, PR China Wu Wen-Tsun Key Laboratory of Mathematics, USTC, Chinese Academy of Sciences, Hefei 230026, Anhui, PR China. URL: http://mail.ustc.edu.cn/∼xwchen
1110.3016	3	1110	2012-06-17T00:43:22	Closure of the cone of sums of 2d-powers in real topological algebras	[ "math.RA" ]	Let $R$ be a unitary commutative real algebra and $K\subseteq Hom(R,\mathbb{R})$, closed with respect to the product topology. We consider $R$ endowed with the topology $\mathcal{T}_K$, induced by the family of seminorms $\rho_{\alpha}(a):=\|\alpha(a)\|$, for $\alpha\in K$ and $a\in R$. In case $K$ is compact, we also consider the topology induced by $\\|a\\|_K:=\sup_{\alpha\in K}\|\alpha(a)\|$ for $a\in R$. If $K$ is Zariski dense, then those topologies are Hausdorff. In this paper we prove that the closure of the cone of sums of 2d-powers, $\sum R^{2d}$, with respect to those two topologies is equal to $Psd(K):=\{a\in R:\alpha(a)\geq 0,\textrm{for all}\alpha\in K\}$. In particular, any continuous linear functional $L$ on the polynomial ring $R=\mathbb{R}[X_1,...,X_n]$ with $L(h^{2d})\ge0$ for each $h\in R$ is integration with respect to a positive Borel measure supported on $K$. Finally we give necessary and sufficient conditions to ensure the continuity of a linear functional with respect to those two topologies.	math.RA	math	CLOSURE OF THE CONE OF SUMS OF 2d-POWERS IN REAL TOPOLOGICAL ALGEBRAS MEHDI GHASEMI1, SALMA KUHLMANN2 Abstract. Let R be a unitary commutative R-algebra and K ⊆ XR ∶= Hom(R, R), closed with respect to the product topology. We consider R endowed with the topology TK , induced by the family of seminorms ρα(a) ∶= α(a), for α ∈ K and a ∈ R. In case K is compact, we also consider the topology induced by aK ∶= supα∈K α(a) for a ∈ R. If K is Zariski dense, then those topologies are Hausdorff. In this paper we prove that the closure of the cone of sums of 2d-powers, ∑ R2d, with respect to those two topologies is equal to Psd(K) ∶= {a ∈ R ∶ α(a) ≥ 0, for all α ∈ K}. In particular, any continuous linear functional L on the polynomial ring R = R[X] ∶= R[X1, . . . , Xn] with L(h2d) ≥ 0 for each h ∈ R[X] is integration with respect to a positive Borel measure supported on K. Finally we give necessary and sufficient conditions to ensure the continuity of a linear functional with respect to those two topologies. 1. Introduction The (real) multidimensional K-moment problem for a given closed set K ⊆ Rn, is the question of when a real valued linear functional L, defined on the real algebra of polynomials R[X], is representable as integration with respect to a positive Borel measure on K. A subset C of R[X] is called a cone, if C + C ⊆ C and R+C ⊆ C, where R+ denotes the non-negative real numbers. Let us denote the cone of non-negative polynomials on K by Psd(K). If L is representable by a measure then clearly, for any polynomial f ∈ Psd(K), L(f) ≥ 0 (i.e. L(Psd(K)) ⊆ R+). Haviland [10, 11], proved that this necessary condition is also sufficient. However, Psd(K) is seldom finitely generated [19, Proposition 6.1]. So in general, there is no practical Date: March 23, 2012. 2010 Mathematics Subject Classification. Primary 13J30, 14P10, 44A60; Secondary 12D15, 43A35, 46B99. Key words and phrases. Positive polynomials, sums of squares, cone of sums of 2d- powers, semialgebraic sets, locally convex topologies, positive semidefinite continuous lin- ear functionals, moment problem. 1 2 M. GHASEMI, S. KUHLMANN decision procedure for the membership problem for Psd(K), and a fortiori for L(Psd(K)) ⊆ R+. We are mainly interested in the solutions of (1) Psd(K) ⊆ C τ , where C is a cone, K is a closed subset of Rn and C denotes the closure of C with respect to a locally convex topology τ on R[X]. It is proved in [8, Proposition 3.1] that (1) holds if and only if for every τ -continuous linear functional L, nonnegative on C, there exists a positive Borel measure µ supported on K, such that τ L(f) = SK f dµ, ∀f ∈ R[X]. Clearly, if the functional L is representable by a measure, then L has to be positive semidefinite, i.e., L(p2) ≥ 0 for all p ∈ R[X]. So in (1), the simplest cone to consider is C = ∑ R[X]2. If τ is the finest locally con- vex topology ϕ on R[X], then any given linear functional is ϕ-continuous In this setting, Schmudgen in [20, Theorem 3.1] and Berg, Chris- [4]. ϕ = ∑ R[X]2. tensen and Jensen in [2, Theorem 3] prove that ∑ R[X]2 ing C = ∑ R[X]2 and τ to be the ℓ1-norm topology, then K = [−1, 1]n solves ⋅1 = Psd([−1, 1]n). This was further generalized in [4] (1), i.e., ∑ R[X]2 Later, in [3, Theorem 9.1] Berg, Christensen and Ressel prove that tak- ⋅1 = Psd([−1, 1]n). denote the smallest preordering, containing S by TS. A preordering T is said and [5] to include commutative semigroup-rings and topologies induced by absolute values. These results has been revisited in [16] with a different approach, and were recently generalized in [8] to weighted ℓp-norms, p ≥ 1. In [9] it is shown that the general result in [5] carries to the even smaller cone of sums of 2d-powers, ∑ R[X]2d ⊆ ∑ R[X]2, where d ≥ 1 is an integer. In particular, ∑ R[X]2d We now discuss (1) for other special cones. A set T ⊆ R[X] is called a preordering, if T + T ⊆ T , T ⋅ T ⊆ T and ∑ R[X]2 ⊆ T . For S ⊆ R[X], we to be finitely generated, if T = TS for some finite set S ⊆ R[X]. The smallest preordering of R[X] is ∑ R[X]2, as considered above. A subset M ⊆ R[X] is a ∑ R[X]2d-module if 1 ∈ M , M + M ⊆ M and ∑ R[X]2d ⋅ M ⊆ M . If d = 1, M = MS for some finite set S ⊆ R[X], and Archimedean if for every f ∈ R[X] there exists n ∈ N such that n ± f ∈ M . The non-negativity set of a subset S ⊂ R[X] will be denoted by KS, and is defined by KS ∶= {x ∈ Rn ∶ ∀f ∈ S f(x) ≥ 0}. If S is finite, KS is called a basic closed semialgebraic set. M is said to be a quadratic module. M is said to be finitely generated, if CLOSURE OF ∑ R2d IN TOPOLOGICAL ALGEBRAS 3 In [21] Schmudgen proves that for a finite S, if KS is compact, then KS ϕ = Psd(KS). Jacobi proved [12] that solves (1) for C = TS and τ = ϕ, i.e., TS ϕ = Psd(KM) (see [18, if M is an Archimedean ∑ R[X]2d-module then M Theorem 1.3 and 1.4] for d = 1). In [15], Lasserre proves that for a specific ⋅w = Psd(KS). In particular, fixed norm ⋅w, and any finite S, MS ⋅w = Psd(Rn). It is worth noting that the latter equality if S = ∅, ∑ R[X]2 could be derived by suitable modifications of Schmudgen [20, Lemma 6.1 and Lemma 6.4]. ⋅w = TS In this paper, we study (1) in a more general context. Throughout the paper the algebras under consideration are unitary and commutative. In Section 2, we recall some standard notations and elementary material which will be needed in the following sections. We consider a Z[ 1 2]-algebra R and a K ⊆ XR ∶= Hom(R, R), closed with respect to the product topology. In Section 3, we associate to K a topology TK on R, making all homomor- phisms in K continuous. When K is compact we define a seminorm ⋅ K on R, which induces a strictly finer topology than TK. If K is Zariski dense, then those topologies are Hausdorff. In section 4, we study (1) in terms of the two topologies TK, ⋅K for the cone C = ∑ R2d of sums of 2d-powers. The two main results are Theorems 4.2 and 4.5: we prove that for K as above, the closure of ∑ R2d with respect to TK is Psd(K). Here Psd(K) ∶= {a ∈ R ∶ α(a) ≥ 0, for all α ∈ K}. When K is compact, we use Stone-Weierstrass to prove that the closure of ∑ R2d In case R = R[X], with respect to the ⋅ K-topology is again Psd(K). K = Rn and d = 1, the first result is a special case of Schmudgen's result for locally multiplicatively convex topologies [20, Proposition 6.2], and the second result is straightforward, as noted in [2, Remark 3.2]. Finally, we apply our results to obtain representation of continuous func- tionals by measures (Corollaries 4.3 and 4.6). In Section 5, we study the case when R is an R-algebra. We define ∑ R2d- modules and archimedean modules exactly as we did for R = R[X]. We prove that the closure of ∑ R2d with respect to any sub-multiplicative norm is Psd(K⋅), where K⋅ is the Gelfand spectrum of (R,⋅) (see Theorem 5.3). Our proof is algebraic and uses a result of T. Jacobi [12, Theorem 4]. Again, we get representation of continuous functionals by measures (Corollary 5.4). Next, we study the case where the cone is a ∑ R2d-module M ⊆ R. We define KM = {α ∈ XR ∶ α(a) ≥ 0 ∀a ∈ M}. We show that M TXR , the closure of M with respect to TXR, is Psd(KM) (see Theorem 5.5). 4 M. GHASEMI, S. KUHLMANN In Section 6, we apply all these results to the ring of polynomials R[X]. Moreover, we study the case when K is not necessarily Zariski-dense. We show that if K is contained in a variety, a locally convex and Hausdorff topology τK can still be defined, as the limit of an inverse family of topologies τK = Psd(K), where Psd(K) is the set of on R[X]. We show that ∑ R[X]2d polynomials which are nonnegative on some open set containing K. Finally, we compare the topologies ⋅K and TK on R[X] to sub-multiplicative norm topologies, and to the Lasserre's topology ⋅ w, considered on [15]. 2. Preliminaries on topological vector spaces and rings. In the following, all vector spaces are over the field of real numbers (unless otherwise specified). A topological vector space is a vector space X equipped with a topology such that the vector space operations (i.e. scalar multipli- cation and vector summation) are continuous. A subset A ⊆ X is said to be convex if for every x, y ∈ A and λ ∈ [0, 1], λx +(1 − λ)y ∈ A. A locally convex (lc for short) topology is a topology which admits a neighborhood basis of convex open sets at each point. Suppose that in addition X is an R-algebra. A subset U ⊆ X is called a multiplicative set, an m-set for short, if U ⋅ U ⊆ U . A locally convex topology on X is said to be locally multiplicatively convex (or lmc for short) if there exists a fundamental system of neighborhoods for 0 consisting of m-sets. It is immediate from the definition that the multiplication is continuous in a lmc-topology. Definition 2.1. A function ρ ∶ X → R≥0 is called a seminorm, if (i) ∀x, y ∈ X ρ(x + y) ≤ ρ(x) + ρ(y), (ii) ∀x ∈ X ∀r ∈ R ρ(rx) = rρ(x). (iii) ∀x, y ∈ X ρ(x ⋅ y) ≤ ρ(x)ρ(y). ρ is called a multiplicative seminorm, if in addition ρ satisfies the following Definition 2.2. Let F be a nonempty family of seminorms on X. The topology generated by F on X is the coarsest topology on X making all seminorms in F continuous. It is a locally convex topology on X. The family of sets of the form ρ1,...,ρk(x) ∶= {y ∈ X ∶ ρi(x − y) ≤ ǫ, i = 1, . . . , k} where ǫ > 0 and ρ1 . . . , ρk ∈ F, forms a basis for this topology. U ǫ We have the following characterization of lc and lmc spaces. CLOSURE OF ∑ R2d IN TOPOLOGICAL ALGEBRAS 5 Theorem 2.3. Let X be an algebra and τ a topology on X. Then (1) τ is lc if and only if it is generated by a family of seminorms on X. (2) τ is lmc if and only if it is generated by a family of multiplicative seminorms on X. Proof. See [13, Theorem 6.5.1] for (1) and [1, 4.3-2] for (2). (cid:3) Let R be a commutative ring with 1 and 1 which is the coarsest topology making all projection functions a ∶ XR → R 2 ∈ R. We always assume that XR = Hom(R, R), the set of unitary homomorphisms, is nontrivial. Clearly, XR ⊂ RR, therefore, it carries a topology as subspace of RR with the product topology which is Hausdorff. For any a ∈ R let U(a) ∶= {α ∈ XR ∶ α(a) < 0}. The family {U(a) ∶ a ∈ R} forms a subbasis for the subspace topology on XR continuous where for a ∈ R, a is defined by a(α) = α(a). XR also can be embedded in Sper(R) equipped with spectral topology [17, Theorem 5.2.5 and Lemma 5.2.6]. Since all projections are continuous, the topology of XR coincides with the subspace topology inherited from RR equipped with product topology. For K be a subset of XR we denote by C(K) the algebra of continuous real valued functions on K. Zariski topology on XR. Definition 2.4. To any subset S of R we associate a subset Z(S) of XR, called the zeros of S or the variety of S by Z(S) ∶= {α ∈ XR ∶ α(S) = {0}}. Denote by ⟨S⟩ the ideal generated by S. Then Z(S) = Z(⟨S⟩), and the family {XR ∖ Z(S) ∶ S ⊆ R} forms a sub-basis for a topology called the Now Suppose that the map (defined in Lemma 3.1 with K = XR, a ↦ a is injective, then a subset K of XR is dense in XR with respect to the Zariski topology (Zariski dense) if and only if K ~⊂ Z(I) for any proper I ideal of R. Example 2.5. Suppose that K ⊆ Rn has nonempty interior, then clearly K is Zariski dense in Rn. Let K be a basic closed semialgebraic set, i.e., K ∶= {x ∈ Rn ∶ fi(x) ≥ 0, i = 1, . . . , m} where fi ∈ R[X], i = 1, . . . , m. We show that if K has an empty interior then K is contained in Z(g), where g = f1 . . . fm. For each x ∈ K, fi(x) = 0 for some 1 ≤ i ≤ m. Otherwise, fi(x) > 0 for each i = 1, . . . , m, the continuity of polynomials implies that fi(y) > 0, i = 1, . . . , m for y sufficiently close to x and hence x is an interior point which is impossible. Therefore g(x) = 0 and hence K ⊆ Z(g). Note that any closed semialgebraic set K, is a finite union of basic closed semialgebraic sets [6, 6 M. GHASEMI, S. KUHLMANN Theorem 2.7.2] i.e., K = ⋃l is so and hence K ⊆ Z(g1 . . . gl), where each gi is the product of generators i=1 Ki. If K has an empty interior, then each Ki of Ki. 2(1, r) = (2, r 2(2, r) = (1, r Remark 2.6. In general, the conclusion of the above example is false. For 2]-module where 1 example, let R = Z3×R[X] with component wise addition and multiplication. 2). Also XR = Rn 2) and 1 R is a Z[ 1 is non-trivial. I = Z3 × {0} is a proper ideal of R and Z(I) = Rn. Hence, every semialgebraic set is contained in Z(I). Definition 2.7. If R is a ring, a function N ∶ R → R+ is called a ring- seminorm if the following conditions hold for all x, y ∈ R: (i) N(0) = 0, (ii) N(x + y) ≤ N(x) + N(y), (iii) N(−x) = N(x), (iv) N(xy) ≤ N(x)N(y). (v) N(x) = 0 only if x = 0. N is called a ring-norm, if in addition We close this section by stating a general version of Haviland's Theorem. Theorem 2.8. Suppose R is an R-algebra, X is a Hausdorff space, and ∶ R → C(X) is an R-algebra homomorphism such that for some p ∈ R, p ≥ 0 on X and the set Xi = p−1([0, i]) is compact for each i = 1, 2, . . . . Then for every linear functional L ∶ R → R satisfying L({a ∈ R ∶ a ≥ 0 on X}) ⊆ R+, there exists a Borel measure µ on X such that ∀a ∈ R L(a) = ∫X a dµ. Proof. See [17, Theorem 3.2.2]. (cid:3) 3. The topologies TK and .K . Throughout we assume that the map ∶ R → C(XR), defined by a(α) = α(a) is injective. Lemma 3.1. Let K be a subset of XR, then (1) The map Φ ∶ R → C(K) defined by Φ(a) = aK is a homomorphism. (2) Im(Φ) contains a copy of Z[ 1 2]. CLOSURE OF ∑ R2d IN TOPOLOGICAL ALGEBRAS 7 Proof. (1) This is clear. Let a, b ∈ R, then for each α ∈ K we have Φ(a + b)(α) = (a + b)K(α) = α(a + b) = α(a) + α(b) = aK(α) + bK(α) = Φ(a)(α) + Φ(b)(α). 2n is the constant function m Similarly Φ(a ⋅ b) = Φ(a) ⋅ Φ(b). (2) Since XR consists of unitary homomorphisms, 1(α) = α(1) = 1, so the constant function 1 ∈ Φ(R). Moreover for each m ∈ Z and n ∈ N, m 2n ∈ R 2] ⊆ and m 2n which belongs to Im(Φ), so Z[ 1 Im(Φ). The topology TK . Let K ⊆ XR. To any α ∈ K we associate a seminorm ρα on C(K) by defining ρα(f) ∶= f(α) for f ∈ C(K). Note that ρα(f g) = ρα(f)ρα(g), so ρα is a multiplicative seminorm. The family of seminorms FK = {ρα ∶ α ∈ K} thus induces an lmc-topology on C(K). Similarly, The restriction of ρα to Φ(R) induces a multiplicative ring- seminorm on R by defining ρα(a) ∶= a(α) = α(a) for a ∈ R. Thus family of ring-seminorms FK induces a topology TK on R. (cid:3) To ease the notation we shall denote the neighborhoods by U ǫ ρα1 ,...,ραm(a) ∶= U ǫ α1,...,αm(a) . Remark 3.2. TK is the coarsest topology on R for which all α ∈ K are continuous. TK is also the coarsest topology on R, for which Φ is continuous. This is clear, because ρα(a) = ρα(Φ(a)) for each a ∈ R. We note for future reference that the topology generated by FK on C(K) Theorem 3.3. Let K ⊆ XR and Φ ∶ R → C(K) be the map defined in Lemma is Hausdorff [4, Proposition 1.8]. 3.1. The following are equivalent: (1) K is Zariski dense, (2) ker Φ = {0}, (3) TK is a Hausdorff topology. Proof. (1)⇒(2) In contrary, suppose that ker Φ ≠ {0} and let 0 ≠ a ∈ ker Φ. Then by definition, a(α) = 0 for all α ∈ K. This implies K ⊆ Z(a) which contradicts the assumption that K is Zariski dense. (2)⇒(3) Since Φ is injective, by Remark 3.2, Φ is a topological embedding. This implies that TK is Hausdorff as well. 8 M. GHASEMI, S. KUHLMANN (3)⇒(1) Suppose that K is not Zariski dense. Than K ⊆ Z(I) for a nontriv- ial ideal of R. Take a, b ∈ I, a ≠ b and let U be an open set in TK, containing a. By definition, there exist α1, . . . , αm ∈ K and ǫ > 0 such that U ǫ α1,...,αm(a) ⊆ U. For each i = 1, . . . , m, ραi(a − b) = αi(b − a) = 0, therefore b ∈ U ǫ α1,...,αm(a) and hence b ∈ U . This shows that TK is not Hausdorff, a contradiction. (cid:3) The topology .K . Assume now that K is compact. In this case, C(K) carries a natural norm topology, the norm defined by fK = supα∈K f(α). The inequalities f + gK ≤ fK +gK and f gK ≤ fKgK implies the continuity of addition and multiplication on (C(K), ⋅ K). Lemma 3.4. If K ⊆ XR is compact, then Φ(R) is dense in (C(K), ⋅ K). Proof. Let A = Φ(R). We make use of Stone-Weierstrass Theorem to show that A = C(K). K is compact and Hausdorff, so once we show that A is an R-algebra which contains all constant functions and separates points of K, we are done (See [22, Theorem 44.7]). Note that A contains all constant functions because Z[ 1 2] ⊂ Φ(R) which is dense in R. Since addition and multiplication are continuous, they extend continuously to A, therefore, A is also closed under addition and multiplication. Moreover, A separates points of K, because Φ(R) does. Hence, by Stone-Weierstrass Theorem A = C(K). (cid:3) Remark 3.5. 1. Corresponding to Remark 3.2, defining the ring-norm ⋅K on R by aK = aK induces a topology which is the coarsest topology such that Φ is continuous. But ⋅ K is not necessarily a norm, unless when Φ is injective which by Theorem 3.3 is equivalent to K being Zariski dense. 2. For any α ∈ K, the evaluation map at α, over C(K,⋅K) satisfies the inequality f(α) = ρα(f) ≤ fK , so it is continuous for each α ∈ K. This observation shows that each TK-open set is also ⋅K -open, i.e., ⋅ K -topology is finer than TK. We show that if K is infinite, then ⋅ K-topology is strictly finer than TK. Proposition 3.6. If K is an infinite, compact subset of XR, then ⋅ K - topology is strictly finer than TK . CLOSURE OF ∑ R2d IN TOPOLOGICAL ALGEBRAS 9 Proof. Let α1, . . . , αm ∈ K and 0 < ǫ < 1. We claim that there exists a ∈ R α1,...,αm(0) and aK > ǫ. Note that XR is Hausdorff and such that a ∈ U ǫ so is K. Compactness of K implies that K is a normal space. Take A = {α1, . . . , αm} and B = {β}, where β ∈ K ∖ A. By Urysohn's lemma, there exists a continuous function f ∶ K → [0, 1] such that f(A) = 0 and f(β) = 1. For δ < min{ǫ, 1 − ǫ}, there exists a ∈ R such that f − aK < δ by Lemma 3.4. Clearly a ∈ U ǫ α1,...,αm(0) and a(β) > ǫ which implies that aK > ǫ which completes the proof of the claim. Let Nǫ(0) = {a ∈ R ∶ aK < ǫ} be an open ball around 0 in ⋅ K for 0 < ǫ < 1. We show that Nǫ(0) does not contain any open neighborhood of 0 in TK. In contrary, suppose that 0 ∈ U δ α1,...,αm(0) ⊆ Nǫ(0). Obviously α1,...,αm(0) such that a ~∈ Nǫ(0) which is a δ ≤ ǫ and so there exists a ∈ U δ contradiction. So, Nǫ(0) is not open in TK and hence, ⋅ K -topology is strictly finer that TK. (cid:3) 4. Closures of ∑ R2d in TK and ⋅ K In this section, we compute the closure ∑ R2d in the two topologies defined in the previous section. In particular, for compact K ⊆ XR, we show that Q R2d ⋅K = Q R2d TK , although for infinite K, the ⋅K -topology is strictly finer than TK on R by Proposition 3.6. Let C +(K) ∶= {f ∶ K → R ∶ f is continuous and ∀α ∈ K f(α) ≥ 0}, denote the set of nonnegative real valued continuous functions over K and Psd(K) ∶= {a ∈ R ∶ a ∈ C +(K)} . Proposition 4.1. Psd(K) is closed in TK . If K is compact, then Psd(K) is also closed in ⋅ K -topology. Proof. For each α ∈ K, let eα(f) = f(α) be the evaluation map. Then α ([0, ∞)) is closed, by continuity of eα in TK. Therefore Psd(K) = e−1 ⋂α∈K e−1 If K is compact, then, again eα is ⋅ K - continuous. Therefore Psd(K) is also closed with respect to ⋅ K . Theorem 4.2. For any compact set K ⊆ XR and integer d ≥ 1, ∑ R2d Psd(K). α ([0, ∞)) is closed. (cid:3) ⋅K = 10 M. GHASEMI, S. KUHLMANN ⋅K Proof. Since ∑ R2d ⊆ Psd(K) and Psd(K) is closed, clearly ∑ R2d ⋅K ⊆ Psd(K). To show the reverse inclusion, let a ∈ Psd(K) and ǫ > 0 be given. 2d√a ∈ C(K). Continuity of multiplication implies the Since a ≥ 0 on K, continuity of the map f ↦ f 2d. Therefore, there exists δ > 0 such that 2d√a − fK < δ implies a − f 2dK < ǫ. Using Lemma 3.4, there is b ∈ R such that 2d√a − bK < δ and so a − b2dK < ǫ. By definition, a − b2d = Φ(a − b2d) and hence a − b2dK < ǫ. Therefore, any neighborhood of a has nonempty intersection with ∑ R2d which proves the reverse inclusion Psd(K) ⊆ ∑ R2d Corollary 4.3. Let K be a compact subset of XR, d ≥ 1 an integer. Assume 2]-linear map, such that L(a2d) ≥ 0 for that L ∶ R → R is ⋅K -continuous, Z[ 1 all a ∈ R, then there exists a Borel Measure µ on K such that ∀a ∈ R L(a) = ∫K a dµ. Proof. Let R ∶= {a ∶ a ∈ R} and define ¯L ∶ R → R by ¯L(a) = L(a). We prove if a ≥ 0, then L(a) ≥ 0. To see this, let ǫ > 0 be given and find δ > 0 such that a − bK < δ implies L(a) − L(b) < ǫ. Take cǫ ∈ R such that a − c2d ǫ K < δ. Then (cid:3) . L(c2d ǫ ) − ǫ < L(a) < L(cǫ) + ǫ, let ǫ → 0, yields L(a) ≥ 0. Note that ¯L is well-defined, since a = 0, implies a ≥ 0 and −a ≥ 0, so ¯L(a) ≥ 0 and ¯L(−a) ≥ 0, simultaneously and hence ¯L(a) = 0. ⋅K -continuity of L on R, implies ⋅ K-continuity of ¯L on R. Let A be the R-subalgebra of C(K), generated by R. Elements of A are of the form r1 a1 + ⋅ ⋅ ⋅ + rk ak, where ri ∈ R and ai ∈ R, for i = 1 . . . , k and k ≥ 1. ¯L is continuously extensible to A by ¯L(ra) ∶= r ¯L(a). By Lemma 3.4, R and hence A is dense in (C(X), ⋅ K). Hahn-Banach Theorem gives a continuous extension of ¯L to C(X). Denoting the extension again by ¯L, an easy verification shows that ¯L(C +(K)) ⊆ R+. Applying Riesz Representation Theorem, the result Remark 4.4. For the special case K = Rn and R = R[X], it follows from [20, Proposition 6.2] that the closure of ∑ R[X]2 with respect to the finest lmc topology η0 on R[X] is equal to Psd(Rn). Since TRn is lmc and Psd(Rn) follows. (cid:3) is closed in TRn we get Psd(Rn) = Q R[X]2 η0 ⊆ Q R[X]2 TRn ⊆ Psd(Rn) TRn = Psd(Rn). CLOSURE OF ∑ R2d IN TOPOLOGICAL ALGEBRAS 11 In the next theorem, we show that a similar result holds for arbitrary K and the smaller set of sums of 2d-powers ∑ R2d ⊂ ∑ R2. TK = Theorem 4.5. Let K ⊆ XR be a closed set and d ≥ 1, then ∑ R2d Psd(K). Proof. Since ∑ R2d ⊆ Psd(K) and by Proposition 4.1, Psd(K) is closed, we have ∑ R2d To get the reverse inclusion, let a ∈ Psd(K) be given. We show that any neighborhood of a in TK has a nonempty intersection with ∑ R2d. Claim. If a > 0 on K then a ∈ ∑ R2d TK ⊆ Psd(K). TK . To prove this, let U be an open set, containing a. There exist α1, . . . , αn ∈ K and ǫ > 0 such that a ∈ U ǫ α1,...,αn(a) ⊆ U . Chose m ∈ N such that 22dm we have 0 < αi(b) < 1. By con- max1≤i≤n αi(a) < 22dm. Now for b = tinuity of f(t) = t2d, for each i = 1, . . . , n there exists δi > 0 such that for any t, if t − αi(b)1~2d < δi, then t2d − αi(b) < 22dm . Take δ = min1≤i≤n δi. j =0 λj tj be the real polynomial satisfying p(αi(b)) = 2dαi(b) Let p(t) = ∑N 2] is dense in R one can choose λj ∈ Z[ 1 for i = 1, . . . , n. Since Z[ 1 2], such that λjαi(b)j < δ, for i = 1, . . . , n. Let c = ∑N λj bj ∈ R. j =1 λjαi(b)j − ∑N ∑N Then αi(b) − αi(c)2d < 22dm . Multiplying by 22dm, αi(a) − αi(2mc)2d < ǫ α1,...,αn(a) ∩ ∑ R2d ≠ ∅ i.e. ραi(a − (2mc)2d) < ǫ for i = 1, . . . , n. Therefore U ǫ and hence a ∈ ∑ R2d 2k) > 0 on K, so For an arbitrary a ∈ Psd(K), and each k ∈ N, (a + 1 ∀k ∈ N, a + 1 2k) → ρα(a), we get a ∈ ∑ R2d 2k ∈ ∑ R2d and hence Psd(K) ⊆ ∑ R2d . Letting k → ∞, ρα(a + 1 which completes the proof of the claim. as desired. j =1 j =1 TK TK TK ǫ TK a ǫ (cid:3) Corollary 4.6. Let K be a closed subset of XR and d ≥ 1 an integer. Assume that there exists p ∈ R, such that p ≥ 0 on K, Ki = p−1([0, i]) is compact for 2]-linear map, and L(a2d) ≥ 0 for each i and L ∶ R → R is TK -continuous, Z[ 1 all a ∈ R, then there exists a Borel Measure µ on K such that ∀a ∈ R L(a) = ∫K a dµ. Proof. Following the argument in the proof of Corollary 4.3, the map ¯L ∶ R → R is well-defined and has a FK-continuous extension to the R-subalgebra A of C(K), generated by R. Applying Theorem 2.8 to ¯L, p and A, the result (cid:3) follows. 12 M. GHASEMI, S. KUHLMANN 5. Results for R-Algebras In this section we assume that R is an R-algebra. First we consider closure of ∑ R2d with respect to any sub-multiplicative norm ⋅ on R. We prove that the closure of ∑ R2d with respect to the norm is equal to nonnegative elements over the global spectrum K⋅ of (R, ⋅ ). Recall that the global spectrum of a topological R-algebra, also known as the Gelfand spectrum, is the set of all continuous elements of XR. Furthermore, in the case of R-algebras, we generalize the conclusion of Theorem 4.5 to an arbitrary ∑ R2d-module M . Normed R-Algebras. Suppose that (R, ⋅ ) is a normed R-algebra, i.e., the norm satisfies the sub-multiplicativity condition xy ≤ xy for all x, y ∈ R. Lemma 5.1. If α ∈ K⋅ then α(x) ≤ x, for all x ∈ R. Proof. In contrary suppose that ∃x ∈ R such that α(x) > x. Then for n ≥ 1, xn ≤ xn ≤ α(x)n = α(xn). xn . Since α(x) x > 1, xn ≤ α(xn) Therefore α(x)n contradicts the fact that α is ⋅ -continuous. So ∀x ∈ R α(x) ≤ x. α(xn) xn → ∞ as n → ∞. This (cid:3) Lemma 5.2. Let d ≥ 1 be an integer, a ∈ R and r > a. Then (r ± a) where R is the completion of (R, ⋅ ). 2d about t = 0. Proof. Let ∑∞ The series has the radius of convergence r. Therefore, it converges for every i=0 λiai < ∞. Note that ai ≤ ai for each t with t < r. Since r > a, ∑∞ i = 0, 1, 2, ⋯, assuming a0 = 1, so i=0 λiti be the power series expansion on (r ± t) 2d ∈ R, 1 1 i=0 λiai i=0 λiai ∑∞ i=0 λiai ≤ ∑∞ ≤ ∑∞ < ∞. i=0 λiai ∈ R. 1 2d = ∑∞ This implies that (1 ± a) (cid:3) CLOSURE OF ∑ R2d IN TOPOLOGICAL ALGEBRAS 13 Let A⋅ ∶= {x ± x ∶ x ∈ R} and M2d be the ∑ R2d-module generated by A⋅. Clearly, M2d is archimedean and hence KM2d is compact. Note that α ∈ KM2d ⇔ α(a) ≥ 0 ∀a ∈ M2d ⇔ x ± α(x) ≥ 0 ∀x ∈ R ⇔ α(x) ≤ x ∀x ∈ R ⇔ α is ⋅ -continuous. Therefore KM2d is nothing but global spectrum of (R, ⋅ ). Theorem 5.3 is the analogue of [9, Theorem 4.3] for normed algebras. Note that the fact that the Gelfand spectrum K⋅ is compact is well-known (under additional assumptions)(see [14, Theorem 2.2.3]). However our proof is algebraic and based on the following result of T. Jacobi. Theorem. Suppose M ⊆ R is an archimedean ∑ R2d-module of R for some integer d ≥ 1. Then, ∀ a ∈ R, a > 0 on KM ⇒ a ∈ M. ⋅ = Psd(K⋅). Proof. See [12, Theorem 4] for every d ≥ 1 or [18, Theorem 1.4] for d = 1. (cid:3) Theorem 5.3. Let (R, ⋅ ) be a normed R-algebra and d ≥ 1 an integer. Then K⋅ is compact and ∑ R2d Proof. Since each α ∈ K⋅ is continuous and Psd(K⋅) = ⋂α∈K⋅ α−1([0, ∞)), ⋅ ⊆ Psd(K⋅) is ⋅ -closed. Clearly ∑ R2d ⊆ Psd(K⋅), therefore ∑ R2d Psd(K⋅). For the reverse inclusion we have to show that if a ∈ Psd(K⋅) and ǫ > 0 are given, then ∃b ∈ ∑ R2d with a − b ≤ ǫ. Note that a + ǫ 2 is strictly positive on K⋅. Since K⋅ = KM2d and M2d is archimedean, KM2d is compact. By i=0 σisi, where σi ∈ ∑ R2d, Jacobi's Theorem, a + ǫ i = 0, . . . , k, s0 = 1 and si ∈ A⋅, i = 1, . . . , k. Choose δ > 0 satisfying i=0σi)δ ≤ ǫ (∑k 2 . By Lemma 5.2 and continuity of the function x ↦ x2d i ≤ δ, on R, there exists ri ∈ R such that δ i = 1, . . . , k. Take b = σ0 +∑k 2 ∈ M2d. So a + ǫ 2 , i.e., si − r2d i ≤ δ 2 = ∑k 2 ǫ 2 ǫ 2 + si − r2d i ∈ ∑ R2d. Then i=1 σir2d kQ a − b = kQ ≤ i=1σi ⋅ si − r2d ≤ ǫ. σir2d σisi − kQ i=1 i=1 i + i − This completes the proof. (cid:3) 14 M. GHASEMI, S. KUHLMANN Corollary 5.4. Let (R, ⋅ ) be a normed R-algebra, d ≥ 1 an integer and L ∶ R Ð→ R a ⋅-continuous linear functional. If for each a ∈ R, L(r2d) ≥ 0, then there exists a Borel measure µ on K⋅ such that L(a) = SK⋅ a dµ ∀a ∈ R. Proof. Since K⋅ is compact by Theorem 5.3, the conclusion follows by ap- plying Theorem 2.8 for X = K⋅ and p = 1. (cid:3) Closures of ∑ R2d-modules in TXR. Theorem 5.5. Let M ⊆ R be a ∑ R2d-module of R-algebra R and d ≥ 1 an integer. Then M TXR = Psd(KM). Proof. Since for each m ∈ M , m ≥ 0 on KM , we have M ⊆ Psd(KM). Each homomorphism α ∈ XR is continuous, so Psd(KM) = α−1([0, ∞)), α∈KM is closed in TXR. Therefore M TXR ⊆ Psd(KM). For the reverse inclusion, we show that for each a ∈ Psd(KM) and any open set U containing a, U ∩M ≠ ∅. Hence a ∈ M For a ∈ Psd(KM) and an open set U containing a, there exist α1, . . . , αn ∈ XR and ǫ > 0 such that U ǫ TXR . α1,...,αn(a) ⊆ U , where U ǫ is a typical basic open set in TXR. α1,...,αn(a) = {b ∈ R ∶ ραi(a − b) < ǫ, i = 1, . . . , n} Claim. For each i = 1, . . . , n there exists ti ∈ M , αi(a) = αi(ti). For each i = 1, . . . , n, either αi(a) ≥ 0, or αi(a) < 0. If αi(a) ≥ 0, take ti = αi(a) ⋅ 1R, which belonges to M . Suppose that αi(a) < 0. There exists si ∈ M such that αi(si) < 0. Otherwise, αi ∈ KM and hence αi(a) ≥ 0 which αi(si) si ∈ M . This completes is a contradiction. So αi(a) the proof of the Claim. For each 1 ≤ i, l ≤ n set αi(si) > 0 and hence ti = αi(a) pil ∶= M αi(tl)≠αj(tl) tl − αj(tl) ⋅ 1 αi(tl) − αj(tl) , if αi(tl) ≠ αj(tl) for some 1 ≤ j ≤ n, and pil = 1, if there is no such j. Then take pi = ∏n l=1 pil. Note that for each 1 ≤ k ≤ n, αk(pi) = ⎧⎪⎪⎨ ⎪⎪⎩ 0 if ∃l ∈ {1, . . . , n} αi(tl) ≠ αk(tl), Otherwise. 1 CLOSURE OF ∑ R2d IN TOPOLOGICAL ALGEBRAS 15 Let λi be the number of elements k ∈ {1, . . . , n} such that αi(tl) = αk(tl), for all 1 ≤ l ≤ n. Take p = ∑n p2d j tj which belonges to M . We have αi(p) = αi(a) for i = 1, . . . , n and hence p ∈ U ǫ α1,...,αn(a). Therefore M ∩ α1,...,αn(a) ≠ ∅, so a ∈ M TXR , which proves the reverse inclusion and hence U ǫ TXR = Psd(KM ). 1 λj M j =1 (cid:3) 6. Application to R ∶= R[X] We are mainly interested in the special case of real polynomials. In this case, R[X] is a free finitely generated commutative R-algebra and hence every α ∈ XR[X] is completely determined by α(Xi), i = 1, . . . , n. So, XR[X] = Rn with the usual euclidean topology. Corollary 6.1. Let K be a closed Zariski dense subset of Rn, (1) The family of multiplicative seminorms FK induces a lmc Hausdorff (2) If K is compact subset then fK = supx∈K f(x) induces a norm on topology TK on R[X] such that ∑ R[X]2d R[X] such that ∑ R[X]2d⋅K = Psd(K). TK = Psd(K). Proof. Apply theorems 4.5 and 4.2 to get the asserted equalities. The fact that TK is Hausdorff and ⋅K is actually a norm, follows from Theorem 3.3 and Remark 3.5 respectively. Remark 6.2. (i) According to [8, Theorem 3.1], the equality ∑ R[X]2d⋅K = Psd(K) is equivalent to the following: If L is a linear functional on R[X] satisfying L(h2d) ≥ 0 for all h ∈ R[X] and ∀f ∈ R[X] L(f) ≤ CfK for some real C > 0, then there exists a positive Borel measure µ on K, representing L: (cid:3) ∀f ∈ R[X] L(f) = SK f dµ. TK = Psd(K), we have the fol- (ii) Reinterpreting the equation ∑ R[X]2d lowing: If L be a linear functional on R[X], satisfying L(h2d) ≥ 0 for all h ∈ R[X] and for every x ∈ K, ∃Cx > 0 such that L(f) ≤ Cxf(x), then there exists a positive Borel measure µ on K representing L: ∀f ∈ R[X] L(f) = SK f dµ. We now discuss the case when K is not Zariski dense. By Theorem 3.3, Φ is not injective and hence the topology TK (or when K is compact the topology induced by aK = supα∈K a(α)) will not be Hausdorff. Let K ⊆ Rn be given, then for ǫ > 0 the set K(ǫ) ∶= ⋃x∈K Nǫ(x) has nonempty 16 M. GHASEMI, S. KUHLMANN interior, in fact K ⊆ (K(ǫ))○ and hence K(ǫ) is Zariski dense in Rn. 0 < ǫ1 ≤ ǫ2, then K ǫ1 ⊆ K ǫ2 and the identity map If idǫ2,ǫ1 ∶ (R[X], TK ǫ2) → (R[X], TK ǫ1) is continuous. Therefore the family {(R[X], T K (ǫ))ǫ>0,(idǫδ)δ≤ǫ} is an in- verse system of lc and Hausdorff vector spaces. The inverse limit of this system exists and is a lc and Hausdorff space [13, Section 2.6]. Let (V, τK) = K (ǫ)). Then V = R[X] and τK is a lc and Hausdorff topology. ǫ>0(R[X], T lim ←Ð Theorem 6.3. For any closed K ⊆ Rn and integer d ≥ 1, ∑ R[X]2d cone Psd(K), consisting of those polynomials which are non-negative over Proof. Since τK = lim ←Ð ǫ>0 some open set, containing K. K (ǫ), we have is the τK T Q R[X]2d τK = lim ←ÐQ R[X]2d⋅K(ǫ) = ǫ Psd(K(ǫ)) (See [7, §4.4]) and ⋂ǫ Psd(K(ǫ)) = Psd(K) by definition. Therefore ⋅ K (ǫ) is defined and and is a norm. Moreover, for 0 < ǫ1 ≤ ǫ2, Remark 6.4. Assuming K is compact implies the compactness of each K(ǫ). (cid:3) the identity map idǫ2,ǫ1 ∶ (R[X], ⋅ K (ǫ2)) → (R[X], ⋅ K (ǫ1)) is continuous by ⋅ K (ǫ1) ≤ ⋅ K (ǫ2) . So {(R[X], ⋅ K (ǫ)),(idǫδ)δ<ǫ} is an inverse limit of normed spaces. The inverse limit topology τK = lim ǫ>0 ⋅ K (ǫ) ←Ð τK = exists and is a lc Hausdorff topology on R[X]. The equality ∑ R[X]2d Psd(K) can be verified similar to Theorem 6.3. 6.1. Comparison with sub-multiplicative norm topologies. Now, let ⋅ be a sub-multiplicative norm on R[X], i.e. f ⋅ g ≤ f ⋅ g ∀f, g ∈ R[X]. Proposition 6.5. The ⋅ -topology is finer than ⋅ K⋅ -topology and Q R[X]2d⋅ = Q R[X]2d⋅K⋅ = Psd(K⋅), for every integer d ≥ 1. CLOSURE OF ∑ R2d IN TOPOLOGICAL ALGEBRAS 17 Proof. To prove that ⋅ -topology is finer than ⋅ K⋅ -topology, we show fK⋅ ≤ f. Note that K⋅ is compact by 5.3, so ⋅K⋅ is defined and by Lemma 5.1, fK⋅ = sup x∈K⋅ f(x) ≤ sup x∈K⋅ f = f. Therefore, the identity map id ∶ (R[X], ⋅ ) → (R[X], ⋅ K⋅) is con- tinuous and hence ⋅ -topology is finer than ⋅ K⋅ -topology. Moreover, ∑ R[X]2d⋅K⋅ = Psd(K⋅) by Corollary 6.1. (cid:3) Definition 6.6. (i) A function φ ∶ Nn → R+ is called an absolute value if φ(0) = 1 and ∀s, t ∈ Nn φ(s + t) ≤ φ(s)φ(t) . (ii) For a polynomial f = ∑s∈Nn fsX s, let fφ ∶= ∑s∈Nn fsφ(s). If φ > 0 on Nn, then ⋅ φ defines a norm on R[X]. Berg and Maserick [4, 5] show that the closure of ∑ R[X]2 with respect to the ⋅ φ-topology is Psd(Kφ), where Kφ ∶= {x ∈ Rn ∶ xs ≤ φ(s), ∀s ∈ Nn} = K⋅φ, the Gelfand spectrum of (R[X], ⋅ φ). If φ > 0 then Kφ has non-empty interior, and hence is Zariski dense. By [9], Kφ is compact. Hence ⋅Kφ is defined and is a norm by Remark 3.5(1). The following corollary to Proposition 6.5 generalizes the result of Berg and Maserick [4, Theorem 4.2.5] to the closure of ∑ R[X]2d. Corollary 6.7. The ⋅ φ-topology is finer than ⋅ Kφ-topology and Q R[X]2⋅φ = Q R[X]2⋅Kφ = Psd(Kφ). 6.2. Comparison with Lasserre's topology. Recently, Lasserre [15] con- sidered the following ⋅ w on R[X]: Q s∈Nn fsX sw = Q s∈Nnfsw(s), where w(s) = (2⌈s~2⌉)! and s = (s1, . . . , sn) = s1 + ⋯ + sn. He proved that for any basic semi-algebraic set K ⊆ Rn, defined by a finite set of polynomials S, the closure of the quadratic module MS and the preordering TS with respect to ⋅ w are equal to Psd(K). Proposition 6.8. Let KS ⊆ Rn be a basic closed semi-algebraic set and d ≥ 1 an integer. 18 M. GHASEMI, S. KUHLMANN (1) If KS is compact, then the ⋅w-topology is finer than ⋅KS -topology and MS⋅w = TS⋅w = Q R[X]2d⋅KS = Psd(KS). (2) ⋅ w-topology is finer than TKS and MS⋅w = TS⋅w = Q R[X]2d TKS = Psd(KS). Proof. (1) To show that ⋅w-topology is finer than ⋅KS -topology, it suffices to prove that the formal identity map id ∶ (R[X], ⋅ w) Ð→ (R[X], ⋅ KS) is continuous. Let pi ∶ Rn → R be the projection on i-th coordinate and M = max 1≤i≤n{pi(x) ∶ x ∈ KS}. So, for each s ∈ Nn we have X sKS ≤ Ms. Also w(s) ≥ s! for all s ∈ Nn. By Stirling's formula s! ∼ √2πe(s+ 1 2) lns−s, we see that X sKS X sw ≤ M s s! ∼ 1√2π s→∞ ÐÐÐ→ 0. es(ln M −lns+1)− 1 2 lns X sw < 1, which shows that Therefore for some N ∈ N, if s > N then X sKS id is bounded and hence continuous. The asserted equality follows from Corollary 4.2 and [15, Theorem 3.3]. (2) It suffices to show that for any x ∈ KS, the evaluation map ex(f) = f(x) is ⋅ w-continuous. Since xs f(x) fw s! is bounded. So, ex and hence ρx is ⋅ w-continuous. Therefore any basic open set in TKS is ⋅ w-open. The asserted equality follows from Theorem 4.5 and [15, Theorem 3.3]. s→∞ ÐÐÐ→ 0, we deduce that supf ∈R[X] (cid:3) References [1] E. Beckenstein, L. Narici and C. Suffel, Topological Algebras, North-Holland Math. Stud., 24,. Elsevier Sci. Publ., Amsterdam, (1977). [2] C. Berg, J. P. R. Christensen, C. U. Jensen, A remark on the multidimensional moment problem, Math. Ann. 243, 163-169, (1979). [3] C. Berg, J.P.R. Christensen, P. Ressel, Positive definite functions on abelian semi- groups, Math. Ann. 223, 253-272, (1976). [4] C. Berg, J. P. R. Christensen, P. Ressel, Harmonic Analysis on Semigroups, Theory of Positive Definite and Related Functions, Springer-Verlag, (1984). [5] C. Berg, P. H. Maserick, Exponentially bounded positive definite functions, Illinois J. Math. 28, 162-179, (1984). CLOSURE OF ∑ R2d IN TOPOLOGICAL ALGEBRAS 19 [6] J. Bochnak, M. Coste, M. F. Roy, G´eom´etrie alg´ebrique r´eelle, Ergeb. Math. 12, Springer, 1987. Real algebraic geometry, Ergeb. Math. 36, Springer, 1998. [7] N. Bourbaki, General Topology, Springer, (1989). [8] M. Ghasemi, S. Kuhlmann, E. Samei, Moment problem for continuous positive semi- definite linear functionals, arXiv:1010.2796v3 [math.AG], submitted. [9] M. Ghasemi, M. Marshall, S. Wagner, Closure results for the cone of sums of 2d- powers, arXiv:1109.0048v1 [math.AC], submitted. [10] E. K. Haviland, On the momentum problem for distribution functions in more than one dimension, Amer. J. Math. 57, 562-572, (1935). [11] E. K. Haviland, On the momentum problem for distribution functions in more than one dimension II, Amer. J. Math. 58, 164-168, (1936). [12] T. Jacobi, A representation theorem for certain partially ordered commutative rings, Math. Z. 237, 259-273, (2001). [13] H. Jarchow, Locally convex spaces, B.G. Teubner Stuttgart, (1981). [14] E. Kaniuth, A Course in Commutative Banach Algebras, Springer, GTM 246, (2009). [15] J. B. Lasserre, The K-Moment problem for continuous functionals, linear arXiv:1102.5763v2 [math.AG]. To appear in Trans. Amer. Math. Soc. [16] J. B. Lasserre, T. Netzer, SOS Approximation of Nonnegative Polynomials Via Simple High Degree Perturbations, Math. Z. 256, 99-112, (2006). [17] M. Marshall, Positive Polynomials and Sum of Squares, Mathematical Surveys and Monographs, Vol. 146, (2007). [18] M. Putinar, Positive Polynomials on Compact Semi-algebraic Sets, Indiana Univ. Math. J. 42, no. 3, 969984, (1993). [19] C. Scheiderer, Sums of squares of regular functions on real algebraic varieties, Trans. AMS 352, 1030-1069, (1999). [20] K. Schmudgen, Positive cones in enveloping algebras, Reports on Mathematical Physics, vol. 14, issue 3, 385-404, (1978). [21] K. Schmudgen, The K-moment problem for compact semialgebraic sets, Math. Ann. 289, 203-206, (1991). [22] S. Willard, General Topology, Addison-Wesley, (1970). 1Department of Mathematics and Statistics, University of Saskatchewan, Saskatoon, SK. S7N 5E6, Canada 2Fachbereich Mathematik und Statistik, Universitat Konstanz 78457 Konstanz, Germany E-mail address: [email protected], [email protected]
1001.4379	3	1001	2011-07-04T16:43:15	Complex and Hypercomplex Discrete Fourier Transforms Based on Matrix Exponential Form of Euler's Formula	[ "math.RA" ]	We show that the discrete complex, and numerous hypercomplex, Fourier transforms defined and used so far by a number of researchers can be unified into a single framework based on a matrix exponential version of Euler's formula $e^{j\theta}=\cos\theta+j\sin\theta$, and a matrix root of -1 isomorphic to the imaginary root $j$. The transforms thus defined can be computed using standard matrix multiplications and additions with no hypercomplex code, the complex or hypercomplex algebra being represented by the form of the matrix root of -1, so that the matrix multiplications are equivalent to multiplications in the appropriate algebra. We present examples from the complex, quaternion and biquaternion algebras, and from Clifford algebras Cl1,1 and Cl2,0. The significance of this result is both in the theoretical unification, and also in the scope it affords for insight into the structure of the various transforms, since the formulation is such a simple generalization of the classic complex case. It also shows that hypercomplex discrete Fourier transforms may be computed using standard matrix arithmetic packages without the need for a hypercomplex library, which is of importance in providing a reference implementation for verifying implementations based on hypercomplex code.	math.RA	math	Complex and Hypercomplex Discrete Fourier Transforms Based on Matrix Exponential Form of Euler's Formula Stephen J. Sangwine∗ Todd A. Ell† November 2, 2018 Abstract We show that the discrete complex, and numerous hypercomplex, Fourier transforms defined and used so far by a number of researchers can be unified into a single framework based on a matrix exponential version of Euler's formula ejθ = cos θ + j sin θ, and a matrix root of −1 isomorphic to the imaginary root j. The transforms thus defined can be computed using standard matrix multiplications and additions with no hypercomplex code, the complex or hypercomplex algebra being represented by the form of the matrix root of −1, so that the matrix multiplications are equivalent to multiplications in the appropriate algebra. We present examples from the complex, quaternion and biquaternion algebras, and from Clifford algebras Cℓ1,1 and Cℓ2,0. The significance of this result is both in the theoretical unification, and also in the scope it affords for insight into the structure of the various transforms, since the formulation is such a simple generalization of the classic complex case. It also shows that hypercomplex discrete Fourier transforms may be computed using standard matrix arithmetic packages without the need for a hypercomplex library, which is of importance in providing a reference implementation for verifying implementations based on hypercomplex code. 1 Introduction The discrete Fourier transform is widely known and used in signal and image processing, and in many other fields where data is analyzed for frequency content [1]. The discrete Fourier transform in one dimension is classically formulated as: F [u] = S f [m] = T f [m] exp(cid:16)−j2π F [u] exp(cid:16) j2π mu M (cid:17) M (cid:17) mu (1) M−1 M−1 Xm=0 Xu=0 where j is the imaginary root of −1, f [m] is real or complex valued with M samples, F [u] is complex valued, also with M samples, and the two scale factors S and T must multiply to 1/M . If the transforms are to be unitary then S must equal T also. In this paper we discuss the formulation of the transform using a matrix exponential form of Euler's formula in which the imaginary square root of −1 is replaced by an isomorphic matrix root. This formulation works for the complex DFT, but more importantly, it works for hypercomplex DFTs (reviewed in § 2). The matrix exponential formulation is equivalent to all the known hypercomplex generalizations of the DFT known to the authors, based on quaternion, biquaternion or Clifford algebras, through a suitable choice of matrix root of −1, isomorphic to a root of −1 in the corresponding hypercomplex algebra. All associative hypercomplex algebras (and indeed the complex algebra) are known to be isomorphic ∗Stephen J. Sangwine is with the School of Computer Science and Electronic Engineering, University of Essex, Wivenhoe Park, Colchester, CO4 3SQ, United Kingdom. Email: [email protected] †Todd A. Ell resides at 5620 Oak View Court, Savage, MN 55378-4695, USA. Email: [email protected] 1 to matrix algebras over the reals or the complex numbers. For example, Ward [2, § 2.8] discusses isomorphism between the quaternions and 4 × 4 real or 2 × 2 complex matrices so that quaternions can be replaced by matrices, the rules of matrix multiplication then being equivalent to the rules of quaternion multiplication by virtue of the pattern of the elements of the quaternion within the matrix. Also in the quaternion case, Ickes [3] wrote an important paper showing how multiplication of quaternions could be accomplished using a matrix-vector or vector-matrix product that could accommodate reversal of the product ordering by a partial transposition within the matrix. This paper, more than any other, led us to the observations presented here. The fact that a hypercomplex DFT may be formulated using a matrix exponential may not be surprising. Nevertheless, to our knowledge, those who have worked on hypercomplex DFTs have not so far noted or exploited the observations made in this paper, which is surprising, given the ramifications discussed later. 2 Hypercomplex transforms The first published descriptions of hypercomplex transforms that we are aware of date from the late 1980s, using quaternions. In all three known earliest formulations, the transforms were defined for two-dimensional signals (that is functions of two independent variables). The two earliest formulations [4, § 6.4.2] and [5, Eqn. 20] are almost equivalent (they differ only in the placing of the exponentials and the signal and the signs inside the exponentials)1: F (ω1, ω2) =Z ∞ −∞Z ∞ −∞ f (t1, t2)eiω1t1 ejω2t2 dt1dt2 In a non-commutative algebra the ordering of exponentials within an integral is significant, and of course, two exponentials with different roots of -1 cannot be combined trivially. Therefore there are other possible transforms that can be defined by positioning the exponentials differently. The first transform in which the exponentials were placed either side of the signal function was that of Ell [6, 7]: F (ω1, ω2) =Z ∞ −∞Z ∞ −∞ eiω1t1f (t1, t2)ejω2t2 dt1dt2 This style of transform was followed by Chernov, Bulow and Sommer [8, 9, 10] and others since. In 1998 the present authors described a single-sided hypercomplex transform for the first time [11] exactly as in (1) except that f and F were quaternion-valued and j was replaced by a general quaternion root of −1. Expressed in the same form as the transforms above, this would be: F (ω1, ω2) =Z ∞ −∞Z ∞ −∞ eµ(ω1t1+ω2t2)f (t1, t2)dt1dt2 where µ is now an arbitrary root of -1, not necessarily a basis element of the algebra. The realisation that an arbitrary root of -1 could be used meant that it was possible to define a hypercomplex transform applicable to one dimension: F (ω) =Z ∞ −∞ eµωtf (t)dt Pei et al have studied efficient implementation of quaternion FFTs and presented a transform based on commutative reduced biquaternions [12, 13]. Ebling and Scheuermann defined a Clifford Fourier transform [14, § 5.2], but their transform used the pseudoscalar (one of the basis elements of the algebra) as the square root of −1. Apart from the works by the present authors [11, 15, 16], the idea of using a root of −1 different to the basis elements of a hypercomplex algebra was not developed further until 2006, with the publication of 1In comparing the various formulations of hypercomplex transforms, we have changed the symbols used by the original authors in order to make the comparisons clearer. We have also made trivial changes such as the choice of basis elements used in the exponentials. 2 a paper setting out the roots of −1 in biquaternions (a quaternion algebra with complex numbers as the components of the quaternions) [17]. This work prepared the ground for a biquaternion Fourier transform [18] based on the present authors' one-sided quaternion transform [11]. More recently, the idea of finding roots of −1 in other algebras has been advanced in Clifford algebras by Hitzer and Ab lamowicz [19] with the express intent of using them in Clifford Fourier transforms, perhaps generalising the ideas of Ebling and Scheuermann [14]. Finally, in this very brief summary of prior work we mention that the idea of applying hypercomplex algebras in signal processing has been studied by other authors apart from those referenced above. For an overview see [20]. In what follows we concentrate on DFTs in one dimension for simplicity, returning to the two dimensional case in § 7. 3 Matrix formulation of the discrete Fourier transform 3.1 Matrix form of Euler's formula The transform presented in this paper depends on a generalization of Euler's formula: exp iθ = cos θ + i sin θ, in which the imaginary root of −1 is replaced by a matrix root, that is, a matrix that squares to give a negated identity matrix. Even among 2×2 matrices there is an infinite number of such roots [21, p16]. In the matrix generalization, the exponential must, of course, be a matrix exponential [22, § 11.3]. The following Lemma is not claimed to be original but we have not been able to locate any published source that we could cite here. Since the result is essential to Theorem 1, we set it out in full. Lemma 1. Euler's formula may be generalized as follows: eJθ = I cos θ + J sin θ where I is an identity matrix, and J 2 = −I. Proof. The result follows from the series expansions of the matrix exponential and the trigonometric func- tions. From the definition of the matrix exponential [22, § 11.3]: J 2θ2 J kθk J 3θ3 ∞ eJθ = Xk=0 k! = J 0 + J θ + + 2! + ··· 3! Noting that J 0 = I (see [23, Index Laws]), and separating the series into even and odd terms: + Iθ2 = I − 2! + Jθ − = I cos θ + J sin θ Iθ4 Iθ6 4! − 6! J θ5 5! − J θ3 3! + + ··· Jθ7 7! + ··· Note that matrix versions of the trigonometric functions are not needed to compute the matrix exponen- tial, because θ is a scalar. In fact, if the exponential is evaluated numerically using a matrix exponential algorithm or function, the trigonometric functions are not even explicitly evaluated [22, § 11.3]. In practice, given that this is a special case of the matrix exponential, (because J 2 = −I), it is likely to be numerically preferable to evaluate the trigonometric functions and to sum scaled versions of I and J . Notice that the matrix eJθ has a structure with the cosine of θ on the diagonal and the (scaled) sine of θ where there are non-zero elements of J . 3 3.2 Matrix form of DFT The classic discrete Fourier transform of (1) may be generalized to a matrix form in which the signals are vector-valued with N components each and the root of −1 is replaced by an N × N matrix root J such that J 2 = −I. In this form, subject to choosing the correct representation for the matrix root of −1, we may represent a wide variety of complex and hypercomplex Fourier transforms. Theorem 1. The following are a discrete Fourier transform pair2: F [:, u] = S f [:, m] = T M−1 M−1 Xm=0 Xu=0 exp(cid:16)−J 2π exp(cid:16) J 2π mu M (cid:17) f [:, m] M (cid:17) F [:, u] mu (2) (3) where J is a N × N matrix root of −1, f and F are N × M matrices with one sample per column, and the two scale factors S and T multiply to give 1/M . Proof. The proof is based on substitution of the forward transform (2) into the inverse (3) followed by algebraic reduction to a result equal to the original signal f . We start by substituting (2) into (3), replacing m by M to keep the two indices distinct, and at the same time replacing the two scale factors by their product 1/M : The exponential of the outer summation can be moved inside the inner, because it is constant with respect to the summation index M: f [:, m] = 1 M M−1 Xu=0 "eJ2π mu M M−1 XM=0 e−J2π Mu M f [:,M]# f [:, m] = 1 M M−1 M−1 Xu=0 XM=0 eJ 2π mu M e−J 2π Mu M f [:,M] The two exponentials have the same root of −1, namely J , and therefore they can be combined: f [:, m] = 1 M M−1 M−1 Xu=0 XM=0 eJ 2π (m−M)u M f [:,M] We now isolate out from the inner summation the case where m = M. In this case the exponential reduces to an identity matrix, and we have: f [:, m] = 1 M + 1 M M−1 M−1 Xu=0 Xu=0 f [:, m]   M−1 XM=0(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)M6=m eJ 2π (m−M)u M f [:,M]  The first line on the right sums to f [:, m], which is the original signal, as required. To complete the proof, we have to show that the second line on the right reduces to zero. Taking the second line alone, and changing the order of summation, we obtain: 2The colon notation used here will be familiar to users of matlab R(cid:13) (an explanation may be found in [22, § 1.1.8]). Briefly, f [:, m] means the m th column of the matrix f . M−1 XM=0(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)M6=m"M−1 Xu=0 eJ 2π (m−M)u M # f [:,M] 4 Using Lemma 1 we now write the matrix exponential as the sum of a cosine and sine term. M−1 XM=0(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)m6=M   I +J M−1 M−1 Xu=0 Xu=0 cos(cid:18)2π sin(cid:18)2π (m − M)u M (m − M)u M  (cid:19) (cid:19)  f [:,M] Since both of the inner summations are sinusoids summed over an integral number of cyles, they vanish, and this completes the proof. Notice that the requirement for J 2 = −I is the only constraint on J . It is not necessary to constrain elements of J to be real. Note that J 2 = −I implies that J −1 = −J , hence the inverse transform is obtained by negating or inverting the matrix root of −1 (the two operations are equivalent). The matrix dimensions must be consistent according to the ordering inside the summation. As written above, for a complex transform represented in matrix form, f and F must have two rows and M columns. If the exponential were to be placed on the right, f and F would have to be transposed, with two columns and M rows. It is important to realize that (2) is totally different to the classical matrix formulation of the discrete Fourier transform, as given for example by Golub and Van Loan [22, § 4.6.4]. The classic DFT given in (1) can be formulated as a matrix equation in which a large M × M Vandermonde matrix containing nth roots of unity multiplies the signal f expressed as a vector of real or complex values. Instead, in (2) each matrix exponential multiplies one column of f , corresponding to one sample of the signal represented by f and the dimensions of the matrix exponential are set by the dimensionality of the algebra (2 for complex, 4 for quaternions etc.). In (2) it is the multiplication of the exponential and the signal samples, dependent on the algebra involved, that is expressed in matrix form, not the structure of the transform itself. Readers who are already familiar with hypercomplex Fourier transforms should note that the ordering of the exponential within the summation (2) is not related to the ordering within the hypercomplex formula- tion of the transform (which is significant because of non-commutative multiplication). The hypercomplex ordering can be accommodated within the framework presented here by changing the representation of the matrix root of −1, in a non-trivial way, shown for the quaternion case by Ickes [3, Equation 10] and called transmutation. We have studied the generalisation of Ickes' transmutation to the case of Clifford algebras, and it appears that there is a more general operation. In the cases we have studied this can be described as negation of the off-diagonal elements of the lower-right sub-matrix, excluding the first row and column3. We believe a more general result is known in Clifford algebra but we have not been able to locate a clear state- ment that we could cite. We therefore leave this for later work, as a full generalisation to Clifford algebras of arbitrary dimension requires further work, and is more appropriate to a more mathematical paper. 4 Examples in specific algebras In this section we present the information necessary for (2) and (3) to be verified numerically. In each of the cases below, we present an example root of −1 and a matrix representation4. We include in the Appendix a short matlab R(cid:13) function for computing the transform in (2). The same code will compute the inverse by negating J . This may be used to verify the results in the next section and to compare the results obtained with the classic complex FFT. In order to verify the quaternion or biquaternion results, the reader will need to install the QTFM toolbox [24], or use some other specialised software for computing with quaternions. 3This gives the same result as transmutation in the quaternion case. 4The matrix representations of roots of -1 are not unique -- a transpose of the matrix, for example, is equally valid. The operations that leave the square of the matrix invariant probably correspond to fundamental operations in the hypercomplex algebra, for example negation, conjugation, reversion. 5 4.1 Complex algebra 1 0 The 2 × 2 real matrix (cid:0) 0 −1 0(cid:1) can be easily verified by eye to be a square root of the negated identity matrix (cid:0) −1 0 −1(cid:1), and it is easy to verify numerically that Euler's formula gives the same numerical results in the classic complex case and in the matrix case for an arbitrary θ. This root of −1 is based on the well-known isomorphism between a complex number a + jb and the matrix representation (cid:0) a −b a(cid:1) [2, Theorem 1.6]5. The structure of a matrix exponential eJθ using the above matrix for J is (cid:0) C −S S C(cid:1) where C = cos θ and S = sin θ. b 4.2 Quaternion algebra The quaternion roots of −1 were discovered by Hamilton [25, pp 203, 209], and consist of all unit pure quaternions, that is quaternions of the form xi + yj + zk subject to the constraint x2 + y2 + z2 = 1. A simple example is the quaternion µ = (i + j + k)/√3, which can be verified by hand to be a square root of −1 using the usual rules for multiplying the quaternion basis elements (i2 = j2 = k2 = ijk = −1). Using the isomorphism with 4 × 4 matrices given by Ward [2, § 2.8], between the quaternion w + xi + yj + zk and the matrix: we have the following matrix representation: µ = 1 √3    w −x −y −z w −z x y  w −x y z z −y w x   0 −1 −1 −1 0 −1 1 1   0 −1 1 1 1 −1 0 1 Notice the structure that is apparent in this matrix: the 2 × 2 blocks on the leading diagonal at the top left and bottom right can be recognised as roots of −1 in the complex algebra as shown in § 4.1 Proposition 1. Any matrix of the form:    0 −x −y −z 0 −z x y  0 −x y z z −y 0 x with x2 + y2 + z2 = 1 is the square root of a negated 4 × 4 identity matrix. Thus the matrix representations of the quaternion roots of −1 are all roots of the negated 4 × 4 identity matrix. Proof. The matrix is anti-symmetric, and the inner product of the ith row and ith column is −x2 − y2 − z2, which is −1 because of the stated constraint. Therefore the diagonal elements of the square of the matrix are −1. Note that the rows of the matrix have one or three negative values, whereas the columns have zero or two. The product of the ith row with the jth column, i 6= j, is the sum of two values of opposite sign and equal magnitude. Therefore all off-diagonal elements of the square of the matrix are zero. The structure of a matrix exponential eJθ using a matrix as in Proposition 1 for J is: C −xS −yS −zS C −zS yS xS C −xS yS zS zS −yS C xS     5We have used the transpose of Ward's representation for consistency with the quaternion and biquaternion representations in the two following sections. 6 where, as before, C = cos θ and S = sin θ. 4.3 Biquaternion algebra The biquaternion algebra [2, Chapter 3] (quaternions with complex elements) can be handled exactly as in the previous section, except that the 4×4 matrix representing the root of −1 must be complex (and the signal matrix must have four complex elements per column). The set of square roots of −1 in the biquaternion algebra is given in [17]. A simple example is i + j + k + I(j − k) (where I denotes the classical complex root of −1, that is the biquaternion has real part i + j + k and imaginary part j − k). Again, this can be verified by hand to be a root of −1 and its matrix representation is: −1 − I −1 + I −1 + I 1 + I −1 0 0 1 0 1 −1 0 1 − I 1 + I 1 − I −1 − I Again, sub-blocks of the matrix have recognizable structure. The upper left and lower right diagonal 2 × 2 blocks are roots of −1, while the lower left and upper right off-diagonal 2 × 2 blocks are nilpotent -- that is their square vanishes.     4.4 Clifford algebras Recent work by Hitzer and Ab lamowicz [19] has explored the roots of −1 in Clifford algebras Cℓp,q up to those with p + q = 4, which are 16-dimensional algebras6. The derivations of the roots of -1 for the 16- dimensional algebras are long and difficult. Therefore, for the moment, we confine the discussion here to lower-order algebras, noting that, since all Clifford algebras are isomorphic to a matrix algebra, we can be assured that if roots of -1 exist, they must have a matrix representation. Using the results obtained by Hitzer and Ab lamowicz, and by finding from first principles the layout of a real matrix isomorphic to a Clifford multivector in a given algebra, it has been possible to verify that the transform formulation presented in this paper is applicable to at least the lower order Clifford algebras. The quaternions and biquaternions are isomorphic to the Clifford algebras Cℓ0,2 and Cℓ3,0 respectively so this is not surprising. Nevertheless, it is an important finding, because until now quaternion and Clifford Fourier transforms were defined in different ways, using different terminology, and it was difficult to make comparisons between the two. Now, with the matrix exponential formulation, it is possible to handle quaternion and Clifford transforms (and indeed transforms in different Clifford algebras) within the same algebraic and/or numerical framework. We present examples here from two of the 4-dimensional Clifford algebras, namely Cℓ1,1 and Cℓ2,0. These results have been verified against the CLICAL package [26] to ensure that the multiplication rules have been followed correctly and that the roots of −1 found by Hitzer and Ab lamowicz are correct. +1, e2 12 = +1 and e1e2 = e12. A possible real matrix representation is as follows: Following the notation in [19], we write a multivector in Cℓ1,1 as α + b1e1 + b2e2 + βe12, where e2 2 = −1, e2 1 =  α b1  b2 −β β −b2 b1 −b2 α −β β b2 α b1 b1 α   In this algebra, the constraints on the coefficients of a multivector for it to be a root of −1 are as follows: 2 + β2 = −1 [19, Table 1]7. Choosing b1 = β = 1 gives b2 = √3 and thus e1 + √3e2 + e12 α = 0 and b2 1 − b2 6 p and q are non-negative integers such that p + q = n and n ≥ 1. The dimension of the algebra (strictly the dimension of the space spanned by the basis elements of the algebra) is 2n . 7We have re-arranged the constraint compared to [19, Table 1] to make the comparison with the quaternion case easier: we see that the signs of the squares of the coefficients match the signs of the squared basis elements. 7 which can be verified algebraically or in CLICAL to be a root of −1. The corresponding matrix is then: 1 −√3 0 1   0 √3 −1 1 −√3 1  −1 √3  0 1 1 0 1 = e2 2 = +1, e2 12 = −1, a possible matrix represen- Following the same notation in algebra Cℓ2,0, in which e2 tation is:  α b1  b2 −β β −b2 The constraints on the coefficients are α = 0 and b2 b2 −β b1 α β −b2 b1 α α b1   and thus e1 + e2 + √3e12 is a root of −1. The corresponding matrix is then: 1 1 + b2 2 − β2 = −1, and choosing b1 = b2 = 1 gives β = √3 1 −√3  0 √3 −1  1 0 0 1 0 1   1 −√3 √3 −1 Notice that in both of these algebras the matrix representation of a root of −1 is very similar to that given for the quaternion case in Proposition 1, with zeros on the leading diagonal, an odd number of negative values in each row and an even number in each column. It is therefore simple to see that minor modifications to Proposition 1 would cover these algebras and the matrices presented above. 5 An example not based on a specific algebra We show here using an arbitrary 2 × 2 matrix root of −1, that it is possible to define a Fourier transform c d(cid:1), then by brute force expansion without a specific algebra. Let an arbitrary real matrix be given as J =(cid:0) a b of J 2 = −I we find the original four equations reduce to but two independent equations. Picking (a, b) and solving for the remaining coefficients we find that any matrix of the form: with finite a and b, and b 6= 0, is a root of −1. Choosing instead (a, c) we get the transpose form: a (cid:18) −(1 + a2)/b −a(cid:19) b (cid:18)a −(1 + a2)/c where c 6= 0. Choosing the cross-diagonal terms (b, c) yields: (cid:18)±κ c ∓κ(cid:19) −a c b (cid:19) (4) where κ = √−1 − bc and bc ≤ −1. In all cases the resulting matrix has eigenvalues of λ = ±i. (This is a direct consequence of the fact that this matrix squares to −1.) Each form, however, has different eigenvectors. The standard matrix representation for the complex operator i is (cid:0) 0 −1 0(cid:1) with eigenvectors v = [1,± i]. In the matrix with (a, b) 1 8 parameters the eigenvectors are v = [1,−b/(a ± i)] whereas the cross-diagonal form with (b, c) parameters has eigenvectors v = [1, (κ ± i)/c]. These forms suggest the interesting question: which algebra, if any, applies here8; and how can the Fourier coefficients (the 'spectrum') be interpreted? We are not able to answer the first question in this paper. The 'interpretation' of the spectrum is relatively simple. Consider a spectrum F containing only one non-zero column at index u0 with value ( x y ) and invert this spectrum using (3). Ignoring the scale factor, the result will be the signal: f [:, m] = exp(cid:16)J 2π mu0 M (cid:17)(cid:18)x y(cid:19) The form of the matrix exponential depends on J . In the classic complex case, as given in § 4.1, the matrix exponential, as already seen, takes the form: (cid:18)cos θ − sin θ cos θ(cid:19) sin θ M . This is a rotation matrix and it maps a real unit vector ( 1 where θ = 2π mu0 0 ) to a point on a circle in the complex plane. It embodies the standard phasor concept associated with sinusoidal functions. Using the same analysis, this time using the matrix in (4) above, one obtains for the matrix exponential the 'phasor' operator: (cid:18)cos θ + κ sin θ c sin θ b sin θ cos θ − κ sin θ(cid:19) Instead of mapping a real unit vector ( 1 0 ) to a point on a circle, this matrix maps to an ellipse. Thus, we see that a transform based on a matrix such as that in (4) has basis functions that are projections of an elliptical, rather than a circular path in the complex plane, as in the classical complex Fourier transform. We refer the reader to a discussion on a similar point for the one-sided quaternion discrete Fourier transform in our own 2007 paper [16, § VI], in which we showed that the quaternion coefficients of the Fourier spectrum also represent elliptical paths through the space of the signal samples. It is possible that the matrices discussed in this section could be transformed by similarity transformations into matrices representing elements of a Clifford algebra9. Note that in the quaternion case, any root of -1 lies on the unit sphere in 3-space, and can therefore be transformed into another root of -1 by a rotation. It is possible that the same applies in other algebras, the transformation needed being dependent on the geometry. Clearly there are interesting issues to be studied here, and further work to be done. 6 Non-existence of transforms in algebras with odd dimension In this section we show that there are no real matrix roots of −1 with odd dimension. This is not unexpected, since the existence of such roots would suggest the existence of a hypercomplex algebra of odd dimension. The significance of this result is to show that there is no discrete Fourier transform as formulated in Theorem 1 for an algebra of dimension 3, which is of importance for the processing of signals representing physical 3-space quantities, or the values of colour image pixels. We thus conclude that the choice of quaternion Fourier transforms or a Clifford Fourier transform of dimension 4 is inevitable in these cases. This is not an unexpected conclusion, nevertheless, in the experience of the authors, some researchers in signal and image processing hesitate to accept the idea of using four dimensions to handle three-dimensional samples or pixels. (This is despite the rather obvious parallel of needing two dimensions -- complex numbers -- to represent the Fourier coefficients of a real-valued signal or image.) Theorem 2. There are no N × N matrices J with real elements such that J 2 = −I for odd values of N . 8It is possible that there is no corresponding 'algebra' in the usual sense. Note that there are only two Clifford algebras of dimension 2, one of which is the algebra of complex numbers. The other has no multivector roots of -1 [19, § 4] and therefore the roots of −1 given above cannot be a root of −1 in any Clifford algebra. 9We are grateful to Dr Eckhard Hitzer for pointing this out, in September 2010. 9 Proof. The determinant of a diagonal matrix is the product of its diagonal entries. Therefore − I = −1 for odd N . Since the product of two determinants is the determinant of the product, J 2 = −1 requires J2 = −1, which cannot be satisfied if J has real elements. 7 Extension to two-sided DFTs There have been various definitions of two sided hypercomplex Fourier transforms and DFTs. We consider here only one case to demonstrate that the approach presented in this paper is applicable to two-sided as well as one-sided transforms: this is a matrix exponential Fourier transform based on Ell's original two-sided two-dimensional quaternion transform [6, Theorem 4.1], [7], [27]. A more general formulation is: M−1 N −1 F [u, v] = S e−J2π mu M f [m, n]e−K2π nv N (5) (6) Xm=0 Xu=0 M−1 Xn=0 Xv=0 N −1 f [m, n] = T e+J2π mu M F [u, v]e+K2π nv N in which each element of the two-dimensional arrays F and f is a square matrix representing a complex or hypercomplex number using a matrix isomorphism for the algebra in use, for example the representations already given in § 4.2 in the case of the quaternion algebra; the two scale factors multiply to give 1/M N , and J and K are matrix representations of two arbitrary roots of −1 in the chosen algebra. (In Ell's original formulation, the roots of −1 were j and k, that is two of the orthogonal quaternion basis elements. The following theorem shows that there is no requirement for the two roots to be orthogonal in order for the transform to invert.) Theorem 3. The transforms in (5) and (6) are a two-dimensional discrete Fourier transform pair, provided that J 2 = K 2 = −I. Proof. The proof follows the same scheme as the proof of Theorem 1, but we adopt a more concise presen- tation to fit the available column space. We start by substituting (2) into (3), replacing m and n by M and N respectively to keep the indices distinct: 1 M−1 N −1 f [m, n] = N −1 M N Xu=0 ×"M−1 XN =0 XM=0 × eK2π nv N eJ2π mu M Xv=0 e−J2π Mu N # M f [M,N ]e−K2π N v The scale factors can be moved outside both summations, and replaced with their product 1/M N ; and the exponentials of the outer summations can be moved inside the inner, because they are constant with respect to the summation indices M and N . At the same time, adjacent exponentials with the same root of −1 can be merged. With these changes, and omitting the scale factor to save space, the right-hand side of the equation becomes: M−1 N −1 M−1 N −1 Xu=0 Xv=0 XM=0 XN =0 eJ2π (m−M)u M f [M,N ]eK2π (n−N )v N We now isolate out from the inner pair of summations the case where M = m and N = n. In this case the exponentials reduce to identity matrices, and we have: 1 M N M−1 N −1 Xu=0 Xv=0 f [m, n] 10 This sums to f [m, n], which is the original two-dimensional signal, as required. To complete the proof we have to show that the rest of the summation, excluding the case M = m and N = n, reduces to zero. Dropping the scale factor, and changing the order of summation, we have the following inner double summation: M−1 N −1 Xu=0 Xv=0 eJ2π (m−M)u M f [M,N ]eK2π (n−N )v N Noting that the first exponential and f are independent of the second summation index v, we can move them outside the second summation (we could do similarly with the exponential on the right and the first summation): M−1 Xu=0 N −1 Xv=0 eJ2π (m−M)u M f [M,N ] eK2π (n−N )v N and, as in Theorem 1, the summation on the right is over an integral number of cycles of cosine and sine, and therefore vanishes. Notice that it was not necessary to assume that J and K were orthogonal: it is sufficient that each be a root of −1. This has been verified numerically using the two-dimensional code given in the Appendix. 8 Discussion We have shown that any discrete Fourier transform in an algebra that has a matrix representation, can be formulated in the way shown here. This includes the complex, quaternion, biquaternion, and Clifford algebras (although we have demonstrated only certain cases of Clifford algebras, we believe the result holds in general). This observation provides a theoretical unification of diverse hypercomplex DFTs. Several immediate possibilities for further work, as well as ramifications, now suggest themselves. Firstly, the study of roots of −1 is accessible from the matrix representation as well as direct representation in whatever algebra is employed for the transform. All of the results obtained so far in hypercomplex algebras, and known to the authors [25, pp 203, 209], [17, 19], were achieved by working in the algebra in question, that is by algebraic manipulation of quaternion, biquaternion or Clifford multivector values. An alternative approach would be to work in the equivalent matrix algebra, but this seems difficult even for the lower order cases. Nevertheless, it merits further study because of the possibility of finding a systematic approach that would cover many algebras in one framework. Following the reasoning in § 5, it is possible to define matrix roots of −1 that appear not to be isomorphic to any Clifford or quaternion algebra, and these merit further study. Secondly, the matrix formulation presented here lends itself to analysis of the structure of the transform, including possible factorizations for fast algorithms, as well as parallel or vectorized implementations for single-instruction, multiple-data (simd) processors, and of course, factorizations into multiple complex FFTs as has been done for quaternion FFTs (see for example [15]). In the case of matrix roots of −1 which do not correspond to Clifford or quaternion algebras, analysis of the structure of the transform may give insight into possible applications of transforms based on such roots. Finally, at a practical level, hypercomplex transforms implemented directly in hypercomplex arithmetic are likely to be much faster than any implementation based on matrices, but the simplicity of the matrix exponential formulation discussed in this paper, and the fact that it can be computed using standard real or complex matrix arithmetic, without using a hypercomplex library, means that the matrix exponential formu- lation provides a very simple reference implementation which can be used for verification of the correctness of hypercomplex code. This is an important point, because verification of the correctness of hypercomplex FFT code is otherwise non-trivial. Verification of inversion is simple enough, but establishing that the spectral coefficients have the correct values is much less so. 11 A matlab R(cid:13) code We include here two short matlab R(cid:13) functions for computing the forward transform given in (2), and (5), apart from the scale factors. The inverses can be computed simply by interchanging the input and output and negating the matrix roots of −1. Neither function is coded for speed, on the contrary the coding is intended to be simple and easily verified against the equations. function F = matdft(f, J) M = size(f, 2); F = zeros(size(f)); for m = 0:M-1 for u = 0:M-1 F(:, u + 1) = F(:, u + 1) ... + expm(-J .* 2 .* pi .* m .* u./M) ... * f(:, m + 1); end end function F = matdft2(f, J, K) A = size(J, 1); M = size(f, 1) ./ A; N = size(f, 2) ./ A; F = zeros(size(f)); for u = 0:M-1 for v = 0:N-1 for m = 0:M-1 for n = 0:N-1 F(Au+1:Au+A, Av+1:Av+A) = ... F(Au+1:Au+A, Av+1:Av+A) + ... expm(-J .* 2pi . m .* u./M) ... * f(Am+1:Am+A, An+1:An+A) ... * expm(-K .* 2pi . n .* v./N); end end end end References [1] R. N. Bracewell. The Fourier Transform and its Applications. McGraw -- Hill, Boston, 3rd edition, 2000. [2] J. P. Ward. Quaternions and Cayley Numbers: Algebra and Applications, volume 403 of Mathematics and Its Applications. Kluwer, Dordrecht, 1997. [3] B. P. Ickes. A new method for performing digital control system attitude computations using quater- nions. AIAA Journal, 8(1):13 -- 17, January 1970. [4] R. R. Ernst, G. Bodenhausen, and A. Wokaun. Principles of Nuclear Magnetic Resonance in One and Two Dimensions. Oxford University Press, Oxford, 1987. [5] M. A. Delsuc. Spectral representation of 2D NMR spectra by hypercomplex numbers. Journal of magnetic resonance, 77(1):119 -- 124, March 1988. [6] T. A. Ell. Hypercomplex Spectral Transformations. PhD thesis, University of Minnesota, 1992. 12 [7] T. A. Ell. Quaternion-Fourier transforms for analysis of 2-dimensional linear time-invariant partial- differential systems. In Proceedings of the 32nd IEEE Conference on Decision and Control, San Antonio, Texas, USA, 15 - 17 December 1993, volume 1-4, pages 1830 -- 1841. IEEE, Control Systems Society, 1993. [8] V. M. Chernov. Discrete orthogonal transforms with data representation in composition algebras. In Proceedings Scandinavian Conference on Image Analysis, pages 357 -- 364, Uppsala, Sweden, 1995. [9] T. Bulow. Hypercomplex Spectral Signal Representations for the Processing and Analysis of Images. PhD thesis, University of Kiel, Germany, 1999. [10] T. Bulow and G. Sommer. Hypercomplex signals -- a novel extension of the analytic signal to the multidimensional case. IEEE Trans. Signal Process., 49(11):2844 -- 2852, November 2001. [11] S. J. Sangwine and T. A. Ell. The discrete Fourier transform of a colour image. In J. M. Blackledge and M. J. Turner, editors, Image Processing II Mathematical Methods, Algorithms and Applications, pages 430 -- 441, Chichester, 2000. Horwood Publishing for Institute of Mathematics and its Applications. Proceedings Second IMA Conference on Image Processing, De Montfort University, Leicester, UK, September 1998. [12] Soo-Chang Pei, Jian-Jiun Ding, and Ja-Han Chang. Efficient implementation of quaternion Fourier transform, convolution, and correlation by 2-D complex FFT. IEEE Trans. Signal Process., 49(11):2783 -- 2797, November 2001. [13] Soo-Chang Pei, Ja-Han Chang, and Jian-Jiun Ding. Commutative reduced biquaternions and their Fourier transform for signal and image processing applications. IEEE Trans. Signal Process., 52(7):2012 -- 2031, 2004. [14] Julia Ebling and Gerik Scheuermann. Clifford Fourier transform on vector fields. IEEE Transactions on Visualization and Computer Graphics, 11(4):469 -- 479, July/August 2005. [15] T. A. Ell and S. J. Sangwine. Decomposition of 2D hypercomplex Fourier transforms into pairs of com- plex Fourier transforms. In Moncef Gabbouj and Pauli Kuosmanen, editors, Proceedings of EUSIPCO 2000, Tenth European Signal Processing Conference, volume II, pages 1061 -- 1064, Tampere, Finland, 5 -- 8 September 2000. European Association for Signal Processing. [16] T. A. Ell and S. J. Sangwine. Hypercomplex Fourier transforms of color images. IEEE Trans. Image Process., 16(1):22 -- 35, January 2007. [17] S. J. Sangwine. Biquaternion (complexified quaternion) roots of -1. Advances in Applied Clifford Algebras, 16(1):63 -- 68, June 2006. [18] Salem Said, N. Le Bihan, and S. J. Sangwine. Fast complexified quaternion Fourier transform. IEEE Trans. Signal Process., 56(4):1522 -- 1531, April 2008. [19] Eckhard Hitzer and Rafa l Ab lamowicz. Geometric roots of −1 in Clifford algebras Cℓp,q with p + q ≤ 4. Advances in Applied Clifford Algebras, 21(1):121 -- 144, 2010. Published online 13 July 2010. [20] Daniel Alfsmann, Heinz Gockler, S. J. Sangwine, and T. A. Ell. Hypercomplex algebras in digital signal processing: Benefits and drawbacks. In Proceedings of EUSIPCO 2007, 15th European Signal Processing Conference, pages 1322 -- 6, Poznan, Poland, 3 -- 7 September 2007. European Association for Signal Processing. [21] Paul J. Nahin. Dr Euler's Fabulous Formula: Cures Many Mathematical Ills. Princeton University Press, 2006. [22] Gene H. Golub and Charles F. van Loan. Matrix Computations. Johns Hopkins studies in the Mathe- matical Sciences. The Johns Hopkins University Press, Baltimore and London, third edition, 1996. 13 [23] E. J. Borowski and J. M. Borwein, editors. Collins Dictionary of Mathematics. HarperCollins, Glasgow, 2nd edition, 2002. [24] Stephen J. Sangwine and Nicolas Le Bihan. http://qtfm.sourceforge.net/, 2005. Software library, lic License. for Matlab R(cid:13). Quaternion Toolbox licensed under the GNU General Pub- [25] W. R. Hamilton. Researches respecting quaternions. First series (1843). In H. Halberstam and R. E. In- gram, editors, The Mathematical Papers of Sir William Rowan Hamilton, volume III Algebra, chapter 7, pages 159 -- 226. Cambridge University Press, Cambridge, 1967. First published as [28]. [26] Pertti Lounesto, Risto Mikkola, and Vesa Vierros. CLICAL user manual. Research Report A248, Helsinki University of Technology, Institute of Mathematics, Espoo, Finland, August 1987. [27] S. J. Sangwine. Fourier transforms of colour images using quaternion, or hypercomplex, numbers. Electronics Letters, 32(21):1979 -- 1980, 10 October 1996. [28] W. R. Hamilton. Researches respecting quaternions. Transactions of the Royal Irish Academy, 21:199 -- 296, 1848. 14
1102.3658	1	1102	2011-02-16T17:28:41	Representations of Each Number Type that Differ by Scale Factors	[ "math.RA", "quant-ph" ]	For each type of number, structures that differ by arbitrary scaling factors and are isomorphic to one another are described. The scaling of number values in one structure, relative to the values in another structure, must be compensated for by scaling of the basic operations and relations (if any) in the structure. The scaling must be such that one structure satisfies the relevant number type axioms if and only if the other structure does.	math.RA	math	Representations of Each Number Type that Differ by Scale Factors Physics Division, Argonne National Laboratory, Paul Benioff Argonne, IL 60439, USA e-mail: [email protected] November 20, 2018 Abstract For each type of number, structures that differ by arbitrary scaling factors and are isomorphic to one another are described. The scaling of number values in one structure, relative to the values in another structure, must be compensated for by scaling of the basic operations and relations (if any) in the structure. The scaling must be such that one structure satisfies the relevant number type axioms if and only if the other structure does. 1 Introduction Numbers play an essential role in many areas of human endeavor. Starting with the natural numbers, N , of arithmetic, one progresses up to integers, I, rational numbers, Ra, real numbers, R, and to complex numbers, C. In mathematics and physics, each of these types of numbers is referred to as the natural numbers, the integers, rational numbers, real numbers, and the complex numbers. As is well known, though, "the" means "the same up to isomorphism" as there are many isomorphic representations of each type of number. In this paper, properties of different isomorphic representations of each num- ber type will be investigated. Emphasis is placed on representations of each number type that differ from one another by arbitrary scale factors. Here math- ematical properties of these representations will be described. The possibility that these representations for complex numbers may be relevant to physics is described elsewhere [1]. Here the mathematical logical description of a representation, as a structure that satisfies a set of axioms relevant to the type of system being considered [2, 3], is used. For the scaled structures considered here, it will be useful in some cases to separate the notion of representation from that of structure, and consider representations as different views of a structure. This will be noted when needed. 1 Each structure consists of a base set, one or more basic operations, basic relations (if any), and constants. Any structure containing a base set, basic operations, relations, and constants that are relevant for the number type, and are such that the structure satisfies the relevant axioms is a model of the ax- ioms. As such it is as good a representation of the number type as is any other representation. The contents of structures for the different types of numbers and the chosen axiom sets are shown below: • N = {N, +, ×, <, 0, 1} Nonnegative elements of a discrete ordered commutative ring with identity [4]. • I = {I, +, −, ×, <, 0, 1} Ordered integral domain [5]. • Ra = {Ra, +, −, ×, ÷, <, 0, 1} Smallest ordered field [6]. • R = {R, +, −, ×, ÷, <, 0, 1} Complete ordered field [7]. • C = {C, +, −, ×, ÷,∗ , 0, 1} Algebraically closed field of characteristic 0 plus axioms for complex conjugation [8, 9]. Here an overline, such as in N , denotes a structure. No overline, as for N , denotes a base set. The complex conjugation operation has been added as a basic operation to C as it makes the development much easier. The same holds for the inclusion of the division operation in Ra, R, and C. For this work, the choice of which axioms are used for each of the number types is not important. For example, an alternate choice for N is to use the axioms of arithmetic [10]. In this case N is changed by deleting the constant 1 and adding a successor operation. There are also other axiom choices for the real numbers [11]. The importance of the axioms is that they will be used to show that, for two structures related by a scale factor, one satisfies the axioms if and only if the other does. This is equivalent to showing that one is a structure for a given number type if and only if the other one is a structure for the same number type. These ideas will be expanded in the following sections. The next section gives a general treatment of fields. This applies to all the number types that satisfy the field axioms (rational, real, complex numbers). However much of the section applies to other numbers also (natural numbers, integers). The following five sections apply the general results to each of the number types. The discussions are mainly limited to properties of the number type that are not included in the description of fields. Section 8 expands the descriptions of the previous sections by considering N , I, Ra, R as substructures of C. In this case the scaling factors relating two structures of the same type are complex numbers. Section 9 concludes the paper with a discussion of some consequences and possible uses of these representations in physics. 2 2 General Description of Fields It is useful to describe the results of this work for fields in general. The results can then be applied to the different number types, even those that are not fields. Let S be a field structure where S = {S, +, −, ×, ÷, 0, 1}. (1) Here S with no overline denotes a base set, +, −, ×, ÷ denote the basic field operations, and 0, 1 denote constants. Denoting S as a field structure implies that S is a structure that satisfies the axioms for a field [12]. Let Sp where Sp = {S, +p, −p, ×p, ÷p, 0p, 1p}. (2) be another structure on the same set S that is in S. The idea is to require that Sp is also a field structure on S where the field values of the elements of S in Sp are scaled by p, relative to the field values in S. Here p is a field value in S. The goal is to show that this is possible in that one can define Sp so that Sp satisfies the field axioms if and only if S does. To this end the notion of correspondence is introduced as a relation between the field values of Sp and S. The field value, ap, in Sp is said to correspond to the field value, pa, in S. As an example, the identity value, 1p, in Sp corresponds to the value p × 1 = p in S. This shows that correspondence is distinct from the concept of sameness. ap is the same value in Sp as a is in S. This differs from pa by the factor p. The distinction between correspondence and sameness is present only if p 6= 1. If p = 1, then the two concepts coincide, and Sp and S are the same structures as far as scaling is concerned. So far a scaling factor has been introduced that relates field values between Sp and S. This must be compensated for by a scaling of the basic operations in Sp relative to those of S. The correspondences of the basic field operations and values in Sp to those in S are given by, ap = pa +p = +, ×p = × p , −p = − ÷p = p ÷ . (3) One can use these scalings to replace the basic operations and constants in Sp and define S p by, p S = {S, +, −, × p , p÷, 0, p}. (4) from Sp. p Both S Here the subscript, p, in Sp, Eq. 2 is replaced by p as a superscript to distinguish S p and Sp can be considered as different representations or views of a structure that differs from S by a scaling factor, p. A useful expression of the relation between Sp and S is referred to either as a representation is that S p p 3 of Sp on S, or as an explicit representation of Sp in terms of the operations and element values of S. Besides changes in the definitions of the basic operations given in Eq. 3 and distinguishing between correspondence and sameness, scaling introduces another change. This is that one must drop the usual assumption that the elements of the base set, S, have fixed values, independent of structure membership. Here the field values of the elements of S, with one exception, depend on the structure containing S. In particular, to say that ap in Sp corresponds to pa in S means that the element of S that has the value ap in Sp has the value pa in S. This is different from the element of S that has the same value, a, in S as ap is in Sp. These relations are shown schematically in Figure 1. The valuations associ- ated with elements in the base set S are shown by lines from S to the structures ¯S and ¯Sp. Figure 1: Relations between Elements in the base set S and their Values in the Structures ¯S and ¯Sp. Here ap is the same value in ¯Sp as a is in ¯S. The lines show that they are values for different elements of S. The lines also show that the S element that has the value a, as ap in ¯Sp, has the value pa in ¯S. The one exception to the structure dependence of valuations is the element of S with the value 0. This value-element association is fixed and is independent of all values of p. In this sense it is the "number vacuum" as it is invariant under all changes p → p′.1 p Another approach to understanding the differences between S, S , and Sp, is to distinguish carefully how structure elements are described, when viewed from inside and from outside the structure.2 Comparison of different structures necessarily is done from outside the structures. p Here there are two structures, A, B, with A represented by S and B by both S and Sp. There is just one representation, S, for A because the inside and outside views coincide for this structure. For A, correspondence and sameness coincide. In mathematical logical terms the properties of the elements of A are absolute [14]. They are the same when viewed inside or outside the struc- 1Like the physical vacuum which is unchanged under all space time translations. 2The importance of distinguishing between inside and outside views of structures is well known in mathematical logic as it plays a role in the resolution of the Skolem paradox [13]. 4 ture. For this reason S will often be referred to as a structure instead of a representation. The situation is quite different for the structure, B. Here S and SP repre- sent, respectively, external and internal views of B. For example, the element of S that has value p in S, also has value p in B when viewed externally. This shown explicitly in the representation S p . p When viewed from inside B, the element of S that has value p when viewed externally, has the value 1 when viewed internally or inside B. This is designated by the identity 1p in Sp. The situation is somewhat different for the basic operations of multiplication and division. The basic operation of multiplication in B, is shown internally by ×p in Sp. It is seen externally as the operation ×/p. Here ×/p = ×(−) ÷ p is the external and internal view of an operation in A. As seen in S, × and ÷ denote the multiplication and division operations in A. One can also define isomorphic maps between the structure representations. Define the maps W p and Wp by Sp = WpS p = WpW pS = FpS. (5) p p W p maps S onto S onto Sp. W p is a map from one structure to another, and Wp is a map between different representations of the same structure. and Wp maps S W p and Wp are defined by W p(a) = pa, W p(a ± b) = W p(a)W p(±)W p(b) = (pa) ± (pb) W p(a × b) = W p(a)W p(×)W p(b) = (pa) × p (pb) W p(a ÷ b) = W p(a)W p(÷)W p(b) = (pa)(p÷)(pb) and Wp(pa) = ap, Wp(pa ± pb) = Wp(pa)Wp(±)Wp(pb) = ap ±p bp W p(pa × p )W p(pb) = ap ×p bp p pb) = W p(pa)W p( × W p(pa(p÷)pb) = W p(pa)W p(p÷)W p(pb) = ap ÷p bp. (6) (7) It is important to emphasize that the definition of S is not just a relabeling of the elements of S. One way to see this is to show that the description of and S by use of isomorphisms is necessary but not the relations between S wyz sufficient. To see this let S denote a structure where p p wyz S = {+, −, × w , (y÷), 0, z}. Here w, y, z are arbitrary field values in S. Define the isomorphism W wyz : S → S wyz by W wyza = za, W wyz(±) = ± W wyz(×) = × w , W wyz(÷) = y ÷ . 5 (8) (9) It is clear from the definition of isomorphisms that S satisfies the field axioms if and only if S does. This follows from the observation that all the field axioms [12] are equations. For example, the existence of a multiplicative identity, a × 1 = a in S gives the equivalences wyz a × 1 = a ⇔ W wyz(a)W wyz(×)W wyz(1) = W wyz(a) ⇔ za × w (z) = za. (10) The righthand equation shows that z is the multiplicative identity in S wyz The equivalences of Eq. 10 show that if one attempts to interpret S as an external view of a different structure whose operations and constants are does not represent a field. Requiring defined in terms of those in S, then S that S and S represent the same structure. In particular, S is just a relabeling of the elements of S with no different valuations implied. Note that this limitation is not present for the case where w = y = z. represents a field, limits one to concluding that S wyz wyz wyz wyz . wyz The following two theorems summarize the relations between S , Sp, and S.. The first theorem shows the invariance of equations under the maps, W p and Wp where p, is a scaling factor. It also shows, that the correspondence between between element values in Sp and those in S, extends to general terms. p Theorem 1 Let t and u be terms in S. Let Sp and S be as defined in Eqs. 2 and 4. Then t = u ⇔ tp = up ⇔ tp = up where tp = W pt, up = W pu, and Wptp = tp, Wpup = up. p Proof: It follows from the properties of W p and Wp, Eqs. 6 and 7, that tp = W pt = pt and up = W pu = pu. Also tp = Wptp and up = Wpup. This gives tp = up ⇔ tp = up ⇔ pt = pu ⇔ t = u. (11) From the left the first equation is in Sp, the second in S and S, and the fourth in S. p , the third in both S p It remains to see in detail that the correspondence between element values in Sp and those in S extends to terms. Let m tp = ( X j,k=1 )p (ap)j (bp)k p. The external view of tp in S p is m tp = ( X j,k=1 )p (pa)j (pb)k p. (12) (13) In the numerator, the j pa factors and j − 1 multiplications contribute factors of pj and p−j+1, respectively, to give a factor p. This is canceled by a similar factor arising from the denominator. A final factor of p arises from the representation of the solidus as shown in Eq. 3 for division. 6 Using this and the fact that addition is not scaled, gives the result that m tp = ( X j,k=1 )p (pa)j (pb)k m p = p( X j,k=1 ) (a)j (b)k = pt. (14) Eq.11 and the theorem follow from the fact that Eq. 14 holds for any term, including up. (cid:4) From this one has Theorem 2 Sp satisfies the field axioms if and only if S axioms if and only if S satisfies the field axioms. p satisfies the field Proof: The theorem follows from Theorem 1 and the fact that all the field axioms are equations for terms. (cid:4) The constructions described here can be iterated. Let p be a number value in S and q ≡ qp be a number value in Sp. Let Sqp be another field structure on the base set, S, and let S be the representation of Sqp using the operations and constants of Sp.3 In more detail qp and Sqp = {S, ±qp, ×qp, ÷qp, 0, 1qp} qp S = {S, ±p, ×p q , q÷p, 0, q1p}. (15) (16) Sqp is related to Sp by the scaling factor q. The goal is to determine the scaling factor for the representation of Sqp on S. To determine this, let aqp be a value in Sqp. This corresponds to a value as ap is qp ×p ap in Sp. Here aqp is the same value in Sqp as qp ×p ap is in S in Sp. qp The value in S that corresponds to aqp in Sqp can be determined from its correspondent, qp ×p ap, in Sp. The value in S that corresponds to qp ×p ap is obtained by use of Eqs. 3 and 7. It is given by (Wp)−1(qp ×p ap) = (pq) × p (pa) = pqa. (17) Here q and a are the same values in S as qp and ap are in Sp. Also (Wp)−1, as the inverse of Wp, maps Sp onto S p . This is the desired result because it shows that two steps, first with p and then with q is equivalent to one step with qp. This result shows that the repre- sentation of Sqp on S is given by qp S = {S, ±, × qp , qp÷, 0, qp1}. (18) 3An equivalent way to define Sqp is as the representation of S qp on S p. 7 Note that the steps commute in that the same result is obtained if one scales first by q and then by p as pq = qp. Here qp is a value in S. Also this is equivalent to determining the scale factor for Sq on S provided one accounts for the fact that q is a value in Sp and not in S. 3 Natural Numbers The natural numbers differ from the generic representation in that they are not fields [4]. This is shown by the structure representation for N , N = {N, +, ×, <, 0, 1}. The structure corresponding to Sn is N n = {Nn, +n, ×n, <n, 0n, 1n}. (19) (20) Here n is any natural number > 0. One can use Eq. 3 to represent N n in terms of the basic operations, relations, and constants of N . It is n N = {Nn, +, × n , <, 0, n}. (21) n and in N . As was the case for fields, N Note that, as is the case for addition, the order relation is the same in N n as in N and N n represent external and internal views of a different structure than that represented by N . For N the external and internal views coincide. The multiplication operator in N , ×/n has the requisite properties. This n n can be seen by the equivalences between multiplication in N n, N an ×n bn = cn ⇔ na × n nb = nc ⇔ a × b = c. n , and N : Note that the simple verification of these equivalences takes place outside the three structures and not within any natural number structure. For this reason division by n can be used to verify the equivalences even though it is not part of any natural number structure. These equivalences show that n is the multiplicative identity in N if and n n only if 1 is the multiplicative identity in N . To see this set b = 1. The structure, N , Eq. 21, and that of Eq. 20, differ from the generic description, Section 2, in that the base set Nn is a subset of N. Nn contains just those elements of N whose values in N are multiples of n. For example, the element with value n in N has value 1 in N n, and the element with value na in N has value a in N n. Elements with values na + l in N where 0 < l < n, are absent from Nn.4 4The exclusion of elements of the base set in different representations occurs only for the natural numbers and the integers. It is a consequence of their not being closed under division. 8 n As noted before, the choice of the representations of the basic operations and , is determined by the requirement relation and constants in N n, as shown in N that N satisfies the natural number axioms [4] if and only if N satisfies the axioms. For the axioms that are equations and do not use the ordering relation, this follows immediately from Theorem 1. n For axioms that use the order relation, the requirement follows from the fact that <n=< and for any pair of terms tn, un, tn <n un ⇔ nt < nu ⇔ t < u. Here Eq. 14 was used with b = 1. Note that n > 0. It follows from these considerations that Theorem 3 N n satisfies the axioms of arithmetic if and only if N only if N does. n does if and . A simple example that illustrates the theorem is the axiom of discreteness for the ordering, 0 < 1V ∀a(a > 0 → a ≥ 1) [4]. One has the equivalences 0 <n 1n V ∀an(an >n 0 → an ≥n 1n) ⇔ 0 < nV ∀a(na > 0 → na ≥ n1) ⇔ 0 < 1V ∀a(a > 0 → a ≥ 1). Subscripts are missing on 0 because the value remains the same in all structures. 4 Integers Integers generalize the natural numbers in that negative numbers are included. Axiomatically they can be characterized as an ordered integral domain [5]. As a structure, I is given by I = {I, +, −, ×, <, 0, 1}. (22) I is a base set, +, −, × are the basic operations, < is an order relation, and 0, 1 are the additive and multiplicative identities. Let j be a positive integer. Let I j be the structure I j = {Ij, +j, −j, ×j, <j, 0j, 1j}. (23) Ij is the subset of I containing all and only those elements of I whose values in I are positive or negative multiples of j or 0. The representation of I j in terms of elements, operations, and relations in I is given by j I = {Ij, +, −, × j , <, 0, j}. (24) This structure differs from that of the natural numbers, Eq. 21, by the presence of the additive inverse, −. The proof that I j and I satisfy the integer axioms if and only if I does, is similar to that for Theorem 3. The only new operation is the additive inverse. j 9 Since the axioms for this are similar to those already present, details of the proof for axioms involving subtraction will be skipped. A new feature enters in the case that j is negative. It is sufficient to consider the case where j = −1 as the case for other negative integers can be described as a combination of j = −1 followed by scaling with a positive j. This would be done by extending the iteration process, described for fields in Section 2, to integers. The integer structure representations for j = −1 that correspond to Eqs. 23 and 24 are given by I −1 = {I, +−1, −−1, ×−1, <−1, 0−1, 1−1}, and −1 I = {I, +, −, × −1 , >, 0, −1}. (25) (26) −1 The main thing to note here is that the order relation, <−1, in I −1 corresponds to > in I. I can also be described as a reflection of the whole structure, I = I 1 through the origin at 0. Not only are the integer values reflected but also the basic operations and order relation are reflected. Also in this case the base set, I−1 =I. Eq. 26 indicates that −1 is the identity and −1 is positive in I follow from and a−1 ×−1 1−1 = a−1 ⇔ (−a) × −1 (−1) = −a 0 <−1 1−1 ⇔ 0 > −1. −1 . These −1 This equivalence shows that the relation, >, which is interpreted as greater than in I, is interpreted as less than in I , 0 > −1 means −1 is greater than 0. As a relation in I, 0 > −1 means −1 is less than 0. . Thus, as a relation in I −1 −1 This is an illustration of the relation of the ordering relation <−1 to < . Integers which are positive in I −1 and I , are negative in I. It follows that 0 <−1 1−1 <−1 2−1 <−1 · · · is true in I −1 if and only if 0 > −1 > −2 > · · · is true in I and in I. −1 −1 These considerations show that I −1 and I satisfy the integer axioms if and only if I satisfies the axioms. For axioms not involving the order relation the proof is similar to that for I j for j > 0. For axioms involving the order relation the proofs proceed by restating axioms for I in terms of > . An example is the axiom for transitivity a < bV b < c ⇒ a < c. For this axiom the validity of the equivalence (a−1 <−1 b−1 V b−1 <−1 c−1 ⇒ a−1 <−1 c−1) ⇔ (−a > −bV −b > −c ⇒ −a > −c) shows that this axiom is true in I −1 if and only if it is true in I if it is true in I. −1 if and only 10 These considerations are sufficient to show that a theorem equivalent to Theorem 3 holds for integers: Theorem 4 For any integer j 6= 0, I j satisfies the integer axioms if and only if I does if and only if I does. j 5 Rational Numbers The next type of number to consider is that of the rational numbers. Let Ra denote a rational number structure Ra = {Ra, +, −, ×, ÷, <, 0, 1}. For each positive rational number r let Rar denote the structure Rar = {Ra, +r, −r, ×r, ÷r, <r, 0r, 1r}. (27) (28) Eqs. 27 and 28 show that Ra and Rar have the same base set, Ra. This is a consequence of the fact that rational numbers are a field. As such Ra nd Rar are special cases of the generic fields described in Section 2. Note also that Ra is the same as Ra1. The definition of Rar is made specific by the representation of its elements in terms of those of Ra. It is r Ra = {Ra, +, −, × r , r÷, <, 0, r}. (29) As was seen for fields in Section 2, the number values of the elements of Ra depend on the structure containing Ra. The element of Ra that has value ra r in Ra has value a in Ra is different from the element that has the same value a in Ra. The only exception is the element with value 0 as this value is the same for all Rar. Also, as noted in Section 2, Ra and Rar represent external and internal views of a structure that differs from that represented by Ra. r . The element of Ra that has the value a in Ra r As was the case for multiplication, the relation between ÷r = r÷ and ÷ is fixed by the requirement that Rar satisfy the rational number axioms [6] if and r only if Ra satisfies the axioms if and only if Ra does. this can be expressed as a theorem: Theorem 5 Let r be any nonzero rational number. Then Ra satisfies the ra- r tional number axioms if and only if Ra satisfies the axioms if and only if Rar does. Proof: Since the axioms for rational numbers include those of an ordered field, the proof contains a combination of that already given for fields in Section 2, Theorem 2, and for the ordering axioms for integers, as in Theorem 4. As a result it will not be repeated here. 11 r It remains to prove that Rar is the smallest ordered field if and only if Ra is if and only if Ra is. To show this, one uses the isomorphisms defined in Section 2. Let S be an ordered field. Let W r and Wr be isomorphisms whose definitions r on Ra and Ra follow that of W p and Wp in Eqs. 6 and 7. That is Rar = WrRa r = WrW rRa. (30) Since these maps, as isomorphisms, are one-one onto and are order preserv- ing, they have inverses which are also isomorphisms. In this case one has r Ra ⊆ S ⇒ (W r)Ra = Ra ⇒ (Wr)Ra r ⊆ S = Rar ⊆ S and Rar ⊆ S ⇒ (Wr)−1Rar = Ra r ⊆ S ⇒ (W r)−1Ra = Ra ⊆ S. r (31) (32) This proves the theorem (cid:4). For rational number terms, Theorem 1 holds here. From Eq. 13 one sees that for rational number structures, m tr = ( X j,k=1 )r (ra)j (rb)k m r = r( X j,k=1 ) (a)j (b)k = rt. (33) 6 Real Numbers The description for real numbers is similar to that for the rational numbers. The structures Ra and Rar, Eqs. 27 and 28, become R = {R, +, −, ×, ÷, <, 0, 1} and Rr = {R, +r, −r, ×r, ÷r, <r, 0r, 1r}. (34) (35) The external representation of the structure, whose internal representation is Rr, is given in terms of the elements, operations, relations and constants of R. It is denoted by R where r r R = {R, +, −, × r , r÷, <, 0, r}. (36) Here r is any positive real number value in R.. If r < 0 then < in Eq. 36 is replaced by > . The axioms for the real numbers [7] are similar to those for the rational numbers in that both number types satisfy the axioms for an ordered field. For r this reason the proof that Rr satisfies the ordered field axioms if and only if R satisfies the axioms if and only if R does will not be given as it is essentially the same as that for the rational numbers. 12 Real numbers are required to satisfy an axiom of completeness. For this axiom, let {(ar)j} = {(ar)j : j = 0, 1, 2, · · · } be a sequence of real numbers in Rr, Eq. 35. Let r be a positive real number value in R. {(ar)j } converges in Rr if For all ǫr >r 0 there exists an h such that for all j, m > h (ar)j −r (ar)mr <r ǫr. (37) Here (ar)j −r(ar)mr denotes the absolute value, in Rr, of the difference between (ar)j and (ar)m. The numerical value in R , Eq. 36, that is the same as (ar)j −r (ar)mr is in Rr, is given by raj − ram. This is the same value as aj − am is in R. raj − ram also corresponds to a number value in R given by r raj − ram = raj − am. (38) Theorem 6 Let r 6= 0 be a real number value in R. The sequence {(ar)j} converges in Rr if and only if {raj} converges in R if and only if {aj} converges in R. r Proof: The proof is in two parts: first r > 0, and then r < 0. r > 0 : Let ǫr be a positive number value in Rr such that (ar)j −r (ar)mr <r and R. Eq. 38 gives the result r ǫr. It follows that raj − ram < rǫ in both R that aj − am < ǫ in R. Conversely Let aj − am < ǫ be true in R. Then raj − ram < rǫ is true in , and (ar)j −r (ar)mr <r ǫr is true in Rr. From this one has the r both R and R equivalences, (ar)j −r (ar)mr <r ǫr ⇔ raj − am < rǫ ⇔ aj − am < ǫ. It follows that {(ar)j } converges in Rr if and only if {raj } converges in R and only if {aj} converges in R. r if r < 0: It is sufficient to set r = −1. In this case R−1 and R Eqs. 35 and 36 with r = −1. In this case R−1 = {R, +−1, −−1, ×−1, ÷−1, <−1, 0−1, 1−1} and −1 R = {R, +, −, × −1 , −1÷, >, 0, −1}. −1 are given by (39) (40) As was the case for integers, and is the case for rational numbers, this structure can also be considered as a reflection of the structure, R, through the origin at 0. As before let {(a−1)j} be a convergent sequence in R−1, Eq. 39. The statement of convergence is given by Eq. 37 where r = −1. The statement 13 (a−1)j −−1 (a−1)m−1 <−1 ǫ−1 says that (a−1)j −−1 (a−1)m−1 is a positive number in R−1 that is less than the positive number ǫ−1 and ≥ 0. The corresponding statement in R , Eq. 40, is −1 − aj − (−am)−1 > −ǫ. (41) −1 is a reflection of R about the origin, the absolute value, − , in Since R , the absolute value, − −1, is always R becomes rx = −x in R positive even though it is always negative in R. For example, −aj −(−am)−1 = −aj − am where − −1 = − − and − are the respective absolute values in R and R. . In R −1 −1 −1 In this case Eq. 41 can be recast as − aj − am > −ǫ. (42) This can be used as the convergence condition in Eq. 37. Note that 0 ≥ −aj − am and that ≥ denotes "less than or equal to" in R −1 . It follows from this that for r = −1 that the sequence {(a−1)j } converges in , Eq. 40, if and R−1, Eq, 39, if and only if the sequence {−aj} converges in R only if the sequence {aj} converges in R. −1 Extension to arbitrary negative r can be done in two steps. One first carries out the reflection with r = −1. This is followed by a scaling with a positive value of r as has already been described in Section 2. (cid:4) Theorem 7 Rr is complete if and only if R complete. r is complete if and only if R is . Proof: Assume that Rr is complete and that the sequence {(ar)j} converges in Rr. Then there is a number value µr in Rr such that limj→∞(ar)j = µr. The properties of convergence, Eq. 37, with (ar)m replaced by µr, and Theorem 6, can be used to show that lim j→∞ (ar)j = µr ⇒ lim j→∞ raj = rµ ⇒ lim j→∞ aj = µ. (43) Here µr is the same number value in Rr as rµ is in R r as µ is in R. Conversely assume that R is complete and that {aj} converges in R to a number value µ. Repeating the above argument gives lim j→∞ (ar)j = µr ⇐ lim j→∞ raj = rµ ⇐ lim j→∞ aj = µ. (44) This shows that Rr is complete if and only if R r is complete if and only if R is complete. (cid:4) Since real number structures are fields, Eq. 14 holds for real number terms. That is m tr = ( X j,k=1 )r (ra)j (rb)k r = r( m X j,k=1 ) (a)j (b)k = rt. (45) 14 j=1(ar)j xj r These terms can be used to give relations between power series in Rr, R , and R. Let Pr(n, xr) = Pn r be a power series in Rr. Then P r(n, rx) and P (n, x) are the same power series in R . This means that for each real number value xr, P r(n, rx) and P (n, x) are the respective r same number values in R and R as Pr(n, xr) is in Rr. Here rx and x are the r and R as xr is in Rr. same number values in R r and R as Pr(n, xr) is in R r in Rr and R r However, the power series in R that corresponds to Pr(n, xr) and P r(n, rx) is obtained from Eq. 14. It is Pr(n, xr) = P r(n, rx) = rP (n, x). (46) This shows that the element of R that has value Pr(n, xr) in Rr has value rP (n, x) in R.5 These relations extend to convergent power series. Theorem 6 gives the result that Pr(n, xr) is convergent in Rr if and only if P r(n, rx) is convergent in R if and only if rP (n, x) is convergent in R. It follows from Theorem 6 and Eqs. 43 and 44 that, r limn→∞ Pr(n, xr) = fr(xr) ⇔ limn→∞ P r(n, rx) = f r(rx) ⇔ limn→∞ rP (n, x) = rf (x). Here fr(xr) is the same analytic function [15] in Rr as f r(rx) is in R is in R. Eqs. 46 and 47 give the result that, for any analytic function f , (47) r as f (x) fr(xr) = f r(rx) = rf (x). (48) Here fr and f are functions in Rr and R and fr(xr) is the same number value in Rr as f r(rx) = rf (x) is in R r = (er)rx = rex for the exponential and sinr(xr) = sinr(rx) = r sin(x) for the sine function. Caution: sin2 as f (x) is in R. Simple examples are exr r(xr) = r sin2(x), not r2 sin2(x). r 7 Complex Numbers The descriptions of structures for complex numbers is similar to that for the real numbers. Let C denote the complex number structure C = {C, +, −, ×, ÷,∗ , 0, 1}. (49) For each complex number c let Cc be the internal representation of another structure where Cc = {C, +c, −c, ×c, ÷c,∗c , 0c, 1c}. (50) 5Recall that Rr and Rr are internal and external views of the same structure. The structure differs from R by the scaling factor, r. Also correspondence is a different concept from sameness unless r = 1. 15 The external representation of the structure, in terms of operations and constants in C, is given by c C = {C, +, −, × c , c÷, c(−)∗, 0, c1}. (51) The relations for the field operations are the same as those for the real numbers except that c replaces r. It follows that Eqs. 45 - 48 hold with c replacing r. These equations show that analytic functions, fc(xc) in C c, have corresponding functions in C given by fc(xc) = cf (x). (52) Here xc denotes the same number in C c as x is in C. The relation for complex conjugation is given by a∗c c = c∗a∗. One way to show this is through the requirement that the relation for complex conjugation must be such that 1c is a real number value in Cc if and only if 1 is a real number value in C. This requires that the equivalences c = ca∗ It is not a∗c c = 1c ⇔ (c1)∗c = c1 ⇔ c(1∗) = c1 ⇔ 1∗ = 1 1∗c be satisfied. These equivalences show that (c1)∗c = c(1∗) or more generally (ac)∗c = (ca)∗c = c(a∗). (53) Note that any value for a is possible including c or powers of c. For example, (cn c )∗c = (c(cn))∗c = c(cn)∗. As values of elements of the base set, C, Eq. 53 shows that the element of C in C c has value ca∗ in C. This is different from the element that has value a∗c c of C that has the same value, a∗, in C as a∗c c is in C c. Another representation of the relation of complex conjugation in Cc to that in C is obtained by writing c = ceiφ. Here c is the absolute value of c. This can be used to write (ac)∗c = (ca)∗c = e2iφc∗a∗. (54) That is (c−)∗c = e2iφc∗(−)∗. For most of the axioms, proofs that Cc satisfies an axiom if and only if C does are similar to those for the number types already treated. However, it is worth discussing some of the new axioms. For example the complex conjugation axiom [9] (x∗)∗ = x has an easy proof. It is based on Eq. 53, which gives (a∗c c )∗c = ((ca)∗c )∗c = (c(a∗))∗c = c(a∗)∗. From this one has the equivalences, (a∗c c )∗c = ac ⇔ c(a∗)∗ = ca ⇔ (a∗)∗ = a. This shows that (x∗)∗ = x is valid for C c if and only if it is valid for C c and C. The other axiom to consider is that for algebraic closure. Theorem 8 C c is algebraically closed if and only if C if and only if C is algebraically closed. c is algebraically closed 16 .Proof: The proof consists in showing that any polynomial equation has a solution in Cc if and only if the same polynomial equations in C and C have the same solutions. Let Pn Cc. Then Pn ac = ca, ×c = ×/c, and +c = + gives the implications j=0 bjxj = 0 denote a polynomial equation that has a solution ac in j=0(bc)j(ac)j = 0 in Cc. Carrying out the replacements (bc)j = cbj, c (Pn j=0)c(bc)j(ac)j = 0 ⇒ (Pn ⇒ Pn j=0 cbjaj = 0 ⇒ cPn j=0)c(cbj(ca)j )c = 0 j=0 bjaj = 0 ⇒ Pn j=0 bjaj = 0. c From the left, the first equation is in Cc, the second is in C , and the other three are in C. This shows that if ac is the solution of a polynomial in Cc then ca is the solution of the same polynomial equation in C , and a is the solution of the same polynomial equation in C. j=0 bjaj = 0 is valid in C for some number a, and reversing the implication directions to obtain The proof in the other direction consists in assuming that Pn j=0)c(bc)j(ac)j = 0 in Cc. (cid:4) c (Pn 8 Number Types as Substructures of C c, C c , and C. As is well known each complex number structure contains substructures for the real numbers, the rational numbers, the integers, and the natural numbers. The structures are nested in the sense that each real number structure contains substructures for the rational numbers, the integers, and the natural numbers, etc. Here it is sufficient to limit consideration to the real number substructures of C c, C , and C. c To this end let c be a complex number and let Rc and R be real number substructures in C c and C. In this case Rc = {Rc, +c, −c, ×c, ÷c, <c, 0c, 1c} and R = {R, +, −, ×, ÷, <, 0, 1}. (55) (56) The field operations in Rc and R are the same as those in Cc and C respec- tively, restricted to the number values of the base sets, Rc and R. Expression of the field operations and constants of Rc in terms of those of C gives a representation of Rc that corresponds to Eq. 51. It is c R = {Rc, +, −, × c , c÷, <c, 0, c1}. (57) Recall that the number values associated with the elements of a base set are not fixed but depend on the structure containing them. R contains just those 17 elements of C that have real values in C. Rc contains just those values of C that that have real values in C and Cc. c Let x be an element of R that has a real value, a, in R. This is different from another element, y, of Rc that has the same real value, ac, in Rc as a is in R. c The element, y, also has the value, ca, in R , which is the same real value in R as a is in R. c This shows that y cannot be an element of R if c is complex. The reason is that ca is a complex number value in C, and R cannot contain any elements of C that have complex values in C. It follows that, if c is complex, then Rc and R have no elements in common, except for the element with value 0, which is the same in all structures. The order relations <c, <c, and < are defined relations as ordering is not a basic relation for complex numbers. A simple definition of < in R is provided by defining a < b in C by a < b if a and b are real number values in C and there exists a positive real number value, d, in C such that a + d = b.  (58) Here d is a positive real number in C if there exists a number g in C such that d = (g∗)g. The definition for <c in Rc is obtained by putting c subscripts everywhere in Eq. 58. c The order relation <c in R is ca <c cb if ca and cb are real numbers in C , and there exists a positive real number cd such that ca + cd = cb.  c Here cd is a positive real number in C that cd = (cg)∗c cg. c if there exists a number cg in C (59) c such One still has to prove that these definitions of ordering are equivalent: Theorem 9 Let a, b be real number values in R. Then a < b ⇔ ca <c cb ⇔ ac <c bc. Proof: Replace the three order statements by their definitions, Eqs. 58 and 59, to get a + d = b ⇔ ca + cd = cb ⇔ ac +c dc = bc. Here d = g∗g, cd = (cg)∗c(cg), and dc = g∗c c gc. Let a, b, g be real numbers values in R. Then ca, cb, cg and ac, bc, gc are the same real number values in R and Rc as a, b, g are in R. c Define the positive number value d by d = g∗g. Then by Eq. 51, g∗g is the same number value in R as (cg)∗c × c (cg) = cg∗g = cd c is in R c values in R as g∗c c gc = dc is in Rc. Thus cd and dc are the same positive number and Rc as d is in R. 18 From this one has a + d = b ⇒ ca + cd = cb ⇒ ac +c dc = bc, which gives a < b ⇒ ca <c cb ⇒ ac <c bc. The reverse implications are proved by starting with ac, bc, and dc = g∗c using an argument similar to that given above. (cid:4) c gc and 9 Discussion So far the existence of many different isomorphic structure representations of each number type has been shown. Some properties of the different represen- tations, such as the fact that number values, operations, and relations, in one representation are related to values in another representation by scale factors have been described. Here some additional aspects of these structure representations and their effect on other areas of mathematics will be briefly summarized. One aspect worth noting is the fact that the rational, real, and complex numbers are a multiplicative group. This can be used to define multiplicative operations on the structures of these three number types and use them to define groups whose elements are the scaled structures. These are based on maps from the number values to scaled representations. The maps are listed below for each type of number: • r → Rar r a rational number • r → Rr r a real number • c → C c c a complex number. These maps are restricted to nonzero values6 of r, r, c. The group properties of the fields for rational, real, and complex numbers, induce corresponding properties in the collection of structures for each of these number types. The discussion will be limited to complex numbers since exten- sion to other number types is similar. Let GC denote the collection of complex number structures that differ by arbitrary nonzero complex scaling factors. Define the operation, ⋄, by C c ⋄ Cd = Ccd. (60) Justification for this definition is provided by the description of iteration that includes Eqs. 15-18. G(C) is defined relative to the identity group structure, C, as c, d and cd are values in C. Every structure C c has an inverse, C1/c as 6It may be useful in some cases to include empty structures associated with the number 0. These are represented by the map extensions 0 → S 0 where S = Ra, R, C. C c ⋄ C1/c = C. (61) 19 This shows that G(C) is a group of complex number structures induced by the multiplicative group of complex number values in C. Its properties mirror the properties of the group of number values. Another consequence of the existence of representations of number types that differ by scale factors is that they affect other mathematical systems that are based on different number types as scalars. Examples include any system type that is closed under multiplication by numerical scalars. Specific examples are vector spaces based on either real or complex scalars, and operator algebras. For vector spaces based on complex scalars, one has for each complex number c a corresponding pair of structures, V c, Cc. The scalars for V c are number values in Cc. The representation of V c as a structure with the basic operations expressed in terms of those in V , with C as scalars, is similar to that for Cc, Eq. 51. For Hilbert spaces, the structure H, based on C, is given by H = {H, +, −, ·, h−, −i, ψ}. (62) Here · and h−, −i denote scalar vector multiplication and scalar product. ψ denotes a general state in H. The representation of another Hilbert space structure that is based on Cc, is given by ¯Hc = {H, +c, −c, ·c, h−, −ic, ψc}. (63) The representation that expresses the basic operations of H c in terms of those for H is given by c H = {H, +, −, · c , h−, −i c , cψ}. (64) Details on the derivation of this for the case in which c is a real number are given in [1]. A possible use of these structures in physics is based on an approach to gauge theories [16, 17] in which a finite dimensional vector space V x is associated with each space time point x. So far just one complex number field, C, serves as the scalars for all the V x. The possibility of generalizing this approach by replacing C with different complex number fields, Cx at each point x has been explored in [1]. If ¯Cx = {Cx, +x, −x, ×x, ÷x, ∗x, 0x, 1x}. (65) and y = x + νdx is a neighbor point of x, then the local representation of Cy at x is given by ¯Cr,x = {Cx, +r,x, −r,x, ×r,x, ÷r,x, ∗r,x, 0r,x, 1r,x}. (66) Number values in Cr,x are denoted by ar.x. C r,x corresponds to C c, Eq. 50 with c restricted to be a real number. The use of r instead of c was done in [1] to keep the presentation as simple as possible. However it not necessary. Here r = ry,x is a space time dependent real number associated with the link from x to y. 20 The local representation of Cy on C x, corresponds to Eq. 51. It is given by C r x = {Cx, +x, −x, ×x r , r÷x, r(−)∗x , 0x, r1x}. (67) At point, x, C x is the complex number structure that is available to an observer, Ox, at x. Also Cy (given by Eq. 65 with y replacing the subscript x) is the complex number structure available to Oy, at point y. The number value, ay, is the same value for Oy as ax is for Ox. However, the value ay at y is seen by Ox as an element of C r x, Eq. 67. This means that Ox sees ay as the value rax in C x. Another way to express this is to refer to rax as the local representation, or correspondent, of ay at x. Then r C x is the local representation of C y on Cx. Recall that rax = ar,x is the same number value in C r x and Cr,x as ax is in Cx. It is proposed in [1] that this space time dependence of complex number structures may play a role in any theory where a comparison is needed of values of a complex valued function or field f (x) at different space time points. An example is the space time derivative in direction µ of f : ∂µ,xf = f (x + dxµ) − f (x) ∂xµ . (68) The subtraction in Eq. 68 has meaning if and only if both terms are in the same complex number structure, such as Cx. This is achieved by defining a covariant derivative Dµ,xf = rµ,xf (x + dxµ)x − f (x) ∂xµ . (69) Here rµ,xf (x + dxµ)x, as a number value in Cx, is the local representation or corresponding value of f (x + dxµ) in C x+dxµ. Also f (x + dxµ)x is the same number value in C x as f (x + dxµ) is in C x+dxµ. Gauge fields are introduced by expressing the real number ry,x by ry,x = e ~A(x)·νdx = eP µ Aµ(x)dxµ = Y µ rµ,x. (70) Use of Eqs. 69 and 70 in gauge theory Lagrangians, expanding exponentials to first order in small terms, and keeping terms that are invariant under local gauge transformations, results in ~A(x) appearing as a gauge boson that can have mass (i.e. a mass term is optional in the Lagrangians.) At present there are no immediate and obvious uses of ~A(x) as a gauge It is suspected that ~A(x) may be relevant to some way to field in physics. relate mathematics and physics at a basic level. As such it might be useful for development of a coherent theory of physics and mathematics together [18, 19]. More work needs to be done to see if these ideas have any merit. 21 Acknowledgement This work was supported by the U.S. Department of Energy, Office of Nuclear Physics, under Contract No. DE-AC02-06CH11357. References [1] P. Benioff, arXiv 1008.3134. [2] J. Barwise, An Introduction to First Order Logic, in Handbook of Math- ematical Logic, J. Barwise, Ed. North-Holland Publishing Co. New York, 1977. pp 5-46. [3] H. J. Keisler, Fundamentals of Model Theory, in Handbook of Mathematical Logic, J. Barwise, Ed. North-Holland Publishing Co. New York, 1977. pp 47-104. [4] R.Kaye, Models of Peano Arithmetic Clarendon Press, Oxford, 1991, pp 16-21. [5] Wikipedia:Integral Domain. [6] A. J. Weir, Lebesgue Integration and Measure Cambridge University Press, New York, NY, 1973, P. 12. [7] J. Randolph, Basic Real and Abstract Analysis, Academic Press, Inc. New York, NY, 1968, P. 26. [8] J. Shoenfield, Mathematical Logic, Addison Weseley Publishing Co. Inc. Reading Ma, 1967, p. 86; Wikipedia: Complex Numbers. [9] Wikipedia: Complex Conjugate. [10] R. Smullyan, Godel's Incompleteness Theorems, Oxford University Press,, New York, NY 1992, p. 29; Wikipedia:Peano axioms, [11] S. Lang, Algebra 3rd Edition, Addison Weseley Publishing Co. Reading, MA, 1993, p. 272. [12] I. Adamson, Introduction to Field Theory, 2nd Edition, Cambridge Univer- sity Press, New York, NY, 1982, Chapter 1. [13] J. Shoenfield,Mathematical Logic, Addison Weseley Publishing Co. Inc., Reading Ma, 1967, p. 79; Wikipedia:LwenheimSkolem theorem [14] J. Burgess, Forcing, in Handbook of Mathematical Logic, J. Barwise, Ed. North-Holland Publishing Co. New York, 1977. pp 403-451; W. Hodges, Model Theory, Cambridge University Press, Cambridge, UK, 1993, pp. 47, 150; Wikipedia:absoluteness 22 [15] W. Rudin,Principles of Mathematical Analysis 3rd Edition, International series in pure and applied Mathematics McGraw Hill Book Co. New York, NY, 1976, page 172. [16] G. Mack, Fortshritte der Physik, 29, 135 (1981). [17] I. Montvay and G. Munster, Quantum Fields on a Lattice, Cambridge Monographs on Mathematical Physics, Cambridge University Press, UK, 1994. [18] P. Benioff, Found. Phys., 35, pp 1825-1856, (2005). [19] P. Benioff, Found. Phys. 32 pp 989-1029 (2002). 23
1212.4689	1	1212	2012-12-19T15:01:29	Hall polynomials for representation-finite cluster-tilted algebras	[ "math.RA", "math.RT" ]	We show the existence of Hall polynomials for representation-finite cluster-tilted algebras.	math.RA	math	HALL POLYNOMIALS FOR REPRESENTATION-FINITE CLUSTER-TILTED ALGEBRAS CHANGJIAN FU Abstract. We show the existence of Hall polynomials for representation-finite cluster- tilted algebras. 1. Introduction 1.1. Let k be a finite field and Λ a locally bounded k-algebra, that is, Λ is an associative al- gebra and Λ has a set of primitive orthogonal idempotents {ei}I such that Λ = ⊕i,j∈IeiΛej, and both dimk Λei and dimk eiΛ are finite for all i ∈ I. Let mod Λ be the category of right Λ-modules with finite length. For L, M, N ∈ mod Λ, we denote by F M N,L the number of submodules U of M such that U ∼= L and M/U ∼= N . Let E be a field extension of k. For any k-space V , we denote by V E the E-space V ⊗k E. Clearly, ΛE naturally becomes a E-algebra. The field E is called conservative [12] for Λ if for any indecomposable M ∈ mod Λ, (End M/ rad End M )E is a field. Set Ω = {EE is a finite field extension of k which is conservative for Λ}. For a given Λ with Ω infinite, the algebra Λ has Hall polynomials provided that for any L, M, N ∈ mod Λ, there exists a polynomial φM N,L ∈ Z[T ] such that for any conservative finite field extension E of k for Λ, N,L(E) = F M E φM N E ,LE . We call φM representation-finite, then Ω is an infinite set. N,L the Hall polynomial associated to L, M, N ∈ mod Λ. Note that if Λ is It has been conjectured by Ringel [12] that any representation-finite algebra has Hall polynomials. This conjecture has been verified for representation-directed algebras by Ringel [12], cyclic serial algebras by Guo [6] and Ringel [13] and some other classes of algebras(cf. eg. [7]). 1.2. Let A be a finite-dimensional hereditary algebra over a field k. Let mod A be the finitely generated right A-modules and Db(mod A) the bounded derived category with sus- pension functor Σ. The cluster category C(A) associated with A was introduced in [4](independently in [5] for An case) as the orbit category Db(mod A)/τ −1 ◦ Σ, where τ is the Auslander- Reiten translation of Db(mod A). A cluster-tilting object T in C(A) is an object such that Ext1 C(A)(T, T ) = 0 and it is maximal with this property. The endomorphism alge- bra EndC(A)(T ) of a cluster-tilting object T is called the cluster-tilted algebra of T , which were first introduced by Buan, Marsh and Reiten in [3]. Among others, they showed that cluster-tilted algebras are Gorenstein of dimension 1. This has been further generalized to a more general setting by Keller-Reiten [9] and Konig-Zhu [10]. Moreover, Keller and Date: January 30, 2020. 2010 Mathematics Subject Classification. 16D90, 16G60. Key words and phrases. Hall polynomial, representation-finite, cluster-tilted algebra. Partially supported by NSF of China(No.11001185). 1 2 CHANGJIAN FU Reiten [9] proved that the stable Cohen-Macaulay category of a given cluster-tilted alge- bra(more generally, 2-Calabi-Yau tilted algebra) is 3-Calabi-Yau. A direct consequence of the stably Calabi-Yau property is that the Auslander-Reiten conjecture holds true for cluster-tilted algebras. Namely, let B be a cluster-tilted algebra over a field k, if M is a finitely generated right B-module such that Exti mod B(M, M ⊕ B) = 0 for all i ≥ 1, then M is a projective B-module. In [14], Zhu has introduced certain Galois coverings for cluster categories and cluster- tilted algebras(cf. also [1]), called repetitive cluster categories and repetitive cluster-tilted algebras respectively( for the precisely definition, cf. Section 2). We refer to [2] for the notions of (Galois) covering functors. The aim of this note is to show that Ringel's con- jecture holds true for representation-finite repetitive cluster-tilted algebras. In particular, representation-finite cluster-tilted algebras have Hall polynomials. Theorem 1.1. Let Λ be a representation-finite repetitive cluster-tilted algebra over a finite field k. Then Λ has Hall polynomials. Let us mention here that a variant proof of this theorem may be applied to generalized cluster-tilted algebras of higher cluster categories of type ADE. In order to make this note concise, we restrict ourselves to the case of cluster-tilted algebras. After recall some basic definitions and properties of repetitive cluster-tilted algebras in Section 2 , we will give the proof of Theorem 1.1 in Section 3. 2. Repetitive cluster categories and repetitive cluster-tilted algebras 2.1. Let D be a k-linear triangulated category with suspension functor Σ and T a func- torially finite subcategory of D. The subcategory T is called a cluster-tilting subcategory, if the followings are equivalent: ◦ X ∈ T ; ◦ Ext1 ◦ Ext1 D(X, T ) = 0; D(T , X) = 0. An object T ∈ D is a cluster-tilting object if and only if add T is a cluster-tilting subcategory. A cluster-tilting object T ∈ D is called basic provided that T = ⊕n i=1Ti, where Ti, i = 1, · · · , n are indecomposable and Ti 6∼= Tj whenever i 6= j. Let T be a cluster-tilting subcategory of D and mod T the category of finitely presented right T -modules. It has been proved by Konig and Zhu in [10](cf. also [9]) that the functor HomD(T , −) : D → mod T induces an equivalence D/ add ΣT ∼= mod T . Moreover, if D has Auslander-Reiten triangles, then the Auslander-Reiten sequences of mod T is induced from the Auslander-Reiten triangles of D. In this case, we have τ −1ΣT = T , where τ is the Auslander-Reiten translation of D. 2.2. Let A be a finite-dimensional hereditary algebra over a field k. Let mod A be the category of finitely generated right A-modules and Db(mod A) the bounded derived cate- gory with suspension functor Σ. Let τ be the Auslander-Reiten translation of Db(mod A). In the following, we fix a positive integer m and set F := τ −1 ◦ Σ. The repetitive cluster category CF m(A) introduced by Zhu [14] is the orbit category of Db(mod A)/ < F m >, which is by definition a k-linear category whose objects are the same as Db(mod A), and whose morphisms are given by CF m(A)(X, Y ) :=Mi∈Z HomDb(mod A)(X, F miY ), where X, Y ∈ Db(mod A). HALL POLYNOMIALS 3 When m = 1, we get the cluster category C(A). By the main theorem of Keller [8], we know that CF m(A) admits a canonical triangle structure such that the canonical projection functor πm : Db(mod A) → CF m(A) is a triangle functor. Moreover, by the universal property of the orbit category CF m(A), we have a triangle functor ρm : CF m(A) → C(A) such that πA = ρm ◦ πm, where πA is the canonical projection functor πA : Db(mod A) → C(A). It has been shown in [14] that there exists bijections between the following three sets: the set of cluster-tilting subcategory of Db(mod A), the set of cluster-tilting subcategories of CF m(A) and the set of cluster-tilting subcategories of C(A), via the triangle functors: πm : Db(mod A) → CF m(A) and ρm : CF m(A) → C(A). In particular, the repetitive cluster categories have cluster-tilting objects. tilted algebra. Then we have the followings: (1) the restriction of πm : Db(mod A) → CF m(A) induces a Galois covering πm : We have the following main results of [14](cf. Theorem 3.7 and Theorem 3.8 in [14]). Theorem 2.1. Let T be a basic cluster-tilting object in C(A) and A = EndC(A(T ) the Let eT be a basic cluster-tilting object in the repetitive cluster category CF m(A), the endomorphism algebra EndCF m (A)(eT ) is called the repetitive cluster-tilted algebra of T . cluster-tilted algebra of T . Let eT be the corresponding basic cluster-tilting object in CF m(A) of T via the triangle functor ρm and eA = EndCF m (A)(eT ) the associated repetitive cluster- A (add T ) → addeT of eA. Moreover, the projection functor πm : Db(mod A) → A (add ΣT ) → mod eA; CF m(A) induces a push-down functor eπm : Db(mod A)/π−1 (2) the functor ρm : CF m(A) → C(A) restricted to the cluster-tilting subcategory addeT = a push-down functor eρm : mod eA → mod A. ρ−1 m (add T ) induces a Galois covering of A. Moreover, the functor ρm also induces Remark 2.2. Set T = π−1 of all indecomposable objects in T . Set A (add T ) and let ind T be a set of representatives of the isoclasses π−1 End(T ) := MTi,Tj ∈ind T HomDb(mod A)(Ti, Tj), which is an associative algebra without units. It is not hard to see that End(T ) is locally bounded. On the other hand, the finitely presented T -modules coincides with End(T )- modules of finite length. Hence, we have an equivalence of categories by [10], which implies that End(T ) is directed. mod End(T ) ∼= Db(mod A)/ΣT Remark 2.3. One can verify that the functor eπm coincides with the restriction of the A (add T ) → addeT . Hence,eπm is push-down functor induced by the Galois covering πm : π−1 an exact functor. On the other hand, any exact sequence of Db(mod A)/ΣT can be lifted to a triangle in Db(mod A) (cf. Lemma 8 of [11]). Then one can also prove the exactness directly in this setting. Note that a Galois covering functor will not induce a Galois covering for the corre- sponding categories of modules in general. However, for the Galois coverings in the above theorem, we have the following observation. Proposition 2.4. Keep the notations in Theorem 2.1, we have the followings: (1) the push-down functoreπm : Db(mod A)/π−1 of mod eA; (2) the push-down functor eρm : mod eA → mod A is a Galois covering of mod A. A (add ΣT ) → mod eA is a Galois covering 4 CHANGJIAN FU Proof. We will only prove the first statement and the second one follows similarly. Let T = π−1 A (add T ). Since ΣT is also a cluster-tilting subcategory of Db(mod A), we have F ΣT = ΣT . One shows that F m : Db(mod A) → Db(mod A) induces a k-linear equivalence F m : Db(mod A)/ΣT → Db(mod A)/ΣT . Let G be the infinite cyclic group generated by F m which is acting freely on the objects of Db(mod A)/ΣT . Note that by Theorem 2.1, we have the following commutative diagram Db(mod A) Q1 Db(mod A)/ΣT πm eπm / CF m(A) Q2 / CF m(A)/ΣeT = mod eA Hom where Q1 and Q2 are natural quotient functors. Then for any indecomposable objects X, Y mod eA(M, N ) =Mi∈Z in Db(mod A)/ΣT , one can show thateπm(X) ∼=eπm(Y ) if and only if there exists i ∈ Z such that Y ∼= F imX in Db(mod A)/ΣT . On the other hand, for any objects M, N in mod eA with preimages cM , bN in Db(mod A)/ΣT respectively. We clearly have HomDb(mod A)/ΣT ((F m)icM , bN ). This particular implies that mod eA identifies the orbit category of Db(mod A)/ΣT by G. It is easy to see that the functoreπm coincides with the quotient functor. HomDb(mod A)/ΣT (cM , (F m)ibN ) =Mi∈Z 3. Poorf of the main theorem (cid:3) To prove our main result, we need the following result of Guo and Peng [7], which gives a sufficient condition for the existence of Hall polynomials Lemma 3.1. Let k be a finite field. Let Λ be a finite-dimensional k-algebra of representation- finite type and there is a locally bounded k-algebra R which is directed, such that there exists a covering funtor F : mod R → mod Λ, and for any M, N ∈ mod Λ, there exist X, Y ∈ mod R with F X = M and F Y = N such that F induces the k-isomorphism Ext1 R(X, Y ) ∼= Ext1 Λ(M, N ). Then Λ has Hall polynomials. Remark 3.2. According to Theorem 5.1 of [7], the last condition in Lemma 3.1 can be weakened to the following: for any M, N ∈ mod Λ with N indecomposable, there ex- ist Xi, Yi ∈ mod R, i = 1, 2 with F Xi ∼= M and F Yi ∼= N such that F induces the k- isomorphisms Ext1 R(Y2, X2) ∼= Ext1 Now we are in a position to prove the main theorem of this note. Λ(M, N ) and Ext1 R(X1, Y1) ∼= Ext1 Λ(N, M ). Proof of Theorem 1.1. By the definition of repetitive cluster-tilted algebras, there is a finite-dimensional hereditary algebra A such that Λ is the endomorphism algebra of a basic cluster-tilting object eT in a repetitive cluster category CF m(A) of A. Note that we have an equivalence of categories HomCF m (A)(eT , −) : CF m(A)/ΣeT → mod Λ. Hence, m (addeT ), Λ is representation-finite implies that A is representation-finite. Let T = π−1 by remark 2.2, we know that End(T ) is locally bounded and directed. By part (1) of Proposition 2.4, we deduce that is a directed Galois covering of mod Λ. eπm : mod End(T ) → mod Λ / / HALL POLYNOMIALS 5 According to Lemma 3.1 and Remark 3.2, it suffices to show that for any M, N ∈ mod Λ with N indecomposable, there exist Xi, Yi ∈ mod End(T ), i = 1, 2 with F Xi ∼= M and F Yi ∼= N such that F induces the k-isomorphisms Ext1 End(T )(X1, Y1) ∼= Ext1 Λ(M, N ) and Ext1 End(T )(Y2, X2) ∼= Ext1 Λ(N, M ). We will only prove the existence for the first isomorphism, where the second one follows similarly. We may and we will assume that M is also indecomposable and Ext1 Λ(M, N ) 6= 0. Recall that we have the following commutative diagram Db(mod A) Q1 mod End(T ) ∼= Db(mod A)/ΣT πm eπm / CF m(A) Q2 / CF m(A)/ΣeT ∼= mod Λ where Q1 and Q2 are natural quotient functors. Sinceeπm is a Galois covering, there exists cM , bN ∈ mod End(T ) such that eπm(cM ) = M and eπm(bN ) = N . Moreover, {F imcM i ∈ Z} forms a complete set of preimages of M in mod End(T ). For simplicity, we set cMi = F imcM in the following. By abuse of notations, we still denote by cMi, bN ∈ Db(mod A) the unique indecomposable preimages of cMi and bN respectively. Without of loss generality, we may assume that bN ∈ mod A. By the definition of covering functor and the fact that eπm is an exact functor preserving projectivity, we have the following k-isomorphism Mi∈Z Ext1 End(T )(cMi, bN ) ∼= Ext1 Λ(M, N ). Since 0 6= dimk Ext1 finitely many i. Let t ∈ Z such that Ext1 j > t. Λ(M, N ) < ∞, the vector space Ext1 0 for i < t. By the Auslander-Reiten translation formula, we have We claim that Ext1 End(T )(cMt, bN ) ∼= Ext1 End(T )(cMt, bN ) 6= 0 and Ext1 Λ(M, N ). It suffices to show that Ext1 End(T )(cMi, bN ) vanishes for all but End(T )(cMj , bN ) = 0 for End(T )(cMi, bN ) = Ext1 End(T )(cMi, bN ) ∼= DHomEnd(T )(bN , τcMi), where τ is the Auslander-Reiten translation of mod End(T ) which is induced by the Auslander- Reiten translation of Db(mod A). On the other hand, we have Hom Db(mod A)(bN , τcMi) . HomEnd(T )(bN , τcMi) = {f : bN → τcMi factoring through add ΣT or add Σ2T } then we have Note that Ext1 End(T )(cMt, bN ) 6= 0 implies that HomDb(mod A)(bN , τcMt) 6= 0. Since bN ∈ mod A, we deduce that τcMt ∈ mod A or τcMt ∈ Σ mod A. Recall thatcMi = F imcM , if τcMt ∈ mod A, HomDb(mod A)(bN , τcMi) = HomDb(mod A)(bN , F (i−t)mτcMt) = 0 for i < t. Now assume that τcMt ∈ Σ mod A and let τcMt = ΣL, where L ∈ mod A. From 0 6= Hom Db(mod A)(bN , τcMt) = Hom Db(mod A)(bN , ΣL), / / 6 CHANGJIAN FU Hom we deduce that L is a predecessor of bN in Db(mod A). In this case, we have Db(mod A)(bN , τcMi) = Hom = HomDb(mod A)(bN , τ (t−i)mΣ(i−t)m+1L) = (0 Db(mod A)(bN , F (i−t)mτcMt) HomDb(mod A)(bN , τ mΣ1−mL) for i < t − 1; for i = t − 1. and Hom of L, which contradicts to the fact that Db(mod A) is directed. We have proved that It is clear that HomDb(mod A)(bN , τ mΣ1−mL) = 0 if m ≥ 2. Now suppose that m = 1 Db(mod A)(bN , τ L) 6= 0, then bN is a predecessor of τ L and hence a predecessor HomDb(mod A)(bN , τcMi) = 0 for i 6= t and hence Ext1 End(T )(cMi, bN ) = 0 for i 6= t, which completes the proof. References [1] I. Assem, T. Brustle and R. Schiffler, On the Galois coverings of a cluster-tilted algebra, J. Pure and App. Algebra 213(2009), no. 7, 1450-1463. [2] K. Bongartz and P. Gabriel, Covering spaces in representation theory, Invent. Math. 65(1982), 331-378. [3] A. Buan, R. Marsh and I. Reiten, Cluster-tilted algebras, Trans. Amer. Math. Soc. 459(2007), no. 1 , 323-332. [4] A. Buan, R. Marsh, M. Reineke, I. Reiten and G. Todorov, Tilting theory and cluster combinatorics, Advances in Math. 204(2006), 572-618. [5] P. Caldero, F. Chapton and R. Schiffler, Quivers with relations arising from clusters (An case), Trans. Aer. Math.Soc. 358(2006), 1347-1364. [6] J. Guo, The calculation of the Hall polynomials for a cyclic quiver, Comm. Algebra 23(1995), No. 2, 743-751. [7] J. Guo and L. Peng, Universal PBW-Basis of Hall-Ringel algebras and Hall polynomials, J. Algebra 198(1997), 339-351. [8] B. Keller, On triangualted orbit categories, Document Math. 10(2005), 551-581. [9] B. Keller and I. Reiten, Cluster-tilted algebras are Gorenstein and stably Calabi-Yau, Advances in Math. 211(2007), 123-151. [10] S. Konig and B. Zhu, From triangulated categories to abelian categories: cluster tilting in a general framework, Math. Zeit.258(2008), 143-160. [11] Y. Palu, Cluster characters for 2-Calabi-Yau triangulated categories, Ann. Inst. Fourier 58(2008), no. 6, 2221-2248. [12] C.M. Ringel, Lie algebras arising in representation theory, in, "London Math. Soc. Lecture Note Ser.," Vol.168, pp, 284-291, Cambridge Univ. Press, Cambridge, UK, 1992. [13] C.M. Ringel, The composition algebra of a cyclic quiver-Towards an explicit description of the quantum group of type eAn, Proc. London Math. Soc.(3)66(1993), 507-537. [14] B. Zhu, Cluster-tilted algebras and their intermediate coverings, Comm. Algebra 39(2011), 2437-2448. Department of Mathematics, SiChuan University, 610064 Chengdu, P.R.China E-mail address: [email protected]
1206.5650	3	1206	2013-01-07T14:01:05	Recollements from partial tilting complexes	[ "math.RA", "math.CT" ]	We consider recollements of derived categories of dg-algebras induced by self orthogonal compact objects obtaining a generalization of Rickard's Theorem. Specializing to the case of partial tilting modules over a ring, we extend the results on triangle equivalences proved in [B2] and [BMT]. In the end we focus on the connection between recollements of derived categories of rings, bireflective subcategories and generalized universal localizations".	math.RA	math	RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES SILVANA BAZZONI AND ALICE PAVARIN Abstract. We consider recollements of derived categories of differen- tial graded algebras induced by self orthogonal compact objects obtain- ing a generalization of Rickard's Theorem. Specializing to the case of partial tilting modules over a ring, we extend the results on triangle equivalences proved in [B] and [BMT]. In the end we focus on the con- nection between recollements of derived categories of rings, bireflective subcategories and "generalized universal localizations". A recollement of triangulated categories is a diagram 1. Introduction T ′ i∗ i∗ i! / T j! j ∗ j ∗ / T ′′ where the six functors involved are the derived version of Grothendieck's functors. In particular, they are paired in two adjoint triples, i∗ is fully faithful and T ′′ is equivalent to a Verdier quotient of T via j∗ so that the straight arrows can be interpreted as an exact sequence of triangulated cat- egories. The notion of recollements was introduced by Beilinson-Bernstein- Deligne [BBD] in a geometric context, where stratifications of varieties in- duce recollements of derived categories of constructible sheaves. The alge- braic aspect of recollements has become more and more apparent. Equivalence classes of recollements of triangulated categories are in bijec- tion with torsion-torsion-free triples, that is triples (X , Y, Z) of full trian- gulated subcategories of the central term T of a recollement, where (X , Y) and (Y, Z) are torsion pairs (see [N2, Section 9.2]). This bijection is studied in details by Nicolas and Saorin in [NS] for rec- ollements of derived categories of small flat differential graded categories (dg categories) and moreover, a parametrization of these recollements is given in terms of homological epimorphisms of dg categories. More precisely, the left end term of a recollement of the derived category D(B) of a small flat dg category B is the derived category of a dg category C linked to B via a Date: October 19, 2018. 1991 Mathematics Subject Classification. 16D90; 16E30; 18E30; Secondary: 16S90; 16G10. Key words and phrases. Recollements, partial tilting complex, dg algebras, homological epimorphism. Research supported by grant CPDA105885/10 of Padova University "Differential graded categories". 1 / b b z z / b b { { 2 S. BAZZONI AND A. PAVARIN homological epimorphism and such that the central term Y of the torsion- torsion-free triple associated to the recollement is the essential image of the functor i∗. In Section 2 we consider the particular case of the derived category D(B) of a k-flat differential graded algebra B and define explicitly a homological epimorphism of differential graded algebras (dg algebras) B → C associated to a torsion-torsion-free triple on D(B). In Section 3 we recall some instances of recollements of derived cate- gories of differential graded algebras as the ones provided by Jorgensen in [J]. There, starting from results in [DG], [Mi] and [N], such recollements are characterized in terms of derived functors associated to two objects, one compact and the other self-compact. But in [J] there is no mention of the connection between the recollements involved and homological epimor- phisms of dg algebras. There is also a strong connection between recollements and tilting theory as shown by Koenig and Ageleri-Koenig-Liu in [K] and [AKL], by considering recollements of derived categories of rings. In Section 4 we specialize the situation to the case of self-orthogonal compact dg modules, which we call partial tilting. Our result (Theorem 4.3), can be viewed as a generalization of the Morita-type theorem proved by Rickard in [R]. In fact, if P is a partial tilting right dg module over a dg algebra B, we can use a quasi-isomorphism between the endomorphism ring L ⊗ B −) induces an equivalence between the quotient of D(B) modulo the full A of P and the dg endomorphism of P , to show that the functor (P triangulated subcategory Ker(P −) and the derived category D(A). Thus, if P is moreover a tilting complex over a ring B with endomorphism ring A, L ⊗ B then Ker(P −) is zero and we recover Rickard's Theorem. L ⊗ B In Section 5 we consider applications to the case of classical partial tilting right modules T over a ring B, that is partial tilting complexes concentrated in degree zero. As examples of this case we start with a possibly infinitely generated left module AT over a ring A, which is self-orthogonal viewed as a complex concentrated in degree zero and finitely generating the ring A in D(A) (see Section 5 for the definitions). Under these assumptions, T , viewed as a right module over its endomorphism ring B, is a faithfully balanced classical partial tilting module and applying Proposition 5.2 we obtain a generalization of the result proved in [BMT] where the stronger assumption on AT to be a "good tilting module" was assumed. Moreover, this setting provides an instance of the situation considered in [Y]. In Section 6, given a classical partial tilting module TB over a ring B, we look for conditions under which the class Y = Ker(T −) is equivalent to the derived category of a ring S for which there is a homological ring epimorphism λ : B → S. We show that this happens if and only if the perpendicular subcategory E consisting of the left B-modules N such that TorB i (T, N ) = 0 for every i ≥ 0, is bireflective and every object of Y is quasi-isomorphic to a complex with terms in E. L ⊗ B RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 3 In Section 7 we show that, if such a ring S exists, it plays the role of a "generalized universal localization" of B with respect to a projective reso- lution of TB and some properties of this localization are illustrated. The situation considered in Sections 6 and 7 is a generalization of a recent In fact, in [CX], completing the results article by Chen and Xi ( [CX]). proved in [B] for "good" 1-tilting modules T over a ring A, it is shown that the derived category D(B) of the endomorphism ring of T is the central term of a recollement with left term the derived category of a ring which is a universal localization in the sense of Schofield. We note that in our approach, instead, we fix a ring B and obtain rec- ollements of D(B) for every choice of classical partial tilting modules over B, while starting with an infinitely generated good tilting module AT over a ring A one obtains a recollement whose central term is the derived cate- gory of the endomorphism ring of AT and this ring might be very large and difficult to handle. In Section 8 we consider examples of classical partial n-tilting modules TB (with n > 1) over artin algebras B showing different possible behaviors In some examples there exists a of the associated perpendicular class E. homological ring epimorphism λ : B → S such that the class Y = Ker(T L ⊗ B −) is triangle equivalent to D(S) (Example 1). But in these cases the modules TB are not arising from good n -tilting modules. Chen and Xi in [CX2] consider the case of a good n-tilting module AT (with n > 1) over a ring A with endomorphism ring B and investigate the problem to decide when the recollement of D(B) induced by T corresponds to a homological epimorphism. They prove some necessary and sufficient conditions and show some counterexamples (see Example 4), but the prob- lem in its full generality, remains open. 2. Preliminaries In this section we recall some definitions and preliminary results that will be useful later on. Let T be a triangulated category admitting small coproducts (also called set indexed coproducts) and let C be a class of objects in T . Then: (1) Tria C denotes the smallest full triangulated subcategory of T con- taining C and closed under small coproducts, (2) tria C denotes the smallest full triangulated subcategory of T con- taining C and closed under finite coproducts and direct summands. Moreover, indicating by [−] the shift functor, we define the following classes: C⊥ = {X ∈ T HomT (C[n], X) = 0, for all C ∈ C, for all n ∈ Z}; ⊥C = {X ∈ T HomT (X, C[n]) = 0, for all C ∈ C, for all n ∈ Z}. An object X ∈ T is called self-orthogonal if HomT (X, X[n]) = 0, for all 0 6= n ∈ Z. 4 S. BAZZONI AND A. PAVARIN The object X ∈ T is called compact if the functor HomT (X, −) commutes with small coproducts. M in T is called self-compact if M is compact in Tria M . Differential graded algebras and differential graded modules. We review the notions of dg algebras, dg modules and of the derived category of dg modules. For more details see [Ke1], [Ke2], [Ke3] or [P]. Let k be a commutative ring. A differential graded algebra over k (dg Bp endowed with a differential d k-algebra) is a Z-graded k-algebra B = ⊕ p∈Z of degree one, such that: d(ab) = d(a)b + (−1)pad(b) for all a ∈ Bp, b ∈ B. In particular, a ring is a dg Z-algebra concentrated in degree 0. Let B be a dg algebra over k with differential dB. A differential graded M p (left) B-module (dg B-module) is a Z-graded (left) B-module M = ⊕ p∈Z endowed with a differential dM of degree 1 such that dM (bm) = bdM (m) + (−1)pdB(b)m for all m ∈ M p, b ∈ B. A morphism between dg B-modules is a morphism of the underlying graded B-modules, homogeneous of degree zero and commuting with the differentials. A morphism f : M → N of dg B-modules is said to be null- homotopic if there exists a morphism of graded modules s : M → N of degree −1 such that f = sdM + dN s. In the sequel we will simply talk about a dg algebra without mentioning the ground ring k. The left dg B-modules form an abelian category denoted by C(B). The homotopy category H(B) is the category with the same objects as C(B) and with morphisms the equivalence classes of morphisms in C(B) modulo the null-homotopic ones. The derived category D(B) is the localization of H(B) with respect to the quasi-isomorphisms, that is morphisms in C(B) inducing isomorphisms in homology. H(B) and D(B) are triangulated categories with shift functor which will be denoted by [−]. We denote by Bop the opposite dg algebra of B. Thus, dg right B-modules will be identified with left dg Bop-modules. Also D(Bop) will denote the derived category of right dg B-modules. We say that M is a dg B-A-bimodule if it is a left dg B-module and a left dg Aop-module, with compatible B and Aop module structure. In this case we also write BMA. Cdg(B) denotes the category of dg B-modules. If M , N are dg B-modules the morphism space HomCdg (B)(M, N ) is the complex HomB(M, N ) with [HomB(M, N )]n = HomB(M, N [n]) (here HomB(M, N ) denotes the group of morphisms of graded B-modules, homogeneous of degree zero) and dif- ferential defined, for each f ∈ [HomB(M, N ]n, by d(f ) = dN ◦ f − (−1)nf ◦ dM . RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 5 Thus, Hn(HomB(M, N )) = HomH(B)(M, N [n]). Observe that, if X is a dg B-module, then HomB(X, X) is a dg algebra called the dg endomorphism ring of X. Thus, the morphism space HomB(M, N ) of dg B-modules M , N in the category C(B) is Z 0(HomCdg (B)(M, N )) and the morphism space in the ho- motopy category H(B) is H0(HomCdg(B)(M, N )). Definition 2.1. A dg B-module is acyclic if it has zero homology and a dg A-module P (resp. I) is called H-projective (resp. H-injective) if HomH(B)(P, N ) = 0 (resp. HomH(B)(N, I) = 0) for all acyclic dg B-modules N . For every dg B-module X there is an H-projective dg B-module pX and an H-injective dg B-module iX such that X is quasi-isomorphic to pX and to iX and such that HomD(B)(X, Y ) = HomH(B)(pX, Y ) = HomH(B)(X, iY ). Definition 2.2. (see e.g. [P]) Let M be a dg B-module. The functor HomB(M, −) induces a total right derived functor RHomB(M, −) : D(B) → D(k) such that RHomB(M, N ) = HomB(M, iN ) = HomB(pM, N ), for ev- ery dg B module N . Moreover, if M is a B-A dg bimodule, then HomB(M, −) induces a to- tal derived functor RHomB(M, −) : D(B) → D(A) and RHomB(M, N ) = HomB(M, iN ) ∈ D(A), for every dg B-module N . If N is a dg Bop-module, then N ⊗B − induces a total left derived functor ⊗B M = pN ⊗BM = N ⊗BpM , for ⊗B − : D(B) → D(k) such that N L L N every dg B-module M . If moreover, N is a dg A-B-bimodule, then N ⊗B − induces a derived L L functor N ⊗B − : D(B) → D(A) where N ⊗B M = N ⊗BpM . Definition 2.3. (see [Ke1, Sec 2.6]) A dg B-module X is called perfect if it is H-projective and compact in D(B). The full subcategory of H(B) consisting of perfect dg B-modules is denoted by per B; it coincides with the subcategory tria B of H(B). By Ravenel-Neeman's result, an object of D(B) is compact if and only if it is quasi-isomorphic to a perfect dg B-module. If B is an ordinary algebra, then the perfect complexes are the bounded complexes with finitely generated projective terms and the category per B is also denoted by Hb p(B). Recollements and TTF. In this subsection we recall the notion of a rec- ollement of triangulated categories and the correspondence with torsion- torsion-free triples. The concept of recollement was introduced by Beilin- son, Bernstein and Deligne in [BBD] to study "exact sequences" of derived categories of constructible sheaves over geometric objects. Definition 2.4. ([BBD]) Let D, D′ and D′′ be triangulated categories. D is said to be a recollement of D′ and D′′, expressed by the diagram 6 S. BAZZONI AND A. PAVARIN (1) D′′ i∗ i∗=i! i! / D j! j!=j ∗ j∗ / D′ if there are six triangle functors satisfying the following conditions: (1) (i∗, i∗), (i!, i!), (j!, j!) and (j∗, j∗) are adjoint pairs; (2) i∗, j∗ and j! are fully faithful functors; (3) j!i! = 0 (and thus also i!j∗ = 0 and i∗j! = 0); (4) for each object C ∈ D, there are two triangles in D : i!i!(C) −→ C −→ j∗j∗(C) −→ i!i!(C)[1], j!j!(C) −→ C −→ i∗i∗(C) −→ j!j!(C)[1]. In the sequel, if F : D → D′ is a triangulated functor between two trian- gulated categories D and D′, we will denote by Im F the essential image of F in D′. We will make frequent use of the following well known properties of rec- ollements (see [BBD, Proposition 1.4.5]. Lemma 2.5. Given a recollement as in Definition 2.4 the following hold true: (1) Im i∗ = Kerj∗. (2) D/Kerj∗ is triangle equivalent to D′. Definition 2.6. Two recollements defined by the data (D, D′, D′′, i∗, i∗, i!, j!, j!, j∗) and (D, T ′, T ′′, i′∗, i′ equality between essential images holds : ∗) are said to be equivalent if the following ∗, i′!, j′!, j′ ! , j′ (Im (j!), Im (i∗), Im (j∗)) = (Im (j′ ! ), Im (i′ ∗), Im (j′ ∗)). Let D be a triangulated category. Definition 2.7. A torsion pair in D is a pair (X , Y) of full subcategories of D closed under isomorphisms satisfying the following conditions: (1) HomD(X , Y) = 0; (2) X[1] ⊆ X and Y [−1] ⊆ Y for each X ∈ X and Y ∈ Y; (3) for each object C ∈ D, there is a triangle XC −→ C −→ YC −→ XC [1] in D with XC ∈ X and YC ∈ Y. In this case X is called a torsion class and Y a torsion free class. Definition 2.8. A torsion-torsion-free triple (TTF triple) in D is a triple (X , Y, Z) of subcategories X , Y and Z of D such that both (X , Y) and (Y, Z) are torsion pairs in D. In this case X , Y, and Z are triangulated subcategories of D. We recall the following important result: / e e y y / e e y y RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 7 Lemma 2.9. ([N2, Section 9.2], [NS, Section 2.1]) Let D be a triangulated category. There is a bijection between: (1) TTF triples (X , Y, Z) in D; (2) equivalence classes of recollements for D. In particular, given a recollement as in Definition 2.4, the triple (Im j!, Im i∗, Im j∗) is a TTF triple in D. Homological epimorphisms of dg algebras. In this subsection we recall the correspondence between bireflective subcategories of module categories and ring epimorphisms. Then we introduce the definition of homological epimorphisms of dg algebras and we state the correspondence between them, TTF triples and recollements. It is well known (see e.g. [GL, Theorem 4.4]) that a ring homomorphism f : R −→ S between two rings R and S is a ring epimorphism if and only if the multiplication map S ⊗R S −→ S is an isomorphism of S-bimodules or, equivalently, if the restriction of scalars functor f∗ : S-Mod → R-Mod is fully faithful. Two ring epimorphisms f : R −→ S and g : R −→ S′ are said to be equivalent if there exists an isomorphism of rings h : S −→ S′ such that hf = g. We recall now the well known bijection existing between equivalence classes of ring epimorphisms and bireflective subcategories of module cate- gories. Definition 2.10. Let E be a full subcategory of R-Mod. A morphism f : M −→ E, with E in E, is called an E-reflection if for every map g : M −→ E′, with E′ in E , there is a unique map h : E −→ E′ such that hf = g. A subcategory E of R-Mod is said to be reflective if every R- module X admits an E-reflection. The definition of coreflective subcategory is given dually. A subcategory that is both reflective and coreflective is called bireflective. Remark 1. It is clear that a full subcategory E of R-Mod is reflective if and only if the inclusion functor i : E −→ R-Mod admits a left adjoint. Dually, a subcategory X is coreflecting if and only if the inclusion functor j : X −→ R-Mod admits a right adjoint. Lemma 2.11. ([GL] and [GP]) Let E be a full subcategory of R-Mod. The following assertions are equivalent: 1) E is a bireflective subcategory of R-Mod; 2) E is closed under isomorphic images, direct sums, direct products, ker- nels and cokernels; 3) there is a ring epimorphism f : R −→ S such that E is the essential image of the restriction of scalars functor f∗ : S-Mod → R-Mod. In particular there is a bijection between the bireflective subcategory of R-Mod and the equivalence classes of ring epimorphisms starting from R. Moreover the map f : R −→ S as in 3) is an E-reflection. A ring epimorphism f : R −→ S is said to be homological if S ⊗R S = S in D(S) or, equivalently, if the restriction of scalars functor f∗ : D(S) → D(R) is fully faithful (see [GL, Section 4]). L 8 S. BAZZONI AND A. PAVARIN Two homological epimorphisms of rings f : R −→ S and g : R −→ S′ are said to be equivalent if there exists an isomorphism of rings h : S −→ S′ such that hf = g. The concept of homological epimorphism of rings can be"naturally" generalized to the setting of dg algebras ([P]) and to the more general setting of dg categories ([NS]). Here we give the definition of homological epimorphism of dg algebras and its characterization at the level of derived categories. Definition 2.12. [P, Theorem 3.9] Let F : C −→ D be a morphism between two dg algebras C and D. Then F is called a homological epimorphism of dg algebras if the the canonical map D equivalently if the induced functor F∗ : D(D) −→ D(C) is fully faithful. ⊗C D → D is an isomorphism, or L Definition 2.13. Two homological epimorphisms of dg algebras F : C −→ D and G : C −→ D′ are said to be equivalent if there exists an isomorphism of dg k-algebras H : D −→ D′ such that HF = G. Remark 2. From the definitions it is clear that a homological epimorphism of rings is exactly a homological epimorphism of dg Z-algebras concentrated in degree 0. In [NS, Theorem 5] it is proved that for a flat small dg category B there are bijections between equivalence classes of recollements of D(B), TTF triples on D(B) and equivalence classes of homological epimorphisms of dg categories F : B → C. Moreover, in [NS, Lemma 5] it is observed that every derived category of a small dg category is triangle equivalent to the derived category of a small flat dg category. To achieve this result one uses construction of a model structure on the category of all small dg categories defined by Tabuada (see [T].) We now state [NS, Theorem 4] for the case of a flat dg k-algebra and we give a proof, since in this case the construction of the homological epimor- phism becomes more explicit and it will also be used later on in Theorem 6.1. Lemma 2.14. Let B be a dg algebra flat as a k-module and (X , Y, Z) be a TTF triple in D(B). Then there is a dg algebra C and a homological epimorphism F : B → C such that Y is the essential image of the restriction of scalars functor F∗ : D(C) −→ D(B). Proof. Since (X , Y, Z) is a TTF triple in D(B) there exists a triangle (2) X −→ B ϕB−→ Y −→ X[1], with X ∈ X and Y ∈ Y. where ϕB is the unit morphism of the adjunction. Without loss of generality, we may assume that Y is an H-injective left dg B-module and that ϕB is a morphism in C(B). Let E = RHomB(Y, Y ) = HomB(Y, Y ), then BYE is a dg B-E-bimodule. Applying the functor RHomB(−, Y ) to the triangle (2) we obtain a triangle in the derived category D(Eop): RHomB(X[1], Y ) −→ RHomB(Y, Y ) β −→ RHomB(B, Y ) −→ RHomB(X, Y ). RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 9 where β = RHomB(ϕB, Y ) = ϕ∗. Since X ∈ X , Y ∈ Y and Y is H-injective, we have, for each i, n ∈ Z: HnRHomB(X[i], Y ) ∼= HomD(B)(X[i], Y [n]) = 0 Therefore we deduce that β is a quasi-isomorphism, so we have (3) E = RHomB(Y, Y ) β ≃ RHomB(B, Y ) γ ≃Y in D(Eop). Let ξ : Y → Y ′ be a quasi isomorphism of dg B-E-bimodules such that Y ′ is an H-injective resolution of Y as a dg B-E-bimodule. Since B is assumed to be k-flat, we have that the restriction functor from dg B-E- bimodules to dg E-modules preserves H-injectivity. In fact, its left adjoint B ⊗k − preserves acyclicity. Then, Y ′ E is an H-injective right dg E-module. Consider the dg algebra C = HomEop(BY ′ E, BY ′ E) and a morphism of dg algebras defined by: E) = RHomEop(BY ′ E, BY ′ F : B −→ C b 7−→ F (b) : y′ 7−→ (−1)by′by′, where · denotes the degree. Since Y ′ E is H-injective we have quasi-isomorphisms: C = RHomEop(BY ′ E, BY ′ E) ξ∗→ RHomEop(BYE, BY ′ E) β∗→ RHomEop(E, BY ′ E) ∼= Y ′. We regard C as a dg B-B-bimodule with the action induced by F , so F is also a morphism of dg B-B-bimodules and the morphism β∗ ◦ ξ∗ : C → Y ′ is a quasi-isomorphism of left dg B-modules; moreover, ξ ◦ ϕB = β∗ ◦ ξ∗ ◦ F . Now define the morphism ε := ξ−1 ◦ β∗ ◦ ξ∗ : C → Y in D(B). Then ε is a quasi isomorphism of left dg B-modules such that ε ◦ F = ϕ and we get an isomorphism of triangles: X X / B F / C ε / X[1] . / B ϕB / / Y / X[1] Consider the restriction of scalars functor F∗ : D(C) −→ D(B). F∗ is a triangulated functor admitting a right adjoint, hence it commutes with small coproducts. Moreover, F∗(C) =B C ∼= Y ′ ∼= Y ∈ Y, hence F∗(Tria C) = F∗(D(C)) is a subcategory of Y, closed under coproducts and containing the generator BY . Now we notice that, F being a morphism of dg B-B- bimodules, one has a triangle of B-B bimodules: (4) X −→ B F −→ C −→ X[1]. Consider the adjunction: C L ⊗ B − F ∗ / D(B) D(C) / / / / / / t t 10 S. BAZZONI AND A. PAVARIN and let M ∈ X and N in D(C), then HomD(C)(C L ⊗ B M, N ) ∼= HomD(B)(M, F ∗(N )) = 0 since F ∗(N ) ∈ Y. Then C L ⊗ B M = 0 for each M ∈ X . Hence, applying the functor C L ⊗ B − to the triangle (4), we obtain C L ⊗ B B ∼= C L C, ⊗ B which shows that F is a homological epimorphism of dg algebras. In par- ticular, Im F∗ is a triangulated subcategory of Y, hence Im F∗ = Y, by the principle of infinite d´evissage. (cid:3) 3. Recollements from compact objects In this section we consider recollements between dg algebras induced by compact objects. Our approach follows the exposition in [J] which general- izes to dg algebras the situation considered in [DG] for derived categories of rings. We collect in the next lemma some self-explanatory facts. Lemma 3.1. Let B be a dg algebra and let Q be an H-projective left dg B-module compact in D(B) (i.e. Q is a perfect left dg B-module (2.3)). Consider the dg endomorphism ring D of Q, that is D = HomB(Q, Q); then Q becomes a dg B-D-bimodule. Let P = Q∗ = RHomB(Q, B), then P is an H-projective right dg B-module and compact in D(Bop); moreover P is a dg D-B-bimodule. The following hold true: (1) ([DG, Sec 2.5] or [J, Sec 2.1]) The functors H = RHomB(Q, −), G = P L ⊗ B − : D(B) → D(D). are isomorphic. (2) The functor HomB(−, B) induces an equivalence HomB(−, B) : per B → per Bop with inverse HomBop(−, B). Thus, P ∗ = RHomB(P, B) is isomorphic to Q. (3) From (2) it follows that the functor HomB(−, B) : C(B) → C(Bop) induces a quasi-isomorphism between the dg algebras HomB(Q, Q) and HomBop (P, P ). Thus we can identify HomBop (P, P ) with D. Setup 3.2. In the notations of Lemma 3.1, we set Y := Ker(P L ⊗ B −). It is well known that the inclusion functor inc : Y −→ D(B) admits both left and right adjoints L, R : D(B) −→ Y and that Y is generated by L(B). Moreover, L(B) is self-compact, since the inclusion functor preserves coproducts. (See also, [NS, Lemma 2.3].) RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 11 The following result appears already in the literature in different forms. In particular, statement (1) can be found in papers by Dwyer Greenless [DG, Sec. 2], Miyachi [Mi, Proposition], Jørgensen [J, Proposition 3.2] and [Y, Theorem 1]. Lemma 3.3. In the notations of Lemma 3.1 and Setup 3.2 the following hold true: (1) There is a recollement Y = Q⊥ L i∗=inc R Q L ⊗ D − / D(B) RHomB(Q,−)∼=P L ⊗B − / D(D) RHomD(P,−) (2) (cid:0)Tria Q, Q⊥, Im RHomD(P, −)(cid:1) is a TTF triple in D(B). Proof. (1) We give an alternative proof following the arguments used by Yang in the proof of [Y, Theorem 1]. We first show that the functor j! = Q L ⊗ D − is fully faithful. By construction we have that Q L ⊗ D − induces an equivalence between In other words the pair (D, BQD) is a standard lift tria D → tria Q. (see [Ke2, Sec.7]). The functor j! commutes with set index coproducts, its restriction to tria D is fully faithful and j!(D) = Q is a compact object. Thus by [Ke2, Lemma 4.2 b] we conclude that j! is fully faithful, since D is a compact generator of D(D). So the functor RHomB(Q, −) ∼= (P L ⊗ D −) has a fully faithful left adjoint and a right adjoint RHomD(P, −). By [Mi, Proposition 2.7], the functor RHomD(P, −) is fully faithful, so the right part of the diagram in the state- ment can be completed to a recollement with left term the kernel of the functor RHomB(Q, −), which coincides with the category BQ⊥, since Q is a compact object. (2) To prove the statement it is enough to show that Tria Q is the essential image of the functor j! = Q L ⊗ D −. This follows from the facts that the fully faithful functor Q L ⊗ D − is a triangle functor which commutes with coproducts and sends the compact generator D of the category D(D) to the object Q of D(B), hence its image is Tria Q. (cid:3) Remark 3. In the notation of Setup 3.2, let E := RHomB(L(B), L(B)). Then by Keller's theorem ([Ke4, Theorem 8.7]) there is a derived equivalence / g g w w / g g w w 12 S. BAZZONI AND A. PAVARIN D(E) ≃ Y which can be illustrated by the following diagram: L L(B) ⊗E − D(E) % Y RHomB (L(B),−) Combining the above remark with Lemmas 2.14 and 3.3 we state in the next proposition the main result of this section. The second part of the statement can be viewed as a generalization of [J, Theorem 3.3], since it characterizes the left term of the recollement as the derived category of a dg algebra obtained by a homological epimorphism. Proposition 3.4. In the notations of Lemma 3.1 and Setup 3.2 there is a recollement RHomB (L(B),−)◦L L inc ◦L(B) ⊗E − D(E) Q L ⊗ D − / D(B) RHomB (Q,−)∼=P L ⊗B − / D(D) RHomB (L(B),−)◦R RHomD(P,−) If B is moreover a k-flat dg algebra, there is a homological epimorphism of dg algebras F : B → C and a recollement: i∗=C L ⊗ B − F∗ D(C) Q L ⊗ D − / D(B) RHomB (Q,−)∼=P L ⊗B− / D(D) i!=RHomB (C,−) RHomD(P,−) such that the essential image of F∗ is Y. In particular, if B ∈ tria Q, then Y vanishes and the functor RHomD(P, −) induces an equivalence between D(D) and D(B) with inverse P L ⊗ B −. 4. Partial Tilting dg modules In this section we specialize the situation illustrated by Proposition 3.4 to the case of self-orthogonal perfect dg modules. Our next result, Theorem 4.3, can be viewed as a generalization of the Morita-type theorem proved by Rickard in [R] in the sense that we consider partial tilting dg modules instead of tilting complexes. Note that some generalizations were obtained also by Koenig in [K] in the case of bounded derived categories of rings. % f f / g g w w / g g w w / g g w w / g g w w RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 13 Definition 4.1. Let B be a dg algebra. A right (left) dg B-module P is called partial tilting if it is perfect and self orthogonal, i.e. HomD(Bop)(P, P [n]) = 0 (HomD(B)(P, P [n]) = 0), for every 0 6= n ∈ Z. A right (left) dg B-module P is called tilting if it is partial tilting and Bop ∈ tria P (B ∈ tria P ). By Lemma 3.1 we have that if BQ is a partial tilting left dg B-module, then P = RHomB(Q, B) is a partial tilting right dg B-module and P ∗ = RHomBop (P, B) is isomorphic to Q. Moreover, D = RHomB(BQ,B Q) ∼= RHomBop (PB, PB). Stalk algebras 4.2. Let P be a partial tilting right dg B module. Let D = RHomBop (PB, PB) and A = HomD(Bop)(P, P ). Then, Hn(D) ∼= HomD(Bop)(P, P [n]) = 0, for every 0 6= n ∈ Z, hence the dg algebra D has homology concentrated in degree zero and H0(D) ∼= A. Thus, by [Ke4, Sec. 8.4] there is a triangle equivalence ρ : D(D) → D(A). For later purposes we give explicitly the functors defining this equivalence and its inverse. Let τ≤0 be the truncation functor and consider the subalgebra D− := τ≤0(D) = · · · → D−2 → D−1 → Z 0(D) → 0 → . . . Then the inclusion f : D− → D and π : D− → H0(D) = A are quasi- isomorphisms of dg algebras, inducing equivalences f∗ and π∗ between the corresponding derived categories. Thus we have the following diagrams: D L ⊗ D− − f∗ D(D) A L ⊗ D− − / D(D−) D(D−) π∗ D(A) RHomD− (D,−) RHomD− (A,−) So ρ = (A L ⊗ D− −) ◦ f∗ and its inverse is ρ−1 = (D L ⊗ D− −) ◦ π∗. Note that f∗ ∼= D−D L ⊗ D − and π∗ ∼= D−A L ⊗ A −. As a special case of Proposition 3.4 we obtain a recollement where the right term is the derived category of a ring. Theorem 4.3. Let B be a dg algebra and let P be a partial tilting right dg B-module. Let A = HomD(Bop)(P, P ), Q = RHomBop (P, B). Then there exists a dg algebra E and a recollement: RHomB(L(B),−)◦L D(E) L(B)⊗L E − / D(B) RHomB (L(B),−) j! j ∗ j∗ / D(A) / b b ! ! = = o o / e e y y / e e y y 14 S. BAZZONI AND A. PAVARIN where, letting D = RHomBop (P, P ) there is a triangle equivalence ρ : D(D) → D(A) such that: L ⊗ D (1) j! = (Q −) ◦ ρ−1. L ⊗ B −). (2) j∗ = ρ ◦ (P (3) j∗ = RHomD(P, −) ◦ ρ−1. (4) D(A) is triangle equivalent to D(B)/Ker (j∗). In particular, if P is a tilting right dg B-module, then Y vanishes and ρ ◦ (P L ⊗ B −) : D(B) → D(A) is a triangle equivalence with inverse RHomD(P, −) ◦ ρ−1. Moreover, if B is k-flat there exists a homological epimorphism of dg algebras F : B −→ C such that the above recollement becomes: D(C) i∗=C L ⊗ B − F∗ i!=RHomB(C,−) / D(B) j! j ∗ j∗ / D(A) Proof. By Lemma 3.1 we can identify P with RHomB(Q, B) and RHomBop (P, P ) with RHomB(Q, Q). The existence of an equivalence ρ : D(D) → D(A) is ensured by Stalk algebras 4.2. An application of Lemma 3.3 (4) proves the statement. If B is k-flat we use Proposition 3.4 to conclude. (cid:3) Remark 4. Rickard's Theorem states that if B is a flat k-algebra over a commutative ring k and PB is a tilting complex of right B-modules with endomorphism ring A, then there is a complex AXB, with terms that are A-B bimodules, isomorphic to PB in D(B), and such that X L ⊗ B − : D(B) → D(A) is an equivalence with inverse the functor RHomA(X, −). Equivalences of this form are called standard equivalences (see [Ke1, Sec. 1.4]). It is still an open problem to decide if all triangle equivalences between derived categories of rings (or dg algebras) are isomorphic to standard equivalence (see [Ke4, Sec.6.1]). In the same assumptions as in Rickard's Theorem, but without any flat- ness condition on B, our Theorem 4.3 provides an equivalence between D(B) and D(A). An analysis of the way in which this equivalence is constructed, shows that it is induced by the composite derived functor A L ⊗ D− (D−P L ⊗ B −) where D− = τ≤0(D) and P is viewed as a dg D−-B-bimodule. 5. Tilting and partial tilting modules The notion of tilting modules goes back to works by Bernstein, Gel'fand and Ponomarev, Brenner and Butler, Happel and Ringel, Auslander, Platzeck and Reiten [BGP, BB, HR, APR], and it has first been considered in the case / g g w w / g g w w RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 15 of finitely generated modules of projective dimension at most one over artin algebras and later generalized to arbitrary rings and to possibly infinitely generated modules of finite projective dimension (see [AC], [CF], [CT]). We recall the definition of (partial) tilting modules over a ring R by using the canonical embedding of the category of R-modules into the derived category D(R) and we restate the various definitions using the terminology of derived categories theory. Definition 5.1. Let R be a ring and T an R-module. Consider the following conditions on T viewed as an object of D(R) under the canonical embedding: (T1) T is isomorphic to a bounded complex with projective terms; (T1') T is a compact object of D(R); (T2) T is orthogonal to coproducts of copies of T , that is HomD(R)(T, T (α)[n]) = 0 for every 0 6= n ∈ Z and every set α. (T2') T is self orthogonal, that is HomD(R)(T, T [n]) = 0 for every 0 6= n ∈ Z. (T3) R ∈ Tria T . (T3') R ∈ tria T . If the projective dimension of T is at most n, then T is called a classical n-tilting module, if it satisfies (T1'), (T2') and (T3') and a classical partial n-tilting module if it satisfies (T1') and (T2'). T is called an n-tilting module (possibly infinitely generated), if it satisfies (T1), (T2) and (T3) and it is called a good n-tilting module if it satisfies (T1), (T2) and (T3'). Classical tilting modules were introduced mainly to generalize Morita theory. They provide equivalences between suitable subcategories of module categories (see [BB], [Miy]). In [H] and [CPS] it was shown that a classical n-tilting module over a ring A with endomorphism ring B induces a triangle equivalence between D(A) and D(B). Infinitely generated tilting modules do not provide equivalences between derived categories of rings, but the first named author proved in [B] that, if T is a good 1-tilting module over a ring A with endomorphism ring B, then the total left derived functor T L ⊗ B − induces an equivalence between D(B)/Ker(T −) and D(A). This result has been generalized in [BMT] to L ⊗ B the case of good n-tilting modules. Recently Chen and Xi [CX] completed the result proved in [B] for good 1-tilting modules T over a ring A, by showing that the derived category of the endomorphism ring B of T is the central term of a recollement with right term D(A) and left term the derived category of a ring C for which there is a homological ring epimorphism λ : B → C where moreover C is a universal localization of a suitable morphism between finitely generated projective B-modules. The disadvantage of starting with an infinitely generated n-tilting module AT over a ring A, is that a good n-tilting module T ′ equivalent to AT is obtained as a summand of a possibly infinite direct sum of copies of T and this procedure produces a very large endomorphism ring B of T ′. So the 16 S. BAZZONI AND A. PAVARIN recollement induced by T ′ concerns the derived category of a ring which is hardly under control. In our approach, instead, we can fix a ring B and obtain recollements of D(B) for every choice of classical partial tilting modules. Here we note that if a (possible infinitely generated) module AT over A satisfies conditions (T2') and (T3'), then, by [Miy, Proposition 1.4 (2)], T is a partial classical n- tilting module over its endomorphism ring B and, moreover, EndB(T ) ∼= A. In particular, we can apply the results of Section 4. This allows to obtain the same conclusion as in [BMT], but with weaker hypotheses on the module AT , namely without asking that it is a good n-tilting A-module, but only that it satisfies conditions (T2') and (T3'). More precisely a direct consequence of Theorem 4.3 yields the following proposition which can be viewed as a generalization of [BMT, Theorem 2.2]. Proposition 5.2. Let B be a ring and let TB be a classical partial n-tilting module with endomorphism ring A. Keeping the notations in Setup 3.2 there is a dg algebra E and a recollement RHomB(L(B),−)◦L L inc ◦L(B) ⊗E − / D(B) D(E) j! L j ∗=T ⊗B − / D(A) RHomB(L(B),−)◦R j∗=RHomA(T,−) where: (1) j∗ = RHomA(T, −) is fully faithful; (2) D(A) is triangle equivalent to D(B)/Ker (T L ⊗ B −). Moreover, if B is k-flat, there is a homological epimorphism of dg algebras F : B → C and the recollement above becomes i∗=C L ⊗ B − F∗ D(C) j! L j ∗=T ⊗B − / D(A) / D(B) i!=RHomB(C,−) j∗=RHomA(T,−) Proof. Let P be a projective resolution of the module T in Mod-B. Then P is a partial tilting complex of D(B) so that we may apply Theorem 4.3 which states that there is a triangle equivalence ρ : D(D) → D(A) where D = RHomBop(P, P ). As shown in Stalk algebras 4.2 we have: ρ = (A L ⊗ D− −) ◦ f∗. where f∗ : D(D) → D(D−) is the restriction of scalar functors induced by the quasi-isomorphism of dg algebras f : D− → D / e e y y / e e y y / g g w w / g g w w RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 17 To conclude the proof we must show that (a) ρ ◦ (P −) ∼= T L ⊗ B L ⊗ B −, (b) RHomD(P, −) ◦ ρ−1 ∼= RHomA(T, −). We first prove (a). Let σ : PB → TB be a morphism of complexes inducing a quasi-isomorphsm in D(B). From the dg algebra morphisms f : D− → D and π : D− → A we have that P and T are left dg D−-modules. Checking the action of the dg algebra D− on P and T we see that σ is a morphism of dg D−-modules. Thus, σ is a quasi isomorphism between P and T as dg D−-B-bimodules. This implies that the functors P L ⊗ B − and T L ⊗ B − from D(B) to D(D−) are isomorphic (see [Ke2, Lemma 6.1 b]). Consequently, , in the notations of Stalk algebras 4.2, we have: j∗ = ρ ◦ (P L ⊗ B −) = (A L ⊗ D− −) ◦ f∗ ◦ (P −) ∼= (A L ⊗ B L ⊗ D− −) ◦ (D−P −) ∼= L ⊗ B ∼= ((A L ⊗ D− −) ◦ (D−T −) ∼= (A L ⊗ B L ⊗ D− −) ◦ π∗ ◦ (AT L ⊗ B −). −) ◦ π∗ is isomorphic to the identity of D(A), we conclude that Since (A j∗ ∼=A T L ⊗ D− L ⊗ B −. Next, from the uniqueness of right adjoints up to isomorphisms, we also get RHomD(P, −) ◦ ρ−1 ∼= RHomA(T, −). (cid:3) Note 5.3. In the assumption of Proposition 5.2 if we let Q = RHomBop(T, B), then, by lemma 3.1 (1) we have RHomB(Q, −) ∼= P −, hence also L ⊗ B f∗ ◦ RHomB(Q, −) ∼= f∗ ◦ (P −) ∼= AT L ⊗ B L ⊗ B −. 6. The case of homological epimorphism of rings As said in the introduction, the situation in which a compactly generated triangulated category is a recollement of triangulated categories compactly generated by a single object has been studied by many authors ( [K], [NS, Corollary 3.4] and [AKL, Sec. 1.5]). We are interested in studying the case in which the subcategory Y = Ker(T −) in Proposition 5.2 is equivalent to the derived category of a ring L ⊗ B via a homological ring epimorphism. The problem is related to the notion of bireflective and perpendicular categories (see Definition 2.10 and Lemma 2.11.) 18 S. BAZZONI AND A. PAVARIN Let TB be a classical partial n-tilting module over a ring B with endo- morphism ring A. Consider the canonical embedding of B-Mod in D(B) and let E be the full subcategory of B-Mod defined by: E = {N ∈ B-Mod N ∈ Y} = {N ∈ B-Mod T L ⊗ B N = 0} Then, E = {N ∈ B-Mod TorB i (T, N ) = 0 for all i ≥ 0}. Remark 5. Note that E is closed under extensions, direct sums and direct products (since T is a classical partial tilting module). So E is bireflective if and only if it is closed under kernel and/or cokernels. Theorem 6.1. Let B be a ring and let TB be a classical partial n-tilting module with endomorphism ring A. Let Y = Ker(T L ⊗ B −), L the left adjoint of the inclusion inc : Y → D(B) and E the subcategory of B-Mod defined above. Then the following conditions are equivalent: (1) Hi(L(B)) = 0 for every 0 6= i ∈ Z. (2) there is a ring S and a homological ring epimorphism λ : B → S inducing a recollement: i∗=S L ⊗ B − λ∗ D(S) j! L j ∗=T ⊗B − / D(A) / D(B) i!=RHomB (S,−) j∗=RHomA(T,−) (3) Every N ∈ Y is quasi-isomorphic to a complex with terms in E and E is a bireflective subcategory of B-Mod. (4) Every N ∈ Y is quasi-isomorphic to a complex with terms in E and the homologies of N belong to E. Proof. Note that the equivalence between (1) and (2) was somehow known to topologists, as shown for instance in [D]. (1) ⇒ (2) Let Y = L(B). First note that, by adjunction, we have HomD(B)(Y, Y [i]) ∼= HomD(B)(B, Y [i]) ∼= Hi(Y ). Thus, by [Ke4, Theorem 8.7], condition (1) implies that the dg algebra E = RHomB(Y, Y ) has ho- mology concentrated in degree zero and H0(E) ∼= HomD(B)(Y, Y ). Consider a triangle (5) X −→ B ϕB−→ Y −→ X[1], with X ∈⊥ Y. where ϕB is the unit of the adjunction morphism and set S = HomD(B)(Y, Y ). As in [AKL, Proposition 1.7], define a ring homomorphism λ : B → S by λ(b) = L(b), where b denotes the right multiplication by b on B. We have BS = HomD(B)(Y, Y ) ∼= HomD(B)(B, Y ) ∼= H0(Y ) ∼= Y . So we have a quasi- isomorphism ε : BS →B Y and from the definition one sees that ε ◦ λ = ϕB. Thus we have an isomorphism of triangles: / d d z z / e e y y RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 19 X X / B λ / S ε / X[1] . / B ϕB / / Y / X[1] Now we can continue arguing as in the last part of the proof of Lemma 2.14 to conclude that λ is a homological epimorphism and that Y is the essential image of λ∗. So condition (2) follows. L ⊗ B (2) ⇒ (3) The subcategory Y = Ker(T −) is the essential image of the functor λ∗, hence the image of S-Mod under λ∗ is the category E. Every object in Y is quasi-isomorphic to a complex with S-modules terms, hence in E. Moreover, since λ is an epimorphism of rings, the differentials are S-module morphisms. Hence, Lemma 2.11 tells us that E is bireflective. (3) ⇒ (4) This follows from the fact that E is closed under kernels and cokernels. (4) ⇒ (1) We first show that condition (4) implies that E is bireflective. f → E1 be a morphism in E. Then, the complex E′ = . . . 0 → E′ = 0. −)-acyclic terms so T f → E1 → 0 → . . . has (T ⊗ B E′ = T ⊗ B L ⊗ B Indeed, let E0 E0 By (4) the kernel and the cokernel of f belong to E. Thus E is bireflective by Remark 5. By Lemma 2.11 there is a ring S and a ring epimorphism λ : R → S such that E = λ∗(S-Mod) where λ∗ : S-Mod → B-Mod is the restriction functor. We show now that L(B) ∼= λ∗(S). For this aim we follows the arguments used in [CX, Proposition 3.6]. Let Y0 be a complex in Y with terms in E and quasi-isomorphic to L(B). ϕ Let B → Y0 be the unit adjunction morphism associated to the adjoint pair (L, j). Since S viewed as a left B-module belongs to Y we have that HomY (Y0, S) ∼= HomD(B)(B, S), so there is a unique morphism f : Y0 → S such that λ = f ◦ ϕ. We have HomH(B)(S, Y0) ∼= H0(HomB(S, Y0)) and, since λ : B → S is a ring epimorphism, HomB(S, Y0) = HomS(S, Y0), and the terms of Y0 are S- modules. Thus, HomH(B)(S, Y0) ∼= H0(Y0) ∼= HomH(B)(B, Y0). Now, every morphism in HomD(B)(S, Y0) is the image under the canonical quotient func- tor of a morphism in HomH(B)(S, Y0), hence going through the construction of the above isomorphisms, we conclude that there is g ∈ HomD(B)(S, Y0) such that g ◦ λ = ϕ. Consequently, g ◦ f ◦ ϕ = ϕ and λ = f ◦ g ◦ λ. Since λ is an E-reflection of B and ϕ is the unit morphism of the adjunction, we conclude that f ◦ g = idS and g ◦ f = idY0. So S ∼= Y0 ∼= L(B), hence (1) follows. Remark 6. Note that if condition (2) of Proposition 6.1 holds, then there is a homological ring epimorphism λ : B → S even without the assumption of flatness on B. The key point is the existence of a quasi-isomorphim between the ring S and the left adjoint of B We add another property related to the situation considered above. (cid:3) / / / / / 20 S. BAZZONI AND A. PAVARIN Proposition 6.2. In the notations of Theorem 6.1 consider the following conditions: (a) a complex of D(B) belongs to Y if and only if all its homologies belong to E. (b) E is bireflective. (c) There is a ring R and a ring epimorphism µ : B → R such that BR ∈ E and Y is contained in the essential image of the restriction functor µ∗ : D(R) → D(B). Then (a) implies (b) and (a) together with (c) is equivalent to any one of the conditions in Theorem 6.1. In particular, if AT is a good n-tilting module with endomorphism ring B, then (a) is equivalent to any one of the conditions in Theorem 6.1 . Proof. Assume that condition (a) holds. Arguing as in the first part of the proof of (4) ⇒ (1) in Theorem 6.1, we see that E is bireflective. Condition (c) imply µ∗(R-Mod) ⊆ E, hence every complex in Y is quasi- isomorphic to a complex with terms in E. Thus, assuming both (a) and (c), we have that condition (4) in Theorem 6.1 is satisfied. Conversely, if condition (2) of Theorem 6.1 is satisfied, then by [AKL, Lemma 4.6] (a) holds; moreover, (c) is satisfied by choosing the ring epi- morphism λ : B → S. To prove the last statement it is enough to show that, if AT is a good n- tilting module then condition (c) holds. This follows as in the proof of [CX, Proposition 4.6], which is stated for the case of 1-good tilting module, but the argument used there works also in case of higher projective dimension. (cid:3) 7. Generalized Universal localizations Chen and Xi in [CX] consider the case of a good 1-tilting module AT with endomorphism ring B. In particular TB becomes a classical partial 1-tilting module over B. They show that the left end term of a recollement as in the statement of Theorem 6.1, is the derived category of a universal localization f of B. Indeed if 0 → P1 → P0 → TB → 0 is a projective resolution of T as right B-module, then λ : B → S is the universal localization, in the sense of Cohn and Schofield, of B at the morphism f . This means that f ⊗ S B is an isomorphism and S satisfies a universal property with respect to this condition, that is for any ring homomorphism µ : B → S′ such that f ⊗ S′ B is an isomorphism, there is a unique ring homomorphism ν : S → S′ such that ν ◦ λ = µ. Note that f ⊗ B S is an isomorphism if and only is the complex · · · → 0 → P1 ⊗ B S f ⊗BS → P0 ⊗ B S → 0 → . . . is acyclic. Inspired by the above interpretation of universal localization, there is a natural way to generalize the notion of universal localization as follows. We give the definition, which was first introduced by Krause under the name homological localization. RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 21 Definition 7.1. (See [Kr, Section 15]) Let B be a ring and Σ a set of perfect complexes P ∈ H(B). A ring S is a generalized universal localization of B at the set Σ if: (1) there is a ring homomorphism λ : B → S such that P ⊗ B (2) for every ring homomorphism µ : B → R such that P ⊗ B S is acyclic; R is acyclic, there exists a unique ring homomorphism ν : S → R such that ν ◦λ = µ. Lemma 7.2. If λ : B → S is a generalized universal localization of B at a set Σ of compact objects P of D(B), then λ is a ring epimorphism. Proof. Let δ : S → R be a ring homomorphism. Then, for every P ∈ Σ we have: Now P ⊗ B S is an acyclic and bounded complex whose terms are projective P ⊗ B R = (P ⊗ B S) ⊗ S R. right S-modules, then the complex (P ⊗ B S) ⊗ S R is still acyclic. By the universal property satisfied by S we conclude that δ is the only possible ring homomorphism extending µ = δ ◦ λ. (cid:3) Now we can relate the result stated in Theorem 6.1 with the notion of generalized universal localization. Proposition 7.3. Let B be a ring and let TB be a classical partial n-tilting module with endomorphism ring A. Let P be a projective resolution of TB in D(B). If condition (2) in Theorem 6.1 is satisfied, then λ : B → S is a generalized universal localization of B at the set {P }. Proof. As usual let Y = Ker(T L ⊗ B −). By assumptions λ∗(S) ∈ Y, thus T L ⊗ B S = 0, so P ⊗ B S is acyclic. Moreover, Y ∩ B-Mod = E is bireflective and, by [GL, Proposition 3.8], we have that λ∗(S) = l(B), where l : B-Mod → E is the left adjoin of the inclusion of i : E → B-Mod. Let µ : B → S′ be a S′ = 0, hence ring homomorphism such that P ⊗ B S′ ∈ E. Thus, HomB(l(B), S′) ∼= HomB(B, S′), hence there is a unique morphism ρ : l(B) → S′ of right B-modules such that ρ ◦ ηB = µ, where ηB : B → l(B) is the unit morphism of the adjunction. Using the fact that S = EndB(l(B)) and the naturality of the maps induced by the adjunction (l, j), it is not hard to see that ρ induces a unique ring homomorphism ν : S → S′ such that ν ◦ λ = µ. (cid:3) S′ is acyclic, then also T L ⊗ B Remark 7. Note that the converse of the above statement does not hold in general. In fact, as shown in [AKL, Example 5.4] even in the case of a classical 1-tilting module over an algebra, the universal localization does not give rise to a homological epimorphism. We now illustrate another property of the "generalized universal localiza- tion" . 22 S. BAZZONI AND A. PAVARIN Proposition 7.4. Let P be a compact complex in D(B). Assume that λ : B → S is a "generalized universal localization" of B at {P }. Let EP = {N ∈ B-Mod P ⊗ B N is acyclic }. Then, the following hold: (1) λ∗(S-Mod) ⊆ EP . (2) λ∗(S-Mod) = EP if and only if EP is a bireflective subcategory of B-Mod. Proof. (1) Let BM ∈ λ∗(S-Mod). We have P ⊗ B M ∼= P ⊗ B (S ⊗ S M ) ∼= (P ⊗ B S) ⊗ S M and (P ⊗ B S) is a complex in D(S) whose terms are finitely generated pro- jective right S-modules and by assumption it is acyclic. Thus, P ⊗ B M is acyclic too, so M ∈ EP . (2) By Lemma 7.2, λ is a ring epimorphism, hence, if λ∗(S-Mod) = EP , then EP is bireflective, by Lemma 2.11. Conversely, assume that EP is bireflective. By Lemma 2.11, there is a ring R and a ring epimorphism µ : B → R such that µ∗(R-Mod) = EP . In particular, µ∗(R) ∈ EP , hence P ⊗ R is an acyclic complex. Thus, by B the universal property satisfied by S, there is a unique ring homomorphism ν : S → R such that ν ◦λ = µ. By Lemma 2.11, µ : B → R is an EP -reflection of B and S ∈ EP by part (1). We infer that there is a unique morphism ρ : R → S such that ρ ◦ µ = λ. By the unicity of the rings homomorphisms ν and ρ it follows that they are inverse to each other. (cid:3) 8. Examples In the notations of Section 6 we give some examples of different behaviors of classical n-partial tilting modules with respect to the class E. In what follows k will indicate an algebraically closed field. Example 1. We exhibit examples of classical partial tilting modules TB of projective dimension two over an artin algebra B such that there exists a "generalized universal localization" S of B at the projective resolution of TB and moreover, there exists a homological ring epimorphism λ : B → S such that the class Y = Ker(T −) is triangle equivalent to D(S). In this L ⊗ B case the classical partial tilting module TB doesn't arise from a good tilting module. Consider a representation-finite type algebra Λ := kQ/I of an acyclic connected quiver Q (with n > 1 vertices) with a unique sink j and the category mod-Λ of the finite dimensional right Λ-modules. Note that we use the just-apposition of arrows for the product in Λ). P (i)) be an APR tilting module over Λ (see Let TΛ = τ −1(S(j)) ⊕ (Li6=j [APR]). Then proj.dim.(TΛ) = 1 and its projective resolution is given by 0 −→ S(j) −→ (Mi6=j P (i)) ⊕ E −→ TΛ −→ 0 RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 23 where 0 −→ S(j) −→ E −→ τ −1(S(j)) −→ 0 is an almost split exact sequence with E a projective Λ-module. Let S(j)d := S(j)d Λ(cid:19) the one point coextension of Homk(S(j), k) and consider B := (cid:18) k Λ by the non injective simple S(j)Λ (see [ASS]). In particular B ≃ kQ′/J where Q′ is the quiver Q with the adjoint of a sink ∗ and of an arrow j −→ ∗. Let I(∗) and S(∗) be the indecomposable injective and simple 0 right B-modules at the vertex ∗, respectively. Then I(∗) = j ∗ and letting P (∗) = I(∗)d = Homk(I(∗), k) be the indecomposable projective at the vertex ∗ ( regarded as right module on Bop), then P (∗) = . ∗ j Every Λ-module can be regarded as a B-module via the natural embed- ding ϕ : mod-Λ ֒→ mod-B. Proposition 8.1. The following hold: (1) TB has projective dimension 2. (2) TB is self orthogonal. (3) EΛ = {M ∈ Λ-Mod TorΛ i (T, M ) = 0, ∀i ≥ 0} = 0 and EB = {M ∈ B-Mod TorB i (T, M ) = 0 ∀i ≥ 0} = Add I(∗)d = Add P (∗) where for every module M , Add M denotes the class of all direct summands of arbitrary direct sums of copies of M . Proof. (1) We have that S(j) regarded as B module is non projective and its projective cover is given by Hence a projective resolution of TB is I(∗) −→ S(j) −→ 0. 0 −→ S(∗) −→ I(∗) −→ (Mi6=j P (i)) ⊕ E −→ τ −1(S(j)) ⊕ (Mi6=j P (i)) −→ 0. (2) To prove the self-orthogonality of TB we can observe that mod-Λ is equivalent to the class mod-BT Ker(HomB(−, I(∗))), then, in particular, it is closed under extensions in mod-B. Hence it is clear that Ext1 B(TB, TB) ≃ Ext1 Λ(TΛ, TΛ) = 0. Moreover Ext2 B(TB, TB) = Ext1 B(S(j)B, TB) = Ext1 Λ(S(j), TΛ) = 0. (3) EΛ = 0 because TΛ is a tilting module. Now, ind-B \ ind-Λ = {I(∗), S(∗)} and I(∗) = j ∗ . We compute the class EB = {M ∈ B-Mod TorB = {M ∈ Mod-Bop Exti i (T, M ) = 0, ∀i ≥ 0} = Bop(M, T d) = 0, ∀i ≥ 0} where T d B := Homk(TB, k). We can regard B-Mod as Mod-Bop and Bop is the one point extension of Λop by the simple S(j)d = S(j). We claim 24 S. BAZZONI AND A. PAVARIN that EB = Add (P (∗)). Note that, as in the previous case, ind-Bop\ind- Λop = {P (∗), S(∗)} and P (∗) = . From the fact that EΛ = 0 and that ∗ j every Λ-module can be regarded as a B-module, only Add {P (∗), S(∗)} could be contained in EB. We prove that S(∗) /∈ KerExt2 B(−, T d the injective resolution of T d B). Since S(j) is the first cosyzygy of I(i)),we show that S(∗) /∈ Λ = τ −1(S(j))d ⊕ (Li6=j KerExt1 B(−, S(j)). Indeed there is the non split short exact sequence 0 −→ S(j) −→ ∗ j −→ S(∗) −→ 0. Hence S(∗) /∈ EB. To show that P (∗) ∈ EB we only have to check that HomBop(P (∗), T d B) = 0, since P (∗) is projective. This is true from the fact that the top of P (∗) = S(∗) does not belongs to any composition series of T d B. Then EB = Add (P (∗)) = Add (I(∗)d). (cid:3) Set now A := EndB(TB) = EndΛ(TΛ), then Λ = EndA(AT ) because TΛ is tilting (hence balanced) over Λ. Hence AT is 1-tilting, but EndA(AT ) 6= B, so AT is not faithfully balanced. In the sequel we will simply write E for the class EB. Lemma 8.2. For each projective left B-module P the unit morphism ηP : P −→ HomA(T, T ⊗B P ) of the adjunction (T ⊗B −, HomA(AT, −), is surjective and KerηP ∈ E = Add (P (∗)). Proof. We can regard ηB : B −→ HomA(AT, TB ⊗B B) ≃ Λ as the projection π : B −→ Λ ≃ B/Be∗B, hence it is surjective and the kernel is the annihilator of TB as right B-module, that is KerηP is the projective B-module P (∗). Now, since E is closed under direct summand, we can prove the statement just for free modules. Let α be a cardinal, then the map ηB(α) : B(α) −→ HomA(T, T ⊗B B(α)) = Λ(α) is exactly π(α) and the kernel of ηB(α) is P (∗)(α). Proposition 8.3. There is a homological ring epimorphism (cid:3) with S = End(P (∗)⊕2). λ : B −→ S Proof. We claim that E = Add (P (∗)) is bireflective. A linear representation of P (∗)(I) for some cardinal I is of the form k(I) ϕ −→ k(I) at the verteces ∗ and j and zero at the other vertices, where ϕ is an iso- morphism of k-vector spaces. An object is in Add P (∗) if and only if it is RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 25 a summand of a k-linear representation of this form. Hence, for every mor- phism f : P (∗)(I) −→ P (∗)(J), Kerf , Cokerf , and Im f belong to Add P (∗), that is, by [GL], E is bireflective. Thus, there exists an object M ∈ E such that S := EndB(M ) is isomorphic to M as B-module and E is equivalent to S-Mod. Since dimk P (∗) = 2, then M = P (∗)⊕2, S ≃ M2(k) and there exists a ring epimorphism λ : B −→ S. We now prove that λ is homolog- ical. In view of Theorem 6.1 we have just to prove that every object in L Y = Ker(T ⊗B −) is quasi-isomorphic to a complex with terms in E. Set H = RHomA(ATB, −) and G = AT B ⊗B −and consider the triangle L ηB−→ HG(B) −→ Y −→ B[1]. B We have L HG(B) = RHomA(ATB,A TB ⊗BB) = RHomA(ATB,A TB) ∼= EndA(ATB) = Λ (because TB ≃ TΛ is self orthogonal in A-Mod, hence it is HomA(AT, −)- acyclic). Then ηB = ηB and, considering the long exact sequence of the homologies, we can conclude that Y is quasi-isomorphic to the stalk complex KerηB[1], that is P (∗)[1]. We now follow [CX, Prop. 4.6]. Let M be an object in D(B), then there is the triangle M ηM−→ HG(M ) −→ YM −→ M [1] where HG(M ) = RHomA(ATB,A TB ⊗B M ) = HomA(ATB,A TB⊗BW ) L with W an H-projective resolution of the complex M . Therefore, the complex HomA(ATB,A TB⊗BW ) has terms of the form HomA(T, Ti) with Ti ∈ Add (AT )). Being AT finitely generated, we have that the module HomA(T, Ti) is in Add (ΛΛ). Regard the triangle in (1) as the triangle W ηM−→ Hom(T, T •) −→ YM −→ W [1] where T • is the complex ATB ⊗B W. Therefore the morphism ηM can be regarded in C(B) as the family (ηi)i∈Z with ηi : W i −→ HomA(T, Ti). Then for Lemma 8.2, noting that (Ker ηM )i = Kerηi ∈ E, we can conclude that YM ≃ Ker ηM [1] has terms in E. Now, for every Y in Y, there is the triangle ηY−→ 0 −→ Y −→ Y [1] Y then Y is Ker ηY which has terms in E. (cid:3) Remark 8. The previous example can be generalized considering a situa- tion similar to [Mi, Corollary 5.5]. Let us point out the key steps used in the previous Example 1. Assume that I is a non-zero projective, idem- potent two-sided ideal of an ordinary k-algebra B. Then the projective dimension of Λ, viewed as a right B-module, is one. By [NS, Example in Section 4] the canonical projection π : B → Λ := B/I is a homological ring epimorphism and ΛB is self-orthogonal. Let now TΛ be a classical n-tilting module over Λ and view T as a right B-module via π. Then I is the annihilator of TB (and of ΛB) and TB is a classical n + 1-partial tilting module, since proj.dim(TΛ) ≤ proj.dim(TB) ≤ proj.dim(TΛ) + 1. Set A := EndΛ(TΛ) = EndB(TB) (where the last equality holds since π 26 S. BAZZONI AND A. PAVARIN ⊗Λ − : D(Λ) → D(A) is a tri- is a ring epimorphism). The functor AT angle equivalence, since TΛ is a classical n-tilting module. Moreover the L L functor AT ⊗B − : D(B) → D(Λ) is given by the composition of functors L L L (AT ⊗Λ −) ◦ (ΛΛ ⊗B −), so the kernel of AT ⊗B − is exactly the kernel of L L (ΛΛ ⊗B −). Thus, Ker(AT ⊗B −) is equivalent to the derived category of a ring via a homological ring epimorphism if and only so is Ker(ΛΛ ⊗B −). But, ΛB is a classical 1-partial tilting module with EndB(Λ) = Λ, so the L class E = Ker(ΛΛ ⊗B −) ∩ B-Mod is bireflective. Now, similarly to the proof of Proposition 8.3, we let G = (ΛΛ ⊗B −) and H = RHomΛ(ΛΛB, −). L L L Then, a complex Y ∈ Ker(ΛΛ ⊗B −) if and only if Y is quasi isomorphic to HG(Y ). Computing HG(Y ) by means of a H-projective resolution of Y in D(B) we obtain that HG(Y ) is a direct summand of complex with terms of the form Λ(I) for some set I, viewed as left B-modules, hence in the class E. By Theorem 6.1, we conclude that the kernel of the functor AT ⊗B − is triangle equivalent to the derived category of a ring via the homological epimorphism π. L Example 2. Now we give a simple example of a finitely generated classical partial tilting module T over a finite dimensional algebra B, such that the class E = ∩ i (T, −) is not bireflective (in particular there are no i≥0 KerTorB homological epimorphism of rings λ : B → S such that Ker(T D(S)). L ⊗B −) ≃ Consider the quiver ◦ 1 −→ ◦ 2 −→ ◦ 3 a b with relation ab = 0 and the right modules over its path algebra B. Consider the simple injective right module S1. The projective dimension of S1 is two and its projective resolution is given by: 0 −→ P3 −→ P2 −→ P1 −→ S1 −→ 0 It is easy to see that S1 is classical partial tilting over B. A calculation 3 2 (cid:27) is not bireflective. In fact there is shows that the class E = Add (cid:26) 2 1 a morphism f : such that the kernel is not in E. Example 3. Consider the quiver , 2 1 −→ 3 2 b a / ◦ 2 ◦ 1 with relation ab = 0 and consider the classical partial tilting module 1 2 of projective dimension 2. Here, as shown in [B2, Example 1], E = 0 then it is bireflective but, obviously, the complexes in Y don't have terms in E. / RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 27 Example 4. [CX2, Section 7.1] The following is an example of a good n- ⊗B −) is not tilting module AT with B = EndA(AT ), such that Ker(TB triangle equivalent to the derived category of a ring via a homological ring epimorphism. Let A be a commutative n-Gorestein ring and consider a minimal injective L resolution of the regular module AA of the form: 0 → A → Mp∈P0 E(A/p) → . . . → Mp∈Pn E(A/p) → 0 where Pi is the set of all prime ideals of A of height i (see [Bas, Theorem 1, Theorem 6.2]). Then, the module AT := M0≤i≤n Mp∈Pi E(A/p) for all 0 ≤ i ≤ n, Ti := Lp∈Pi is an n-tilting module by [GT, Example 5.16]] and it is moreover good. Set, E(A/p), then we have HomA(Tj, Ti) = 0 for all 0 ≤ i ≤ j ≤ n. Assume that n ≥ 2 and that the injective dimension of A is exactly n; then T has projective dimension n (see [B2, Proposition 3.5]). Note that Ti 6= 0 for every 2 ≤ i ≤ n so T satisfies the hypotheses of [CX2, Corollary 1.2], hence Ker(TB ⊗B −) cannot be realized as the derived category D(S) of a ring S linked to B via a homological ring epimorphism B → S. L Acknowledgements. We are particularly indebted to the referees for pointing out some incorrectness in a first version of the paper and for giving valuable comments and suggestions. The second named author of this paper would like to thank Pedro Nicolas and Manolo Saorin for their kind hospitality in Murcia and for the very useful explanations on dg categories. In addition we thank also Gabriella D'Este and Nicola Sambin for several discussions on tilting modules over artin algebras. References [AC] L. Angeleri Hugel and F. U. Coelho. Infinitely generated tilting modules of finite projective dimension. Forum Math. 13, no. 2, 239250, 2001. [AKL] L. Angeleri Hugel, S. Koenig and Q. Liu. Recollements and tilting objects. J. Pure Appl. Algebra, 215(2): 420 -- 438, 2011. [APR] M. Auslander, M.I. Platzeck, I. Reiten. Coxeter functors without diagrams. Trans. Amer. Math. Soc, 250, 146, 1979. [ASS] I. Assem, D. Simson and A. Skowro´nski. Elements of the representation theory of associative algebras. Vol. 1. Techniques of representation theory. London Mathemat- ical Society Student Texts 65. Cambridge University Press, Cambridge, 2006. [Bas] H. Bass. On the ubiquity of Gorenstein rings. Math. Z. 82 1963 8 -- 28. [B] S. Bazzoni. Equivalences induced by infinitely generated tilting modules. Proc. Amer. Math. Soc., 138(2): 533-544, 2009. [B2] S. Bazzoni. A characterization of n-cotilting and n-tilting modules. J. Algebra, 273, no. 1, 359-372, 2004 [BB] S. Brenner and M. C. R. Butler. Generalizations of the Bernstein-Gel'fand- Ponomarev reflection functors. In Representation theory, II (Proc. Second Internat. Conf., Carleton Univ., Ottawa, Ont., 1979), volume 832 of Lecture Notes in Math., pages 103 -- 169. Springer, Berlin, 1980. 28 S. BAZZONI AND A. PAVARIN [BMT] S. Bazzoni, F. Mantese and A. Tonolo. Derived equivalence induced by infinitely generated n-tilting modules. Proc. Amer. Math. Soc., 139(12): 4225 -- 4234, 2011. [BBD] A.A. Beilinson, J. Bernstein and P. Deligne. Faisceaux pervers. Ast´erisque, 100, 1982. [BGP] I.N. Bernstein, I. M. Gel'fand and V. A. Ponomarev. Coxeter functors, and Gabriel's theorem. Uspehi Mat. Nauk, 28(2(170)):19 -- 33, 1973. [CX] H.X. Chen and C.C. Xi. Good tilting modules and recollements of derived module categoires. Proc. Lond. Math. Soc., 104:959 -- 996, 2012. [CX2] H.X. Chen and C.C. Xi. Ringel modules and homological subcategoires. preprint, arXiv:1206.0522, 2012. [CPS] E. Cline, B. Parshall and L. Scott. Derived categories and Morita theory. J. Algebra, 104(2):397 -- 409, 1986. [CF] R.R. Colby, K.R. Fuller. Tilting, cotilting and serially tilted rings. Comm. Algebra, 18(5), 1585 -- 1615, 1990. [CT] R. Colpi and J. Trlifaj. Tilting modules and tilting torsion theories. J. Algebra, 178(2):614 -- 634, 1995. [D] W.G. Dwyer. Noncommutative localization in homotopy theory. Noncommutative localization in algebra and topology, 24 -- 39 London Math. Soc. Lec- ture Note, 330, 2006. [DG] W.G. Dwyer and J.P.C. Greenlees. Complete modules and torsion modules. Amer. J. Math., 124: 199 -- 220, 2002. [GP] P. Gabriel and A.J. de la Pena. Quotients of representation-finite algebras. Comm. Algebra, 15: 2235 -- 2250, 1987. [GL] W. Geigle and H. Lenzing. Perpendicular categories with applications to represen- tations and sheaves. J. Algebra, 144(2): 273 -- 343, 1991. [GT] R. Gobel J and J. Trlifaj, Approximations and endomorphism algebras of modules. De Grutyter Expositions in Mathematics, de Gruyter, Berlin, 14, 2006. [H] D. Happel. On the derived category of a finite-dimensional algebra. Comment. Math. Helv., 62(3): 339 -- 389, 1987. [HR] D. Happel and C. M. Ringel. Tilted algebras. Trans. Amer. Math. Soc., 274(2): 399 -- 443, 1982. [J] P. Jørgensen. Recollements for differential graded algebras. J. Algebra, 299: 589 -- 601, 1991. [Ke1] B. Keller. On the construction of triangle equivalences. Lectures Workshop Pappen- heim 1994, 1994. [Ke2] B. Keller. Deriving DG categories. Ann. Sci. ´Ecole Norm. Sup., 27(4): 63 -- 102, 1994. [Ke3] B. Keller. On differential graded categories. International Congress of Mathemati- cians, vol. II, Eur. Math. Soc., Zrich, pp. 151190, 2006. [Ke4] B. Keller. Derived categories and tilting. Handbook of tilting theory, London Math. Soc. Lecture Note Ser., Cambridge Univ. Press, Cambridge, 332: 49 -- 104, 2007 [K] S. Koenig. Tilting complexes, perpendicular categories and recollements of derived categories of rings. J. Pure Appl. Algebra, 73: 211 -- 232, 1991. [Kr] H. Krause. Cohomological quotients and smashing localizations. Amer. J. Math. 127(6) (2005) 1191-1246. [Mi] J. Miyachi. Localization of triangulated categories and derived categories. J. Algebra, 141(2): 463 -- 483, 1991 [Miy] Y. Miyashita. Tilting modules of finite projective dimension. Math. Z., 193(1):113 -- 146, 1986. [N] A. Neeman. The Grothendieck duality theorem via Bousfield's techniques and Brown representability. J. Am. Math. Soc., 9: 205 -- 236, 1996. [N2] A. Neeman. Triangulated Categories. Ann. of Math. Stud., vol. 148, Princeton Uni- versity Press, 2001. [NS] P. Nicol´as and M. Saor´ın. Parametrizing recollement data from triangulated cate- gories. J. Algebra, 322:1220 -- 1250, 2009. [P] D. Pauksztello. Homological epimorphisms of differential graded algebras. Comm. Algebra, 37:2337 -- 2350, 2009. RECOLLEMENTS FROM PARTIAL TILTING COMPLEXES 29 [R] J. Rickard. Morita theory for derived categories. J. London Math. Soc. (2), 39(3):436 -- 456, 1989. [T] G. Tabuada. Une structure de cat´egorie de mod`eles de Quillen sur la cat´egorie des dg-cat´egories. C. R. Math. Acad. Sci. Paris , 340(1):15 -- 19, 2005. [Y] D. Yang. Recollements from generalized tilting. Proc. Amer. Math. Soc., 140(1):83 -- 91, 2012. (Silvana Bazzoni) Dipartimento di Matematica Pura ed Applicata, Universit`a di Padova, via Belzoni 7, I-35131 Padova - Italy E-mail address, Silvana Bazzoni: [email protected] (Alice Pavarin) Dipartimento di Matematica Pura ed Applicata, Universit`a di Padova, via Belzoni 7, I-35131 Padova - Italy E-mail address, Alice Pavarin: [email protected]
1611.00988	1	1611	2016-11-03T12:56:13	Commutative rings whose finitely generated ideals are quasi-flat	[ "math.RA", "math.AC" ]	A definition of quasi-flat left module is proposed and it is shown that any left module which is either quasi-projective or flat is quasi-flat. A characterization of local commutative rings for which each ideal is quasi-flat (resp. quasi-projective) is given. It is also proven that each commutative ring R whose finitely generated ideals are quasi-flat is of $\lambda$-dimension $\le$ 3, and this dimension $\le$ 2 if R is local. This extends a former result about the class of arithmetical rings. Moreover, if R has a unique minimal prime ideal then its finitely generated ideals are quasi-projective if they are quasi-flat. In [1] Abuhlail, Jarrar and Kabbaj studied the class of commutative fqp-rings (finitely generated ideals are quasi-projective). They proved that this class of rings strictly contains the one of arithmetical rings and is strictly contained in the one of Gaussian rings. It is also shown that the property for a commutative ring to be fqp is preserved by localization. It is known that a commutative ring R is arithmetical (resp. Gaussian) if and only if R M is arithmetical (resp. Gaussian) for each maximal ideal M of R. But an example given in [6] shows that a commutative ring which is a locally fqp-ring is not necessarily a fqp-ring. So, in this cited paper the class of fqf-rings is introduced. Each local commutative fqf-ring is a fqp-ring, and a commutative ring is fqf if and only if it is locally fqf. These fqf-rings are defined in [6] without a definition of quasi-flat modules. Here we propose a definition of these modules and another definition of fqf-ring which is equivalent to the one given in [6]. We also introduce the module property of self-flatness. Each quasi-flat module is self-flat but we do not know if the converse holds. On the other hand, each flat module is quasi-flat and any finitely generated module is quasi-flat if and only if it is flat modulo its annihilator. In Section 2 we give a complete characterization of local commutative rings for which each ideal is self-flat. These rings R are fqp and their nilradical N is the subset of zerodivisors of R. In the case where R is not a chain ring for which N = N 2 and R N is not coherent every ideal is flat modulo its annihilator. Then in Section 3 we deduce that any ideal of a chain ring (valuation ring) R is quasi-projective if and only if it is almost maximal and each zerodivisor is nilpotent. This complete the results obtained by Hermann in [11] on valuation domains. In Section 4 we show that each commutative fqf-ring is of $\lambda$-dimension $\le$ 3. This extends the result about arithmetical rings obtained in [4]. Moreover it is shown that this $\lambda$-dimension is $\le$ 2 in the local case. But an example of a local Gaussian ring R of $\lambda$-dimension $\ge$ 3 is given.	math.RA	math	COMMUTATIVE RINGS WHOSE FINITELY GENERATED IDEALS ARE QUASI-FLAT FRANC¸ OIS COUCHOT Abstract. A definition of quasi-flat left module is proposed and it is shown that any left module which is either quasi-projective or flat is quasi-flat. A characterization of local commutative rings for which each ideal is quasi-flat (resp. quasi-projective) is given. It is also proven that each commutative ring R whose finitely generated ideals are quasi-flat is of λ-dimension ≤ 3, and this dimension ≤ 2 if R is local. This extends a former result about the class of arithmetical rings. Moreover, if R has a unique minimal prime ideal then its finitely generated ideals are quasi-projective if they are quasi-flat. In [1] Abuhlail, Jarrar and Kabbaj studied the class of commutative fqp-rings (finitely generated ideals are quasi-projective). They proved that this class of rings strictly contains the one of arithmetical rings and is strictly contained in the one of Gaussian rings. It is also shown that the property for a commutative ring to be fqp is preserved by localization. It is known that a commutative ring R is arithmetical (resp. Gaussian) if and only if RM is arithmetical (resp. Gaussian) for each maximal ideal M of R. But an example given in [6] shows that a commutative ring which is a locally fqp-ring is not necessarily a fqp-ring. So, in this cited paper the class of fqf-rings is introduced. Each local commutative fqf-ring is a fqp-ring, and a commutative ring is fqf if and only if it is locally fqf. These fqf-rings are defined in [6] without a definition of quasi-flat modules. Here we propose a definition of these modules and another definition of fqf-ring which is equivalent to the one given in [6]. We also introduce the module property of self-flatness. Each quasi-flat module is self-flat but we do not know if the converse holds. On the other hand, each flat module is quasi-flat and any finitely generated module is quasi-flat if and only if it is flat modulo its annihilator. In Section 2 we give a complete characterization of local commutative rings for which each ideal is self-flat. These rings R are fqp and their nilradical N is the subset of zerodivisors of R. In the case where R is not a chain ring for which N = N 2 and RN is not coherent every ideal is flat modulo its annihilator. Then in Section 3 we deduce that any ideal of a chain ring (valuation ring) R is quasi- projective if and only if it is almost maximal and each zerodivisor is nilpotent. This complete the results obtained by Hermann in [11] on valuation domains. In Section 4 we show that each commutative fqf-ring is of λ-dimension ≤ 3. This extends the result about arithmetical rings obtained in [4]. Moreover it is shown that this λ-dimension is ≤ 2 in the local case. But an example of a local Gaussian ring R of λ-dimension ≥ 3 is given. 2010 Mathematics Subject Classification. 13F05, 13B05, 13C13, 16D40, 16B50, 16D90. Key words and phrases. quasi-flat module, chain ring, arithmetical ring, fqf-ring, fqp-ring, λ-dimension. 1 2 FRANC¸ OIS COUCHOT In this paper all rings are associative and commutative (except in the first sec- tion) with unity and all modules are unital. 1. quasi-flat modules: generalities Let R be a ring, M a left R-module. A left R-module V is M -projective if the natural homomorphism HomR(V, M ) → HomR(V, M/X) is surjective for every submodule X of M . We say that V is quasi-projective if V is V -projective. A ring R is said to be a left fqp-ring if every finitely generated left ideal of R is quasi-projective. We say that V is M -flat1 if for any epimorphism p : M → M ′, for any homo- morphism u : V → M ′ and for any homomorphism v : G → M , where M ′ is a left R-module and G a finitely presented left R-module, there exists a homomorphism q : G → M such that pq = uv. We call V quasi-flat (resp. self-flat) if V is V n-flat for each integer n > 0 (resp. n = 1). Clearly each quasi-flat module is self-flat but we do not know if the converse holds. An exact sequence S of left R-modules 0 → F → E → G → 0 is pure if it remains exact when tensoring it with any right R-module. Then, we say that F is a pure submodule of E. Recall that S is pure if and only if HomR(M, S) is exact for each finitely presented left R-module M ([15, 34.5]). When E is flat, then G is flat if and only if S is pure ([15, 36.5]). Proposition 1.1. Let R be a ring. Then: (1) each quasi-projective left R-module is quasi-flat; (2) each flat left R-module is quasi-flat. Proof. (1). V n-projective for each integer n > 0. If V is a quasi-projective left R-module then, by [15, 18.2(2)], V is (2). By [15, 36.8.3] a left R-module is flat if and only if it is M -flat for each left (cid:3) R-module M . Proposition 1.2. Let R be a ring, 0 → A t−→ B → C → 0 an exact sequence of left R-modules and V a left module. If V is B-flat then V is A-flat and C-flat. Proof. Clearly V is C-flat. Let p : A → A′ be an epimorphism of left R-modules. Consider the following pushout diagram of left R-modules: 0 ↓ A t ↓ B 0 ↓ A′ → 0 t′ ↓ p −→ p′ −→ B′ → 0 Let G be a finitely presented R-module and V u−→ A′ and G v−→ V be homomor- phisms. Since V is B-flat there exists a linear map G d−→ B such that t′uv = p′d. By [15, 10.7] the above diagram is also a pullback diagram of left R-modules, so there exists a homomorphism G (cid:3) q −→ A such that pq = uv. Hence V is A-flat. Corollary 1.3. Let R be a ring, V a finitely generated left module and I its anni- hilator. Then V is flat over R/I if and only if V is quasi-flat. 1the module property M -flat is generally used to define flat module FQF-RINGS 3 Proof. If V is flat over R/I then, from Proposition 1.1 we deduce that it is quasi- flat. Conversely, if V is generated by n elements then R/I is isomorphic to a submodule of V n. It follows that F = (R/I)n is isomorphic to a submodule of V n2 . By Proposition 1.2 V is F -flat. Since there exists an epimorphism p : F → V , we get that ker(p) is a pure submodule of F . Hence V is flat over R/I. (cid:3) In section 2 (Corollary 2.14 and Example 2.15) an example of a quasi-flat module (over a commutative ring) which is not flat modulo its annihilator is given. We say that a ring R is a left fqf-ring if each finitely generated left ideal is quasi-flat. By Corollary 1.3 this definition is equivalent to the one given in [6, Section 3]. 2. quasi-flat ideals over local fqp-rings In this section R is a commutative ring. A module U is uniserial if its lattice of submodules is totally ordered by inclu- sion. A ring R is a chain ring (or a valuation ring) if it is a uniserial R-module. A chain ring which is an integral domain is a valuation domain. Recall that R is an IF-ring if each injective R-module is flat. When R is a chain ring, we denote by P its maximal ideal, by Z its subset of zero-divisors which is a prime ideal, by N its nilradical and by Q its quotient ring RZ. Lemma 2.1. Let R be a chain ring and U an R-module. If U is quasi-flat (resp. quasi-projective) then aU is quasi-flat (resp. quasi-projective) too for each a ∈ R. Proof. We consider the following homomorphisms: p : (aU )n → U ′, u : aU → U ′ and v : G → aU where U ′ is an R-module, p is surjective, n an integer > 0 and G a finitely presented R-module. By [14, Theorem 1] G is a direct sum of cyclic submodules. It is easy to see that we may assume that G is cyclic. So G = R/bR for some b ∈ R. If x = v(1 + bR) then bx = 0 and there exists y ∈ U such that x = ay. So, bay = 0. Let v′ : R/baR → U , u′ : U → U ′ and p′ : U n → U ′ be the homomorphisms defined by v′(r + baR) = ry for each r ∈ R, u′(z) = u(az) for each z ∈ U and p′(w) = p(aw) for each w ∈ U n. The quasi-flatness of U implies that there exists a morphism q′ : R/baR → U n such that p′q′ = u′v′. If we put q(r + bR) = aq′(r + baR) for each r ∈ R then the equalities bq(1 + bR) = baq′(1 + baR) = 0 imply that q : G → (aU )n is a well defined homomorphism, and we get pq = uv. Now, suppose that n = 1 and U is quasi-projective. There exists t′ : U → U (cid:3) such that p′t′ = u′. Let t = t′aU . Then pt = u. Let I be a non-zero proper ideal of a chain ring R. Then I ♯ = {r ∈ R rI ⊂ I} is a prime ideal which is called the top prime ideal associated to I. It is easy to check that I ♯ = {r ∈ R I ⊂ (I : r)}. It follows that I ♯/I is the inverse image of the set of zerodivisors of R/I by the natural map R → R/I. So, Z = 0♯. Proposition 2.2. Let R be a chain ring. Then each proper ideal I satisfying Z ⊂ I ♯ is flat modulo its annihilator. Proof. First assume that Z ⊂ I. In this case I is a direct limit of free modules of rank one. So, it is flat. Now suppose that I ⊆ Z and let t ∈ I ♯ \ Z and a ∈ I \ tI. Then a = ts for some s ∈ Z \ I and t ∈ (I : s). So, Z ⊂ (I : s). It is easy to check that I = s(I : s), I ∼= (I : s)/(0 : s), (0 : I) ⊇ (0 : s) and (I : s)/(0 : s) strickly contains Z/(0 : s) the subset of zerodivisors of R/(0 : s) (see [3, Lemma 21]). (cid:3) 4 FRANC¸ OIS COUCHOT Remark 2.3. If P = Z then by [10, Lemma 3] and [12, Proposition 1.3] we have (0 : (0 : I)) = I for each ideal I which is not of the form P t for some t ∈ R. In this case R is self FP-injective and the converse holds. So, if A is a proper ideal such that A♯ = P then R/A is self FP-injective and it follows that (A : (A : I)) = I for each ideal I ⊇ A which is not of the form P t for some t ∈ R. Proposition 2.4. Let R be a chain ring. Then any proper ideal I satisfying I ♯ ⊂ Z is not self-flat. Proof. Let s ∈ Z \ I ♯. Since s /∈ s2Q, by applying the above remark to Q we get that there exists a ∈ (0 : s2) \ (0 : s). The multiplication by s in I induces an isomorphism σ : I/(0 : s) → I. Let u = σ−1, p : I → I/(0 : s) be the natural epimorphism and v : R/sR → I the homomorphism defined by v(r + sR) = rsa. Then uv(1 + sR) = a + (0 : s) and sb 6= 0 for each b ∈ a + (0 : s). So, there is no homomorphism q : R/sR → I such that pq = uv. (cid:3) Lemma 2.5. Let R be a chain ring and I a nonzero proper ideal. Assume that P = Z and I 6= aP for each a ∈ R. Then I is FP-injective over R/A where A = (0 : I). Proof. By Remark 2.3 we have I = (0 : A). Let x ∈ I and c ∈ R \ A such that (A : c) ⊆ (0 : x). Then (0 : c) ⊆ (0 : x). Since R is self FP-injective there exists y ∈ R such that x = cy. We have (0 : y) = c(0 : x) ⊇ c(A : c) = A (the first equality holds by [3, Lemma 2]). Hence y ∈ I. (cid:3) Theorem 2.6. Let R be a chain ring. Assume that either Z 6= Z 2 or Q is not coherent. Then the following conditions are equivalent: (1) Z = N ; (2) each ideal I is flat over R/A where A = (0 : I). Proof. By Proposition 2.4 (2) ⇒ (1). (1) ⇒ (2). Let I be an ideal and A = (0 : I). By Proposition 2.2 it remains to examine the case where I ♯ = Z. If Z 6= Z 2 then Z is principal over Q. It follows that Q is Artinian. Since I is a principal ideal of Q then I is flat over Q/A and R/A. Now suppose that Z = Z 2 and Q is not coherent. By [3, Theorem 10] Z is flat, and we easily deduce that aZ is flat over R/(0 : a) for each a ∈ R. Now suppose that I is neither principal over Q nor of the form aZ for each a ∈ R. By Lemma 2.5 I is FP-injective over R/A. From Q no coherent we deduce that (0 : r) is not principal over Q for each 0 6= r ∈ I. By [7, Theorem 15(4)(c)] I is flat over R/A. (cid:3) Remark 2.7. If R is a chain ring such that either Z is principal over Q or Q is not coherent then each ideal I satisfying Z ⊆ I ♯ is flat modulo its annihilator. Lemma 2.8. Let R be a chain ring and M a finitely generated R-module. Then, for each proper ideal A which is not of the form rP for any r ∈ P , we have AM = ∩s∈P \AsM . Proof. By [8, Theorem 15] there is a finite sequence of pure submodules of M , 0 = M0 ⊂ M1 ⊂ · · · ⊂ Mn−1 ⊂ Mn = M. such that Mk/Mk−1 is cyclic for each k = 1, . . . , n. We proceed by induction on n. When n = 1 M is cyclic and we use [3, Lemma 29] to conclude. Now suppose that FQF-RINGS 5 n > 1. Let x ∈ ∩s∈P \AsM . We may assume that x /∈ Mn−1. Since M/Mn−1 is cyclic there exist y ∈ M and a ∈ A such that (x − ay) ∈ Mn−1. Moreover, by using the fact that Mn−1 is a pure submodule of M we have that (x−ay) ∈ ∩s∈P \AsMn−1. From the induction hypothesis we deduce that x = ay + bz for some z ∈ Mn−1 and b ∈ A. (cid:3) Proposition 2.9. Let R be a chain ring. Then, for each a ∈ R, aZ is quasi-flat. Proof. We may assume that Z = Z 2 6= 0. By Lemma 2.1 it is enough to study the case a = 1. First suppose that Z = P . We consider the following homomorphisms: p : Z n → Z ′, u : Z → Z ′ and v : G → Z where Z ′ is an R-module, p is surjective, n an integer > 0 and G = R/aR for some a ∈ R. If r = v(1 + aR) then ar = 0. Let s ∈ Z \Rr. Then r = ss′ for some s′ ∈ P and u(r) = su(s′). So, u(r) ∈ ∩s∈Z\RrsZ ′. Consider the following commutative pushout diagram: 0 ↓ Z n t ↓ Rn 0 ↓ Z ′ → 0 t′ ↓ p −→ p′ −→ R′ → 0 where t is the canonical inclusion. Clearly R′ is finitely generated. So, by Lemma 2.8 u(r) = rx′ for some x′ ∈ R′. Let x ∈ Rn such that p′(x) = x′. Let q : G → Z n be the homomorphism defined by q(1 + aR) = rx. Then q : G → Z n is well defined because ar = 0 and we have pq = uv. Hence P is quasi-flat. Now assume that Z 6= P and Z is faithful. Let a ∈ P and t ∈ Z \ (0 : a). Let K = ker(p) and G1 = R/Rta. Then G ∼= Rt/Rta ⊆ G1. Since Q is FP-injective v extends to v1 : G1 → Q. But v1(1 + Rta) ∈ Z because it is annihilated by ta. There exists a homomorphism q′ : (G1)Z → Z n such that pZq′ = uZ(v1)Z . Let q1 be the composition of the natural map G1 → (G1)Z with q′, r = v1(1 + atR) and x = q1(1 + atR). Then u(r) − p(x) ∈ KZ/K. Let q = q1G. We have v(t + taR) = tr and q(t + taR) = tx. So, u(tr) − p(tx) = t(u(r) − p(x)) = 0. Hence uv = pq. (cid:3) Proposition 2.10. Let R be a chain ring. Then each ideal I satisfying Z ⊆ I ♯ is self-flat. Proof. By Remark 2.7 we may assume that 0 6= Z = Z 2 and Q is coherent, and by Proposition 2.2 that I ♯ = Z. We may suppose that I is neither principal over Q nor of the form aZ for any a ∈ R. We consider the following homomorphisms: p : I → I ′, u : I → I ′ and v : G → I where I ′ is an R-module, p is surjective and G = R/aR for some a ∈ P . By Lemma 2.5 I is FP-injective over R/A, where A = (0 : I) and by [7, Theorem 15(4)(c)] Z ⊗R I is flat over R/A because (0 : r) is principal over Q for each r ∈ I. Since I = ZI the canonical homomorphism φ : Z ⊗R I → I is surjective. Let r = v(1 + aR). Then r = φ(s ⊗ b) where s ∈ Z and b ∈ I. Since ar = 0 then a(s ⊗ b) ∈ ker(φ) ∼= TorQ 1 (Q/Z, I). So, aZ ⊆ (0 : s ⊗ b). Let v′ : R/taR → Z ⊗R I be the homomorphism by v′(1 + taR) = s ⊗ b where t ∈ Z. From the flatness of Z ⊗R I we deduce there exists q′ t : R/taR → I such that pq′ t(1 + taR). Then taxt = 0. Let t′ another element of Z. Thus p(xt) = p(xt′ ) = u(r), whence (xt′ − xt) ∈ ker(p) ⊂ Qxt since I is a uniserial Q-module. So, Qxt = Qxt′ and we can choose xt = xt′ = x. Hence aZ ⊆ (0 : x). t = uφv′. Let xt = q′ 6 FRANC¸ OIS COUCHOT But (0 : x) is a principal ideal of Q, whence ax = 0. If we put q(c + aR) = cx for each c ∈ R then pq = uv. (cid:3) Theorem 2.11. Let R be a chain ring. Then each ideal I is self-flat if and only if Z = N . The following can be proven by using [1, Lemmas 3.8, 3.12 and 4.5]. Theorem 2.12. [6, Theorem 4.1]. Let R a local ring and N its nilradical. Then R is a fqp-ring if and only if either R is a chain ring or R/N is a valuation domain and N is a divisible torsionfree R/N -module. Corollary 2.13. Let R be a local fqf-ring which is not a chain ring and N its nilradical. Then each ideal of R is flat modulo its annihilator. Proof. Let I be an ideal. If I ⊆ N then I is a torsionfree module over the valuation domain R/N . Hence it is a flat R/N -module. If I * N then each finitely subideal of I is principal and free. So, I is flat. (cid:3) The following corollary and example allow us to see there exist quasi-flat modules which are not flat modulo their annihilator. Corollary 2.14. Let R a chain ring. Assume that P is not principal and R is an IF-ring. Then, for each a ∈ R, aP is quasi-flat but it is not flat over R/(0 : aP ). Proof. Since R is coherent and P is not finitely generated we get that P is faithful. By [3, Theorem 10] P is not flat. Let 0 6= a ∈ P . There exists b ∈ P such that (0 : a) = Rb. So, aP ∼= P/Rb, and Rb = (0 : aP ) because P is faithful. By [3, Theorem 11] R/Rb is an IF-ring and consequently R/Rb satisfies the same conditions as R. Hence aP is quasi-flat but not flat over R/Rb. (cid:3) Example 2.15. Let R = D/dD, where D is a valuation domain with a non- principal maximal ideal and d a non-zero element of D which is not invertible. Then R satisfies the assumptions of Corollary 2.14. 3. Quasi-projective ideals over local fqp-rings An R-module M is said to be linearly compact if every finitely solvable set of congruences x ≡ xα (mod Mα) (α ∈ Λ, xα ∈ M and Mα is a submodule of M for each α ∈ Λ) has a simultaneous solution in M . A chain ring R is maximal if it is linearly compact over itself and R is almost maximal if R/A is maximal for each non-zero ideal A. Theorem 3.1. Let R be a chain ring. The following conditions are equivalent: (1) R is almost maximal and Z = N ; (2) each ideal is quasi-projective. Proof. (1) ⇒ (2). Let I be a non-zero proper ideal of R, p : I → I ′ an epimorphism, K = ker(p) and u : I → I ′ a homomorphism. First suppose that I ⊆ Z. By Theorem 2.11 I is self-flat. So, for each r ∈ I and b ∈ (0 : r) there exists yr,b ∈ I such that byr,b = 0 and u(r) = p(yr,b). Even if K 6= 0 we can take yr,b = yr,c = yr if c is another element of (0 : r). So, (0 : r) ⊆ (0 : yr). Since Q is FP-injective then yr = rxr where xr ∈ Q. We put R′ = Q/K, p′ : Q → R′ the canonical epimorphism and x′ r. If s ∈ I \ Rr then r) ∈ R′[r] = {y ∈ R′ ry = 0}. If R is almost maximal we easily check that (x′ r = p′(xr) for each r ∈ I. So, for each r ∈ I, u(r) = rx′ s − x′ FQF-RINGS 7 then the family of cosets (x′ r + R′[r])r∈I has a non-empty intersection. Let x′ be an element of this intersection. Then u(r) = rx′ for each r ∈ I. Let x ∈ Q such that p′(x) = x′. For each r ∈ I, rx ∈ rxr + K ⊆ I. If q is the multiplication by x in I then pq = u. Hence I is quasi-projective. Now suppose that Z ⊂ I. Then for each r ∈ I \ Z there exists yr ∈ I such that u(r) = p(yr). But yr = r(r−1yr) = rxr where xr ∈ Q. We do as above to show that I is quasi-projective. (cid:3) Proposition 3.2. Let R a chain ring. Assume that P = N . Then R is almost maximal if each ideal I is quasi-projective. Proof. If P is finitely generated then R is Artinian. In this case R is maximal. Now assume that P is not finitely generated. Let (aλ + Iλ)λ∈Λ be a totally ordered family of cosets such that I = ∩λ∈ΛIλ 6= 0. By [3, Lemma 29] I 6= aP for each a ∈ R. Let A = P (0 : I). First we assume that I is different of the minimal non-zero ideal when it exists. So, A ⊂ P . We have I = (0 : A) = ∩r∈A(0 : r) (if (0 : I) is not principal then A = (0 : I). If not, either I is not principal and from I = P I we deduce that (0 : (0 : I)) = (0 : A), or I is principal which implies that P is faithful and (0 : (0 : I)) = (0 : A)). Let r ∈ A. We may assume that I ⊂ Iλ for each λ ∈ Λ. Hence there exists λ ∈ Λ such that Iλ ⊆ (0 : r). We put a(r) = aλr. If Iµ ⊂ Iλ then (aµ − aλ) ∈ Iλ, whence aµr = aλr. So, in this manner, we define an endomorphism of A. Since P = N there exists c ∈ P \ A such that c2 ∈ A. Let B = (A : c). Then A = cB and c ∈ B. Let p : B → A be the homomorphism defined by p(r) = cr and u : B → A be the homomorphism defined by u(r) = a(cr), for each r ∈ B. The quasi-projectivity of B implies there exists an endomorphism q of B such that pq = u. Since (0 : c) ⊆ (0 : q(c)) and R is self FP-injective we deduce that q(c) = ca′ for some a′ ∈ R and a(cr) = cq(r) = q(cr) = rq(c) = a′cr for each r ∈ B. Let λ ∈ Λ. We have I = ∩r∈B(0 : rc). Since I ⊂ Iλ then (0 : rc) ⊆ Iλ for some r ∈ B. From I = ∩µ∈ΛIµ we deduce there exists µ ∈ Λ such that Iµ ⊆ (0 : rc). It follows that (a′ − aµ) ∈ (0 : rc). But (aµ − aλ) ∈ Iλ, so a′ ∈ (aλ + Iλ) for each λ ∈ Λ. Now we assume that I is the minimal non-zero ideal of R. In this case A = P . Let s, t ∈ P such that I = Rst. There exists λ0 ∈ Λ such that Iλ0 ⊆ Rt ⊂ P . Let Λ′ = {λ ∈ Λ Iλ ⊆ Iλ0 } . Put Jλ = (Iλ : t) and J = ∩λ∈Λ′ Jλ. Since s ∈ J \ I then J is not minimal. From (aλ − aλ0 ) ∈ Iλ0 we deduce there exists bλ ∈ R such that (aλ − aλ0 ) = tbλ for all λ ∈ Λ′. If λ, µ ∈ Λ′ such that Iµ ⊆ Iλ then we easily check that bµ ∈ bλ + Jλ. From above it follows that there exists b ∈ ∩λ∈Λ′ bλ + Jλ and it is easy to see that (aλ0 + tb) ∈ ∩λ∈Λaλ + Iλ. Hence R is almost maximal. (cid:3) Proof of (2) ⇒ (1) in Theorem 3.1. Since each ideal I is self-flat we have Z = N by Theorem 2.11. From the previous proposition we deduce that Q is almost maximal and we may assume that P 6= Z. When Z = 0 R is almost maximal by [11, Theorem 3.3]. Now suppose Z 6= 0. We shall prove that R/Z is maximal and we will conclude that R is almost maximal by using [5, Theorem 22]. Let 0 6= x ∈ Z and I a proper ideal of R such that Z ⊂ I. Since I is quasi-projective then I is (Qx/Zx)-projective by [15, 18.2]. Let q : I → Q/Z be a homomorphism. If z ∈ Z and t ∈ I \ Z then z = z′t for some z′ ∈ Z. So, q(z) = z′q(t) = 0, whence q factors through I/Z. It follows that I/Z is (Q/Z)-projective for each ideal I containing Z. By [11, Theorem 3.3] R/Z is almost maximal. Suppose that Z 2 = Z. We have that Qx is (Qx/Zx)-projective. Let q : Q → Q/Z be a homomorphism and z ∈ Z. There exist z′, t ∈ Z such that z = z′t. So, q(z) = z′q(t) = 0 whence q 8 FRANC¸ OIS COUCHOT factors through Q/Z. It follows that Q/Z is quasi-projective. If Z 6= Z 2 then Z is principal over Q and there exists x ∈ Z such that Z = (0 : x). Hence Q/Z ∼= Qx is quasi-projective. From R/Z almost maximal and [11, Theorem 3.4] we get that R/Z is maximal. (cid:3) Theorem 3.3. Let R be a local fqp-ring which is not a chain ring and N its nilradical. Consider the following conditions: (1) R is a linearly compact ring; (2) each ideal is quasi-projective; (3) N is of finite rank over R/N . Then (1) ⇔ ((2) and (3)). Proof. (1) ⇒ ((2) and (3)). Let R′ = R/N and Q′ its quotient field. Then R′ is a maximal valuation domain. Since N is a direct sum of modules isomorphic to Q′ and a linearly compact module then N is of finite rank by [15, 29.8]. Let I be an ideal contained in N . By [11, Lemma 4.4] I is quasi-projective. Now suppose that I * N . In this case N ⊂ I and since I/N is uniserial, by a similar proof as the one of Theorem 3.1 we show that I is quasi-projective. ((2) and (3))⇒ (1). Since Q′ is isomorphic to a direct summand of N , Q′ and its submodules are quasi-projective. Hence, by [11, Theorems 3.3 and 3.4] R′ is maximal. It follows that N is linearly compact. Whence R is linearly compact by [15, 29.8]. (cid:3) Remark 3.4. In the previous theorem: (1) if N is the maximal ideal of R then each ideal is quasi-projective even if N is not of finite rank over R/N ; (2) if N is not the maximal ideal and if Q/N is countably generated over R/N then (1) ⇔ (2) because (2) ⇒ (3) by [11, Lemma 4.3(b)]. 4. λ-dimension of commutative fqf-rings In this section R is a commutative ring. We say that R is arithmetical if RP is a chain ring for each maximal ideal P . An R-module E is said to be of finite n-presentation if there exists an exact sequence: with the Fi's free R-modules of finite rank. We write Fn → Fn−1 → · · · F1 → F0 → E → 0 λR(E) = sup{n there is a finite n−presentation of E}. If E is not finitely generated we also put λR(E) = −1. The λ-dimension of a ring R (λ−dim(R) is the least integer n (or ∞ if none such exists) such that λR(E) ≥ n implies λR(E) = ∞. See [13, chapter 8]. Recall that R is Noetherian if and only if λ-dim(R) = 0 and R is coherent if and only if λ-dim(R) ≤ 1. The rings of λ-dimension ≤ n are also called n-coherent by some authors. This notion of λ-dimension of a ring was formulated in [13, chapter 8] to study the rings of polynomials or power series over a coherent ring. Theorem 4.1. Let R be a local fqp-ring. Then λ−dim(R) ≤ 2. Proof. By [4, Theorem II.11] λ−dim(R) ≤ 2 if R is a chain ring. Theorem 2.12 and the following proposition complete the proof. (cid:3) FQF-RINGS 9 Proposition 4.2. Let R be a local fqp-ring which is not a chain ring and N its nilradical. Then: (1) either R is Artinian or λ−dim(R) = 2 if N is maximal; (2) λ−dim(R) = 2 if N is not maximal. Moreover, if G is a finitely 2-presented module then: (3) G is free if N is maximal and not finitely generated; (4) G is of projective dimension ≤ 1 if N is not maximal. Proof. (1) and (3). If N is an R/N -vector space of finite dimension then R is Artinian. Assume that N is not of finite dimension over R/N . Let G be an R- module of finite 2-presentation. So, there exists an exact sequence F2 u2−→ F1 u1−→ F0 → G → 0, where Fi is free of finite rank for i = 0, 1, 2. Let Gi be the image of ui for i = 1, 2. Since R is local we may assume that Gi ⊆ N Fi−1 for i = 1, 2. Then Gi is a module of finite length for i = 1, 2. It follows that F1 is of finite length too. This is possible only if F1 = 0. So, G is free. Hence λR(G) = ∞ and λ−dim(R) ≤ 2. Let 0 6= r ∈ N . Since (0 : r) = N , λR(R/rR) = 1, whence λ−dim(R) = 2. (2) and (4). Let P be the maximal ideal of R. Each r ∈ P \ N is regular. So, Rr is free and since each finitely generated ideal which is not contained in N is principal, P is flat. Let G, G1 and G2 be as in (1). Since R is local we may assume that Gi ⊆ P Fi−1 for i = 1, 2. Then TorR 2 (G, R/P ) = 0. So, the following sequence is exact: 1 (G1, R/P ) ∼= TorR 0 → G2/P G2 → F1/P F1 v−→ G1/P G1 → 0, where v is induced by u1. Since v is an isomorphism it follows that G2/P G2 = 0, and by Nakayama Lemma G2 = 0. So, G1 is free. Now, we do as in (1) to conclude (N is not finitely generated because it is divisible over R/N ). (cid:3) Let A be a ring and E an A-module. The trivial ring extension of A by E (also called the idealization of E over A) is the ring R := A ∝ E whose underlying group is A × E with multiplication given by (a, e)(a′, e′) = (aa′, ae′ + a′e). Let R be a ring. For a polynomial f ∈ R[X], denote by c(f ) (the content of f ) the ideal of R generated by the coefficients of f . We say that R is Gaussian if c(f g) = c(f )c(g) for any two polynomials f and g in R[X]. The following example shows that we cannot replace "fqf-ring" with "Gaussian ring" in Theorem 4.1. Example 4.3. Let D be a non almost maximal valuation domain and M its max- imal ideal. Let 0 6= d ∈ M such that D/Dd is not maximal and E the injective hull of D/Dd. Consider R = D ∝ E the trivial ring extension of D by E. Then R is a Gaussian local ring and λ−dim(R) ≥ 3. Proof. By [3, Theorem 17] E is not uniserial. By [6, Corollary 4.3] R is Gaussian but not a fqp-ring because E is neither uniserial nor torsionfree. Let e ∈ E such that (0 : e) = Dd. We put a = (0, e) and b = (d, 0). Then (0 : a) = Rb and (0 : b) = {(0, x) dx = 0} = 0 ∝ E[d], where E[d] = {x ∈ E dx = 0}. If D′ = D/Dd then E[d] is isomorphic to the injective hull of D′ over D′ and D′ 6= E[d] because D′ is not maximal. By [3, Theorem 11] D′ is an IF-ring and consequently E[d] and E[d]/D′ are flat over D′. Then E[d] is not finitely generated, else E[d]/D′ 10 FRANC¸ OIS COUCHOT is a free D′-module and this contradicts that E[d] is an essential extension of D′. So, (0 : b) is not finitely generated, λR(R/Ra) = 2 and λ−dim(R) ≥ 3. (cid:3) Proposition 4.4. Let R be a fqf-ring with a unique minimal prime ideal N . The following assertions hold: (1) RP is not a chain ring for each maximal ideal P if R is not arithmetical; (2) R is a fqp-ring. Proof. (1). There exists a maximal ideal L such that RL is a local fqp-ring which is not a chain ring. So, NL is torsionfree and divisible over RL/NL. Moreover, since NL is not uniserial over RL, by [1, Lemma 3.8] there exist a, b ∈ NL such that aRL ∩ bRL = 0. It follows that NL = NN and it is a vector space over RN /NN of dimension > 1. Let P be a maximal ideal. Then NN is a localization of NP . Consequently NP is not uniserial. Hence, RP is not a chain ring. (2). It follows that N is a torsionfree divisible module over R/N . So, if I is a finitely generated ideal contained in N then I is a finitely generated flat module over the Prufer domain R/N . So, I is projective over R/N . Now, if I * N then IP is a free RP -module of rank 1. We conclude by [2, Chap.2, §5, 3, Th´eor`eme 2] that I is projective. (cid:3) Corollary 4.5. Let R be a fqf-ring with a unique minimal prime ideal N . Assume that R is not arithmetical. Then either R is Artinian or λ−dim(R) = 2. Proof. When N is maximal we use Proposition 4.2. Now assume that N is not maximal. Let G be a R-module such that λR(E) ≥ 2. We use the same notations as in the proof of Proposition 4.2. This proposition implies that G1 is locally free. Since G1 is a finitely presented flat module, we successively deduce that G1 is projective, G2 is projective and ker(u2) is finitely generated. (cid:3) An integral domain D is almost Dedekind is DP is a Noetherian valuation domain for each maximal ideal P . The following example shows that we cannot remove the assumption "R is not arithmetical" in Corollary 4.5. Example 4.6. Let D be an almost Dedekind domain which is not Noetherian (see [9, Example III.5.5]), Q its quotient field, P ′ a maximal ideal of D which is not finitely generated and E = Q/DP ′. Let R = D ∝ E and N = {(0, y) y ∈ E}. Then R is an arithmetical ring, N is its unique minimal prime ideal and λ−dim R = 3. Moreover, RP is IF where P is the maximal ideal of R satisfying P ′ = P/N , and RL is a valuation domain for each maximal ideal L 6= P . Proof. For each maximal ideal L of R let L′ = L/N . Let p ∈ P ′ such that P ′DP ′ = pDP ′, x = 1/p + DP ′, a = (p, 0) and b = (0, x). Since 0 is the sole prime ideal of D contained in P ′ ∩ L′ for each maximal ideal L′ 6= P ′, then EL′ = 0. So, RL = DL′. Since E is uniserial and divisible over DP ′, RP is a chain ring by [6, Proposition 1.1]. So, R is arithmetical. We have (0 :RP b) = aRP = P RP . By [3, Theorem 10] RP is IF. Clearly Dx is the minimal submodule of E. So, P ′ = (0 : x) and DP ′x = Dx. If q ∈ Q \ DP ′ then q = spn/t where s, t ∈ D \ P ′ and n an integer > 0. So, pq ∈ DP ′ if and only if n = 1. It follows that Dx = {y ∈ E py = 0}. Let a and b be the respective multiplications in R by a and b. Then ker(a) = Rb and ker(b) = P which is not finitely generated. So, λR(R/Ra) = 2 and λ−dim(R) = 3 by [4, Theorem II.1]. (cid:3) FQF-RINGS 11 Theorem 4.7. Let R be a fqf-ring. Then λ−dim(R) ≤ 3. Proof. Let G be an R-module of finite 3-presentation. So, there exists an exact sequence F3 u3−→ F2 u2−→ F1 u1−→ F0 → G → 0, where Fi is free of finite rank for i = 0, 1, 2, 3. Let Gi be the image of ui for i = 1, 2, 3. We do as in the proof of [4, Theorem II.1]. For each maximal ideal P we shall prove that there exist tP ∈ R \ P such that λRtP (GtP ) ≥ 4. We end as in the proof of [4, Theorem II.1] to show that ker(u3) is finitely generated, by using the fact that Max R is a quasi-compact topological space. Let P be a maximal ideal. First assume that RP is a chain ring. As in the proof of [4, Theorem II.1] we show there exists tP ∈ R \ P such that λRtP (GtP ) ≥ 4. Now assume that RP is not a chain ring. We suppose that either P is not minimal or P is minimal but P RP is not finitely generated over RP . In this case (G1)P is free over RP by Proposition 4.2. Since G1 is finitely presented, there exists tP ∈ R \ P such that (G1)tP is free over RtP by [2, Chapitre 2, §5, 1, Corollaire de la proposition 2]. It follows that (G2)tP and (G3)tP are projective. So, ker((u3)tP ) is finitely generated. Finally assume that RP is not a chain ring, P is minimal and P RP is finitely generated over RP . We have P 2RP = 0. Since P 2RL = RL for each maximal ideal L 6= P , P 2 is a pure ideal of R. It follows that R/P 2 is flat. Clearly R/P 2 is local . So, RP = R/P 2. If P 2 is finitely generated then P 2 = Re where e is an idempotent of R. So, if tP = 1 − e then D(tP ) = {P }, RtP = RP and ker((u3)tP ) is finitely generated. If P 2 is not finitely generated then P = I + P 2 where I is finitely generated but not principal because so is P/P 2. Since I 2 is a finitely generated subideal of the pure ideal P 2 there exists a ∈ P 2 such that r = ar for each r ∈ I 2. It follows that (1 − a)I 2 = 0. Hence I 2Rt = 0 where t = (1 − a) and IRt 6= 0 because for each s ∈ I \ P 2, s 6= sa. Since Gt is finitely generated, after possibly multiplying t with an element in R \ P , we may assume that Gt has a generating system {g1, . . . , gp} whose image in (Gt)P is a minimal generating system of (Gt)P containing p elements. Let F ′ 0 be a free Rt-module with basis {e1, . . . , ep}, π : F ′ 0 → Gt be the homomorphism defined by π(ek) = gk for k = 1, . . . , p and G′ 1 = ker(π). We get the following commutative diagram with exact rows and columns: 0 ↓ 0 ↓ 0 ↓ 0 → P 2G′ 1 → P 2F ′ 0 π−→ P 2Gt → 0 π−→ Gt → 0 πP−−→ (Gt)P → 0 ↓ ↓ 1 → F ′ 0 → G′ 0 ↓ ↓ 1)P → (F ′ 0 → (G′ 0)P ↓ ↓ 0 0 1 ⊆ P F ′ 0)P , we have that G′ ↓ ↓ ↓ 0 1)P ⊆ P (F ′ Since (G′ after possibly multiplying t with and element of R\P , we may assume that G′ and that G′ 1 is finitely generated, 1 ⊆ IF ′ 0 1)P is a minimal 1 has a generating system {g′ 0. But, since G′ 1, . . . , g′ q} whose image in (G′ 12 FRANC¸ OIS COUCHOT 1, . . . , e′ q}, u′ 1 : F ′ k for k = 1, . . . , q and G′ 2 = ker(u′ 2 is contained in IF ′ 1)P with q elements. Let F ′ 0 defined by u′ k) = g′ 1(e′ generating system of (G′ {e′ 1 → F ′ Again, for a suitable t ∈ R \ P we may assume that G′ has a generating system whose image in (G′ the same cardinal. Since I 2 generated by {r1, . . . , rn} then G′ F ′ the homomorphism defined by u′ G′ 3 = ker(u′ get that G′ let F ′ defined like u′ 3) = IF ′ for a suitable tP ∈ R \ P we have λRtP (GtP ) ≥ 4. 1 a free Rt-module with basis 1). 1 and 2)P is a minimal generating system with 2 = IF ′ 1. Consequently, if It is k 1 ≤ i ≤ n, 1 ≤ k ≤ q}. Let 1 be k for i = 1, . . . , n and k = 1, . . . , q and 3 is finitely generated, as above, for a suitable t ∈ R \ P , we 2. We easily deduce that It = (0 :Rt ri) for each i = 1, . . . , n. Now, 2 be the homomorphism 3 and it is finitely generated. So, (cid:3) 2 be a free Rt-module with basis {ǫi,k 1 ≤ i ≤ n, 1 ≤ k ≤ q}, u′ t = 0, it follows that G′ 2 is generated {rie′ 3 be a free Rt-module of rank qn2 and u′ 2. Then we get that ker(u′ 2). Since G′ 3 = ItF ′ 3 : F ′ 3 → F ′ 2(ǫi,k) = rie′ 2 : F ′ 2 → F ′ With a similar proof as the one of [4, Corollary II.13], and by using Proposition 4.2 we get the following theorem. Theorem 4.8. Let R be a fqf-ring. Assume that RP is either an integral domain or a non-coherent ring for each maximal ideal P which is not an isolated point of Max R. Then λ−dim(R) ≤ 2. Proof. Let G be an R-module of finite 2-presentation. So, there exists an exact sequence F2 u2−→ F1 u1−→ F0 → G → 0, where Fi is free of finite rank for i = 0, 1, 2. Let Gi be the image of ui for i = 1, 2. We do as in the proof of the previous theorem. First suppose that P is a non- isolated point of Max R. In this case (G1)P is free over RP by Proposition 4.2. Since G1 is finitely presented, there exists tP ∈ R \ P such that (G1)tP is free over RtP by [2, Chapitre 2, §5, 1, Corollaire de la proposition 2]. It follows that (G2)tP is projective. So, ker((u2)tP ) is finitely generated. Now assume that P is isolated. There exists tP ∈ R\ P such that RP ∼= RtP . By Theorem 4.1 ker((u2)tP ) is finitely generated. (cid:3) Example 4.6 and the following show that the assumption "RP is a non-coherent ring" cannot be removed in Theorem 4.8. Example 4.9. Let A be a von Neumann regular ring which is not self-injective, H the injective hull of A, x ∈ H \ A, E = A + Ax and R = A ∝ E. Then: (1) R is a fqf-ring which is not an fqp-ring; (2) for each maximal ideal P , RP is Artinian; (3) λ−dim(R) = 3. Proof. Let N = {(0, y) y ∈ E}, a = (0, 1) and b = (0, x). (1). See [6, Example 4.6]. (2). If P is a maximal ideal of R then RP is the trivial ring extension of the field AP ′ by the finite dimensional vector space EP ′ where P ′ = P/N . Hence RP is Artinian. (3). Consider the following free resolution of R/aR: R2 u2−→ R u1−→ R → R/Ra → 0 where u2((r, s)) = ra + sb for each (r, s) ∈ R2 and u1(r) = ra for each r ∈ R. We easily check that this sequence of R-modules is exact. The A-module E is not FQF-RINGS 13 finitely presented, else E/A is finitely presented and, since each exact sequence of A-modules is pure, A is a direct summand of E which contradicts that A is essential in E. Consequently, N , which is the image of u2, is not finitely presented. So, λR(R/Ra) = 2 and λ−dim(R) = 3. (cid:3) References [1] J. Abuhlail, V. Jarrar, and S. Kabbaj. Commutative rings in which every finitely generated ideal is quasi-projective. J. Pure Appl. Algebra, 215:2504 -- 2511, (2011). [2] N. Bourbaki. Alg`ebre commutative, Chapitres 1 et 2. Hermann, Paris, (1961). [3] F. Couchot. Injective modules and fp-injective modules over valuation rings. J. Algebra, 267:359 -- 376, (2003). [4] F. Couchot. The λ-dimension of commutative arithmetic rings. Comm. Algebra, 31(7):3143 -- 3158, (2003). [5] F. Couchot. Pure-injective hulls of modules over valuation rings. J. Pure Appl. Algebra, 207:63 -- 76, (2006). [6] F. Couchot. Trivial ring extensions of Gaussian rings and fqp-rings. Comm. Algebra, 43(7):2863 -- 2874, (2015). [7] F. Couchot. Weak dimension of fp-injective modules over chain rings. Comm. Algebra, 44:381 -- 389, (2016). [8] L. Fuchs and L. Salce. Modules over valuation domains, volume 97 of Lecture Notes in Pure and Appl. Math. Marcel Dekker, New York, (1985). [9] L. Fuchs and L. Salce. Modules over Non-Noetherian Domains. Number 84 in Mathematical Surveys and Monographs. American Mathematical Society, Providence, (2001). [10] D.T. Gill. Almost maximal valuation rings. J. London Math. Soc., 4:140 -- 146, (1971). [11] P. Hermann. Self-projective modules over valuation rings. Arch. Math., 43:332 -- 339, (1984). [12] G.B. Klatt and L.S. Levy. Pre-self injectives rings. Trans. Amer. Math. Soc., 137:407 -- 419, (1969). [13] W.V. Vasconcelos. The rings of dimension two, volume 22 of Lecture Notes in pure and applied Mathematics. Marcel Dekker, (1976). [14] R. Warfield. Decomposability of finitely presented modules. Proc. Amer. Math. Soc., 25:167 -- 172, (1970). [15] R. Wisbauer. Foundations of Module and Ring Theory. (1991). Normandie Univ, UNICAEN, CNRS, LMNO, 14000 Caen, France E-mail address: [email protected]
1106.0350	1	1106	2011-06-02T00:44:39	On realizing zero-divisor graphs of po-semirings	[ "math.RA", "math.AC", "math.CO" ]	In this paper, we determine bipartite graphs and complete graphs with horns, which are realizable as zero-divisor graphs of po-semirings. As applications, we classify commutative rings $R$ whose annihilating-ideal graph $\mathbb {AG}(R)$ are either bipartite graphs or complete graphs with horns.	math.RA	math	On realizing zero-divisor graphs of po-semirings ∗ Houyi Yu† and Tongsuo Wu‡ Department of Mathematics, Shanghai Jiaotong University, Shanghai, 200240, China 1 1 0 2 n u J 2 ] . A R h t a m [ 1 v 0 5 3 0 . 6 0 1 1 : v i X r a Abstract. In this paper, we determine bipartite graphs and complete graphs with horns, which are realizable as zero-divisor graphs of po-semirings. As applications, we classify commutative rings R whose annihilating-ideal graph AG(R) are either bipartite graphs or complete graphs with horns. Key Words: Po-semirings; Graph properties; Zero-divisors; Annihilating- ideals of a commutative ring MSC(2000): 13A15; 05C75. 1 Introduction Throughout this paper, all rings are assumed to be commutative with identity. For a ring R, let Z(R) be its set of zero-divisors. The zero-divisor graph of R, denoted by Γ(R), is a simple graph (i.e., an undirected graph without loops and multiple edges) with vertices Z(R)∗ = Z(R)\{0}, such that distinct vertices x and y are adjacent if and only if xy = 0. The concept for a ring was first introduced and studied by Beck in [8] and further investigated by many authors, see [6, 5, 4]. Later, DeMeyer, McKenzie and Schneider [12] extended the notion to commutative semigroups S with 0 in a similar manner. The idea establishes a connection between graph theory and algebraic theory and will be beneficial for those two branches of mathematics. For a ring R, let I(R) be the set of ideals of R, A(R) the set of annihilating-ideals of R, where a nonzero ideal I of R is called an annihilating-ideal if there exists a non-zero ideal J of R such that IJ = 0. The annihilating-ideal graph AG(R) of R, first introduced and ∗The first author is partly supported by Shanghai Jiaotong University Innovation Fund for Postgrad- uates (NO. AE071202). †[email protected] (H.Y. Yu) ‡Corresponding author, [email protected] (T.S. Wu) 1 studied in [9], is a graph with vertex set A(R)∗ = A(R)\{0}, such that distinct vertices I and J are adjacent if and only if IJ = 0. The graph provides an excellent setting for studying some aspects of algebraic property of a commutative ring, especially, the ideal structure of a ring. Clearly, the graph AG(R) is an empty graph if and only if R is an integral domain. In fact, I(R) admits a natural algebraic structure, called a po-semiring by Wu, Lu and Li [18]. Recall that a commutative semiring is a set A which contains at least two elements 0, 1 and is equipped with two binary operations, + and ·, called addition and multiplication respectively, such that the following conditions hold: (1) (A, +, 0) is a commutative monoid with zero element 0. (2) (A, ·, 1) is a commutative monoid with identity element 1. (3) Multiplication distributes over addition. (4) 0 annihilates A with respect to multiplication, i.e., 0a = 0, ∀a ∈ A. Recall the following definition from [18]. Definition 1.1. A partially-ordered semiring is a commutative semiring (A, +, ·, 0, 1), together with a compatible partial order ≤, i.e., a partial order ≤ on the underlying set A that is compatible with the semiring operations in the sense that it satisfies the following conditions: (5) x ≤ y implies x + z ≤ y + z, and (6) 0 ≤ x and y ≤ z imply that xy ≤ xz for all x, y, z in A. If A satisfies the following additional condition, then A is called a po-semiring: (7) The partially ordered set (A, ≤, 0, 1) is bounded, i.e., 1 is the largest element and 0 is the least element of A. Note that condition (7) is a very strong assumption. Under the assumption, a po- semiring A is in fact a dioid, where a semiring is called a dioid if its addition is idempotent (a + a = a, ∀a ∈ A). Furthermore, the above defined partial order ≤ for a po-semiring A is identical with the new partial order ≤1 defined by the following a ≤1 b if and only if a + b = b. In other words, (A, +, 0, 1) is a bounded join-semilattice. Clearly, any bounded, distribu- tive lattice is a po-semiring under join and meet, where a ≤ b if and only if a ∧ b = a. We remark that the class of po-semirings is much smaller than the class of semigroups. For example, the five-element lattice M5 depicted in Figure 1 is not a distributive lattice, so it is not a po-semiring. But (M5, ∧) is clearly a semigroup. For a po-semiring A, denote by Z(A) the set of all multiplicative zero-divisors. The zero-divisor graph of the multiplicative semigroup (A, ·, 1), denoted by Γ(A), is called the 2 zero-divisor graph of the po-semiring A. Clearly, all known results on zero-divisor graphs of semigroups hold for Γ(A). A nonzero element x ∈ A is called minimal, if 0 < y ≤ x implies x = y for any y ∈ A. Note that each minimal element of A is a zero divisor, if Z(A) ≥ 2. Refer to [18] for more details on po-semirings. The prototype of a po-semiring is the po-semiring I(R) of a commutative ring R. The multiplication is the ideal multiplication, the addition is the addition of ideals, the partial order is the usual inclusion. Therefore, the annihilating-ideal graph of R is the zero-divisor graph of the po-semiring I(R), i.e., AG(R) = Γ(I(R)). All throughout, let G be a finite or an infinite simple graph. The vertex set of G is denoted by V (G). The core of G, denoted by C(G), is the largest induced subgraph of G in which every edge is an edge of a cycle in G. For a vertex x of G, let N(x) be the set of vertices adjacent to x, and call it the neighborhood of x. A vertex is called an end vertex if its degree is 1. All end vertices which are adjacent to a same vertex of G together with the edges is called a horn. We adopt more graph theoretic notations from [11]. We recall some notation used in [14]. Let X and Y be disjoint nonempty subsets of the vertex set of a graph. We use the notation X − Y to represent the complete bipartite graph with parts X and Y . In particular, if u 6∈ X, then u − X represents a star graph, that is, u is adjacent to every vertex in X and no two distinct vertices in X are adjacent. The graph depicted in Figure 2 is called a complete bipartite graph with a horn, where the induced subgraphs on nonempty sets X, Y, U are discrete. We denote the graph by X − U − v − Y . In particular, G is called a two-star graph if U = 1. a b 1 0 c X v U Y Figure 1 Figure 2 a3 X Y a1 a2 Figure 3 Let Λ be an index set, and let m, n be two finite or infinite cardinal numbers such that n = Λ ≥ 1, 0 ≤ m ≤ n, and let Kn be a complete graph with V (Kn) = {ai i ∈ Λ}. We denote the complete graph Kn together with m horns X1, X2, · · · , Xm by Kn(m), where a1 − X1, a2 − X2, · · · , am − Xm. For example, the graph in Figure 3 above is K3(2). Clearly, K1(0) is an isolated vertex, K1(1), K2(0) (i.e., K2) and K2(1) are star graphs, K2(2) is a two-star graph, K3(0) is a triangle, while K3(m) is a triangle with m horns (1 ≤ m ≤ 3). In this paper, we mainly study the case of m ≤ 3. So we always assume 3 a1 − X, a2 − Y, a3 − Z for brevity. As in [13], a graph will be called realizable (for po-semirings) if it is isomorphic to Γ(A) for some po-semiring A. In this paper, we investigate the realization problem of graphs as zero-divisor graphs of po-semirings. In Section 2, it is shown that a bipartite graph G is realizable if and only if G is either a complete bipartite graph or a complete bipartite graph with a horn. Realizable complete graphs with horns are then completely determined in Section 3. In particular, it is shown that Kn(m) is the zero-divisor graph of some po-semiring if and only if either 0 ≤ m ≤ min{2, n} or m = n = 3. The final section is devoted to the study of annihilating-ideal graph AG(R) of a commutative ring It is proved that AG(R) is a complete bipartite graph with a horn if and only if R. R ∼= D × S, where D is an integral domain and S is a ring with a unique non-trivial ideal. AG(R) ∼= K3(3) if and only if R ∼= F1 × F2 × F3, where F1, F2, F3 are fields. We also show that there exists no ring R such that AG(R) ∼= Kn(2) for any n ≥ 3. Throughout the paper, set X = {xi i ∈ Γ}, Y = {yk k ∈ Θ}, Z = {zp p ∈ Ω}, U = {us s ∈ Φ}. 2 Bipartite graphs which are realizable for po-semirings In this section, we give a complete classification of all bipartite graphs which can be realized as po-semiring graphs. Lemma 2.1. Any complete bipartite graph G is realizable for po-semirings. P roof. Assume that G has X and Y as its two parts. Set A = {0, 1, w} ∪ X ∪ Y and define a partial order ≤ by 0 < x1 < x2 < · · · < w < 1, 0 < y1 < y2 < · · · < w < 1. Define a commutative addition by b + c =( max{b, c} w if b and c are comparable in A, otherwise. Define a commutative multiplication on A by 0a = 0, 1a = a (∀a ∈ A), w2 = w, xixj = x1, xiyk = 0, xiw = x1, ykyl = y1, ykw = y1 (i, j ∈ Γ, k, l ∈ Θ). Clearly, A is a po-semiring such that Γ(A) ∼= G. (1) (cid:3) 4 Lemma 2.2. Let G be a complete bipartite graph with a horn and set G : X − U − v − Y . Then there exists a po-semiring A such that Γ(A) ∼= G. P roof. Let A = {0, 1, v, w} ∪ X ∪ Y ∪ U. We define a partial order ≤ on A by 0 < v < x1 < x2 < · · · < w < 1, 0 < v < y1 < y2 < · · · < w < 1, 0 < u1 < u2 < · · · < y1 < y2 < · · · < w < 1, Define a commutative addition by ui + v = y1 for any ui ∈ U, and other additions are defined by (1). Clearly, (A, +) is a commutative semigroup. Now we define a commutative multiplication on A such that (A, +, ·, 0, 1) is a po- semiring. For any a ∈ A, let 0a = 0, 1a = a. Furthermore, for any i, j ∈ Γ, k, l ∈ Θ and s, t ∈ Φ, let xixj = xiw = x1, xiyk = xiv = v, xius = 0, ykyl = ykus = u1, ykv = 0, ykw = y1, usut = usw = u1, usv = v2 = 0, vw = v, w2 = w. A direct checking shows that the associativity holds for (A, ·). Since the multiplication is commutative, we only need to verify that the left distributivity holds for (A, +, ·, 0, 1). In fact, take any a, b, c ∈ A and it is obvious, by the addition and multiplication defined above, that b < c implies b + c = c and ab ≤ ac, so a(b + c) = ac = ab + ac. Thus, we only need to assume that b and c are incomparable and show that a(b + c) = ab + ac. If a = us, then us(ut + v) = usy1 = u1 = usut + usv, us(ut + xi) = usw = u1 = usut + usxi, us(xi + yk) = usw = u1 = usxi + usyk. Similarly, we can show that a(b + c) = ab + ac always holds when a ∈ {v, xi, yk, w} and b, c ∈ A. Hence (A, +, ·, 0, 1) is a posemiring. Clearly, Γ(A) ∼= G and the result follows. (cid:3) In order to check that (A, +, ·, 0, 1) defined above is a po-semiring, what is the most difficult is to check the two associative laws and one distributive law hold. In fact, each examination concerns three elements at most. By the definition of addition and multiplication in Lemma 2.2, it is enough to take X = Y = U = 3 so that A = 13, and hence we can do the examination by taking the advantage of a computer. 5 In general, a po-semiring corresponding to a given graph need not be unique up to isomorphism. For example, for the given complete bipartite graph with a horn X − U − v − Y , we can define another po-semiring A′ such that A′ is not isomorphic to the one defined in Lemma 2.2. Example 2.3. Set X + Y = {x + y x ∈ X, y ∈ Y }, and set A′ = {0, 1, v} ∪ X ∪ Y ∪ U ∪ (X + Y ). Let (A′, +) be an upper semilattice with the Hasse diagram in Figure 4. In 1 xi + yk yk xi y2 y1 us u1 x2 x1 v 0 Figure 4 particular, x1, y1 and u1 are the unique minimal elements in X, Y , and U, respectively. Now we define a commutative multiplication on A such that (A′, +, ·, 0, 1) is a po-semiring. For any i, j ∈ Γ, k, l ∈ Θ, s, t ∈ Φ, we define a multiplication on A′ by 0a = 0, 1a = a (∀a ∈ A′), xixj = min{xi, xj}, xiyk = xiv = v, xius = 0, xi(xj + yk) = xixj, ykyl = ykus = u1, ykv = 0, yk(xi + yl) = y1, usut = us(xi + yk) = u1, usv = 0, v2 = 0, v(xi + yk) = v, (xi + yk)(xj + yl) = xixj + y1. Then (A′, +, ·, 0, 1) is a po-semiring with Γ(A′) is isomorphic to the given graph. Clearly, A′ is not isomorphic to the po-semiring A constructed in Lemma 2.2, if X ≥ 2. The following is the main result of this section. 6 Theorem 2.4. Let G be the zero-divisor graph of a po-semiring with V (G) ≥ 2. Then the following conditions are equivalent. (i) G is triangle-free. (ii) G is a bipartite graph. (iii) G is either a complete bipartite graph or a complete bipartite graph with a horn. P roof. Note that any star graph is a complete bipartite graph and any two-star graph is a complete bipartite graph with a horn, so if G = Γ(S) for some zero-divisor semigroup S, then the equivalence of (i) and (ii) follows from [14, Theorem 2.1] while the equivalence of (ii) and (iii) follows from [14, Theorem 2.10]. Therefore, we only need to show that both complete bipartite graphs and complete bipartite graphs with a horn can be realized for po-semiring, which follows from Lemmas 2.1 and 2.2. (cid:3) Lemma 2.5. Any isolated vertex G is realizable for po-semirings. P roof. Let G = {a}. Set A = {0, 1, a}, and define a partial order < by 0 < a < 1. Define a commutative addition by x + y = max{x, y} for any x, y ∈ A. Define a commutative multiplication by 0x = 0, 1x = x (∀ x ∈ A), a2 = 0. Then it is routine to check that A is a po-semiring such that Γ(A) ∼= G. (cid:3) Recall that a tree is a simple connected graph G without a cycle, i.e., the core of G is empty. By Theorem 2.4 and Lemma 2.5 , we have the following corollary. Corollary 2.6. Let G be a tree. Then G is the graph of a po-semiring if and only if G is one of the following graphs: an isolated vertex, a star graph, a two-star graph. 3 Complete graphs with horns which are realizable for po-semirings In this section we classify all po-semiring graphs which are complete graphs or complete graphs with horns. Note that the following facts were already obtained in section 2: Kn(m) is realizable for po-semirings, for any m, n with 1 ≤ n ≤ 2, 0 ≤ m ≤ n. Thus we only need to consider the case of n ≥ 3 in this section. Lemma 3.1. For any n ≥ 3 and m ∈ {0, 1, 2}, there exists a po-semiring A such that Kn(m) ∼= Γ(A). 7 It suffices to show that each graph of the given type has a corresponding po- P roof. semiring. Case 1. Suppose that m = 0. Then G ∼= Kn is a complete graph where V (Kn) = {ai i ∈ Λ}. Set A = {0, 1} ∪ {ai i ∈ Λ}, and define a partial order ≤ on A by 0 < a1 < a2 < · · · < 1. Define an addition on A by x + y = max{x, y} for any x, y ∈ A. Define a commutative multiplication on A by 0x = 0, 1x = x (∀x ∈ A) and aiaj = 0 (∀i, j ∈ Λ). Then it is routine to check that A is a po-semiring with Γ(A) ∼= Kn. Case 2. Suppose that m = 1. Set A = {0, 1} ∪ {ai i ∈ Λ} ∪ X and define a partial order ≤ on A by 0 < a1 < a2 < · · · < x1 < x2 < · · · < 1. Define an addition by x + y = max{x, y} for any x, y ∈ A. Define a commutative multiplication by 0a = 0, 1a = a (∀a ∈ A), aiaj = a1xk = 0, arxk = ar, xkxl = x1 (∀i, j, r ∈ Λ, r 6= 1, ∀k, l ∈ Γ). Then (A, +, ·, 0, 1) is a po-semiring such that Γ(A) ∼= Kn(1), where a finite or an infinite horn X is adjacent to the vertex a1. Case 3. Suppose that m = 2. Denote X + Y = {x + y x ∈ X, y ∈ Y }, and set A = {0, 1} ∪ {ai i ∈ Λ} ∪ X ∪ Y ∪ (X + Y ). Let (A, +) be a upper semilattice with the Hasse diagram in Figure 5. In particular, an is the unique maximal element of {ai i ∈ Λ}. Now we define a commutative multiplication on A such that (A, +, ·, 0, 1) is a po-semiring. For any i, j ∈ Λ, k, l ∈ Γ, p, q ∈ Θ, we define a multiplication on A by 0a = 0, 1a = a (∀a ∈ A), aiaj = 0 (j 6= n), a2 n = an, a1xk = 0, a1yp = a1(xk + yp) = a1, a2xk = a2, a2yp = 0, a2(xk + yp) = a2, anxk = anyp = an(xk + yp) = an, aixk = a2, aiyp = a1, ai(xk + yp) = a3, (i 6= 1, 2, n), xkxl = x1, xkyp = an, xk(xl + yp) = x1, ypyq = yp(xk + yq) = y1, (xk + yp)(xl + yq) = x1 + y1. Then (A, +, ·, 0, 1) is a po-semiring with Γ(A) ∼= Kn(2). (cid:3) 8 1 xk + yp xk x2 x1 yp y2 y1 an ai a3 a1 a2 0 Figure 5 Lemma 3.2. There exists a posemiring A such that Γ(A) ∼= K3(3). P roof. Set A = {0, 1, a1, a2, a3, w} ∪ X ∪ Y ∪ Z. Define a partial order on A by 0 < a1, a2 < z1 < z2 < · · · < w < 1, 0 < a2, a3 < x1 < x2 < · · · < w < 1, 0 < a1, a3 < y1 < y2 < · · · < w < 1. Then A is a partially-ordered set with a unique maximal element w 6= 1, where a1, a2 and a3 are the only nonzero minimal elements. Define a commutative addition by 0 + a = a + a = a, 1 + a = 1, (∀a ∈ A), a1 + a2 = z1, a1 + a3 = y1, a2 + a3 = x1, and for any b, c ∈ A such that {b, c} * {a1, a2, a3}, we define b + c =( max{b, c} w if b and c are comparable in A, otherwise. (2) For any k ∈ Γ, p ∈ Θ and s ∈ Ω, define a commutative multiplication by 0a = 0, 1a = a (∀a ∈ A), 9 a2 1 = a1, a1a2 = a1a3 = a1xk = 0, a1yp = a1zs = a1w = a1, a2 2 = a2, a2a3 = a2yp = 0, a2xk = a2zs = a2w = a2, a2 3 = a3, a3zs = 0, a3xk = a3yp = a3w = a3, X 2 = {x1}, XY = {a3}, XZ = {a2}, wX = {x1}, Y 2 = {y1}, Y Z = {a1}, wY = {y1}, Z 2 = {z1}, wZ = {z1}, w2 = w. It is easy to check that (A, +, ·, 0, 1) is a po-semiring such that Γ(A) ∼= K3(3). (cid:3) The following is an improvement of [16, Theorem 2.2]. Lemma 3.3. For any n ≥ 4 and any m with n ≥ m ≥ 3, Kn(m) has no corresponding semigroups. P roof. Assume to the contrary that S = {0} ∪ {ai i ∈ Λ} ∪ X1 ∪ X2 ∪ · · · ∪ Xm is a zero-divisor semigroup such that Γ(S) ∼= Kn(m), where n ≥ 4 and n ≥ m ≥ 3. First we claim that for any xj ∈ Xj, we must have aixj = ai, where 1 ≤ i 6= j ≤ m. In fact, if aixj = xk for some xk ∈ Xk, then ajxk = ai(ajxj) = 0 so that j = k whence xixj = xi(aixj) = 0, a contradiction. Thus, aixj ∈ {ai i ∈ Λ}. Let aixj = ar, then for any x′ i)xj = 0 and hence r = i, that is, aixj = ai. i ∈ Xi, arx′ i = (aix′ Take x2 ∈ X2, x3 ∈ X3. Then a1(x2x3) = (a1x2)x3 = a1x3 = a1 so that x2x3 6∈ {ai i ∈ Λ, i 6= 1}. On the other hand, a2(x2x3) = 0 = a3(x2x3) which means that x2x3 ∈ {ai i ∈ Λ} and hence x2x3 = a1. So (a4x2)x3 = a4a1 = 0, which implies that a4x2 = a3. Substituting a3, x3, a1, x1 by a1, x1, a3, x3 respectively, we can obtain that a4x2 = a1, a contradiction. (cid:3) Note that a po-semiring is certainly a multiplicative commutative semigroup. So we have Corollary 3.4. For any n ≥ 4 and any m with n ≥ m ≥ 3, Kn(m) has no corresponding po-semirings. In view of Lemmas 2.5, 3.1, 3.2 and 3.3, we obtain the main result of this section. Theorem 3.5. Let G = Kn(m) be a complete graph Kn together with m horns, where n and m could be any cardinal numbers such that 0 ≤ m ≤ n and n ≥ 1. Then G is a po-semiring graph if and only if either 0 ≤ m ≤ min{2, n} or n = m = 3. 10 4 Some results on annihilating-ideal graphs of com- mutative rings The annihilating-ideal graph AG(R) of a commutative ring R was introduced in [9] and further studied in [10, 1, 2, 3]. In this section, we add more results on the graph. Lemma 4.1. Let A be a po-semiring and let x − u − y be a path in Γ(A). If it is contained in neither triangle nor quadrilateral, then u is a minimal element of A. P roof. Assume to the contrary that u is not a minimal element of A. Assume further 0 < v < u for some v ∈ A. If v = x, then xy = vy ≤ uy = 0 whence there is a triangle x − u − y − x, a contradiction. Similarly, v 6= y. If v 6= x, y, then there is a square x − u − y − v − x, another contradiction. (cid:3) Lemma 4.2. Let R be a ring, I a minimal ideal of R. Then ann(I) is a maximal ideal of R. P roof. Since I is minimal, I is a principal ideal. Suppose that I = Rx for some x ∈ I. Then ann(x) = ann(I) whence Rx ∼= R/ann(I) is a simple R-module, so ann(I) is a maximal ideal. (cid:3) In view of [10, Theorem 2.3] and [3, Corollary 23], we known that AG(R) is a complete bipartite graph if and only if either AG(R) is a star graph or R is a reduced ring with Min(R) = 2. For a complete bipartite graph with a horn, we have Theorem 4.3. The graph AG(R) is a complete bipartite graph with a horn if and only if R ∼= D × S, where D is an integral domain and S is a ring with a unique non-trivial ideal. In this case, AG(R) ∼= K1 + Dr + K1 + Dr where r is either 1 or an infinite cardinal number. P roof. (⇒) Let AG(R) be a complete bipartite graph with a horn X − U − c − Y . Set p = Ppi ∈X pi, q = Pqj ∈Y qj and u = Puk ∈U uk. Clearly, cq = 0. But for any pi ∈ X, uk ∈ U, qpi 6= 0, quk 6= 0, which means that q ∈ Y . Similarly, we have p ∈ X and u ∈ U. If U = 1, then AG(R) is a two-star graph. By [2, Theorem 2], AG(R) ∼= P4 and R ∼= F × S where F is a field and S is a ring with a unique non-trivial ideal. So we assume that U ≥ 2 in next discussion. Now, we show the following claims: Claim 1. q is a maximal ideal. By Lemma 4.1, c is a minimal ideal. Note that cq = 0, so we only need to show that ann(c) = q by Lemma 4.2. In fact, first we have 11 pq 6= 0. On the other hand, for any uj ∈ U, uj(pq) = 0 = c(pq) which together with U ≥ 2 yield that pq = c. Thus, c2 = (cq)p = 0 whence c(c + q) = 0, which together with u(c + q) = uq 6= 0 and p(c + q) 6= 0 imply that c + q ∈ Y and hence c ⊆ q. Since c(u + q) = 0, but p(u + q) = pq = c 6= 0 so that u + q ∈ Y , which means that u ⊆ q. Therefore, we obtain that ann(c) = c ∪ u ∪ q = q. Claim 2. p is a prime ideal and p + q2 = R. Take any x ∈ u. Then p(Rx) ⊆ pu = 0 = c(Rx), so Rx ∈ U. Clearly, x2 6= 0. Otherwise, (Rx)(Rx + c) = 0 = c(Rx + c). Since c is minimal, it follows that Rx + c 6= c, so Rx + c ∈ U and hence Rx + c = Rx. Thus c ⊆ Rx and hence cp ⊆ p(Rx) = 0, a contradiction. So x2 6= 0 whence ann(x) = c ∪ p. Since c is minimal, it follows that cp = c whence c ⊆ p and hence ann(x) = p. If p is not a maximal ideal contained in Z(R), then there exists y ∈ Z(R)\p such that p + Ry ⊆ Z(R). By the proof of Claim 1, we obtain that u ⊆ q and c ⊆ q, so we have Z(R) = p ∪ u ∪ c ∪ q = p ∪ q whence y ∈ q\p. Take any z ∈ p such that cz 6= 0. Then Rz ∈ X and hence c(y +z) = cz 6= 0 so that y +z 6∈ q, at the same time, u(y +z) = uy 6= 0 which implies that y + z 6∈ p, that is, y + z 6∈ Z(R), a contradiction. This proves that p is maximal among all ideals of R that are annihilators of elements, so p is a prime ideal. Clearly, p and q are incomparable. Otherwise, we have p ⊆ q, so c = pc ⊆ qc = 0, a contradiction. Therefore, p + q2 = R, as required. Claim 3. p ∩ q2 = 0. Note that for any uk ∈ U, we have uk(p ∩ q2) ⊆ ukp = 0 and c(p ∩ q2) ⊆ cq = 0 which together with U ≥ 2 yield that p ∩ q2 = c or p ∩ q2 = 0. If p ∩ q2 = c, then c ⊆ q2, hence c = pc ⊆ pq2 = cq = 0, a contradiction. So, p ∩ q2 = 0. By Chinese Remainder Theorem, we have R ∼= R/p × R/q2. Clearly, R/p is an integral domain, R/q2 is an Artinian local ring. Thus all non-trivial ideals of R/q2 are annihilating-ideals. If there exist two non-trivial ideals, say m, n, in R/q2, then m, n ⊆ q/q2 and hence mn = 0. So we have a triangle (R/p, 0) − (0, m) − (0, n) − (R/p, 0) in AG(R), which contradicts the fact that AG(R) is a bipartite graph. This complete the proof of necessity. (⇐) Let R = D × S where D is an integral domain, S has a unique non-trivial ideal m. If D is not a field, then D has infinitely many ideals, and AG(R) is the complete bipartite graph with a horn X − U − c − Y , where c = (0, m), X = {(0, S)}, Y = {(b, m) b ∈ I(D)∗}, U = {(b, 0) b ∈ I(D)∗}. So AG(R) ∼= K1 + Dr + K1 + Dr, where r is the cardinality of I(D)∗. In this case, r = ∞. If D is a field, then r = 1 and AG(R) ∼= P4. This complete the proof of the theorem. (cid:3) 12 Remark 4.4. Let R be a ring with only one non-trivial ideal. Then, by [15], either R ∼= K[x]/(x2) where K is a field or R ∼= V /p2V , where V is a discrete valuation ring of characteristic zero and residue field of characteristic p, for some prime number p. In the latter case, the characteristic of R is p2. By [1, Theorem 3], the graph AG(R) is a complete graph if and only if either AG(R) ∼= K2 or Z(R) is an ideal of R with Z(R)2 = 0. Moreover, in the first case, either R ∼= F1×F2, where F1, F2 are fields, or (R, m) is a local ring with exactly two non-trivial ideals m and m2. For complete graphs with horns, K1(0) is an isolated vertex, K1(1), K2(0) and K2(1) are star graphs, while K2(2) is a two-star graph. By [1, Theorem 2] or Theorem 4.3 above, AG(R) is a two-star graph if and only if AG(R) ∼= P4 , if and only if R ∼= F × S, where F is a field and S is a ring with a unique non-trivial ideal. For Kn(m) where n ≥ 3 and 1 ≤ m ≤ 3, we have the following results. Lemma 4.5. Let AG(R) ∼= Kn(1), where n ≥ 3. Then Z(R) is a maximal ideal. If further R is Artinian, then Z(R)5 = 0. P roof. Let c be the only center of AG(R), that is, the only horn is adjacent to c. Then, by Lemma 4.1, c is a minimal ideal. Take distinct a, b ∈ V (Kn)\{c}. Then a(c + b) = 0 so that c + b ∈ V (Kn). Obviously, c + b 6= c and hence c(c + b) = 0, that is, c2 = 0. Put m = Z(R). Clearly, m = ann(c), which means that m is maximal by Lemma 4.2. If R is an Artinian ring, then m is a nilpotent ideal. If the nilpotency index is n ≥ 6, then {mn−2, mn−1} ⊆ N(m2) ∩ N(m3), so m2 and m3 can not be end vertices, Hence m2, m3 ∈ V (Kn) so that m5 = 0, a contradiction. Therefore, n ≤ 5, as required. (cid:3) The structure of R with AG(R) ∼= Kn(1) seems to be rather complicated and hard to determine completely. For n = 1 or n = 2 where the unique horn contains exactly one vertex, see [17] for the detailed structure theorems on commutative rings with at most three nontrivial ideals. We will discuss the problem for n ≥ 3 in a separate paper. Theorem 4.6. There exists no ring R such that AG(R) ∼= Kn(2) for any n ≥ 3. P roof. Let R be a ring such that AG(R) ∼= Kn(2) for some n ≥ 3. Suppose that Kn is the complete graph with V (Kn) = {ai i ∈ Λ} where n = Λ ≥ 3, and the two horns are a1 − X, a2 − Y . Put a =Pi∈Λ ai, p =Ppj ∈X pj and q =Pqk ∈Y qk. Then p ∈ X, q ∈ Y . Now, we have the following claims: Claim 1. p and q are maximal ideals. In view of Lemma 4.1, we get that a1, a2 are minimal, so a1 + a2 6∈ {a1, a2}. Since a3(a1 + a2) = 0, it follows that a1 + a2 ∈ {ai i ∈ Λ, i 6= 1, 2} and hence a1(a1 + a2) = 0, so a2 1 = 0, which yields that a1a = 0. Similarly, a2 2 = 0, a2a = 0, which means that a ∈ {ai i ∈ Λ}. Note that a1(a + p) = 0, it follows 13 that a + p ∈ A(R). On the other hand, a2(a + p) = a2p 6= 0, so a + p ∈ X and hence a + p ⊆ p whence a ⊆ p. Thus ann(a1) = a ∪ p = p. Consequently, p is a maximal ideal of R by Lemma 4.2. Similarly, q is also maximal. Claim 2. a3 = 0. By the proof of Claim 1, a ⊆ p ∩ q. On the other hand, a1(p ∩ q) = a2(p ∩ q) = 0 implies that p ∩ q ∈ V (Kn) and hence p ∩ q ⊆ a so that p ∩ q = a. Now, we show that a2 6= a. Assume to the contrary that a2 = a. Since a1 + a2 ⊆ a, it follows that a is not minimal. So we can take 0 6= x ∈ a such that a 6= Rx. Then there exist u, v ∈ a such that x = uv. This implies that (Ru)(Rv) = Rx 6= 0. If a = Ru = Rv, then Rx = a2 = a, a contradiction. If at least one of Ru, Rv is not equal to a, without loss of generality, suppose that Ru 6= a, then Ru ∈ V (Kn) whence (Ru)(Rv) ⊆ (Ru)a = 0, that is, Rx = 0, another contradiction. Thus, a2 6= a. Clearly, a2 ∈ V (Kn), hence a3 = aa2 = 0. By the proof of Claim 2, p ∩ q = a, so pq ⊆ a and hence p3q3 = 0. By Claim 1, p, q are maximal, hence R is an Artinian ring with exactly two maximal ideals. By [7, Theorem 8.7], there exist two Artinian local rings, say (R1, m) and (R2, n), such that R = R1 × R2. Hence we can assume that p = (m, R2), q = (R1, n) and hence a = (m, n). Note that (R1, 0) − (0, R2) − (m, 0) is a path in AG(R), it follows that (0, R2) is not an end vertices, that is, (0, R2) ∈ V (Kn) and hence a(0, R2) = (0, n) = 0 so that n = 0. Similarly, m = 0. This implies that R1 and R2 are fields, and AG(R) ∼= K2, a contradiction. The contradiction followed from the assumption that there exists a ring R such that AG(R) ∼= Kn(2) for some n ≥ 3. (cid:3) Theorem 4.7. AG(R) is K3(3) if and only if R ∼= F1 × F2 × F3, where F1, F2, F3 are fields. P roof. Let AG(R) ∼= K3(3). Suppose that V (K3) = {a, b, c}, the three horns are a − X, b − Y and c − Z. By Lemma 4.1, a, b, c are minimal ideals. Denote m = Pmi∈X mi, n =Pnj ∈Y nj, p =Ppk ∈Z pk. Clearly, am = 0, but bm 6= 0, cm 6= 0, so m ∈ X. Similarly, n ∈ Y , p ∈ Z. Now, we show the following claims: Claim 1. a2 = a, b2 = b, c2 = c. Since a is minimal, it follows that a2 ∈ {0, a}. If a2 = 0, then a(a + b) = 0, which together with c(a + b) = 0 yields that a + b ∈ {a, b, c}, which contradicts the fact a, b, c are minimal ideals. Similarly, b2 = b, c2 = c. Claim 2. m, n, p are maximal ideal of R. Since am = 0, so, by Lemma 4.2, we only need to show that m = ann(a). Note that a(m + b) = 0, b(m + b) 6= 0, c(m + b) 6= 0, hence m + b ∈ X whence m + b ⊆ m, that is, b ⊆ m. Similarly, c ⊆ m. Therefore, ann(a) = b ∪ c ∪ m = m so that m is a maximal ideal. The other two results can be obtained similarly. Claim 3. m ∩ n ∩ p = 0. Take any x ∈ m ∩ n ∩ p. If x 6= 0, then Rx ∈ N(a) ∩ N(b) ∩ N(c) = ∅, a contradiction. 14 By Chinese Remainder Theorem, we have R ∼= R/m × R/n × R/p, that is, R ∼= (cid:3) F1 × F2 × F3, where F1, F2, F3 are fields. The converse is trivial. Corollary 4.8. If AG(R) is K3(3), then AG(R) = 6. It is interesting to make a comparison between the results of sections 2, 3 and those in section 4. Lemma 2.2 shows that for any complete bipartite graph with a horn, there exists a corresponding po-semiring. However, in view of Theorem 4.3, only a few of complete bipartite graphs with a horn can ba realized as annihilating-ideal graph. We can obtain a similar result for complete graphs with horns by Theorems 3.5, 4.6 and 4.7. This observation indicates that the class of po-semirings I(R) of rings R is a very small subclass of po-semirings. So, we have the following question. Question 4.9. Classify the po-semirings A such that A ∼= I(R) for some commutative ring R. References [1] G. Aalipour, S. Akbari, M. Behboodi, R. Nikandish, M.J. Nikmehr and F. Shaveisi. The classification of the annihilating-ideal graph of a commutative ring, Algebra Coll. (accepted). [2] G. Aalipour, S. Akbari, R. Nikandish, M.J. Nikmehr and F. Shaveisi. Minimal prime ideals and cycles in annihilating-ideal graphs, Rocky Mountain J. Math. (ac- cepted). [3] G. Aalipour, S. Akbari, R. Nikandish, M.J. Nikmehr and F. Shaveisi. On the coloring of the annihilating-ideal graph of a commutative ring, Preprint. [4] S. Akbari, H.R. Maimani, S. Yassemi, When a zero-divisor graph is plannar or a complete r-partite graph, J. Algebra, 270(2003) 169 − 180. [5] D.F. Anderson and P.S. Livingston. The zero-divisor graph of a commutative ring, J. Algebra 217(1999) 434 − 447. [6] D.D. Anderson and M. Naseer. Beck's coloring of a commutative ring, J. Algebra 159(1993) 500 − 514. [7] M.F. Atiyah and I.G. MacDonald. Introduction to Commutative Algebra, Addison- Wesley, Reading, MA, 1969. 15 [8] I. Beck. Coloring of commutative rings, J. Algebra, 116(1988) 208 − 226. [9] M. Behboodi and Z. Rakeei. The annihilating-ideal graph of commutative rings I, J. Algebra Appl. (accepted). [10] M. Behboodi and Z. Rakeei. The annihilating-ideal graph of commutative rings II, J. Algebra Appl. (accepted). [11] F. Buckley and M. Lewinter. A Friendly Introduction to Graph Theory, Prentic Hall, 2003. [12] F.R. DeMeyer, T. McKenzie and K. Schneider. The zero-divisor graph of a commu- tative semigroup, Semigroup Forum 65(2002) 206-214. [13] J. D. LaGrange. On realizing zero-divisor graph, Comm. Algebra 36(2008) 4509- 4520. [14] D.C. Lu and T.S. Wu. On bipartite zero-divisor graphs, Discrete Math 309(2009) 755-762. [15] J. Reineke. Commutative rings in which every proper ideal is maximal, Fund. Math 97(1977) 229-231. [16] T.S. Wu and D.C. Lu. Zero-divisor semigroups and some simple graphs, Comm. Algebra 34(8)(2006) 3043-3052. [17] T.S. Wu and D.C. Lu. Finite local rings with at most three nontrivial ideals, Preprint. [18] T.S. Wu, D.C. Lu and Y. L. Li. On zero-divisors and prime elements of po-semirings, Preprint. 16
1910.00954	1	1910	2019-10-02T13:51:45	Nilpotent varieties of some finite dimensional restricted Lie algebras	[ "math.RA" ]	In the late 1980s, A. Premet conjectured that the variety of nilpotent elements of any finite dimensional restricted Lie algebra over an algebraically closed field of characteristic $p>0$ is irreducible. This conjecture remains open, but it is known to hold for a large class of simple restricted Lie algebras, e.g. for Lie algebras of connected algebraic groups, and for Cartan series $W, S$ and $H$. In this thesis we start by proving that Premet's conjecture can be reduced to the semisimple case. The proof is straightforward. However, the reduction of the semisimple case to the simple case is very non-trivial in prime characteristic as semisimple Lie algebras are not always direct sums of simple ideals. Then we consider some semisimple restricted Lie algebras. Under the assumption that $p>2$, we prove that Premet's conjecture holds for the semisimple restricted Lie algebra whose socle involves the special linear Lie algebra $\mathfrak{sl}_2$ tensored by the truncated polynomial ring $k[X]/(X^{p})$. Then we extend this example to the semisimple restricted Lie algebra whose socle involves $S\otimes \mathcal{O}(m; \underline{1})$, where $S$ is any simple restricted Lie algebra such that $\text{ad} S=\text{Der} S$ and its nilpotent variety $\mathcal{N}(S)$ is irreducible, and $\mathcal{O}(m; \underline{1})=k[X_1, \dots, X_m]/(X_1^p, \dots, X_m^p)$ is the truncated polynomial ring in $m\geq 2$ variables. In the final chapter we assume that $p>3$. We confirm Premet's conjecture for the minimal $p$-envelope $W(1; n)_{p}$ of the Zassenhaus algebra $W(1; n)$ for all $n\geq 2$. This is the main result of the research paper [3] (see thesis) which was published in the Journal of Algebra and Its Applications.	math.RA	math	NILPOTENT VARIETIES OF SOME FINITE DIMENSIONAL RESTRICTED LIE ALGEBRAS 9 1 0 2 t c O 2 ] . A R h t a m [ 1 v 4 5 9 0 0 . 0 1 9 1 : v i X r a A thesis submitted to the University of Manchester for the degree of Doctor of Philosophy in the Faculty of Science and Engineering 2019 Cong Chen School of Natural Sciences Department of Mathematics Contents Abstract Declaration Copyright Statement Acknowledgements 1 Introduction 1.1 Restricted Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 p-envelopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Gradations and standard filtrations . . . . . . . . . . . . . . . . . . . . 1.4 Graded Lie algebras of Cartan type . . . . . . . . . . . . . . . . . . . . 1.5 Nilpotent varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6 Some useful theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.7 Overview of results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Reduction of Premet's conjecture to the semisimple case 3 Nipotent varieties of some semisimple restricted Lie algebras 3.1 Socle involves sl2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Socle involves S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Nilpotent elements of D . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Nilpotent elements of L . . . . . . . . . . . . . . . . . . . . . . 3.2.3 The irreducibility of N (L) . . . . . . . . . . . . . . . . . . . . . 4 5 6 7 8 8 12 13 15 27 38 41 44 48 48 59 63 84 98 4 The nilpotent variety of W (1; n)p is irreducible 104 2 4.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 4.1.1 W (1; n) and W (1; n)p . . . . . . . . . . . . . . . . . . . . . . . . 105 4.1.2 The automorphism group G . . . . . . . . . . . . . . . . . . . . 107 4.2 The variety N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 4.2.1 Some elements in N . . . . . . . . . . . . . . . . . . . . . . . . 109 4.2.2 An irreducible component of N . . . . . . . . . . . . . . . . . . 127 4.2.3 The irreducibility of N . . . . . . . . . . . . . . . . . . . . . . . 131 Bibliography 140 Word count 23,388 3 The University of Manchester Cong Chen Doctor of Philosophy Nilpotent varieties of some finite dimensional restricted Lie algebras October 3, 2019 In the late 1980s, A. Premet conjectured that the variety of nilpotent elements of any finite dimensional restricted Lie algebra over an algebraically closed field of charac- teristic p > 0 is irreducible. This conjecture remains open, but it is known to hold for a large class of simple restricted Lie algebras, e.g. for Lie algebras of connected algebraic groups, and for Cartan series W, S and H. In this thesis we start by proving that Premet's conjecture can be reduced to the semisimple case. The proof is straightforward. However, the reduction of the semisim- ple case to the simple case is very non-trivial in prime characteristic as semisimple Lie algebras are not always direct sums of simple ideals. Then we consider some semisim- ple restricted Lie algebras. Under the assumption that p > 2, we prove that Premet's conjecture holds for the semisimple restricted Lie algebra whose socle involves the spe- cial linear Lie algebra sl2 tensored by the truncated polynomial ring k[X]/(X p). Then we extend this example to the semisimple restricted Lie algebra whose socle involves S ⊗ O(m; 1), where S is any simple restricted Lie algebra such that ad S = Der S and its nilpotent variety N (S) is irreducible, and O(m; 1) = k[X1, . . . , Xm]/(X p 1 , . . . , X p m) is the truncated polynomial ring in m ≥ 2 variables. In the final chapter we assume that p > 3. We confirm Premet's conjecture for the minimal p-envelope W (1; n)p of the Zassenhaus algebra W (1; n) for all n ∈ N≥2. This is the main result of the research paper [3] which was published in the Journal of Algebra and Its Applications. 4 Declaration No portion of the work referred to in the thesis has been submitted in support of an application for another degree or qualification of this or any other university or other institute of learning. 5 Copyright Statement i. The author of this thesis (including any appendices and/or schedules to this thesis) owns certain copyright or related rights in it (the "Copyright") and s/he has given The University of Manchester certain rights to use such Copyright, including for administrative purposes. ii. Copies of this thesis, either in full or in extracts and whether in hard or electronic copy, may be made only in accordance with the Copyright, Designs and Patents Act 1988 (as amended) and regulations issued under it or, where appropriate, in ac- cordance with licensing agreements which the University has from time to time.This page must form part of any such copies made. iii. The ownership of certain Copyright, patents, designs, trade marks and other intel- lectual property (the "Intellectual Property") and any reproductions of copyright works in the thesis, for example graphs and tables ("Reproductions"), which may be described in this thesis, may not be owned by the author and may be owned by third parties. Such Intellectual Property and Reproductions cannot and must not be made available for use without the prior written permission of the owner(s) of the relevant Intellectual Property and/or Reproductions. iv. Further information on the conditions under which disclosure, publication and com- mercialisation of this thesis, the Copyright and any Intellectual Property and/or Re- productions described in it may take place is available in the University IP Policy (see http://documents.manchester.ac.uk/DocuInfo.aspx?DocID=487), in any relevant- Thesis restriction declarations deposited in the University Library, The University Library's regulations (see http://www.manchester.ac.uk/library/aboutus/regulations) and in The University's Policy on Presentation of Theses. 6 Acknowledgements I would like to thank Professor Alexander Premet for his knowledge and guidance during my research. I feel lucky to have such a supervisor who always promptly answers my questions, generously shares ideas with me, and patiently explains maths in great detail to me. Thanks for giving me the freedom to choose a project that I like. Thanks for helping me to prepare a research paper and guiding me through the publication process. Overall, it has been my great pleasure to be his student and my PhD life is enjoyable. I would also like to thank Dr Rudolf Tange and Professor Mike Prest for their extremely useful comments to help me improving an earlier version of the thesis. Thanks to Professor Helmut Strade for giving me a copy of the textbook Simple Lie Algebras over Fields of Positive Characteristic. Thanks to the University of Manchester for awarding me a scholarship to support my research. Thanks to the School of Mathematics for providing me a nice working environment. Thanks for the support from all friendly staff and students at Alan Turing Building. To my parents, thank you for providing me the opportunity to study in the UK. Without your constant support, I would not have been gone this far. 7 Chapter 1 Introduction The theory of modular Lie algebras begins with E. Witt who discovered a new non- classical simple Lie algebra W (1; 1) sometime before 1937. It is now called the Witt algebra. Then H. Zassenhaus generalized the Witt algebra and got a new Lie algebra W (1; n), called the Zassenhaus algebra. In 1937, N. Jacobson introduced the concept of "restricted Lie algebras". Later, more nonclassical Lie algebras were constructed. In this introductory chapter we first review some basic concepts in the theory of modular Lie algebras. Then we explain the construction process of a class of nonclas- sical Lie algebras, namely the Lie algebras of Cartan type. In the end, we introduce Premet's conjecture on the variety of nilpotent elements of any finite dimensional re- stricted Lie algebra over an algebraically closed field of characteristic p > 0. We shall discuss what have been done so far. Throughout the thesis we assume that all Lie algebras are finite dimensional, and k is an algebraically closed field of characteristic p > 0 (unless otherwise specified). We denote by k∗ the multiplicative group of k. We denote by Fp the finite field with p elements. We denote by N0 the set of all nonnegative integers, i.e. N0 = {0, 1, 2, . . . }. 1.1 Restricted Lie algebras Let us first introduce the notion of restricted Lie algebras. Definition 1.1.1. [12, Definition 4, Sec. 7, Chap. V; 31, Sec. 2.1, Chap. 2] Let g be a Lie algebra over k. A mapping [p] : g → g, x 7→ x[p] is called a [p]-th power map if it satisfies 8 1.1. Restricted Lie algebras 1. ad x[p] = (ad x)p, 2. (λx)[p] = λpx[p], 3. (x + y)[p] = x[p] + y[p] +Pp−1 i=1 si(x, y), where the terms si(x, y) ∈ g are such that (ad(tx + y))p−1(x) = p−1 Xi=1 isi(x, y)ti−1 for all x, y ∈ g, λ ∈ k and t a variable. The pair (g, [p]) is referred to as a restricted Lie algebra. The third condition in the definition is known as Jacobson's formula for p-th powers. A more general form of this formula is given by Lemma 1.1.1. [30, Lemma 1.1.1] Let (g, [p]) be a restricted Lie algebra over k. Then for all x1, . . . , xm ∈ g, the following holds: (cid:18) m Xi=1 xi(cid:19)[p]n = m Xi=1 x[p]n i + n−1 Xl=0 v[p]l l , where vl is a linear combination of commutators in xi, 1 ≤ i ≤ m. By Jacobi identity, we can rearrange each vl so that vl is in the span of [wt, [wt−1, [. . . , [w2, [w1, w0] . . . ], where t = pn−l − 1 and each wj, 0 ≤ j ≤ t, is equal to some xi, 1 ≤ i ≤ m. Let us give some examples of restricted Lie algebras. Example 1.1.1. 1. Any associative algebra A over k is a Lie algebra with the Lie bracket given by [x, y] := xy − yx for all x, y ∈ A. Denote this Lie algebra by A(−). Then A(−) is a restricted Lie algebra with the [p]-th power map given by x 7→ xp for all x ∈ A; see [31, Sec. 2.1, Chap. 2] for details. In particular, if A = Matn(k), the algebra of n × n matrices with entries in k, then the general linear Lie algebra gln(k) := Matn(k)(−) is a restricted Lie algebra. 2. Let A be any algebra over k (not necessarily associative). The derivation algebra Der A is a restricted Lie algebra with the [p]-th power map given by D 7→ Dp for all D ∈ Der A; see [12, Sec. 7, Chap. V] for details. As an example, the Lie alge- bra g of an algebraic k-group G is a restricted Lie algebra as g is identified with the subalgebra of left invariant derivations of k[G]; see [10, Sec. 9.1, Chap. III]. 9 1.1. Restricted Lie algebras If a [p]-th power map exists on a Lie algebra g, we may ask how many different [p]-th power maps there are. It turns out that Lemma 1.1.2. [31, Proposition 2.1, Sec. 2.2, Chap. 2] In a restricted Lie algebra (g, [p]), every [p]-th power map is of the form x 7→ x[p] + f (x), where f is a map of g into its centre z(g) satisfying f (λx + y) = λpf (x) + f (y) for all x, y ∈ g, λ ∈ k. An immediate corollary is that Corollary 1.1.1. [31, Corollary 2.2, Sec. 2.2, Chap. 2] If a Lie algebra g is centreless, then g has at most one [p]-th power map. Definition 1.1.2. [31, Sec. 2.1, Chap. 2] Let (g, [p]) be a restricted Lie algebra over k. A subalgebra (respectively an ideal) S of g is called a p-subalgebra (respectively a p-ideal) if x[p] ∈ S for all x ∈ S. Examples of p-ideals include the centre z(g) and the radical Rad g. Moreover, if I and J are p-ideals of g, then so are I + J, I ∩ J, and (I + J)/J ∼= I/(I ∩ J). Note that if I is a p-ideal of g, then the quotient Lie algebra g/I carries a natural [p]-th power map given by (x + I)[p] := x[p] + I for all x ∈ g; see [31, Proposition 1.4, Sec. 2.1, Chap. 2]. Definition 1.1.3. [31, p. 65] Let S be a subset of a restricted Lie algebra (g, [p]). (i) The intersection of all p-subalgebras of g containing S, denoted Sp, is a p- subalgebra of g and is referred to as the p-subalgebra generated by S in g. Note that Sp is the smallest p-subalgebra of g containing S. (ii) Let i ∈ N0. The image of S under the iterated application of the [p]-th power map, denoted S[p]i, is defined by S[p]i := {x[p]i x ∈ S}. Let us give an explicit characterization of Sp in the following case. Lemma 1.1.3. [31, Proposition 1.3(1), Sec. 2.1, Chap. 2] Let (g, [p]) be a restricted Lie algebra over k, and let H be a subalgebra of g with basis {ej j ∈ J}. Then the p-subalgebra of g generated by H is given by Hp = Xi≥0 hH [p]i i = Xj∈J,i≥0 ke[p]i j . 10 1.1. Restricted Lie algebras Definition 1.1.4. [31, Sec. 2.3, Chap. 2] Let (g, [p]) be a restricted Lie algebra over k. An element x ∈ g is called semisimple (or p-semisimple) if x ∈ (kx[p])p = Pi≥1 kx[p]i. If x[p] = x, then x is called toral. Lemma 1.1.4. [31, Proposition 3.3, Sec. 2.3, Chap. 2] Let (g, [p]) be a restricted Lie algebra over k. Then the following statements hold: (i) If x is toral, then x is semisimple. (ii) If x and y are semisimple and [x, y] = 0, then x + y is semisimple. (iii) If x is semisimple, then x[p]i is semisimple for every i ∈ N. Moreover, y is semisimple for every y ∈ (kx)p. Definition 1.1.5. [31, Sec. 2.4, Chap. 2] Let (g, [p]) be a restricted Lie algebra over k. A subalgebra t ⊂ g is called a torus if t is an abelian p-subalgebra of g consisting of semisimple elements. Lemma 1.1.5. [31, Theorem 3.6(1), Sec. 2.3, Chap. 2] Let (g, [p]) be a restricted Lie algebra over k. Then any torus in g has a basis consisting of toral elements. Definition 1.1.6. [30, Notation 1.2.5] Let (g, [p]) be a restricted Lie algebra over k. Set MT(g) := max{dim t t is a torus of g}, the maximal dimension of tori in g. Lemma 1.1.6. [30, Lemma 1.2.6(2)] Let (g, [p]) be a restricted Lie algebra over k and let I be a p-ideal of g. Then the following holds: MT(g) = MT(g/I) + MT(I). Note that in a restricted Lie algebra (g, [p]), Cartan subalgebras can be described by maximal tori. Definition 1.1.7. [31, Sec. 1.3, Chap. 1] Let L be a Lie algebra over k (not necessarily restricted). The lower central series of L is the sequence of ideals of L defined as follows: L1 = L, Li+1 = [L, Li] for i ≥ 1. Then L = L1 ⊇ L2 ⊇ · · · ⊇ Ln ⊇ . . . We say that L is nilpotent if Lm = 0 for some m ∈ N. 11 1.2. p-envelopes Definition 1.1.8. [31, Sec. 1.4, Chap. 1] Let L be a Lie algebra over k (not necessarily restricted). A subalgebra h of L is called a Cartan subalgebra if it is nilpotent and equal to its own normalizer, i.e. h = NL(h) =: {x ∈ L [x, h] ∈ h for all h ∈ h}. Theorem 1.1.1. [31, Theorem 4.1, Sec. 2.4, Chap. 2] Let (g, [p]) be a restricted Lie algebra over k. Let h be a subalgebra of g. The following statements are equivalent: (i) h is a Cartan subalgebra. (ii) There exists a maximal torus t in g such that h = cg(t), the centralizer of t in g. Definition 1.1.9. [31, Sec. 2.1, Chap. 2] Let (g, [p]) be a restricted Lie algebra over k. An element x ∈ g is called nilpotent (or p-nilpotent) if there is n ∈ N such that x[p]n = 0 . We denote by N (g) the variety of all nilpotent elements in g. It is well known that N (g) is a Zariski closed, conical subset of g. Theorem 1.1.2 (Jordan-Chevalley Decomposition [31, Theorem 3.5, Chap. 2]). Let (g, [p]) be a restricted Lie algebra over k. For any x ∈ g, there exist a unique semisimple element xs ∈ g and a unique nilpotent element xn ∈ g such that x = xs +xn and [xs, xn] = 0. It follows from the above theorem that g = N (g) if and only if MT(g) = 0. 1.2 p-envelopes Let L be a Lie algebra over k. It is useful to embed L into a restricted Lie algebra. Definition 1.2.1. [31, Sec. 2.5, Chap. 2] Let L be a Lie algebra over k. A triple (L, [p], i) consisting of a restricted Lie algebra (L, [p]) and a Lie algebra homomorphism i : L → L is called a p-envelope of L if i is injective and the p-subalgebra generated by i(L), denoted (i(L))p, coincides with L. We often identify L with i(L) ⊂ L. Let us review some properties of p-envelopes. 12 1.3. Gradations and standard filtrations Theorem 1.2.1. [30, Theorem 1.1.7] Let (L1, [p]1, i1) and (L2, [p]2, i2) be two p-envelopes of L. Then there exists an isomorphism ψ of restricted Lie algebras ψ : L1/z(L1) ∼−→ L2/z(L2) such that ψ ◦ π1 ◦ i1 = π2 ◦ i2, where π1 : L1 → L1/z(L1) and π2 : L2 → L2/z(L2) are the canonical homomorphisms of restricted Lie algebras. Definition 1.2.2. [31, Sec. 2.5, Chap. 2] A p-envelope (L, [p], i) of L is called minimal if z(L) ⊂ z(i(L)). Theorem 1.2.2. [30, Theorem 1.1.6 and Corollary 1.1.8; 31, Theorem 5.8, Sec. 2.5, Chap. 2] (i) If (L, [p], i) is a p-envelope of L, then there exists a minimal p-envelope (H, [p]1, i1) of L and an ideal J ⊂ z(L) such that L = H ⊕ J and i1 = i (i.e. H ⊂ L). (ii) Any two minimal p-envelopes of L are isomorphic as ordinary Lie algebras. (iii) Suppose L is semisimple. Then every minimal p-envelope of L is semisimple, and all minimal p-envelopes of L are isomorphic as restricted Lie algebras. Remark 1.2.1. [30, p. 22; 31, p. 97] If L is semisimple, then we can easily describe its minimal p-envelope. Since L is semisimple, there is an embedding L ∼= ad L ֒→ Der L via the adjoint representation. Then the minimal p-envelope of L is the p-subalgebra of Der L generated by ad L, i.e. (ad L)p. To compute (ad L)p, we often identify L with ad L. 1.3 Gradations and standard filtrations Definition 1.3.1. [31, Sec. 3.2, Chap. 3] Let L be a Lie algebra over k. A Z-grading of L is a collection of subspaces (Li)i∈Z such that (i) L = Li∈Z Li and (ii) [Li, Lj] ⊂ Li+j for all i, j ∈ Z. If there exist r, s ∈ Z such that L = Ls (respectively height) of this gradation. −r Li, then r (respectively s) is called the depth 13 1.3. Gradations and standard filtrations Note that L0 is a Lie subalgebra of L and each subspace Li obtains an L0-module structure via the adjoint representation. Definition 1.3.2. [31, Sec. 3.2, Chap. 3] Let (g, [p]) be a restricted Lie algebra over k. A gradation (gi)i∈Z of g is called restricted if g[p] i ⊂ gpi for all i ∈ Z. Definition 1.3.3. [31, Sec. 1.9, Chap. 1] Let L be a Lie algebra over k. A descending filtration of L is a collection of subspaces (L(i))i∈Z such that (i) L(i) ⊃ L(j) if i ≤ j. (ii) [L(i), L(j)] ⊂ L(i+j) for all i, j ∈ Z. A filtration is called separating if ∩i∈ZL(i) = {0} and exhaustive if ∪i∈ZL(i) = L. The notion of ascending filtration is defined similarly. It is common to use descending filtrations for Lie algebras. Note that L(0) is a Lie subalgebra of L and each subspace L(i) obtains an L(0)-module structure via the adjoint representation. If the filtration is exhaustive, then ∩i∈ZL(i) is an ideal of L. If in addition that L is simple, then either L(i) = L for all i, or the filtration is separating; see [31, p. 100]. By a result of Weisfeiler [35], we can define standard filtrations. Definition 1.3.4. [24, Sec. 2.4; 30, Definition 3.5.1] Let L be a Lie algebra over k and L(0) be a maximal subalgebra of L. Let L(−1) be an L(0)-invariant subspace of L which contains L(0). Moreover, assume that L(−1)/L(0) is an irreducible L(0)-module. Set L(i+1) := {x ∈ L(i) [x, L(−1)] ⊂ L(i)}, i ≥ 0, L(−i−1) := [L(−i), L(−1)] + L(−i), i ≥ 1. The sequence of subspaces (L(i))i∈Z defines a standard filtration on L. Since L(0) is a maximal subalgebra of L this filtration is exhaustive. If L is simple, then this filtration is separating. So there are s1 > 0 and s2 ≥ 0 such that L = L(−s1) ⊃ · · · ⊃ L(0) ⊃ · · · ⊃ L(s2+1) = (0). (1.1) Theorem 1.3.1. [31, Theorem 1.3, Sec. 3.1, Chap. 3] Let L be a simple Lie algebra over an algebraically closed field of characteristic p > 3. If there is x ∈ L such that (ad x)p−1 = 0, then there exists a standard filtration as above (1.1). 14 1.4. Graded Lie algebras of Cartan type It was proved by A. Premet that such x 6= 0 with (ad x)p−1 = 0 always exists in simple Lie algebras over algebraically closed fields of characteristic p > 3; see [17, Theorem 1] . Hence they admit a standard filtration. We can define restricted filtrations in a similar way. Definition 1.3.5. [31, Sec. 3.1, Chap. 3] Let (g, [p]) be a restricted Lie algebra over k. A filtration (g(i))i∈Z of g is called restricted if g[p] (i) ⊂ g(pi) for all i ∈ Z. It is useful to note the interrelation between gradations and filtrations; see [31, Sec. 3.3, Chap. 3]. Given any Z-graded Lie algebra L = Li∈Z Li, set L(j) := Li≥j Li. Then this Z-grading induces a filtration on L. Conversely, suppose L has a descending filtration (L(i))i∈Z. We can define the graded Lie algebra gr L associated with L. Put Lj := L(j)/L(j+1) for all j ∈ Z. Then gr L := Lj∈Z Lj. The Lie bracket in gr L is given by [x + L(j+1), y + L(l+1)] := [x, y] + L(j+l+1) for all x ∈ L(j) and y ∈ L(l). 1.4 Graded Lie algebras of Cartan type In a series of papers, A. Premet and H. Strade have completed the classification of finite dimensional simple modular Lie algebras and proved the following: Theorem 1.4.1 (Classification Theorem [25, Theorem 1.1]). Any finite dimensional simple Lie algebra over an algebraically closed field of charac- teristic p > 3 is of classical, Cartan or Melikian type. The classical simple modular Lie algebras include both classical simple (modulo its centre for slmp) and exceptional Lie algebras over C. They were constructed using a Chevalley basis by reduction modulo p [4]. The Melikian algebras M(m, n) depend on two parameters m, n ∈ N, and they only occur in characteristic 5 [16]. The Lie algebras of Cartan type provide a large class of nonclassical simple Lie algebras. They are finite dimensional modular analogues of the four families Witt, special, Hamiltonian, contact of infinite dimensional complex Lie algebras. Their construction was motivated by Cartan's work on pseudogroups. The formal power series algebras over C were 15 1.4. Graded Lie algebras of Cartan type replaced by divided power algebras over k; see [14] and [15]. Our work relates to Lie algebras of Cartan type, particularly the general Cartan type Lie algebras. So let us give a detailed description of these Lie algebras. All definitions and theorems can be found in [1], [30] and [31]. Notation 1.4.1. [30, Sec. 2.1, Chap. 2; 31, Sec. 3.5, Chap. 3] Let Nm 0 denote the set of all m-tuples of nonnegative integers. For a = (a1, . . . , am), b = (b1, . . . , bm) ∈ Nm 0 , we write x(ai) i := 1 ai! xai i , (cid:18)a + b b (cid:19) := a := m Yi=1 Xi=1 m bi (cid:19), (cid:18)ai + bi ai. x(a) := a! := x(ai) i , m Yi=1 Yi=1 m ai!, Definition 1.4.1. [30, Sec. 2.1, Chap. 2] Let O(m) denote the commutative associa- tive algebra with unit element over k defined by generators x(r) , 1 ≤ i ≤ m, r ≥ 0, and i relations x(0) i = 1, x(r) i x(s) i = (cid:18)r + s r (cid:19)x(r+s) i , 1 ≤ i ≤ m, r, s ≥ 0. Then {x(a) a ∈ Nm 0 } forms a basis of O(m), and O(m) is called the divided power algebra. For simplicity, we write x(1) i as xi. Some results on binomial coefficients may be useful. Lemma 1.4.1. [30, Lemma 2.1.2(1)] For a, b ∈ N, let a = Pi≥0 aipi, b = Pi≥0 bipi, 0 ≤ ai, bi ≤ p − 1, be the p-adic expansions of a and b. Then the following congruence holds: b(cid:19) ≡ Yi≥0 (cid:18)a bi(cid:19) (mod p). (cid:18)ai Sketch of proof. Let a and b be as in the lemma. Let Y be a variable. Note that for any integer n such that 1 ≤ n ≤ p − 1, n(cid:19) ≡ 0 (cid:18)p (mod p). 16 1.4. Graded Lie algebras of Cartan type Hence (1 + Y )p ≡ 1 + Y p (mod p). In general, one can show by induction that for any i ≥ 1, (1 + Y )pi ≡ 1 + Y pi (mod p). Consider (1 + Y )a and expand it over Z, we get (1 + Y )a = Yi≥0 (1 + Y )aipi ≡ Yi≥0 (1 + Y pi )ai (mod p). Compare the coefficients of Y b on both sides, we get b(cid:19) ≡ Yi≥0 (cid:18)a bi(cid:19) (mod p). (cid:18)ai This completes the sketch of proof. Applying the above result, we can prove that Corollary 1.4.1. Let r ∈ N be such that r ≥ 2 and 1 ≤ s ≤ r − 1. Then for any 0 ≤ i ≤ ps − 1, the following congruence holds: (cid:18)pr − ps + i pr − ps (cid:19) ≡ 1 (mod p). Proof. Let r ∈ N be such that r ≥ 2 and 1 ≤ s ≤ r − 1. Then the p-adic expansion of pr − ps is pr − ps = r−1 Xj=s (p − 1)pj. (1.2) Since 0 ≤ i ≤ ps − 1 and the p-adic expansion of ps − 1 is ps − 1 = Ps−1 follows that the p-adic expansion of i is j=0(p − 1)pj, it i = s−1 Xj=0 ajpj, (1.3) where 0 ≤ aj ≤ p − 1 and 0 ≤ Ps−1 expansion of pr − ps + i is j=0 aj ≤ s(p − 1). By (1.2) and (1.3), the p-adic pr − ps + i = s−1 Xj=0 ajpj + r−1 Xj=s (p − 1)pj. It follows from Lemma 1.4.1 that (cid:18)pr − ps + i pr − ps (cid:19) ≡ s−1 Yj=0 0(cid:19) · (cid:18)aj r−1 Yj=s p − 1(cid:19) (mod p) = 1 (mod p). (cid:18)p − 1 This completes the proof. 17 1.4. Graded Lie algebras of Cartan type Note that there is a Z-grading on O(m) given by O(m)i := span{x(a) a = i}. Hence O(m) = L∞ i=0 O(m)i. Put O(m)(j) induces a descending filtration on O(m), called the standard filtration. := Li≥j O(m)i. Then this Z-grading Definition 1.4.2. [30, Definition 2.1.1] A system of divided powers on O(m)(1) is a sequence of maps γr : O(m)(1) → O(m), f 7→ f (r) ∈ O(m), where r ≥ 0, satisfying (i) f (0) = 1, f (r) ∈ O(m)(1) for all f ∈ O(m)(1), r > 0, (ii) f (1) = f for all f ∈ O(m)(1), (iii) f (r)f (s) = (r+s)! r!s! f (r+s) for all f ∈ O(m)(1), r, s ≥ 0, (iv) (f + g)(r) = Pr l=0 f (l)g(r−l) (v) (f g)(r) = f rg(r) for all f, g ∈ O(m)(1), r ≥ 0, for all f ∈ O(m), g ∈ O(m)(1), r ≥ 0, (vi) (f (s))(r) = (rs)! r!(s!)r f (rs) for all f ∈ O(m)(1), r ≥ 0, s > 0. Definition 1.4.3. [30, Definition 2.1.1(2)] A derivation D of O(m) is called special if D(f (r)) = f (r−1)D(f ) for all f ∈ O(m)(1) and r > 0. For 1 ≤ i ≤ m, set ǫi = (δi1, . . . , δim). Let ∂i denote the ith partial derivative defined by ∂i(x(a)) = x(a−ǫi) if ai > 0 and 0 otherwise; see [31, p. 132]. We denote by W (m) the set of all special derivations of O(m). This is a Lie subalgebra of Der O(m) and it obtains an O(m)-module structure via (f D)(g) := f D(g) for all f, g ∈ O(m) and D ∈ W (m). Since each D ∈ W (m) is uniquely determined by its effects on x1, . . . , xm, the Lie algebra W (m) is a free O(m)-module of rank m generated by the partial derivatives ∂1, . . . , ∂m; see [30, Proposition 2.1.4]. By [31, Lemma 2.1(1), Sec. 4.2, Chap. 4], we know that for any f, g ∈ O(m) and D, E ∈ W (m), [f D, gE] = f D(g)E − gE(f )D + f g[D, E]. (1.4) Since {x(a) ∂i a ∈ Nm 0 , 1 ≤ i ≤ m} forms a basis for W (m) and [∂i, ∂j] = 0 for any 1 ≤ i, j ≤ m, it follows from (1.4) that the Lie bracket in W (m) is given by [x(a) ∂i, x(b) ∂j] = (cid:18)a + b − ǫi a (cid:19)x(a+b−ǫi) ∂j −(cid:18)a + b − ǫj b (cid:19)x(a+b−ǫj ) ∂i . (1.5) 18 1.4. Graded Lie algebras of Cartan type Note that W (m) inherits a grading and descending filtration from O(m): W (m)i := m Mj=1 O(m)i+1 ∂j, W (m)(i) := m Mj=1 O(m)(i+1) ∂j for i ≥ −1. Both are called standard. For any m-tuple n := (n1, . . . , nm) ∈ Nm, define O(m; n) := span{x(a) 0 ≤ ai < pni}. It is easy to see that O(m; n) is a subalgebra of O(m) invariant under ∂i for all 1 ≤ i ≤ m. Moreover, dim O(m; n) = pn. The general Cartan type Lie algebra W (m; n) is the Lie subalgebra of W (m) which normalizes O(m; n). Since O(m; n) is a subalgebra of O(m), the grading and filtration on O(m; n) induce a grading and filtration on W (m; n): W (m; n)i := m Mj=1 O(m; n)i+1 ∂j, W (m; n)(i) := m Mj=1 O(m; n)(i+1) ∂j (1.6) for i ≥ −1. In particular, W (m; n) = s Mi=−1 W (m; n)i, where s = (Pm i=1 pni) − m − 1; see [31, Proposition 2.2(3), Sec. 4.2, Chap. 4]. Theorem 1.4.2. [31, Proposition 5.9, Sec. 3.5, Chap. 3; Proposition 2.2 and Theorem 2.4, Sec. 4.2, Chap. 4] (i) W (m; n) is a free O(m; n)-module with basis {∂1, . . . , ∂m}. (ii) The set {x(a) ∂i 0 ≤ ai < pni, 1 ≤ i ≤ m} forms a basis for W (m; n). Hence dim W (m; n) = mpn. (iii) W (m; n) is simple unless m = 1 and p = 2. (iv) W (m; n) is a subalgebra of the restricted Lie algebra Der O(m; n). (v) W (m; n) is restricted if and only if n = (1, . . . , 1), and in that case D[p] = Dp for all D ∈ W (m; n) and the gradation is restricted. 19 1.4. Graded Lie algebras of Cartan type We refer to the Lie algebras W (m) or W (m; n) as Lie algebras of Witt type. In Chapter 3 we will spell out W (m; 1) in more details. By [31, Lemma 2.1(3), Sec. 4.2, Chap. 4], we know that O(m; 1) is isomorphic to the truncated polynomial ring m) in m variables. Hence W (m; 1) ∼= Der O(m; 1), and it k[X1, . . . , Xm]/(X p is called the mth Jacobson-Witt algebra. In Chapter 4 we will study the Zassenhaus 1 , . . . , X p algebra W (1; n). Note that if char k = p > 2 and n = 1, then W (1; n) coincides with the Witt algebra W (1; 1), a simple and restricted Lie algebra. If char k = p > 2 and n ≥ 2, then W (1; n) provides the first example of a simple, non-restricted Lie algebra. In this case it is useful to consider its minimal p-envelope. Let us determine the minimal p-envelope of the simple, non-restricted Witt algebra W (m; n). Since W (m; n) is simple, it follows from Theorem 1.2.2(iii) that all its minimal p-envelopes are isomorphic as restricted Lie algebras. Moreover, there is an embedding W (m; n) ∼= ad W (m; n) ֒→ Der W (m; n) via the adjoint representation. By Remark 1.2.1, the minimal p-envelope of W (m; n), denoted W (m; n)[p], is the p-subalgebra (ad W (m; n))p of Der W (m; n) generated by ad W (m; n), i.e. W (m; n) ∼= ad W (m; n) ֒→ W (m; n)[p] = (ad W (m; n))p ֒→ Der W (m; n); see Definition 1.1.3 and Lemma 1.1.3 for notations. In [30, Sec. 7.1 and 7.2], H. Strade computed W (m; n)[p]. He first proved the following: Theorem 1.4.3. [30, Theorems 7.1.2(1)] Der W (m; n) ∼= W (m; n) + m Xi=1 X0<ji<ni k ∂pji i . The isomorphism is given by the adjoint representation, W (m; n) ∼= ad W (m; n) ֒→ Der W (m; n). Then H. Strade computed W (m; n)[p] = (ad W (m; n))p. He identified W (m; n) with ad W (m; n). By Theorem 1.4.2(iv), we know that W (m; n) is a subalgebra of the restricted Lie algebra Der O(m; n). So instead of computing the p-subalgebra (ad W (m; n))p of Der W (m; n) generated by ad W (m; n), H. Strade computed the p- subalgebra (W (m; n))p of Der O(m; n) generated by W (m; n). By Definition 1.1.3 and Lemma 1.1.3, (W (m; n))p = Xi≥0 hW (m; n)pii, 20 1.4. Graded Lie algebras of Cartan type where W (m; n)pi := {Dpi D ∈ W (m; n)} is the image of W (m; n) under the iterated application of the [p]-th power map of Der O(m; n). H. Strade first observed that Lemma 1.4.2. [30, Lemma 7.1.1(3)] W (m; n)(0) is a restricted Lie subalgebra of Der O(m; n). It follows that W (m; n)[p] = (ad W (m; n))p contains W (m; n) and all iterated p-th powers of the partial derivatives ∂1, . . . , ∂m. Applying Theorem 1.4.3, we get Theorem 1.4.4. [30, Theorem 7.2.2(1)] The minimal p-envelope W (m; n)[p] of W (m; n) in Der W (m; n) is given by W (m; n)[p] = W (m; n) + m Xi=1 X0<ji<ni k ∂pji i . Since W (m; n) is the Lie subalgebra of W (m), the Lie bracket in W (m; n) is given by (1.5). By (1.4), we have that for any 1 ≤ i, j ≤ m, 0 < r < ni and 0 ≤ ai < pni, the brackets [∂pr i , x(a) ∂j] = x(a−pr ǫi) ∂j if ai ≥ pr and 0 otherwise. It remains to describe special, Hamiltonian and contact Lie algebras of Cartan type. Consider the divergence map div : W (m; n) → O(m; n) m Xi=1 fi ∂i 7→ m Xi=1 ∂i(fi). It is easy to check that div([D, E]) = D(div(E)) − E(div(D)) for all D, E ∈ W (m; n). As a result, the set S(m; n) := (cid:8)D ∈ W (m; n) div(D) = 0(cid:9) (1.7) is a Lie subalgebra of W (m; n) [31, Lemma 3.1, Sec. 4.3, Chap. 4]. It is not simple. But its derived subalgebra S(m; n)(1) is simple. We refer to the Lie algebra S(m; n)(1) as the simple special Lie algebra of Cartan type. More generally, S(m; n) or S(m; n)(1) is referred to as a special Cartan type Lie algebra. Let us describe the structure of S(m; n)(1) in more detail. Define Di,j : O(m; n) → W (m; n) f 7→ ∂j(f ) ∂i − ∂i(f ) ∂j 21 1.4. Graded Lie algebras of Cartan type Theorem 1.4.5. [31, Lemma 3.2, Proposition 3.3, Theorems 3.5 and 3.7, Sec. 4.3, Chap. 4] Suppose m ≥ 3. (i) Di,j is a linear map of degree −2 satisfying Di,i = 0 and Di,j = −Dj,i for all 1 ≤ i, j ≤ m. (ii) Di,j(O(m; n)) ⊂ S(m; n) for all 1 ≤ i, j ≤ m. (iii) S(m; n)(1) is the subalgebra of S(m; n) generated by (cid:8)Di,j(x(a)) 0 ≤ al < pnl for 1 ≤ l ≤ m and 1 ≤ i < j ≤ m(cid:9). (iv) S(m; n)(1) is a simple Lie algebra of dimension (m − 1)(pPm i=1 ni − 1). (v) S(m; n)(1) is a graded subalgebra of W (m; n), i.e. S(m; n)(1) = s1 (S(m; n)(1))i, Mi=−1 where s1 = (Pm i=1 pni) − m − 2 and (S(m; n)(1))i = S(m; n)(1) ∩ W (m; n)i. (vi) S(m; n)(1) is restricted if and only if n = (1, . . . , 1), and in that case S(m; n)(1) is a p-subalgebra of W (m; n) with restricted gradation. Alternatively, we can define special Lie algebras of Cartan type using differential forms on O(m; n); see [24, Sec. 3.2] and [30, Sec. 4.2]. Set Ω0(m; n) := O(m; n), Ω1(m; n) := HomO(m;n)(W (m; n), O(m; n)). Then Ω1(m; n) admits an O(m; n)-module structure via (f α)(D) := f α(D) for all f ∈ O(m; n), α ∈ Ω1(m; n), D ∈ W (m; n), and a W (m; n)-module structure via (Dα)(E) = D(α(E)) − α([D, E]) for all D, E ∈ W (m; n), α ∈ Ω1(m; n). Since W (m; n) is a free O(m; n)-module with basis ∂1, . . . , ∂m, every α ∈ Ω1(m; n) is determined by its effects on ∂1, . . . , ∂m. It is easy to check that α = Pm This implies that Ω1(m; n) is a free O(m; n)-module with basis dx1, . . . , dxm. i=1 α(∂i)dxi. 22 1.4. Graded Lie algebras of Cartan type Define d : Ω0(m; n) → Ω1(m; n) by df (D) = D(f ) for all f ∈ O(m; n), D ∈ W (m; n). Then d is a homomorphism of W (m; n)-modules. Set Ωr(m; n) := r ^ Ω1(m; n), the r-fold exterior power algebra over O(m; n). It is a free O(m; n)-module with basis {dxi1 ∧ · · · ∧ dxir 1 ≤ i1 < · · · < ir ≤ m}. Let Ω(m; n) := M1≤r≤m Ωr(m; n). Then elements of Ω(m; n) are called differential forms on O(m; n). We can extend the above linear operator d to Ω(m; n) by setting d(α1 ∧ α2) := d(α1) ∧ α2 + (−1)deg(α1)α1 ∧ d(α2) for all α1, α2 ∈ Ω(m; n). Then d is a linear operator of degree 1 satisfying d2(α) = 0, D(dα) = dD(α), d(f α) = (df ) ∧ α + f d(α), D(df ) = dD(f ) for all f ∈ O(m; n), D ∈ W (m; n), α ∈ Ω(m; n). Note that D(f α) = (Df )α + f D(α) for every D ∈ W (m; n). Hence D extends to a derivation of Ω(m; n). Recall the volume form Then ωS := dx1 ∧ · · · ∧ dxm, m ≥ 3. (cid:8)D ∈ W (m; n) D(ωS) = 0(cid:9) coincides with S(m; n) (1.7); see [31, p. 161]. The simple special Lie algebra of Cartan type is the derived subalgebra of S(m; n). Suppose now char k = p > 2 and m = 2r ≥ 2. The Hamiltonian form ωH := r Xi=1 dxi ∧ dxi+r, m = 2r ≥ 2 gives rise to a Lie subalgebra H(2r; n) := (cid:8)D ∈ W (2r; n) D(ωH) = 0(cid:9) (1.8) 23 1.4. Graded Lie algebras of Cartan type of W (2r; n). The second derived subalgebra H(2r; n)(2) is simple. We refer to H(2r; n)(2) as the simple Hamiltonian Lie algebra of Cartan type. More generally, H(2r; n) or H(2r; n)(2) is a Hamiltonian Cartan type Lie algebra. Alternatively, we can define H(2r; n) using a linear map. Let us first introduce some notations. Set σ(j) :=   j′ :=   1 −1 if 1 ≤ j ≤ r, if r < j ≤ 2r, j + r if 1 ≤ j ≤ r, j − r if r < j ≤ 2r. (1.9) (1.10) fi ∂i ∈ W (2r; n) σ(i) ∂j ′(fi) = σ(j) ∂i′(fj), 1 ≤ i, j ≤ 2r(cid:27). Consider the set (cid:26)D = 2r Xi=1 One can check that this is equivalent to H(2r; n) (1.8). To describe H(2r; n)(2), we define DH : O(2r; n) → W (2r; n) f 7→ 2r Xi=1 σ(i) ∂i(f ) ∂i′ . Denote the image of DH by H(2r; n). Note that H(2r; n) is a proper subset of H(2r; n). Indeed derivations x(pnj −1) ∂j ′ for 1 ≤ j ≤ 2r lie in H(2r; n), but do not lie in H(2r; n); j see [31, p. 163]. Theorem 1.4.6. [31, Lemma 4.1, Proposition 4.4 and Theorem 4.5, Sec. 4.4, Chap. 4] (i) DH is a linear map of degree −2 with Ker DH = k. (ii) [DH(f ), DH(g)] = DH(DH(f )(g)) for all f, g ∈ O(2r; n). (iii) H(2r; n)(1) ⊆ H(2r; n). (iv) H(2r; n)(2) is a simple Lie algebra with basis (cid:8)DH(x(a)) (0, . . . , 0) < a < (pn1 − 1, . . . , pn2r − 1)(cid:9). Hence dim H(2r; n)(2) = pP2r i=1 ni − 2. 24 1.4. Graded Lie algebras of Cartan type (v) H(2r; n)(2) is a graded subalgebra of W (2r; n), i.e. H(2r; n)(2) = s2 (H(2r; n)(2))i, Mi=−1 where s2 = (P2r i=1 pni) − 2r − 3 and (H(2r; n)(2))i = H(2r; n)(2) ∩ W (2r; n)i. (vi) H(2r; n)(2) is restricted if and only if n = (1, . . . , 1), and in that case H(2r; n)(2) is a p-subalgebra of W (2r; n) with restricted gradation. Remark 1.4.1. [31, p. 168] Note that DH defines a Lie bracket on O(2r; n). For any f, g ∈ O(2r; n), define {f, g} := 2r Xi=1 σ(i) ∂i(f ) ∂i′(g) = DH(f )(g). It follows from Theorem 1.4.6 that (O(2r; n), {, }) is a Lie algebra with centre k, and (O(2r; n)/k)(1) ∼= H(2r; n)(2); see [1, p. 54]. The Lie bracket {, } is referred to as the Poisson bracket. Suppose char k = p > 2 and m = 2r + 1 ≥ 3. Consider the contact form r ωK := dxm + (xidxi+r − xi+rdxi), m = 2r + 1 ≥ 3. Xi=1 Set (cid:8)D ∈ W (2r + 1; n) D(ωK) ∈ O(2r + 1; n)ωK(cid:9). (1.11) This gives a Lie subalgebra of W (2r + 1; n), denoted K(2r + 1; n), called the contact Cartan type Lie algebra. The derived subalgebra K(2r + 1; n)(1) is simple. Similarly, we can describe the structure of K(2r + 1; n) using a linear map. Let σ(j) and j′ be as in (1.9) and (1.10). Define DK : O(2r + 1; n) → W (2r + 1; n) 2r f 7→ Xj=1 (cid:0)σ(j) ∂j(f ) + xj ′ ∂2r+1(f )(cid:1) ∂j ′ +(cid:0)2f − xj ∂j(f )(cid:1) ∂2r+1 . 2r Xj=1 Then the image DK(O(2r+1; n)) gives a Lie subalgebra of W (2r+1; n) which coincides with K(2r + 1; n) (1.11); see [30, p. 189]. 25 Theorem 1.4.7. [1, Theorem 2.6.2; 31, Proposition 5.3, Theorems 5.5 and 5.6, 1.4. Graded Lie algebras of Cartan type Sec. 4.5, Chap. 4] (i) DK is an injective linear map of degree −2. (ii) K(2r + 1; n) is graded, i.e. s3 where s3 = (P2r K(2r + 1; n) = Mi=−2 i=1 pni) + 2pn2r+1 − 2r − 3 and K(2r + 1; n)i = span(cid:26)DK(x(a)) K(2r + 1; n)i, 2r+1 Xi=1 ai + a2r+1 − 2 = i(cid:27). Note that this grading has depth 2. (iii) K(2r + 1; n)(1) is simple, and K(2r + 1; n) if 2r + 4 6≡ 0 (mod p), span(cid:8)DK(x(a)) 0 ≤ a < τ (n)(cid:9) if 2r + 4 ≡ 0 (mod p), where τ (n) := (pn1 − 1, . . . , pn2r+1 − 1). Then K(2r + 1; n)(1) =  dim K(2r + 1; n)(1) =  pP2r+1 i=1 ni if 2r + 4 6≡ 0 (mod p), pP2r+1 i=1 ni − 1 if 2r + 4 ≡ 0 (mod p). (iv) K(2r + 1; n)(1) is restricted if and only if n = (1, . . . , 1), and in that case K(2r + 1; n)(1) is a p-subalgebra of W (2r + 1; n) with restricted gradation. Remark 1.4.2. [30, p. 191(4.2.17); 31, p. 172] As in the case of Hamiltonian Lie algebras, the linear map DK also defines a Lie bracket on O(2r + 1; n). For any f, g ∈ O(2r + 1; n), define hf, gi := DK(f )(g) − 2g ∂2r+1(f ). By [31, Proposition 5.2, Sec. 4.5, Chap. 4], DK(hf, gi) = [DK(f ), DK(g)]. Moreover, DK is injective. Hence (O(2r + 1; n), h, i) is a Lie algebra over k, and O(2r + 1; n) ∼= K(2r + 1; n). The Lie bracket h, i is referred to as the contact bracket. These are the four families of Lie algebras of Cartan type. We finish this section by emphasizing that Theorem 1.4.8. [30, Corollary 7.2.3] The simple restricted Lie algebras of Cartan type are W (m; 1), m ≥ 1, S(m; 1)(1), m ≥ 3, H(2r; 1)(2), r ≥ 1 and K(2r+1; 1)(1), r ≥ 1. 26 1.5. Nilpotent varieties 1.5 Nilpotent varieties In this section we review Premet's results on nilpotent variety of any finite dimensional restricted Lie algebra over k. Then we introduce Premet's conjecture and discuss what have been done so far. Let (g, [p]) be a finite dimensional restricted Lie algebra over k. By Jacobson's formula (see Definition 1.1.1(3)), we see that the [p]-th power map is a morphism given by homogeneous polynomial functions on g of degree p. Recall the nilpotent variety N (g) which is the set of all x ∈ g such that x[p]N = 0 for N ≫ 0. It is well known that N (g) is a Zariski closed, conical subset of g. Let gss denote the set of all semisimple elements of g. By Theorem 1.1.2, it is straightforward to see that g[p]N = gss for N ≫ 0. Define e = e(g) to be the smallest nonnegative integer such that W [p]e ⊆ gss for some nonempty Zariski open subset W of g. By [18, Theorem 2], we know that e = 0 if and only if g possesses a toral Cartan subalgebra. Let s = MT(g) denote the maximal dimension of tori in g. In Sec. 1.1, we already observed that s = 0 if and only if g coincides with N (g). We present the theorem on N (g) which was proved by A. Premet. Theorem 1.5.1. [19, Theorem 2 and Corollary 2; 21, Theorem 4.2] There exist ho- mogeneous polynomials ψ0, . . . , ψs−1 ∈ k[g] with deg ψi = pe+s − pe+i such that for any x ∈ g, x[p]e+s = Ps−1 i=0 ψi(x)x[p]e+i. Moreover, the following are true: (i) ψi(x[p]) = ψi(x)p for all x ∈ g and i ≤ s − 1. (ii) N (g) = Z(ψ0, . . . , ψs−1), the set of all common zeros of ψ0, . . . , ψs−1 in g. (iii) All irreducible components of N (g) have dimension dim g−s, i.e. N (g) is equidi- mensional. (iv) For any x ∈ N (g), x[p]e+s = 0. The above theorem gives useful information on N (g) and plays an important role in Chapters 3 and 4. It is well known that the nilpotent variety of any finite dimensional algebraic Lie algebra over C is irreducible. Then Premet conjectured that Conjecture 1. [19, p. 563] For any finite dimensional restricted Lie algebra g over k the variety N (g) is irreducible. 27 1.5. Nilpotent varieties This conjecture is still open, but it is known that if g is the Lie algebra of a con- nected algebraic group G′, and n is the set of nilpotent elements in a Borel subalgebra of g, then N (g) = {g.n g ∈ G′, n ∈ n}. Since G′ is connected and n is irreducible, the variety N (g) is irreducible; see [13, p. 64]. Moreover, this conjecture holds for the Jacobson-Witt algebras W (n; 1) [20, Lemma 6], for the Special Lie algebras S(n; 1) and S(n; 1)(1) [34, Theorem A], for the Poisson Lie algebras (O(2n; 1), {, }) [29, Theo- rem 6.4] and for the Hamiltonian Lie algebras H(2n; 1)(2) [33, Theorem A]. There are no known counterexamples. A good understanding of the proof in the case W (n; 1) is important. This is because some of the results will be used in Sec. 3.2, Chapter 3. Moreover, a similar idea works for the minimal p-envelope W (1; n)p of the Zassenhaus algebra W (1; n); see Chapter 4. So let us first introduce some notations and state a few preliminary results. Then we show that N (W (n; 1)) is irreducible. Let k be an algebraically closed field of characteristic p > 2. Let O(n; 1) = k[X1, . . . , Xn]/(X p 1 , . . . , X p n) denote the truncated polynomial ring in n variables. Note that O(n; 1) is a local ring with its unique maximal ideal denoted m. Let W (n; 1) denote the derivation algebra of O(n; 1). This is a simple restricted Lie algebra. Moreover, W (n; 1) obtains an O(n; 1)-module structure via (f D)(g) = f D(g) for all f, g ∈ O(n; 1) and D ∈ W (n; 1). Since each D ∈ W (n; 1) is uniquely determined by its effects on x1, . . . , xn, it is easy to see that W (n; 1) is a free O(n; 1)-module of rank n generated by the partial derivatives ∂1, . . . , ∂n such that ∂i(xj) = δij for all 1 ≤ i, j ≤ n. Hence dim W (n; 1) = npn. Note that there is a standard filtration {W (n; 1)(i)}−1≤i≤m(p−1)−1 defined on W (n; 1); see (1.6). In particular, the subalgebra W (n; 1)(0) := (cid:26) n Xi=1 fi ∂i fi ∈ m for all i(cid:27) is referred to as the standard maximal subalgebra of W (n; 1). Note that for any D1 ∈ W (n; 1)(0), the following holds: D1(O(n; 1)) ⊆ m. Let G denote the automorphism group of O(n; 1). Each σ ∈ G is uniquely de- termined by its effects on x1, . . . , xn. An assignment σ(xi) = fi extends to an au- tomorphism of O(n; 1) if and only if fi ∈ m, and the Jacobian Jac(f1, . . . , fn) := (cid:12)(cid:12)(cid:0) ∂ fi ∂ xj(cid:1)1≤i,j≤n(cid:12)(cid:12) /∈ m. It follows that G is a connected algebraic group of dimension dim W (n; 1) − n. It is well known that any automorphism of W (n; 1) is induced by an 28 1.5. Nilpotent varieties automorphism of O(n; 1) via the rule Dσ = σ ◦ D ◦ σ−1 for all σ ∈ G and D ∈ W (n; 1). So we can identify G with the automorphism group of W (n; 1). Note that for any f ∈ O(n; 1), D ∈ W (n; 1) and σ ∈ G, (f D)σ = f σDσ, (1.12) where f σ = f (σ(x1), . . . , σ(xn)); see [20, p. 154]. It follows from (1.12) that if D1 = Pn i=1 gi ∂i is an element of W (n; 1), then Dσ 1 = n Xi, j=1 gσ i (cid:18) ∂i(cid:0)σ−1(xj)(cid:1)(cid:19)σ ∂j; (1.13) see [7, Sec. 3]. Note also that G respects the standard filtration of W (n; 1). It is known that MT(W (n; 1)) = n and e(W (n; 1)) = 0. By Theorem 1.5.1, we know that dim N (W (n; 1)) = dim W (n; 1)−n and the nilpotency index of any element in N (W (n; 1)) is at most pn. We want to show that N (W (n; 1)) is irreducible. Let us start with the following result: Lemma 1.5.1. [20, Lemma 3] Let D = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 n−1 ∂n. Then (i) D pl = (−1)l(∂l+1 +xp−1 l+1 ∂l+2 + · · · + xp−1 l+1 · · · xp−1 n−1 ∂n) for all 0 ≤ l ≤ n − 1 and D pn = 0. (ii) D, D p, . . . , D pn−1 forms a basis of the O(n; 1)-module W (n; 1). (iii) D pn−1(xp−1 1 · · · xp−1 n ) = (−1)n. Hence the matrix of the endomorphism D : O(n; 1) → O(n; 1) is similar to a Jordan block of size pn with zeros on the main diagonal. (iv) The stabilizer of D in G is trivial. Sketch of proof. (i) We prove (i) by induction on l. For l = 0, the result is clear. For l = 1, let D ′ = xp−1 n−1 ∂n. Recall Jacobson's formula, ∂2 + · · · + xp−1 · · · xp−1 1 1 (a + b)p = ap + bp + p−1 Xi=1 si(a, b), 29 1.5. Nilpotent varieties where the terms si(a, b) ∈ W (n; 1) are such that (ad(ta + b))p−1(a) = p−1 Xi=1 isi(a, b)ti−1. Set a = D ′ and b = ∂1. Then for any s ≤ p − 2, [a, (ad b)s(a)] = (−1)ss![xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 n−1 ∂n, xp−1−s 1 ∂2 + · · · + xp−1−s 1 xp−1 2 · · · xp−1 n−1 ∂n] = 0. Hence p−1 D p = ∂p 1 +(D ′)p + si(D ′, ∂1) = s1(D ′, ∂1) = (ad ∂1)p−1(D ′) Xi=1 ∂3 + · · · + xp−1 2 = −(∂2 +xp−1 2 · · · xp−1 n−1 ∂n). So the result holds for l = 1. Continuing in this way and by induction, one can show that D pl = (−1)l(∂l+1 +xp−1 l+1 ∂l+2 + · · · + xp−1 l+1 · · · xp−1 n−1 ∂n) for all 0 ≤ l ≤ n − 1. In particular, D pn−1 = (−1)n−1 ∂n. Then D pn = 0. This proves (i). (ii) Let M : W (n; 1) → W (n; 1) be the endomorphism such that M(∂i) = D pi−1 for 1 ≤ i ≤ n. By (i), it is easy to check that the matrix of M with respect to ∂1, . . . , ∂n is an upper triangular matrix with 1 or −1 on the diagonal. Hence M is invertible. Therefore, D, D p, . . . , D pn−1 forms a basis of the O(n; 1)-module W (n; 1). This proves (ii). (iii) By (i), it is easy to check that for 0 ≤ l ≤ n − 2, (D pl )p−1(xp−1 l+1 · · · xp−1 n ) =(cid:0)(−1)l(∂l+1 +xp−1 =(−1)l(p−1)(−1)xp−1 l+1 ∂l+2 + · · · + xp−1 l+2 · · · xp−1 n l+1 · · · xp−1 n−1 ∂n)(cid:1)p−1(xp−1 l+1 · · · xp−1 n ) = − xp−1 l+2 · · · xp−1 n . For l = n − 1, (D pn−1 )p−1(xp−1 n ) = (cid:0)(−1)n−1 ∂n(cid:1)p−1(xp−1 n ) = (−1)(n−1)(p−1)(−1) = −1. 30 1.5. Nilpotent varieties Then D pn−1(xp−1 1 · · · xp−1 n ) =D pn−pD p−1(xp−1 1 · · · xp−1 n ) = −D pn−p(xp−1 2 · · · xp−1 n ) = − (D p)pn−1−1(xp−1 2 · · · xp−1 n ) = − (D p)pn−1−p(D p)p−1(xp−1 2 · · · xp−1 n ) =(−1)2(D p)pn−1−p(xp−1 3 · · · xp−1 n ) =(−1)2(D p2 )pn−2−1(xp−1 3 · · · xp−1 n ) = · · · = (−1)n−1(D pn−1 )p−1(xp−1 n ) = (−1)n−1(−1) = (−1)n. Hence D pn−1 6= 0. Since D pn = 0, this implies that the matrix of D is similar to a Jordan block of size pn with zeros on the diagonal. This proves (iii). (iv) Let C be the stabilizer of D in G and c ∈ C. By (iii), we know that O(n; 1) is an indecomposable kD-module, i.e. pn−1 kD i(xp−1 1 · · · xp−1 n ). (1.14) Mi=0 Since G acts on the straight line kxp−1 O(n; 1) = · · · xp−1 n , we have that 1 c(xp−1 1 · · · xp−1 n ) = λcxp−1 1 · · · xp−1 n for some λc ∈ k∗. Moreover, c commutes with D. Applying (1.14), we get c(cid:0)D i(xp−1 1 · · · xp−1 n )(cid:1) = D i(cid:0)c(xp−1 1 · · · xp−1 n )(cid:1) = λcD i(xp−1 1 · · · xp−1 n ). Hence c = λc Id. Since c(1) = 1, this implies that λc = 1. So c = Id. This proves (iv). Set O := G.D. Since G is connected, it follows from the last result that the Zariski closure of O is an irreducible variety of dimension dim G = dim N (W (n; 1)). Since D is nilpotent, it follows that O is an irreducible component of N (W (n; 1)). We want to describe O explicitly. For that, we need the following result: Lemma 1.5.2. [7, Lemma 6] Let z be an element of W (n; 1) such that z /∈ W (n; 1)(0). Then z is conjugate under G to z1 = ∂1 +xp−1 ϕi ∈ k[X2, . . . , Xn]/(X p n) for all i. Moreover, zp 1 = −(1 + xp−1 i=1 ϕi ∂i, where 1 Pn 2 , . . . , X p 1 ϕ1)Pn i=1 ϕi ∂i. 31 1.5. Nilpotent varieties Sketch of proof. In [7, Sec. 3], S. P. Demushkin used the convention that for any D ∈ W (n; 1) and Φ ∈ G, DΦ = Φ−1 ◦ D ◦ Φ. Our convention is that DΦ = Φ ◦ D ◦ Φ−1. So we need to slightly modify his proof. Let z = Pn i=1 fi ∂i be an element of W (n; 1) such that z /∈ W (n; 1)(0). Then fµ(0, . . . , 0) 6= 0 for some 1 ≤ µ ≤ n. If µ 6= 1, then we show that z is conjugate under G to Pn i=1 fi ∂i with f1(0, . . . , 0) 6= 0. Let Φ ∈ G be such that Φ(x1) = xµ, Φ(xµ) = x1 and Φ(xj) = xj for j 6= 1, µ. Then Φ−1(x1) = xµ, Φ−1(xµ) = x1 and Φ−1(xj) = xj for j 6= 1, µ. By (1.13), zΦ = n Xi, j=1 f Φ i (cid:18) ∂i(cid:0)Φ−1(xj)(cid:1)(cid:19)Φ ∂j = f Φ µ ∂1 +f Φ 1 ∂µ + Xi6=1, µ f Φ i ∂i . Since f Φ µ (0, . . . , 0) = fµ(0, . . . , 0) 6= 0, we may assume from the beginning that z = Pn i=1 fi ∂i with f1(0, . . . , 0) 6= 0. Let Φ2 ∈ G be such that Φ2(x1) = f1(0, . . . , 0)x1 and Φ2(xj) = xj for 2 ≤ j ≤ n. Then Φ−1 2 (x1) = f −1 1 (0, . . . , 0)x1 and Φ−1 2 (xj) = xj for 2 ≤ j ≤ n. Applying Φ2 to z, we get zΦ2 = n Xi, j=1 f Φ2 i (cid:18) ∂i(cid:0)Φ−1 2 (xj)(cid:1)(cid:19)Φ2 ∂j = f Φ2 1 f −1 1 (0, . . . , 0) ∂1 + n Xi=2 f Φ2 i ∂i . Since f Φ2 1 (0, . . . , 0) = f1(0, . . . , 0), we may assume from the beginning that z = Pn i=1 fi ∂i with f1(0, . . . , 0) = 1. Let Φ3 ∈ G be such that Φ3(x1) = x1 and Φ3(xj) = xj + αjx1, where αj = fj(0 . . . , 0) for 2 ≤ j ≤ n. Then Φ−1 3 (x1) = x1 and Φ−1 3 (xj) = xj − αjx1 for 2 ≤ j ≤ n. Applying Φ3 to z, we get n f Φ3 zΦ3 = i (cid:18) ∂i(cid:0)Φ−1 3 (xj)(cid:1)(cid:19)Φ3 Xi, j=1 Xi=2 Hence we may assume from the beginning that z = Pn i=1 fi ∂i with f1(0, . . . , 0) = 1 and fi(0, . . . , 0) = 0 for all 2 ≤ i ≤ n. Then we can write i − αif Φ3 ∂j = f Φ3 1 ∂1 + 1 ) ∂i . n (f Φ3 f1 = 1 + m(p−1) Xl=1 f1,l and fi = m(p−1) Xl=1 fi,l (1.15) for some f1,l, fi,l ∈ m with deg f1,l = deg fi,l = l. Let Φ4 ∈ G be such that Φ−1 4 (xj) = xj +gj for 1 ≤ j ≤ n, where gj are elements of m with the same degree ν ≥ 2. Applying 32 Φ4 to z, we get zΦ4 = n Xi, j=1 f Φ4 i (cid:18) ∂i(cid:0)Φ−1 4 (xj)(cid:1)(cid:19)Φ4 1.5. Nilpotent varieties ∂j = n Xi, j=1 f Φ4 i (cid:18) ∂i(cid:0)xj + gj(cid:1)(cid:19)Φ4 ∂j . (1.16) Substituting (1.15) into (1.16) and expanding out, we get n zΦ4 ≡ Xi=1 (cid:0)fi + ∂1(gi)(cid:1) ∂i (mod W (n; 1)(ν−1)). Since fi(0, . . . , 0) = δi1, we can rewrite the above as zΦ4 ≡ ∂1 +(cid:0)f1 − 1 + ∂1(g1)(cid:1) ∂1 + Xi=2 (cid:0)fi + ∂1(gi)(cid:1) ∂i (mod W (n; 1)(ν−1)), n where f1 − 1 + ∂1(g1), fi + ∂1(gi) ∈ m with degrees strictly less than ν. It follows that z is conjugate under G to z1 = ∂1 +xp−1 1 n Xi=1 ϕi ∂i, where ϕi ∈ k[X2, . . . , Xn]/(X p It remains to show that zp n). 2 , . . . , X p 1 = −(1 + xp−1 1 ϕ1)Pn i=1 ϕi ∂i. Note that z1(x1) = 1 + xp−1 1 ϕ1, 1(x1) = (p − 1)xp−2 z2 1 ϕ1, . . . 1 (x1) = (p − 1) · · · (p − η + 1)xp−η zη 1(x1) = (p − 1)!(1 + xp−1 zp 1 ϕ1)ϕ1 = −(1 + xp−1 1 ϕ1)ϕ1. 1 ϕ1 for 2 ≤ η ≤ p − 1, and Similarly, one can show that zp 1(xi) = −(1 + xp−1 1 ϕ1)ϕi for 2 ≤ i ≤ n. Hence 1 = −(1 + xp−1 zp 1 ϕ1) n Xi=1 ϕi ∂i . This completes the sketch of proof. Recall that D = ∂1 +xp−1 1 ∂2 + · · ·+xp−1 1 · · · xp−1 n−1 ∂n. We can now describe O = G.D as: Lemma 1.5.3. [20, Lemma 4] O = (cid:8)D ∈ N (W (n; 1)) Dpn−1 6∈ W (n; 1)(0)(cid:9). 33 Sketch of proof. Set S = (cid:8)D ∈ N (W (n; 1)) Dpn−1 6∈ W (n; 1)(0)(cid:9). It follows from Lemma 1.5.1(i) that D pn = 0 and D pn−1 = (−1)n−1 ∂n /∈ W (n; 1)(0). Hence O ⊆ S. To 1.5. Nilpotent varieties show that S ⊆ O, we proceed by induction on n. Let z be an element of W (n; 1) such that z /∈ W (n; 1)(0). By Lemma 1.5.2, z is conjugate under G to ∂1 +xp−1 where ϕi ∈ k[X2, . . . , Xn]/(X p 1 Pn i=1 ϕi ∂i, n). Moreover, 2 , . . . , X p n (cid:0) ∂1 +xp−1 1 Xi=1 ϕi ∂i(cid:1)p = −(1 + xp−1 1 ϕ1) n Xi=1 ϕi ∂i . If zp = 0, then (1 + xp−1 1 ϕ1)ϕi = 0 for all 1 ≤ i ≤ n. Since 1 + xp−1 1 ϕ1 is invertible in O(n; 1), we have that ϕi = 0 for all 1 ≤ i ≤ n. Hence any D ∈ W (n; 1) with D 6∈ W (n; 1)(0) and Dp = 0 is conjugate under G to ∂1. Thus, the result is true for n = 1. Suppose n > 1. Let y ∈ N (W (n; 1)) be such that ypn−1 6∈ W (n; 1)(0). Set z = ypn−1. Since z 6∈ W (n; 1)(0) and zp = 0, there exists Φ ∈ G such that ΦzΦ−1 = (ΦyΦ−1)pn−1 = ∂n. So we may assume that z = ypn−1 = ∂n. Since [y, ypn−1] = 0, we have that y ∈ cW (n;1)(∂n). One can check that the centralizer cW (n;1)(∂n) is isomorphic to the semidirect product of W (n − 1; 1) and an abelian ideal J = {fn ∂n fn ∈ O(n − 1; 1)}. Hence y = f1 ∂1 + · · · + fn−1 ∂n−1 +fn ∂n for some fi ∈ O(n − 1; 1). Set y1 = f1 ∂1 + · · · + fn−1 ∂n−1 and y2 = fn ∂n. We show that y1 is conjugate under G to D0 = ∂1 +xp−1 ∂2 + · · · + xp−1 n−2 ∂n−1. Since · · · xp−1 1 1 y1 ∈ cW (n;1)(∂n), it follows from (1.4) and Jacobson's formula that for any j ≥ 1, ypj = ypj (fn) ∂n. In particular, 1 + ypj−1 1 ypn−1 = ypn−1 1 + ypn−1−1 1 (fn) ∂n . Since ypn = 0, this implies that y1 ∈ N (W (n − 1; 1)). Then ypn−1 ypn−1−1 (fn) ∂n. By our assumption, ypn−1 6∈ W (n; 1)(0). Hence ypn−1−1 1 1 1 = 0 and ypn−1 = (fn) is invertible in O(n − 1; 1). Note that if D1 ∈ W (n; 1)(0), then D1(O(n; 1)) ⊆ m. that ypn−2 (y1)pn−2(f ′) ∈ m, a contradiction. So ypn−2 ∈ W (n − 1; 1)(0), then ypn−1−1 Indeed, if ypn−2 /∈ W (n − 1; 1)(0). /∈ W (n − 1; 1)(0). 1 1 1 1 It follows (fn) = By the induction hypothesis, y1 is conjugate under Aut(W (n − 1; 1)) to D0 = ∂1 +xp−1 1 ∂2 + · · ·+ xp−1 1 · · · xp−1 n−2 ∂n−1. Since W (n−1; 1) is a Lie subalgebra of W (n; 1), we may identify Aut(W (n − 1; 1)) with a subgroup of Aut(W (n; 1)) by letting σ(xn) = 34 1.5. Nilpotent varieties xn for all σ ∈ Aut(W (n − 1; 1)). Hence we may assume that y = D0 + ψ ∂n for some ψ ∈ O(n − 1; 1) with D pn−1−1 that Ker(D pn−1−1 O(n−1;1)) = D0(O(n − 1; 1)). By Lemma 1.5.1(iii), we have that (ψ) invertible in O(n − 1; 1). Since D pn−1 = 0, we have 0 0 0 D pn−1−1 0 (O(n − 1; 1)) = k and D pn−1−1 0 (xp−1 1 · · · xp−1 n−1) = (−1)n−1. Hence there exist ϕ ∈ O(n − 1; 1) ∩ m and α ∈ k∗ such that ψ = D0(ϕ) + αxp−1 1 · · · xp−1 n−1. Next we show that y = D0 + ψ ∂n is conjugate under G to D. Let Φ1 ∈ G be such that Φ1O(n−1;1) = Id and Φ1(xn) = xn + ϕ. It is easy to check that Φ−1 1 ≤ i ≤ n − 1 and Φ−1 1 (xn) = xn − ϕ. Then for 1 ≤ i ≤ n − 1, 1 (xi) = xi for Φ1yΦ−1 1 (xi) = Φ1(D0 + ψ ∂n)(xi) = D0(xi), and Φ1yΦ−1 1 (xn) = Φ1(D0 + ψ ∂n)(xn − ϕ) = Φ1(cid:0) − D0(ϕ) + ψ(cid:1) = αxp−1 1 = D0 + αxp−1 n−1 ∂n. Let Φ2 ∈ G be such that Φ2O(n−1;1) = Id and · · · xp−1 n−1. 1 So Φ1yΦ−1 Φ2(xn) = αxn. Then one can show that Φ2(Φ1yΦ−1 · · · xp−1 1 1 )Φ−1 2 = D. It follows that y ∈ O. Therefore, we proved by induction that S ⊆ O. As a result, O = S. This completes the sketch of proof. Now we consider the complement of O in N (W (n; 1)) and show that this com- plement has dimension strictly less than dim N (W (n; 1)) = dim W (n; 1) − n; see Lemma 1.5.5. For that, we need to introduce the notion of admissible n-tuples and define a lexicographic ordering on them; see also Definition 3.2.3 in Sec. 3.2.1. Definition 1.5.1. [20, Lemma 5] (i) We say that an n-tuple A = (a1, . . . , an) ∈ Nn 0 is admissible if 0 ≤ ai ≤ p − 1 for any 1 ≤ i ≤ n. We write xA := xa1 1 · · · xan n . (ii) We define a lexicographic ordering ≺lex on the set of admissible n-tuples by extending the ordering (1, 0, . . . , 0) ≺lex (0, 1, . . . , 0) ≺lex · · · ≺lex (0, 0, . . . , 1). 35 Explicitly, for any two non-equal admissible n-tuples A = (a1, . . . , an) and A′ = 1.5. Nilpotent varieties (a′ 1, . . . , a′ n), A = (a1, . . . , an) ≺lex (a′ 1, . . . , a′ n) = A′ if and only if ai < a′ i, where i is the largest number in {1, . . . , n} for which ai 6= a′ i. Lemma 1.5.4. [20, Lemma 5] The action of D = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 n−1 ∂n on O(n; 1) is compatible with the lexicographic ordering ≺lex defined in Definition 1.5.1. Explicitly, for any admissible n-tuple A 6= (0, . . . , 0), we have that D(x A) = λAx A′ , where λA ∈ k∗ and A′ ≺lex A. It follows from the explicit formulas for A′ in the proof that if B and C are admissible n-tuples such that B, C 6= (0, . . . , 0) and B ≺lex C, then B′ ≺lex C ′. Sketch of proof. Let A = (a1, . . . , an) 6= (0, . . . , 0) be any admissible n-tuple. We first show that D(xA) = λAxA′, where λA ∈ k∗ and A′ ≺lex A. If a1 ≥ 1, then D(x A) = a1xa1−1 1 xa2 2 · · · xan n = a1x A′ , where A′ = (a1 − 1, a2, . . . , an). It is easy to see that A′ = (a1 − 1, a2, . . . , an) ≺lex (a1, . . . , an) = A. If a1 = · · · = as−1 = 0 and as ≥ 1, then D(x A) = asxp−1 1 xp−1 2 · · · xp−1 s−1xas−1 s xas+1 s+1 · · · xan n = asx A′ , where A′ = (p − 1, . . . , p − 1, as − 1, as+1, . . . , an). It is easy to see that A′ = (p − 1, . . . , p − 1, as − 1, as+1, . . . , an) ≺lex (0, . . . , 0, as, as+1, . . . , an) = A. Hence D(xA) = λAxA′ 6= (0, . . . , 0), then applying D to xA′, we get D(xA′) = λA′ xA′′, where λA′ ∈ k∗ and A′′ ≺lex A′. It follows for some λA ∈ k∗ and A′ ≺lex A. If A′ from the explicit formulas for A′ that if B and C are admissible n-tuples such that B, C 6= (0, . . . , 0) and B ≺lex C, then B′ ≺lex C ′. Hence the action of D is compatible with the lexicographic ordering ≺lex. This completes the sketch of proof. Lemma 1.5.5. [20, Lemma 5] Set N0 := (cid:8)D ∈ N (W (n; 1)) Dpn−1 ∈ W (n; 1)(0)(cid:9). Then dim N0 < dim W (n; 1) − n. 36 1.5. Nilpotent varieties Sketch of proof. By Theorem 1.6.8, it is enough to construct an (n + 1)-dimensional subspace V in W (n; 1) such that V ∩ N0 = {0}. Set V := Tn ⊕ kD, where Tn is a maximal torus in W (n; 1) with basis {x1 ∂1, . . . , xn ∂n}, and D is the nilpotent element in Lemma 1.5.1. We show that V ∩ N0 = {0}. Suppose the contrary, i.e. V ∩ N0 6= {0}. Then t + D ∈ N0 for some 0 6= t ∈ Tn. Note that for any admissible n-tuple A = (a1, . . . , an), the straight line kxA is invariant under Tn and corresponds to the weight ¯c1θ1 + · · · + ¯cnθn, where {θ1, . . . , θn} is a basis of T ∗ n dual to the basis {x1 ∂1, . . . , xn ∂n} and ¯ci ≡ ci (mod p). By Lemma 1.5.4, we know the action of D on O(n; 1) is compatible with the lexicographic ordering ≺lex. Hence for the admissible n-tuple δ = (p − 1, . . . , p − 1), we have that (t + D)pn−1(x δ) = D pn−1(x δ) + XA≻lex(0,...,0) λAx A for some λA ∈ k∗. By Lemma 1.5.1(iii), D pn−1(xδ) = (−1)n. Hence (t + D)pn−1(xδ) is invertible in O(n; 1). But t + D ∈ N0 by our assumption. Hence (t + D)pn−1 ∈ W (n; 1)(0). Note that if D1 ∈ W (n; 1)(0), then D1(O(n; 1)) ⊆ m. In particular, (t + D)pn−1(xδ) ∈ m, which is not invertible. This is a contradiction. Hence V ∩ N0 = {0}. Applying Theorem 1.6.8, we get the desired result. This completes the sketch of proof. We are now ready to prove that Theorem 1.5.2. [20, Lemma 6] The variety N (W (n; 1)) is irreducible. Sketch of proof. By Theorem 1.5.1, we know that N (W (n; 1)) is equidimensional of dimension dim W (n; 1) − n. By Lemma 1.5.1, we know that O = G.D is an irreducible component of N (W (n; 1)). Let Z1, . . . , Zt be pairwise distinct irreducible components of N (W (n; 1)), and set Z1 = O. Suppose t ≥ 2. Observe that Z2 ∩ O = ∅. Hence Z2 ⊆ N (W (n; 1)) \ O = N0; see Lemma 1.5.3. Then dim W (n; 1) − n = dim Z2 ≤ dim N0 < dim W (n; 1) − n by Lemma 1.5.5. This is a contradiction. Hence t = 1 and N (W (n; 1)) is irreducible. This completes the sketch of proof. By a similar argument, the nilpotent variety of S(n; 1) (respectively S(n; 1)(1)) is proved to be irreducible. By Remark 1.4.1, we see that the Poisson Lie algebra 37 1.6. Some useful theorems (O(2n; 1), {, }) is closely related to H(2n; 1)(2). In fact, (O(2n; 1)/k)(1) ∼= H(2n; 1)(2). So the proof for H(2n; 1)(2) relies heavily on Skryabin's work for (O(2n; 1), {, }); see [29, Lemma 1.5 and Theorem 6.4]. 1.6 Some useful theorems We present three Block's theorems which will be used in Chapter 3. The first one is useful for part (b) of the sketch proof of Theorem 3.2.2 which characterizes all regular elements of the mth Jacobson Witt algebra W (m; 1). Let us begin with some definitions. Definition 1.6.1. [2, p. 433] Let B be a ring (not necessarily associative or has a unit element). A derivation of B is an additive mapping d : B → B such that d(ab) = d(a)b + ad(b) for all a, b ∈ B. Let D be a set of derivations of B. By a D-ideal of B we mean an ideal of B which is invariant under D. The ring B is called D-simple ( d-simple if D consists of a single derivation d) if B2 6= 0 and if B has no proper D-ideals. Also B is called differentiably simple if it is D-simple for some set of derivations D of B, and hence for the set of all derivations of B. Note that the above definitions are also used for algebras over a ring C and the derivations are assumed to be C-linear. Suppose B is a differentiably simple commu- tative associative ring. At characteristic 0, B is an integral domain. In particular, if B has a minimal ideal then B is a field; see [2, Sec. 4, p. 441]. Now suppose B has prime characteristic. The following theorem determines B: Theorem 1.6.1. [2, Theorem 4.1] Let B be a differentiably simple commutative as- sociative ring of prime characteristic p, and let R = {x ∈ B xp = 0}. If Rx = 0 for some x 6= 0 in B (this will hold, e.g. if B has a minimal ideal), then there is a subfield E of B and an r ≥ 0 such that B ∼= O(r; 1) as E-algebras. Here E may be taken to be any maximal subfield of B containing the subfield F of differential constants (i.e. elements of B which are annihilated by all derivations). The second theorem describes the derivation algebra of a Lie algebra in the follow- ing form: 38 1.6. Some useful theorems Theorem 1.6.2. [30, Corollary 3.3.4] Let S be a finite dimensional simple algebra such that S2 6= (0). Then Der(cid:0)S ⊗ O(m; n)(cid:1) = (cid:0)(Der S) ⊗ O(m; n)(cid:1) ⋊(cid:0) IdS ⊗ Der O(m; n)(cid:1) ∼= (cid:0)(Der S) ⊗ O(m′; 1)(cid:1) ⋊(cid:0) IdS ⊗W (m′; 1)(cid:1) for some m′ ∈ N. The last line follows from the fact that considered just as an algebra, O(m; n) is isomorphic to the truncated polynomial ring O(m′; 1) in m′ = n1 + · · · + nm variables; see [30, p. 64]. Hence Der O(m; n) ∼= Der O(m′; 1) = W (m′; 1). The third theorem describes the structure of any finite dimensional semisimple Lie algebras. Recall the socle of a finite dimensional semisimple Lie algebra L, denoted Soc(L), is the direct sum ⊕jIj of minimal ideals Ij of L. In particular, these ideals Ij are irreducible L-modules. Theorem 1.6.3. [2, Theorem 9.3; 30, Corollary 3.3.6] Let L be a finite dimensional semisimple Lie algebra. Then there are simple Lie algebras Si and truncated polynomial rings O(mi; 1) such that Soc(L) = Lt Moreover, i=1 Si ⊗ O(mi; 1) and L acts faithfully on it. t Mi=1 t Si ⊗ O(mi; 1) ⊂ L ⊂ Mi=1 (cid:0)(Der Si) ⊗ O(mi; 1)(cid:1) ⋊(cid:0) IdSi ⊗W (mi; 1)(cid:1). We will see in Chapter 2 that Premet's conjecture can be reduced to the case where the Lie algebra is semisimple. The above theorem leads us to the study of nilpotent varieties for a particular class of semisimple restricted Lie algebras, namely the ones that are sandwiched between t Mi=1 Si ⊗ O(mi; 1) and t Mi=1 (cid:0)(Der Si) ⊗ O(mi; 1)(cid:1) ⋊(cid:0) IdSi ⊗W (mi; 1)(cid:1). We finish this section with some useful theorems from algebraic geometry. Definition 1.6.2. [9, p. 91, 3.7] A morphism ψ : V → W of affine varieties is called dominant if the image ψ(V ) is dense in W , i.e. ψ(V ) = W . 39 1.6. Some useful theorems Given any morphism of irreducible affine varieties, proving directly its dominance may be difficult. However, we can show its differential map is surjective. More pre- cisely, Theorem 1.6.4 (Differential Criterion for Dominance [8, Proposition 1.4.15]). Let ψ : V → W be a morphism of irreducible affine varieties. Let v be a smooth point in V such that ψ(v) is a smooth point in W . If the differential of ψ at v, dvψ : TvV → Tψ(v)W , is surjective, then the morphism ψ is dominant. Once we know that a morphism is dominant, we can get a nonempty open set from the image of the morphism. Specifically, Theorem 1.6.5. [8, Corollary 2.2.8] Let ψ : V → W be a dominant morphism of irreducible affine varieties. Then the image of any nonempty open subset U ⊆ V contains a nonempty open subset of W . Finally, we present some theorems on dimensions. The first two relate to the dimension of fibres. Theorem 1.6.6. [6, Sec. 4.4, Chap. 2, II; 28, Theorem 1.25] Let ψ : V → W be a dominant morphism of irreducible varieties. Suppose that dim V = m and dim W = n. Then m ≥ n, and (i) dim F ≥ m − n for any w ∈ W and for any component F of the fibre ψ−1(w); (ii) there exists a nonempty open subset U ⊂ W such that dim ψ−1(u) = m − n for all u ∈ U. Theorem 1.6.7 (Chevalley's Semi-continuity Theorem [6, Sec. 4.5, Chap. 2, II]). Let ψ : V → W be a morphism of affine varieties. Then for every r ∈ N0, the set Vr = (cid:8)v ∈ V dim ψ−1(ψ(v)) ≥ r(cid:9) is Zariski closed in V . The last theorem in this section relates to the dimension of intersections in An. Theorem 1.6.8 (Affine Dimension Theorem [9, Proposition 7.1, Chap. I]). Let V, W be varieties of dimensions r, s in An. Then every irreducible component U of V ∩ W has dimension ≥ r + s − n. 40 1.7. Overview of results 1.7 Overview of results Chapter 2. Let k be an algebraically closed field of characteristic p > 0. We start with Premet's conjecture which states that the nilpotent variety of any finite dimensional restricted Lie algebra over k is irreducible. We prove that this conjecture can be reduced to the semisimple case. Theorem 1 (see Theorem 2.0.1). Let (g, [p]) be a finite dimensional restricted Lie algebra over k. Let Rad g denote the radical of g. Then N (g) is irreducible if and only if N (g/ Rad g) is irreducible. The proof is done by induction on dim g and it relies on a result from algebraic geometry (see Lemma 2.0.1). Since semisimple Lie algebras are not always direct sums of simple ideals in prime characteristic, the reduction of Premet's conjecture to the simple case is very non-trivial. Chapter 3. We start to look at a particular class of semisimple restricted Lie algebras and verify Premet's conjecture in that case. By Theorem 1.6.3, we know that any finite dimensional semisimple Lie algebra is sandwiched between and t t Mi=1 Si ⊗ O(mi; 1) Mi=1 (cid:0)(Der Si) ⊗ O(mi; 1)(cid:1) ⋊(cid:0) IdSi ⊗W (mi; 1)(cid:1) for some simple Lie algebras Si and truncated polynomial rings O(mi; 1) with Der O(mi; 1) = W (mi; 1). Thus, to verify Premet's conjecture, we begin with the simplest example, g = (sl2 ⊗ O(1; 1)) ⋊ (Idsl2 ⊗k ∂), where sl2 is the special linear Lie algebra, O(1; 1) is the truncated polynomial ring k[X]/(X p), and ∂ = d dx which acts on sl2 ⊗ O(1; 1) in the natural way. We assume further that char k = p > 2. So sl2 is a simple restricted Lie algebra over k with all its derivations inner. We prove that the maximal dimension of tori in g is 1, and the nilpotency index of any element in N (g) is at most p2. It follows from Theorem 1.5.1 that dim N (g) = 3p. After gathering these pieces of information, we are ready to prove that Theorem 2 (see Theorem 3.1.1). The variety N (g) is irreducible. 41 1.7. Overview of results We will see that the argument in the proof is quite general. It works if we replace sl2 by any Lie algebra g2 = Lie(G2), where G2 is a reductive algebraic group. Then we extend the example g to the semisimple restricted Lie algebra L := (S ⊗ O(m; 1)) ⋊ (IdS ⊗D), where S is a simple restricted Lie algebra over k such that ad S = Der S and N (S) is irreducible, O(m; 1) = k[X1, . . . , Xm]/(X p m) is the truncated polynomial 1 , . . . , X p ring in m ≥ 2 variables, and D is a restricted transitive subalgebra of W (m; 1) = Der O(m; 1) such that N (D) is irreducible. We split our study on N (L) into three sections. In the first section, we study nilpotent elements of D. Then we study nilpotent elements of L and carry out some calculations using Lemma 1.1.1. We finally prove that Theorem 3 (see Theorem 3.2.3). The variety N (L) is irreducible. As a remark, we see from the above that Premet's conjecture holds for t Mi=1 (Si ⊗ O(mi; 1)) ⋊ (IdSi ⊗Di), where each Si is a simple restricted Lie algebra over k such that ad Si = Der Si and N (Si) is irreducible, O(mi; 1) are truncated polynomial rings, and each Di is a restricted transitive subalgebra of W (mi; 1) = Der O(mi; 1) such that N (Di) is irreducible. Chapter 4. This is our final chapter. It corresponds to a paper [3] of the author which was published in the Journal of Algebra and Its Applications. We assume that char k = p > 3 and n ∈ N≥2. Then the Zassenhaus algebra, denoted L = W (1; n), provides the first example of a simple, non-restricted Lie algebra. We can embed L into its minimal p-envelope Lp = W (1; n)p. This restricted Lie algebra is semisimple. By [37, Theorem 4.8(i)], the variety N (L) := N (Lp) ∩ L is reducible. So investigating the variety N (Lp) becomes critical. Let N denote the nilpotent variety of Lp. We split our study on N into three sections. In the first section, we focus on nilpotent elements of Lp and carry out some calculations using Jacobson's formula. This work enables us to identify an irreducible component Nreg of N . Moreover, we can explicitly describe it as: 42 1.7. Overview of results Proposition 1 (see Proposition 4.2.1 and Lemma 4.2.7). Nreg = G.(∂ +k ∂p + · · · + k ∂pn−1 ), where G = Aut(Lp). In the final section, we show that the complement of Nreg in N , denoted Nsing, has dim Nsing < dim N . The proof is similar to Premet's proof for the Jacobson-Witt algebra W (n; 1); see Lemma 1.5.5. But we have to construct a new subspace V of Lp such that dim V = n + 1 and V ∩ Nsing = {0}; see Proposition 4.2.2. Combining all these results, we are able to prove the last theorem in the thesis: Theorem 4 (see Theorem 4.2.1). The variety N = N (Lp) coincides with the Zariski closure of and hence is irreducible. Nreg = G.(∂ +k ∂p + · · · + k ∂pn−1 ) 43 Chapter 2 Reduction of Premet's conjecture to the semisimple case Let (g, [p]) be a finite dimensional restricted Lie algebra over k. Recall Premet's conjecture which states that the variety N (g) = {x ∈ g x[p]N = 0 for N ≫ 0} is irreducible. In this short chapter we show that this conjecture can be reduced to the case where g is semisimple. For that, we need the following result from algebraic geometry. It gives a criterion for a variety to be irreducible. Lemma 2.0.1. Let ψ : X → Y be a surjective morphism of algebraic varieties such that (i) Y is irreducible, (ii) all fibres of ψ are irreducible and have the same dimension d, and (iii) X is equidimensional. Then X is irreducible. Proof. Let X = X1 ∪ · · · ∪ Xt be the decomposition of X into pairwise distinct irreducible components Xi. Suppose t ≥ 2. Then for any y ∈ Y , ψ−1(y) = t [i=1(cid:0)ψ−1(y) ∩ Xi(cid:1). Since ψ−1(y) is irreducible, we have that ψ−1(y) = ψ−1(y) ∩ Xi for some i. 44 For every i, define Oi := Xi \Sj6=i(Xi ∩ Xj). Then Oi is a nonempty open subset of Xi. If y ∈ ψ(Oi), then the fibre ψ−1(y) is not contained in ψ−1(y) ∩ Xj for every j 6= i. Hence for any y ∈ ψ(Oi), ψ−1(y) = ψ−1(y) ∩ Xi. (2.1) Next we show that there is a unique irreducible component Xj of X such that Y = ψ(Xj). Since ψ is surjective, we have that Y = St so that Y = ψ(Xj) for some j. By definition of Oj, we also have that Y = ψ(Oj), i=1 ψ(Xi). But Y is irreducible, i.e. ψ(Oj) is dense in Y . Thus, ψ(Oj) contains a nonempty open subset Z of Y . By (2.1), we have that for any y ∈ Z, ψ−1(y) = ψ−1(y) ∩ Xj. This means that if i 6= j, then Xi \ (Xi ∩ Xj) ⊆ ψ−1(Y \ Z). As a result, Xi = Xi \ (Xi ∩ Xj) is contained in ψ−1(Y \ Z). This implies that ψ(Xi) 6= Y for every i 6= j. So there is a unique irreducible component Xj of X such that Y = ψ(Xj). We now show that for every i, dim Xi = dim ψ(Xi) + d. Consider the restriction of ψ to Xi, ψ : Xi → ψ(Xi). By Theorem 1.6.6, there exists a nonempty open subset U ⊂ ψ(Xi) such that for every y ∈ U, dim Xi = dim ψ(Xi) + dim(ψ−1(y) ∩ Xi). Since U ∩ ψ(Oi) 6= ∅, then (2.1) implies that for every i, dim Xi = dim ψ(Xi) + d. (2.2) By the previous argument, we know that there is a unique Xj that dominates Y . So if i 6= j, then ψ(Xi) is a proper subset of Y . This and (2.2) imply that dim Xi = dim ψ(Xi) + d < dim Y + d = dim ψ(Xj) + d = dim Xj. But this contradicts the fact that X is equidimensional. Hence t = 1 and X is irreducible. This completes the proof. Theorem 2.0.1. Let (g, [p]) be a finite dimensional restricted Lie algebra over k. We denote by Rad g the radical of g. Then Conjecture 1 can be reduced to the case where g is semisimple. Explicitly, N (g) is irreducible if and only if N (g/ Rad g) is irreducible. Proof. We proceed by induction on dim g. If dim g = 0 or 1, the result is trivially true. Suppose dim g > 1 and "N (g) is irreducible if and only if N (g/ Rad g) is irreducible" 45 holds for all g of dimension less than m. Let dim g = m. If g is semisimple, then there is nothing to prove. If g is not semisimple, i.e. Rad g 6= 0, then there exists a nonzero abelian ideal. Let R be such a p-ideal. Case 1 . If R contains a non-nilpotent element, say y, then it follows fromTheo- rem 1.1.2 that there exist a unique semisimple element ys ∈ g and a unique nilpotent element yn ∈ g such that y = ys + yn and [ys, yn] = 0. Since y is non-nilpotent, then ys 6= 0. Replace y by its semisimple part ys. We show that y[p]N lies in the centre z(g) for N ≫ 0. Indeed, since R is a p-ideal we have that [y[p]N , g] ⊆ R. Moreover, R is abelian, then [y[p]N , [y[p]N , g]] = 0. Since y is semisimple, so is y[p]N by Lemma 1.1.4(iii). This implies that [y[p]N , g] = 0. Hence y[p]N ∈ z(g) and z(g) 6= 0. Let z(g) = zsL zn, where zs is a subalgebra with a one-to-one [p]-th power map and zn is the subspace in z(g) consisting of nilpotent elements of z(g). By above, we already know that y[p]N ∈ zs and so zs 6= 0. We prove that the canonical homomorphism ψ : g → g/zs, x 7→ x + zs induces a bijective morphism ψ : N (g) → N (g/zs). If x + zs ∈ N (g/zs), then x[p]N ∈ zs for N ≫ 0. Since the [p]-th power map on zs is one-to-one, there exists z ∈ zs such that x[p]N = z[p]N . Since z is an element of zs, it commutes with x. As a result, (x − z)[p]N = 0. This implies that x − z ∈ N (g). So the morphism ψ is surjective. For injectivity, suppose ψ(x) = ψ(y) for some x, y ∈ N (g). Then y = x + z1 for some z1 ∈ zs. Since x and y are nilpotent, then for N ≫ 0 we get 0 = y[p]N = (x + z1)[p]N = x[p]N + z[p]N 1 = z[p]N 1 . Since the [p]-th power map on zs is one-to-one, we have that z1 = 0 and so y = x. Therefore, the morphism ψ is bijective. Next we claim that N (g) is irreducible if and only if N (g/zs) is irreducible. The "if" part is trivial because surjective morphism preserves irreducibility. For the "only if" part, suppose N (g/zs) is irreducible. By Theorem 1.5.1(iii), we know that N (g) is equidimensional. Since ψ is bijective, it follows that all fibres of ψ are single points. Applying Lemma 2.0.1 with d = 0 we get N (g) is irreducible. Case 2 . If R is contained in N (g), then we may assume that R[p]r = 0 for some r ∈ N. Note that the canonical homomorphism ψ : g → g/R, x 7→ x = x + R induces a surjective mapping ψ : N (g) → N (g/R). We show that N (g) is irreducible if and only if N (g/R) is irreducible. The "if" part is trivial. For the "only if" part, suppose N (g/R) is irreducible. We show that N (g) = ψ−1(cid:0)N (g/R)(cid:1), i.e. x is nilpotent in g if 46 and only if x + R is nilpotent in g/R. By Jacobson's formula and that R is a p-ideal of g, we have that for any v ∈ R, (x + v)[p] = x[p] + v1 for some v1 ∈ R. By Jacobson's formula again, we have that (x + v)[p]N = x[p]N + vN for some vN ∈ R. If x ∈ N (g), then we can choose N sufficiently large so that (x + v)[p]N = 0. Hence x + v ∈ N (g/R). Conversely, if x + R ∈ N (g/R), then for N ≫ 0, x[p]N ∈ R. By our assumption, R[p]r = 0. Hence x[p]N +r = 0 and so x ∈ N (g). Therefore, for any x ∈ N (g/R), the fibre ψ−1(x) consists of elements of the form x + R. But x + R ∼= R as an affine space, we conclude that all fibres of ψ are irreducible and have the same dimension. It follows from Theorem 1.5.1(iii) and Lemma 2.0.1 that N (g) is irreducible. To sum up, if Rad g 6= 0, then there exists an abelian p-ideal A, i.e. A = zs or A = R with R[p]r = 0 for some r ∈ N. We proved that N (g) is irreducible if and only if N (g/A) is irreducible. Since dim(g/A) < dim g, the induction hypothesis implies that N (g/A) is irreducible if and only if N(cid:0)(g/A)/ Rad(g/A)(cid:1) ∼= N (g/ Rad g) is irreducible. Therefore, we proved that N (g) is irreducible if and only if N (g/ Rad g) is irreducible. This completes the proof. It is natural to ask if we could reduce Premet's conjecture to the simple case. Unfortunately, this is very non-trivial in prime characteristic as semisimple Lie algebras are not always direct sums of simple ideals. Hence to prove Premet's conjecture, we shall focus on the semisimple case. 47 Chapter 3 Nipotent varieties of some semisimple restricted Lie algebras In this chapter, we assume that char k = p > 2. We prove Premet's conjecture holds for a class of semisimple restricted Lie algebras whose socle involves any simple restricted Lie algebra with all its derivations inner. We start with the semisimple restricted Lie algebra whose socle involves sl2 tensored by the truncated polynomial ring k[X]/(X p). Then we extend it to the semisimple restricted Lie algebra whose socle involves S ⊗ O(m; 1), where S is any simple restricted Lie algebra such that ad S = Der S and N (S) is irreducible, and O(m; 1) = k[X1, . . . , Xm]/(X p 1 , . . . , X p m) is the truncated polynomial ring in m ≥ 2 variables. 3.1 Socle involves sl2 By Block's theorem (Theorem 1.6.3) on finite dimensional semisimple Lie algebras, the first example (perhaps the easiest one) we shall consider is (sl2⊗O(1; 1))⋊(Idsl2 ⊗k ∂)1. Let k be an algebraically closed field of characteristic p > 2. Recall the special linear Lie algebra sl2 of 2 × 2 matrices with trace 0, and its standard basis e =   0 1 0 0   , f =   0 0 1 0   , h =   1 0 0 −1   1This problem was introduced by A. Premet in the workshop Lie Theory and Representation Theory in Pisa, Italy, January-February 2015. In [23], A. Premet and D. I. Stewart studied semisimple restricted Lie subalgebras of this type (referred to as exotic semidirect products) in the Lie algebra g1 = Lie(G1), where G1 is a simple algebraic group of exceptional type over k. 48 3.1. Socle involves sl2 such that [e, f ] = h, [h, e] = 2e, [h, f ] = −2f. Since p > 2, we know that sl2 is simple and all its derivations are inner [27, Sec. 5, Chap. V]. By Example 1.1.1 and Corollary 1.1.1, we know that sl2 is a restricted Lie algebra with the unique [p]-th power map given by p-th power of matrices. An easy calculation shows that ep = f p = 0 and hp = h. Hence e, f are nilpotent and h is toral. Let O(1; 1) denote the p-dimensional truncated polynomial ring k[X]/(X p). We write x for the image of X in O(1; 1). Note that O(1; 1) is a local ring. Let m denote its unique maximal ideal. For every g ∈ O(1; 1), gp = g(0)p. Then gp = 0 for all g ∈ m. Consider the algebra sl2 ⊗ O(1; 1). The restricted Lie algebra structure on sl2 induces a restricted Lie algebra structure on sl2 ⊗ O(1; 1). Explicitly, the Lie bracket is given by [y ⊗ g1, z ⊗ g2] := [y, z] ⊗ g1g2, and the [p]-th power map is given by (y ⊗ g1)[p] := yp ⊗ gp 1 for all y ⊗ g1, z ⊗ g2 ∈ sl2 ⊗ O(1; 1). Note that dim(sl2 ⊗ O(1; 1)) = 3p. Let W (1; 1) denote the derivation algebra of O(1; 1). It is known as the Witt algebra. By Theorem 1.4.2 and Example 1.1.1, we know that W (1; 1) is a simple restricted Lie algebra with the [p]-th power map given by D[p] = Dp for all D ∈ W (1; 1). Moreover, W (1; 1) is a free O(1; 1)-module of rank 1 generated by ∂ = d dx . Note that ∂ acts on sl2 ⊗ O(1; 1) by differentiating truncated polynomials in O(1; 1), i.e. ∂(y ⊗ g) = y ⊗ ∂(g) for all y ⊗ g ∈ sl2 ⊗ O(1; 1). Let g := (sl2 ⊗ O(1; 1)) ⋊ (Idsl2 ⊗k ∂). To ease notation we identify Idsl2 ⊗k ∂ with k ∂. It is easy to check that g is a (3p + 1)-dimensional Lie algebra with the Lie bracket defined in the natural way: [y ⊗ g, ∂] := −y ⊗ ∂(g) for all y ⊗ g ∈ sl2 ⊗ O(1; 1). By Theorem 1.6.2, we see that g embeds in the restricted Lie algebra Der(sl2 ⊗O(1; 1)) = (cid:0)sl2 ⊗O(1; 1)(cid:1)⋊(cid:0) Idsl2 ⊗W (1; 1)(cid:1). Then g is restricted 49 3.1. Socle involves sl2 with the [p]-th power map given by p-th power of derivations. Note that ∂p = 0. Let us look at the structure of g in more detail. Lemma 3.1.1. (i) [g, g] = sl2 ⊗ O(1; 1) which is not semisimple. (ii) sl2 ⊗ O(1; 1) is the unique minimal ideal of g. (iii) The maximal dimension of tori in sl2 ⊗ O(1; 1) is 1, i.e. MT(sl2 ⊗ O(1; 1)) = 1. (iv) All irreducible components of N (sl2 ⊗ O(1; 1)) have dimension 3p − 1. Proof. (i) Write sl2 ⊗ O(1; 1) = (sl2 ⊗ 1) ⊕ (sl2 ⊗ m). Since sl2 is simple, it is easy to check that [sl2 ⊗ O(1; 1), sl2 ⊗ O(1; 1)] = sl2 ⊗ O(1; 1). As sl2 ⊗ O(1; 1) is a Lie subalgebra of g, we have that sl2 ⊗ O(1; 1) = [sl2 ⊗ O(1; 1), sl2 ⊗ O(1; 1)] ⊆ [g, g]. On the other hand, [y ⊗ g1 + λ1 ∂, z ⊗ g2 + λ2 ∂] = [y, z] ⊗ g1g2 + λ1z ⊗ ∂(g2) − λ2y ⊗ ∂(g1) for all y, z ∈ sl2, g1, g2 ∈ O(1; 1) and λ1, λ2 ∈ k. Clearly, this is an element of sl2 ⊗ O(1; 1). Hence [g, g] ⊆ sl2 ⊗ O(1; 1). As a result, [g, g] = sl2 ⊗ O(1; 1). Now suppose sl2 ⊗ O(1; 1) is semisimple. This means the only abelian ideal in sl2 ⊗ O(1; 1) is the zero ideal. Since sl2 is simple, any ideal I of sl2 ⊗ O(1; 1) has the form sl2 ⊗J for some ideal J of O(1; 1). Take J = (xp−1), the principal ideal generated by xp−1 in O(1; 1). Then we find a nonzero ideal I = sl2 ⊗ (xp−1) in sl2 ⊗ O(1; 1) such that [I, I] = 0, a contradiction. Hence sl2 ⊗ O(1; 1) is not semisimple. This proves (i). (ii) Observe that if w = y ⊗ g1 + λ1 ∂ is an element of g such that [w, z ⊗ g2] = 0 for all z ⊗ g2 in sl2 ⊗ O(1; 1), then we must have that either y = 0 or g1 = 0, and λ1 = 0. As a result, w = 0. This shows that g acts faithfully on sl2 ⊗ O(1; 1) via the adjoint representation, and embeds sl2 ⊗ O(1; 1) into its derivation algebra Der(sl2 ⊗ O(1; 1)). Hence for any nonzero ideal J of g we must have that [J, sl2 ⊗ O(1; 1)] 6= 0. In particular, for any minimal ideal I of g, [I, sl2 ⊗ O(1; 1)] 6= 0. By (i) of this lemma, sl2 ⊗ O(1; 1) is an ideal of g, so is [I, sl2 ⊗ O(1; 1)]. Since [I, sl2 ⊗ O(1; 1)] ⊆ I, the minimality of I implies that [I, sl2 ⊗ O(1; 1)] = I. As I = [I, sl2 ⊗ O(1; 1)] ⊆ I ∩ (sl2 ⊗ O(1; 1)), we have that I ⊆ sl2 ⊗ O(1; 1). Let v be any nonzero element of I. Since I ⊆ sl2 ⊗ O(1; 1), we can write v = y ⊗ (a0 + a1x + · · · + ajxj + · · · + ap−1xp−1) for some 0 6= y ∈ sl2 and aj ∈ k∗. Then (ad ∂)j(v) = y ⊗ (j!aj + xg3) 50 3.1. Socle involves sl2 for some g3 ∈ O(1; 1). This shows that I ∩ (sl2 ⊗ 1) 6= {0}. By the simplicity of sl2, we have that sl2 ⊗ 1 ⊂ I. Now consider the adjoint endomorphisms from sl2 ⊗ x to sl2 ⊗ 1, one can deduce that sl2 ⊗ x ⊂ I. Continuing in this way and by induction, we get sl2 ⊗ xl ⊂ I for all 0 ≤ l ≤ p − 1. Hence I = sl2 ⊗ O(1; 1). This proves (ii). (iii) Since any element y⊗g in sl2⊗m satisfies (y⊗g)p = 0, it follows that sl2⊗m is a p-ideal of sl2 ⊗ O(1; 1) consisting of nilpotent elements. As a results, MT(sl2 ⊗ m) = 0. By Lemma 1.1.6, we have that MT(sl2 ⊗ O(1; 1)) = MT(sl2 ⊗ 1). It is known that the maximal dimension of tori in sl2 is 1. Hence MT(sl2 ⊗ O(1; 1)) = 1. This proves (iii). (iv) By Theorem 1.5.1 and (iii) of this lemma, we know that all irreducible compo- nents of N (sl2⊗O(1; 1)) have dimension dim(sl2⊗O(1; 1))−MT(sl2⊗O(1; 1)) = 3p−1. This proves (iv). Before we are going to prove a similar result for g, we need to construct some automorphisms of g. Lemma 3.1.2. Let y ⊗ q be an element of sl2 ⊗ O(1; 1). For any σ ∈ Aut(sl2), define σ(y ⊗ q) := σ(y) ⊗ q. Then σ extends to an automorphism of g. Proof. We first show that σ extends to an automorphism of sl2 ⊗ O(1; 1). For any y1 ⊗ g1, y2 ⊗ g2 ∈ sl2 ⊗ O(1; 1), we have that σ([y1 ⊗ g1, y2 ⊗ g2]) = σ([y1, y2] ⊗ g1g2) = [σ(y1), σ(y2)] ⊗ g1g2. On the other hand, [σ(y1 ⊗ g1), σ(y2 ⊗ g2)] = [σ(y1) ⊗ g1, σ(y2) ⊗ g2] = [σ(y1), σ(y2)] ⊗ g1g2. Hence σ is an automorphism of sl2 ⊗ O(1; 1). Then it induces an automorphism of the derivation algebra Der(sl2 ⊗ O(1; 1)) = (cid:0)sl2 ⊗ O(1; 1)(cid:1) ⋊(cid:0) Idsl2 ⊗W (1; 1)(cid:1) via conjugation. Note that g ⊂ Der(sl2 ⊗ O(1; 1)). It remains to check σ ◦ ∂ ◦σ−1. For any y1 ⊗ g1 ∈ sl2 ⊗ O(1; 1), we have that σ ◦ ∂ ◦σ−1(y1 ⊗ g1) = σ ◦ ∂(cid:0)σ−1(y1) ⊗ g1(cid:1) = σ(cid:0)σ−1(y1) ⊗ ∂(g1)(cid:1) = σσ−1(y1) ⊗ ∂(g1) = y1 ⊗ ∂(g1). 51 3.1. Socle involves sl2 Thus, σ ◦ ∂ ◦σ−1 = ∂. As a result, σ preserves g and it extends to an automorphism of g. This completes the proof. Lemma 3.1.3. Let u = y ⊗g be an element of sl2 ⊗ m. Then exp(ad u) is an automor- phism of g. Similarly, if v = z ⊗ q is an element of N (sl2) ⊗ O(1; 1), then exp(ad v) is an automorphism of g. Proof. For any u = y ⊗ g ∈ sl2 ⊗ m, we first show that exp(ad u) is an automorphism of sl2 ⊗O(1; 1). Suppose that L is a Lie algebra over k and D ∈ Der L satisfies Dp = 0. i! Di is an automorphism of L it suffices to show i=0 1 that In order to show that exp(D) = Pp−1 X0≤i,j≤p, i+j≥p 1 i!j! [Di(w1), Dj(w2)] = 0 for all w1, w2 ∈ L. Note that ad u with u above is a derivation of sl2 ⊗O(1; 1) such that (ad u)p = ad up = ad(yp ⊗ gp) = 0; see Definition 1.1.1(1). Set D = ad u = ad(y ⊗ g). Then for any w1 = y1 ⊗ g1 and w2 = y2 ⊗ g2 in sl2 ⊗ O(1; 1), we have that Di(w1) = (ad y)i(y1) ⊗ gig1, and Dj(w2) = (ad y)j(y2) ⊗ gjg2. Hence X0≤i,j≤p, i+j≥p = X0≤i,j≤p, i+j≥p 1 i!j! 1 i!j! [Di(w1), Dj(w2)] [(ad y)i(y1), (ad y)j(y2)] ⊗ gi+jg1g2. Since g ∈ m and i + j ≥ p, we have that gi+j = 0. As a result, X0≤i,j≤p, i+j≥p 1 i!j! [Di(w1), Dj(w2)] = 0. Therefore, exp(ad u) is an automorphism of sl2 ⊗ O(1; 1). Then it induces an auto- morphism of the derivation algebra Der(sl2 ⊗ O(1; 1)) = (cid:0)sl2 ⊗ O(1; 1)(cid:1) ⋊(cid:0) Idsl2 ⊗W (1; 1)(cid:1) via conjugation. Note that g = (sl2 ⊗ O(1; 1)) ⋊ k ∂ ⊂ Der(sl2 ⊗ O(1; 1)). 52 3.1. Socle involves sl2 To conclude that exp(ad u) is an automorphism of g, we need to check that exp(ad u) preserves g, i.e. exp(ad u) ◦ ∂ ◦ exp(ad(−u)) ∈ g. Recall that u = y ⊗ g ∈ sl2 ⊗ m. We show that for any y1 ⊗ g1 ∈ sl2 ⊗ O(1; 1), (cid:18) exp(ad(y ⊗ g)) ◦ ∂ ◦ exp(ad(−y ⊗ g))(cid:19)(y1 ⊗ g1) =y1 ⊗ ∂(g1) −(cid:0) ad(cid:0)y ⊗ ∂(g)(cid:1)(cid:1)(y1 ⊗ g1) −(cid:0) ad(cid:0)yp ⊗ gp−1 ∂(g)(cid:1)(cid:1)(y1 ⊗ g1). (3.1) We first compute ∂ ◦ exp(ad(−y ⊗ g))(y1 ⊗ g1). Then we apply exp(ad(y ⊗ g)) to it. We will use the following in our computations: (i) Since ∂ is a derivation of O(1; 1), we have that ∂(gig1) = gi ∂(g1) + igi−1 ∂(g)g1 for all 0 ≤ i ≤ p − 1. (ii) exp(ad(y ⊗ g)) exp(ad(−y ⊗ g)) = Id. (iii) Since g ∈ m, we have that gp = 0. (iv) (p − 1)! ≡ −1(mod p). Let us compute ∂ ◦ exp(ad(−y ⊗ g))(y1 ⊗ g1). ∂ ◦ exp(ad(−y ⊗ g))(y1 ⊗ g1) 1 i! p−1 (ad(−y))i(y1) ⊗ gig1(cid:19) = ∂(cid:18) p−1 Xi=0 (ad(−y))i(y1) ⊗(cid:18)gi ∂(g1) + igi−1 ∂(g)g1(cid:19) (by (i)) 1 i! = Xi=0 = exp(ad(−y ⊗ g))(y1 ⊗ ∂(g1)) +(cid:18) p−1 Xi=1 = exp(ad(−y ⊗ g))(y1 ⊗ ∂(g1)) +(cid:18) p−1 Xi=1 1 i! (ad(−y))i(y1) ⊗ igi−1 ∂(g)g1(cid:19) 1 (i − 1)! (ad(−y))i(y1) ⊗ gi−1 ∂(g)g1(cid:19) = exp(ad(−y ⊗ g))(y1 ⊗ ∂(g1)) +(cid:18) ad(−y ⊗ ∂(g)) p−1 Xi=1 1 (i − 1)! (ad(−y ⊗ g))i−1(cid:19)(y1 ⊗ g1) = exp(ad(−y ⊗ g))(y1 ⊗ ∂(g1)) +(cid:18) ad(−y ⊗ ∂(g))(cid:18) exp(ad(−y ⊗ g)) − 1 (p − 1)! (ad(−y ⊗ g))p−1(cid:19)(cid:19)(y1 ⊗ g1). 53 3.1. Socle involves sl2 Applying exp(ad(y ⊗ g)) to the above, we get (cid:18) exp(ad(y ⊗ g)) ◦ ∂ ◦ exp(ad(−y ⊗ g))(cid:19)(y1 ⊗ g1) =y1 ⊗ ∂(g1) +(cid:18) ad(−y ⊗ ∂(g))(cid:18) Id −(cid:0) 1 (p − 1)! (ad(−y ⊗ g))p−1(cid:1) exp(ad(y ⊗ g))(cid:19)(cid:19)(y1 ⊗ g1) (by (ii)) (by (iii) and (iv)) =y1 ⊗ ∂(g1) +(cid:18) ad(−y ⊗ ∂(g))(cid:18) Id +(ad(−y ⊗ g))p−1(cid:19)(cid:19)(y1 ⊗ g1) =y1 ⊗ ∂(g1) +(cid:18) ad(−y ⊗ ∂(g)) + (ad(−y ⊗ ∂(g))(ad(−y ⊗ g))p−1(cid:19)(y1 ⊗ g1) =y1 ⊗ ∂(g1) +(cid:0) ad(−y ⊗ ∂(g))(cid:1)(y1 ⊗ g1) + (ad(−y ⊗ ∂(g))(ad(−y ⊗ g))p−1(y1 ⊗ g1) =y1 ⊗ ∂(g1) −(cid:0) ad(y ⊗ ∂(g))(cid:1)(y1 ⊗ g1) + (−1)p(ad y)p(y1) ⊗ gp−1 ∂(g)(g1) =y1 ⊗ ∂(g1) −(cid:0) ad(cid:0)y ⊗ ∂(g)(cid:1)(cid:1)(y1 ⊗ g1) −(cid:0) ad(cid:0)yp ⊗ gp−1 ∂(g)(cid:1)(cid:1)(y1 ⊗ g1). The last line follows from Definition 1.1.1(1) that (ad y)p = ad yp. Hence we get (3.1) as desired. It follows that exp(ad(y ⊗ g)) ◦ ∂ ◦ exp(ad(−y ⊗ g)) = ∂ − ad(cid:0)y ⊗ ∂(g)(cid:1) − ad(cid:0)yp ⊗ gp−1 ∂(g)(cid:1) = ∂ −y ⊗ ∂(g) − yp ⊗ gp−1 ∂(g). (3.2) The last line follows from that sl2 ∼= ad(sl2) via the adjoint representation and hence we may identify y ⊗ ∂(g) (respectively yp ⊗ gp−1 ∂(g)) with its image ad(cid:0)y ⊗ ∂(g)(cid:1) (respectively ad(cid:0)yp ⊗ gp−1 ∂(g)(cid:1)) in gl(sl2 ⊗ O(1; 1)) under ad. As a result, we see that exp(ad(y ⊗ g)) ◦ ∂ ◦ exp(ad(−y ⊗ g)) ∈ g. Therefore, exp(ad(y ⊗ g)) preserves g and it is an automorphism of g. Now let v = z ⊗ q be an element of N (sl2) ⊗ O(1; 1). Note that sl2 contains a self-centralizing maximal torus, namely kh. Hence e(sl2) = 0. Also, MT(sl2) = 1. It follows from Theorem 1.5.1 that any nilpotent element z of sl2 satisfies zp = 0. As a result, ad v is a derivation of sl2 ⊗ O(1; 1) such that (ad v)p = ad vp = ad(zp ⊗ qp) = 0. Since (ad z)p = 0, one can show similarly that exp(ad v) is an automorphism of sl2 ⊗ O(1; 1). Then it induces an automorphism of the derivation algebra Der(sl2 ⊗ O(1; 1)) via conjugation. By (3.2), we have that exp(ad(z ⊗ q)) ◦ ∂ ◦ exp(ad(−z ⊗ q)) = ∂ −z ⊗ ∂(q) ∈ g. 54 Hence exp(ad(z ⊗ q)) preserves g and it is an automorphism of g. This completes the 3.1. Socle involves sl2 proof. Lemma 3.1.4. (i) g is semisimple. (ii) The maximal dimension of tori in g equals MT(sl2 ⊗ O(1; 1)) which is 1. (iii) Let gss denote the set of all semisimple elements of g. Then 1 = e(g) := min(cid:8)r ∈ Z≥0 W pr ⊆ gss for some nonempty Zariski open W ⊂ g(cid:9). (iv) All irreducible components of N (g) have dimension 3p. (v) Any a ∈ N (g) satisfies ap2 = 0. Proof. (i) By Lemma 3.1.1(ii), if g is not semisimple, then any nonzero abelian p-ideal would have to contain the unique minimal ideal sl2 ⊗ O(1; 1). But sl2 ⊗ O(1; 1) is nonabelian, we get a contradiction. This proves (i). (ii) Since sl2 ⊗ O(1; 1) is a p-ideal of g and ∂p = 0, it follows from Lemma 1.1.6 and Lemma 3.1.1(iii) that MT(g) = MT(sl2 ⊗ O(1; 1)) + MT(k ∂) = 1 + 0 = 1. This proves (ii). (iii) By Lemma 3.1.3, we know that exp(t ad(z ⊗ q)) is an automorphism of g for any t ∈ k and z ⊗ q ∈ N (sl2) ⊗ O(1; 1). Let G denote the group generated by these exp(t ad(z ⊗ q)). Then G is a connected algebraic group. Consider the set G.(h ⊗ O(1; 1)∗ ⋊ k ∂). Here O(1; 1)∗ denotes the set of all invertible elements of O(1; 1). We want to show that this set contains a nonempty Zariski open subset of g. This can be done by showing that the morphism θ : G × (h ⊗ O(1; 1) ⋊ k ∂) → g (g, v) 7→ g(v) is dominant. Consider the differential of θ at (1G, h ⊗ 1) (dθ)(1G,h⊗1) : Lie(G) ⊕ (h ⊗ O(1; 1) ⋊ k ∂) → g (X, Y ) 7→ [X, h ⊗ 1] + Y. It is easy to see that Im((dθ)(1G,h⊗1)) contains h⊗O(1; 1)⋊k ∂ and [Lie(G), h⊗1]. Since exp(t ad(e⊗q)) is in G for any t ∈ k and q ∈ O(1; 1), we have that e⊗O(1; 1) ⊂ Lie(G). 55 3.1. Socle involves sl2 Here we identify sl2 ⊗ O(1; 1) with its image in gl(sl2 ⊗ O(1; 1)) under ad. By our assumption, p > 2. So [e ⊗ O(1; 1), h ⊗ 1] = −2e ⊗ O(1; 1) ⊂ [Lie(G), h ⊗ 1]. Similarly, one can show that f ⊗ O(1; 1) ⊂ [Lie(G), h ⊗ 1]. Hence Im((dθ)(1G,h⊗1)) contains h⊗O(1; 1)⋊k ∂ +e⊗O(1; 1)+f ⊗O(1; 1) which is g. Thus, Im((dθ)(1G,h⊗1)) = g and so (dθ)(1G,h⊗1) is surjective. It follows from Theorem 1.6.4 that the morphism θ is dominant. Note that for any g = Pp−1 i=0 aixi ∈ O(1, 1) and λ ∈ k, (h ⊗ g + λ ∂)p = h ⊗ (ap 0 − λp−1ap−1) by Jacobson's formula. It follows that there is a nonempty Zariski open subset U0 = h⊗O(1, 1)∗⋊k ∂ in h⊗O(1, 1)⋊k ∂ consisting of elements u0 such that up 0 is semisimple. By Theorem 1.6.5, G.U0 contains a nonempty Zariski open subset U of g. We have that up ∈ gss for all u ∈ U. Therefore, e(g) ≤ 1. If e(g) = 0, then it follows from the definition of e(g) that gss contains a nonempty Zariski open subset V of g. Then there is an element v ∈ V \ sl2 ⊗ O(1, 1) such that v ∈ span(cid:8)vpi i ≥ 1(cid:9). This is impossible as wpi ∈ sl2 ⊗ O(1, 1) for all w ∈ g. Hence e(g) = 1. This proves (iii). (iv) By (ii) of this lemma and Theorem 1.5.1, all irreducible components of N (g) have dimension dim g − MT(g) = 3p. This proves (iv). (v) By Theorem 1.5.1 and the fact that e(g) = 1 = MT(g), we have that ap2 = 0 for all a ∈ N (g). This proves (v). We need another result which shows that applying suitable automorphisms of g, some nilpotent elements of g can be reduced to a nice form. Lemma 3.1.5. Let a = λ ∂ +v be an element of g, where λ ∈ k∗ and v ∈ sl2 ⊗ O(1; 1). Then there exist automorphisms of g such that a is conjugate to a′ = λ ∂ +v′, where v′ ∈ sl2 ⊗ mp−1. Moreover, if a is nilpotent, then v′ = b ⊗ xp−1 for some b ∈ N (sl2) (possibly 0). Proof. Let a = λ ∂ +v be an element of g, where λ ∈ k∗ and v ∈ sl2 ⊗ O(1; 1). Suppose v = z1 ⊗ Pp−1 i=0 aixi for some z1 ∈ sl2 and ai ∈ k. By Lemma 3.1.3, we know that exp(ad(y ⊗ g)) with y ⊗ g ∈ sl2 ⊗ m is an automorphism of g. Let H 56 3.1. Socle involves sl2 denote the subgroup of Aut(g) generated by these exp(ad(y ⊗ g)). Then H is a connected algebraic group. We show that choosing suitable element y ⊗ g ∈ sl2 ⊗ m, (cid:0) exp(ad(y ⊗ g))(cid:1)(a) = λ ∂ +v1 for some v1 ∈ sl2 ⊗ m. If a0 = 0, then a is of the desired form and there is nothing to do. Suppose a0 6= 0. By (3.2), (cid:0) exp(ad(y ⊗ g))(cid:1)(a) = exp(ad(y ⊗ g)) ◦ λ ∂ ◦ exp(ad(−y ⊗ g)) +(cid:0) exp(ad(y ⊗ g))(cid:1)(v) =λ ∂ −λy ⊗ ∂(g) − λyp ⊗ gp−1 ∂(g) + z1 ⊗ p−1 Xi=0 aixi + p−1 Xj=1 1 j! (ad y)j(z1) ⊗ gj p−1 Xi=0 aixi. Choose y = z1/λ ∈ sl2 and g = a0x ∈ m. Then −λy ⊗ ∂(g) + z1 ⊗ a0 = −z1 ⊗ a0 + z1 ⊗ a0 = 0, −λyp ⊗ gp−1 ∂(g) = −λyp ⊗ ap 0xp−1, p−1 Xj=1 1 j! (ad y)j(z1) ⊗ gj p−1 Xi=0 aixi = 0. and Hence Since −λyp ⊗ ap (cid:0) exp(ad(y ⊗ g))(cid:1)(a) = λ ∂ −λyp ⊗ ap 0xp−1 + z1 ⊗ Pp−1 0xp−1 + z1 ⊗ p−1 Xi=1 aixi. λ ∂ +v1 for some v1 ∈ sl2 ⊗ m. Continuing in this way and by induction we get a is i=1 aixi ∈ sl2 ⊗ m, we see that (cid:0) exp(ad(y ⊗ g))(cid:1)(a) = conjugate under H to λ ∂ +vp−2 for some vp−2 ∈ sl2 ⊗ mp−2. Suppose vp−2 = zp−2 ⊗ (ap−2xp−2 + ap−1xp−1) for some zp−2 ∈ sl2 and ap−2, ap−1 ∈ k. If ap−2 = 0, then vp−2 ∈ sl2 ⊗ mp−1 and there is nothing to do. Suppose ap−2 6= 0. Then for any exp(ad(y ⊗ g)) ∈ H, we have that (cid:0) exp(ad(y ⊗ g))(cid:1)(λ ∂ +vp−2) = exp(ad(y ⊗ g)) ◦ λ ∂ ◦ exp(ad(−y ⊗ g)) +(cid:0) exp(ad(y ⊗ g))(cid:1)(vp−2) =λ ∂ −λy ⊗ ∂(g) − λyp ⊗ gp−1 ∂(g) + zp−2 ⊗ (ap−2xp−2 + ap−1xp−1) + p−1 Xj=1 1 j! 57 (ad y)j(zp−2) ⊗ gj(ap−2xp−2 + ap−1xp−1). 3.1. Socle involves sl2 Choose y = zp−2/λ ∈ sl2 and g = ap−2xp−1/(p − 1) ∈ m. Then −λy ⊗ ∂(g) + zp−2 ⊗ ap−2xp−2 = −zp−2 ⊗ ap−2xp−2 + zp−2 ⊗ ap−2xp−2 = 0. Moreover, −λyp ⊗ gp−1 ∂(g) = 0, and p−1 Xj=1 1 j! (ad y)j(zp−2) ⊗ gj(ap−2xp−2 + ap−1xp−1) = 0. Hence (cid:0) exp(ad(y ⊗ g))(cid:1)(λ ∂ +vp−2) = λ ∂ +zp−2 ⊗ ap−1xp−1. Therefore, we proved that if a = λ ∂ +v ∈ g with λ ∈ k∗ and v ∈ sl2 ⊗ O(1; 1), then a is conjugate under H to a′ = λ ∂ +v′, where v′ ∈ sl2 ⊗ mp−1. By above, we may assume that a = λ ∂ +b ⊗ xp−1 for some λ ∈ k∗ and b ∈ sl2. If a is nilpotent, then Lemma 3.1.4(v) implies that ap2 = 0. By Jacobson's formula, ap = (p − 1)!λp−1b. So ap2 = 0 implies that bp = 0. As a result, b ∈ N (sl2). This completes the proof. We are now ready to prove that Theorem 3.1.1. The variety N (g) is irreducible. Proof. Let X be any irreducible component of N (g). Put X0 := X ∩ (sl2 ⊗ O(1; 1)). Then X0 ⊂ N (sl2 ⊗ O(1; 1)). By Lemma 3.1.1(iv) and Lemma 3.1.4(iv), we have that dim N (sl2 ⊗ O(1; 1)) = 3p − 1 and dim X = 3p. It follows that X0 is a proper Zariski closed subset of X. Then the complement X \ X0 is Zariski open and nonempty in X. As a result, X equals the Zariski closure of X \ X0. Let us describe X \X0 explicitly. If a ∈ X \X0, then a = λ ∂ +v for some λ ∈ k∗ and v ∈ sl2 ⊗ O(1; 1). By Lemma 3.1.5, we know that there exists a subgroup H of Aut(g) generated by all exp(ad u) with u ∈ sl2 ⊗ m such that a is H-conjugate to λ ∂ +b⊗xp−1 for some b ∈ N (sl2) (possibly 0). Since dim H = dim Lie(H) = dim(sl2⊗m) = 3(p−1), X\X0 is not contained in the union of H.(λ ∂). As a result, there are some λ ∂ +b⊗xp−1 with b 6= 0 in X \ X0. Let G = Aut(sl2)H. Note that H is a normal subgroup of G and G ⊂ Aut(g) is connected. Moreover, it is known that all nonzero nilpotent elements of sl2 are conjugate under Aut(sl2). By Lemma 3.1.2, we know that Aut(sl2) fixes ∂. Hence any 58 3.2. Socle involves S a ∈ X \ X0 is conjugate under G to λ ∂ +µe ⊗ xp−1 for some λ ∈ k∗ and µ ∈ k. This shows that X \ X0 ⊆ G.(λ ∂ +µe ⊗ xp−1 λ ∈ k∗, µ ∈ k). It is clear that G.(λ ∂ +µe ⊗ xp−1 λ ∈ k∗, µ ∈ k) ⊆ X \ X0. Thus X \ X0 = G.(λ ∂ +µe ⊗ xp−1 λ ∈ k∗, µ ∈ k). This quasi-affine variety is the image of a morphism from the irreducible variety G × k∗ × k, hence X \ X0 is irreducible. The above result holds for any irreducible component X of N (g) and shows that X \ X0 is independent of the choice of X. Therefore, there is only one X and so N (g) is irreducible. This completes the proof. Remark 3.1.1. 1. We see that the above proof relies heavily on the structure and automorphisms of g, and Premet's theorem on N (g) (Theorem 1.5.1). Since we are familiar with the special linear Lie algebra sl2 and the truncated polynomial ring O(1; 1), it is easy to do calculations using Jacobson's formula and construct such a proof. But for other semisimple restricted Lie algebras with little known structural information, it may be more difficult to verify Premet's conjecture. 2. Another useful thing to point out is that the above proof works if we replace sl2 by any Lie algebra g2 = Lie(G2), where G2 is a reductive algebraic group. This is because (i) g2 is a restricted Lie algebra by Example 1.1.1, so is (g2 ⊗ O(1; 1)) ⋊ k ∂. (ii) One can show similarly that MT(g2 ⊗ O(1; 1)) = MT(cid:0)(g2 ⊗ O(1; 1)) ⋊ k ∂(cid:1). Hence dim N (g2 ⊗ O(1; 1)) < dim N(cid:0)(g2 ⊗ O(1; 1)) ⋊ k ∂(cid:1). (iii) We know that N (g2) is irreducible and it is the Zariski closure of a single nilpotent G2-orbit. Hence all nonzero nilpotent elements of g2 are conjugate under G2; see [13, Theorem 1, Sec. 2.8 and Sec. 6.3-6.4]. 3.2 Socle involves S We would like to extend the previous example by replacing O(1; 1) with the trun- cated polynomial ring O(m; 1) in m variables. Then we need to replace ∂. A recent result proved by A. Premet and D. I. Stewart shows that ∂ is one of the representa- tives of transitive subalgebras of the derivation algebra Der O(1; 1) under the action 59 3.2. Socle involves S of Aut(O(1; 1)); see [23, Lemma 2.1]. This suggests that a transitive subalgebra of Der O(m; 1) will do the job. Let us properly explain the set up. Assume that k is an algebraically closed field of characteristic p > 2. Suppose m ≥ 2. Let O(m; 1) = k[X1, . . . , Xm]/(X p 1 , . . . , X p m) denote the truncated polynomial ring in m variables. We write xi for the image of Xi in O(m; 1). Note that O(m; 1) is a local ring with dim O(m; 1) = pm. The degree function on the polynomial ring k[X1, . . . , Xm] induces a grading on O(m; 1). For each i ∈ N0, set O(m; 1)i := {f ∈ O(m; 1) deg f = i}. (3.3) Then O(m; 1) = Lm(p−1) i=0 O(m; 1)i. For each j ∈ N0, set O(m; 1)(j) := Mi≥j O(m; 1)i = {f ∈ O(m; 1) deg f ≥ j}. Then the Z-grading on O(m; 1) induces a descending filtration on O(m; 1). Note that each O(m; 1)(j) is an ideal of O(m; 1). If j > m(p − 1), then O(m; 1)(j) = 0. The unique maximal ideal of O(m; 1), denoted m, is O(m; 1)(1). Since f p = f (0)p for any f ∈ O(m; 1), we have that f p = 0 for all f ∈ m. Let W (m; 1) denote the derivation algebra of O(m; 1). It is called the mth Jacobson- Witt algebra. By Theorem 1.4.2, this restricted Lie algebra is simple. Note that any derivation D of O(m; 1) is uniquely determined by its effects on x1, . . . , xm. It is easy to see that W (m; 1) is a free O(m; 1)-module of rank m generated by the partial deriva- tives ∂1, . . . , ∂m such that ∂i(xj) = δij for all 1 ≤ i, j ≤ m. Hence dim W (m; 1) = mpm. Put W (m; 1)l := m Xi=1 O(m; 1)l+1 ∂i (3.4) for −1 ≤ l ≤ m(p − 1) − 1. Then the Z-grading on O(m; 1) induces a Z-grading on W (m; 1), i.e. W (m; 1) = W (m; 1)−1 ⊕ W (m; 1)0 ⊕ · · · ⊕ W (m; 1)m(p−1)−1. Note that W (m; 1)−1 = Pm i=1 k ∂i. Similarly, put W (m; 1)(l) := O(m; 1)(l+1) ∂i (3.5) m Xi=1 60 3.2. Socle involves S for −1 ≤ l ≤ m(p−1)−1. Then the natural filtration on O(m; 1) induces a descending filtration on W (m; 1), i.e. W (m; 1) = W (m; 1)(−1) ⊃ W (m; 1)(0) ⊃ · · · ⊃ W (m; 1)(m(p−1)−1) ⊃ 0. It is called the standard filtration; see Definition 1.3.4. Note that each W (m; 1)(l) is a Lie subalgebra of W (m; 1). The subalgebra W (m; 1)(0) = Pm i=1 m ∂i is often referred to as the standard maximal subalgebra of W (m; 1). This is because it can be characterized as the unique proper subalgebra of smallest codimension in W (m; 1). By Lemma 1.4.2, we know that W (m; 1)(0) is a restricted subalgebra of W (m; 1). Moreover, for any D1 ∈ W (m; 1)(0), it is easy to check that D1(O(m; 1)) ⊆ m. It is also easy to see that W (m; 1)(1) = Pm i=1 m2 ∂i is the nilradical of W (m; 1)(0). A restricted subalgebra Q′ of W (m; 1) is called transitive if it does not preserve any proper nonzero ideals of O(m; 1). Equivalently, Q′ + W (m; 1)(0) = W (m; 1); see [23, Sec. 2.1] and [30, Definition 2.3.1]. Let G denote the automorphism group of O(m; 1). Each σ ∈ G is uniquely de- termined by its effects on x1, . . . , xm. An assignment σ(xi) = fi extends to an au- tomorphism of O(m; 1) if and only if fi ∈ m, and the Jacobian Jac(f1, . . . , fm) := It follows that G is a connected algebraic group of dimension (cid:12)(cid:12)(cid:0) ∂ fi ∂ xj(cid:1)1≤i,j≤m(cid:12)(cid:12) /∈ m. mpm − m. By [11, Theorem 10], every automorphism of W (m; 1) is induced by a unique automorphism of O(m; 1) via the rule Dσ = σ ◦ D ◦ σ−1 for all σ ∈ G and D ∈ W (m; 1). So we can identify G with the automorphism group of W (m; 1). Note that for any f ∈ O(m; 1), D ∈ W (m; 1) and σ ∈ G, (f D)σ = f σDσ, where f σ = f (σ(x1), . . . , σ(xm)); see (1.12) in Sec. 1.5. It follows that if D2 = i=1 gi ∂i ∈ W (m; 1), then Pm Dσ 2 = m Xi, j=1 gσ i (cid:18) ∂i(cid:0)σ−1(xj)(cid:1)(cid:19)σ ∂j; see (1.13) in Sec. 1.5. Note also that G respects the standard filtration of W (m; 1), and Lie(G) = W (m; 1)(0). Let us introduce the Lie algebra that we are going to work with. Let D be any restricted transitive subalgebra of W (m; 1) such that N (D) is irreducible. Let S be 61 3.2. Socle involves S any simple restricted Lie algebra over k such that all its derivations are inner. We assume further that N (S) is irreducible. It is natural to form the semidirect product L := (S ⊗ O(m; 1)) ⋊ (IdS ⊗D). Then L is known to be a semisimple restricted Lie algebra over k; see also [23, Sec. 2.1]. To ease notation we identify IdS ⊗D with D. Let us explain the reason that we assumed that D is a transitive subalgebra of W (m; 1). Note that if D is not transitive, then O(m; 1) contains a proper nonzero D-invariant ideal, say n. Since O(m; 1) is a local ring, this ideal is contained in the maximal ideal m of O(m; 1). Hence f p = 0 for all f ∈ n. Now consider S ⊗n, a nonzero subspace of L. Since O(m; 1)n ⊆ n, we have that [S ⊗ O(m; 1), S ⊗ n] ⊆ S ⊗ n. Since n is D-invariant, it follows that S ⊗ n is a nonzero ideal of L consisting of nilpotent elements of L. Hence L is not semisimple and we are not interested in such L in this chapter. Therefore, for L to be semisimple, it is necessary that D is transitive. This is the reason that we assumed that D is transitive when we introduced the Lie algebra L. We want to prove Theorem 3.2.3 which states that the variety N (L) is irreducible. We will consider the surjective morphism ψ : N (L) → N (D) and need some prelimi- nary results. Let us outline these results in each section. In Sec. 3.2.1 we consider nilpotent elements of D. Since D is a restricted transitive subalgebra of W (m; 1), there exists d ∈ D such that d /∈ W (m; 1)(0). If d is also nilpotent, then we show that d is conjugate under G = Aut(W (m; 1)) to an element in a nice form. For that, we need a few results. Thanks to Lemma 1.5.2 in Sec. 1.5 and Lemma 3.2.1 which state that any z ∈ W (m; 1) such that z /∈ W (m; 1)(0) is conjugate under G to an element in a nice form. Then we show that any such nilpotent element z is conjugate under G to d0 + u, where d0 = ∂1 +xp−1 ∂2 + · · · + xp−1 1 · · · xp−1 s−1 ∂s with 1 1 ≤ s ≤ m, u ∈ (I ∂1 + · · ·+I ∂m)∩W (m; 1)(p−1) and I is the ideal of O(m; 1) generated by xs+1, . . . , xm; see Lemma 3.2.7 for notations. For the proof, we need Theorem 3.2.2 which was proved by A. Premet in [22]. Then we show that (d0 +u)ps ∈ W (m; 1)(0); see Lemma 3.2.8. We show in Lemma 3.2.9 that any d ∈ N (D) such that d /∈ W (m; 1)(0) is conjugate under G to d0 + u. Moreover, dps ∈ W (m; 1)(0) by Lemma 3.2.8. We shall denote d0 + u ∈ N (D) by z. 62 3.2. Socle involves S Next we define a monomial ordering DegLex on O(m; 1); see Definition 3.2.4. Then we consider the subspace M = I + z(m), where m is the maximal ideal of O(m; 1), I is the ideal of O(m; 1) generated by xs+1, . . . , xm with s ≥ 1, and z = d0 + u as above. We show M has the property that M ⊕ kxp−1 s = O(m; 1); see Lemma 3.2.10 and Lemma 3.2.11. · · · xp−1 1 In Sec. 3.2.2 we consider nilpotent elements of L. Let D′ = Pm(p−1) be an element of L, where si ∈ S, fi ∈ O(m; 1) with deg fi = i, and d ∈ N (D) si ⊗ fi + d i=0 with d ∈ W (m; 1)(0). In Lemma 3.2.12, we compute p-th powers of D′ and show that D′ ∈ N (L) if and only if s0 ∈ N (S). We want to prove a similar result for the element D = Pm(p−1) i=0 where si ∈ S, fi ∈ O(m; 1) with deg fi = i, and z = d0 + u ∈ N (D) as above. We first si ⊗ fi + z of L, construct some automorphisms of L, i.e. we show that exp(ad(s⊗f )) with s⊗f ∈ S⊗m is an automorphism of L; see Lemma 3.2.13. Let H denote the subgroup of Aut(L) generated by these exp(ad(s ⊗ f )). Then we show that D = Pm(p−1) i=0 conjugate under H to si ⊗ fi + z is D1 = s′ 0 ⊗ xp−1 1 · · · xp−1 s + v′ + z, where s′ 0 ∈ S (possibly 0), v′ ∈ S⊗I, I is the ideal of O(m; 1) generated by xs+1, . . . , xm with s ≥ 1, and z = d0 + u ∈ N (D) as above; see Lemma 3.2.14. Then D is nilpotent if and only if D1 is nilpotent. In Lemma 3.2.15, we show that D1 ∈ N (L) if and only if s′ 0 ∈ N (S). In Sec. 3.2.3, we prove that the variety N (L) is irreducible. 3.2.1 Nilpotent elements of D Since D is a restricted transitive subalgebra of W (m; 1), there are elements in D which are not in W (m; 1)(0). In this section we consider these elements. If they are nilpotent, we show that they are conjugate under G = Aut(W (m; 1)) to an element in a nice form; see Lemma 3.2.9. For that, we start with elements of W (m; 1) which are not in W (m; 1)(0). 63 Lemma 3.2.1. [20, p. 154, line 17] Let z be an element of W (m; 1) such that z /∈ W (m; 1)(0). Then there exists an 1 ≤ s ≤ m such that z is conjugate under G to 3.2. Socle involves S s Xi=1 xp−1 1 · · · xp−1 i−1 (1 + xp−1 i ψi) ∂i +xp−1 1 · · · xp−1 s m Xi=s+1 ψi ∂i, where ψi ∈ k[Xi+1, . . . , Xm]/(X p ideal of k[Xs+1, . . . , Xm]/(X p s+1, . . . , X p m), for s + 1 ≤ i ≤ m. i+1, . . . , X p m) for 1 ≤ i ≤ s and ψi ∈ mm−s, the maximal Sketch of proof. Recall that if σ ∈ G and D = Pm m i=1 gi ∂i ∈ W (m; 1), then Dσ = σ ◦ D ◦ σ−1 = Xi, j=1 gσ i (cid:0) ∂i(σ−1(xj))(cid:1)σ ∂j, (3.6) where gσ i = gi(σ(x1), . . . , σ(xm)); see (1.12) and (1.13) in Sec. 1.5. Let z be an element of W (m; 1) such that z /∈ W (m; 1)(0). By Lemma 1.5.2, there exists Φ1 ∈ G such that m zΦ1 = ∂1 +xp−1 Xi=1 where ϕi ∈ k[X2, . . . , Xm]/(X p 2 , . . . , X p 1 ϕi ∂i = (1 + xp−1 1 ϕ1) ∂1 +xp−1 1 m Xi=2 ϕi ∂i, If z2 ∈ W (m; 1)(0), then we find an s = 1 such that zΦ1 is of the desired form. If m) for all 1 ≤ i ≤ m. Set z2 = Pm i=2 ϕi ∂i. z2 /∈ W (m; 1)(0), then Lemma 1.5.2 implies that there exists Φ2 ∈ G such that 2 = (1 + xp−1 zΦ2 2 ϕ′ 2) ∂2 +xp−1 2 m Xi=3 ϕ′ i ∂i, i ∈ k[X3, . . . , Xm]/(X p where ϕ′ Φ2(x1) = x1 and Φ2(xj) ∈ k[X2, . . . , Xm]/(X p k[X2, . . . , Xm]/(X p 3 , . . . , X p 2 , . . . , X p m) for all 2 ≤ i ≤ m. Here we may assume that 2 , . . . , X p m) ∩ m, i.e. the intersection of m) with the maximal ideal m of O(m; 1), for 2 ≤ j ≤ m. By (3.6), it is easy to check that zΦ1Φ2 = (zΦ1)Φ2 = (1 + xp−1 1 ϕΦ2 1 ) ∂1 +xp−1 1 (1 + xp−1 2 ϕ′ 2) ∂2 +xp−1 1 xp−1 2 m Xi=3 ϕ′ i ∂i . Set z3 = Pm i=3 ϕ′ i ∂i. If z3 ∈ W (m; 1)(0), then we find an s = 2 such that zΦ1Φ2 is of the desired form. If z3 /∈ W (m; 1)(0), then Lemma 1.5.2 implies that there exists Φ3 ∈ G such that 3 = (1 + xp−1 zΦ3 3 ϕ′′ 3) ∂3 +xp−1 3 m Xi=4 ϕ′′ i ∂i, 64 i ∈ k[X4, . . . , Xm]/(X p where ϕ′′ that Φ3(x1) = x1, Φ3(x2) = x2 and Φ3(xj) ∈ k[X3, . . . , Xm]/(X p 4 , . . . , X p m) for all 3 ≤ i ≤ m. Here we may assume 3 , . . . , X p m) ∩ m for 3.2. Socle involves S 3 ≤ j ≤ m. By (3.6), it is easy to check that zΦ1Φ2Φ3 = (zΦ1Φ2)Φ3 =(1 + xp−1 1 (ϕΦ2 1 )Φ3) ∂1 +xp−1 1 (1 + xp−1 2 (ϕ′ + xp−1 1 xp−1 2 (1 + xp−1 3 ϕ′′ 3) ∂3 +xp−1 1 xp−1 2 xp−1 3 ϕ′′ i ∂i . m 2)Φ3) ∂2 Xi=4 Continue doing the above process until we find an 1 ≤ s ≤ m and Φ1, . . . , Φs ∈ G such that zΦ1Φ2Φ3...Φs =(1 + xp−1 1 ψ1) ∂1 +xp−1 1 (1 + xp−1 2 ψ2) ∂2 +xp−1 (1 + xp−1 m 3 ψ3) ∂3 + . . . + xp−1 1 · · · xp−1 s−1(1 + xp−1 s ψs) ∂s +xp−1 1 · · · xp−1 ψi ∂i = s Xi=1 xp−1 1 · · · xp−1 i−1 (1 + xp−1 i ψi) ∂i +xp−1 1 · · · xp−1 ψi ∂i, s 2 1 xp−1 Xi=s+1 Xi=s+1 m s where ψi ∈ k[Xi+1, . . . , Xm]/(X p ideal of k[Xs+1, . . . , Xm]/(X p i+1, . . . , X p m) for 1 ≤ i ≤ s and ψi ∈ mm−s, the maximal s+1, . . . , X p m), for s + 1 ≤ i ≤ m. Note that if s = m, then the above process shows that z is G-conjugate to Pm where ψi ∈ k[Xi+1, . . . , Xm]/(X p i+1, . . . , X p m) for 1 ≤ i ≤ m. Note that ψm = 0. This i=1 xp−1 1 · · · xp−1 i−1 (1 + xp−1 i ψi) ∂i, completes the sketch of proof. Next we assume that z is a nilpotent element of W (m; 1) such that z /∈ W (m; 1)(0). We want to prove that z is conjugate under G to an element in a nicer form; see Lemma 3.2.7. For that, we need Theorem 3.2.2 which was proved by A. Premet in [22]. This theorem characterizes all regular elements of W (m; 1). To state that theorem, we need to introduce some notations used in [22, Sec. 2 and 3]. Then we state a few preliminary results and give a sketch proof of that theorem. Let g be any finite dimensional restricted Lie algebra over k. Given x ∈ g, let g0 x denote the set of all y ∈ g for which (ad x)N (y) = 0, where N ≫ 0. It is known that g0 x is a restricted subalgebra of g containing the centralizer cg(x). Define rk(g) := minx∈g dim g0 x. We say that x ∈ g is regular if its centralizer cg(x) has the smallest possible dimension. Note that dim g0 x = rk(g) if and only if x is regular and in this case g0 x is a Cartan subalgebra of minimal dimension in g. It is known that the set of all regular elements of g is Zariski open in g. We consider the restricted Lie algebra W (m; 1). In [20], A. Premet proved that 65 3.2. Socle involves S Lemma 3.2.2. [20, Lemma 9] The set of singular elements of N (W (m; 1)) coincides with N0 = {D1 ∈ N (W (m; 1)) Dpm−1 1 ∈ W (m; 1)(0)}; see also Lemma 1.5.5. In [22], A. Premet characterized all regular elements of W (m; 1). The following result on tori will be used in part (a) of the proof. Theorem 3.2.1. [30, Theorem 7.5.1] Let t ⊂ W (m; 1) be a torus and set t0 := t ∩ W (m; 1)(0). Let t1, . . . , ts be toral elements of t linearly independent (mod t0). Then there exists Φ ∈ G = Aut(W (m; 1)) such that Φ(ti) = (1 + xi) ∂i for 1 ≤ i ≤ s and Φ(t0) ⊂ Pm i=s+1 kxi ∂i. Let t be any torus in W (m; 1). We denote by (ttor)∗ the set of all linear functions α : t → k such that α(t) ∈ Fp for all toral elements t ∈ t. It is known that (ttor)∗ is an Fp-form of the dual space t∗ with Card((ttor)∗) = pl, where l = dim t. Note that O(m; 1) is tautologically a W (m; 1)-module. Then O(m; 1) decomposes as O(m; 1) = Lλ∈t∗ O(m; 1)λ, where O(m; 1)λ = {f ∈ O(m; 1) t.f = λ(t)f for all t ∈ t}. If O(m; 1)λ 6= {0}, then we say that λ ∈ t∗ is a weight of O(m; 1) with respect to t or a t-weight. We denote the set of all t-weights of O(m; 1) by Λ(O(m; 1)). Note that Λ(O(m; 1)) ⊆ (ttor)∗. Lemma 3.2.3. [22, Lemma 1] Let t be an r-dimensional torus in W (m; 1). Then Λ(O(m; 1)) = (ttor)∗ and dim O(m; 1)λ = pm−r for all λ ∈ Λ(O(m; 1)). Next we consider a particular torus tD which is defined as follows: for any D ∈ W (m; 1), there exist a unique semisimple element Ds and a unique nilpotent element Dn such that D = Ds + Dn and [Ds, Dn] = 0; see Theorem 1.1.2. Let tD denote the torus of W (m; 1) generated by the semisimple part Ds of D. For any D ∈ W (m; 1), we also know that there exist homogeneous polynomials ϕ0, . . . , ϕm−1 in k[W (m; 1)] with deg ϕi = pm − pi such that Dpm +Pm−1 i=0 ϕi(D)Dpi = 0; see Theorem 1.5.1. Define r = r(D) := min{0 ≤ i ≤ m − 1 ϕi(D) 6= 0} for D /∈ N (W (m; 1)), and (3.7) r = r(D) := m for D ∈ N (W (m; 1)). Then the following hold: 66 3.2. Socle involves S Lemma 3.2.4. [22, Lemma 2(i)] dim tD = m − r(D) = m − r and Dpr n = 0. In [22, Sec 3.3], A. Premet first observed that Lemma 3.2.5. [22, Remark 1] If D is a regular element of W (m; 1) and r = r(D), then (ad D)pr−1 maps W (m; 1)0 the centralizer cW (m;1)(D) = Pm−1 linearly independent. D = {D1 ∈ W (m; 1) (ad D)N (D1) = 0 for N ≫ 0} onto i=0 kDpi, and the derivations D, Dp, . . . , Dpm−1 are Then A. Premet proved the following theorem which characterizes all regular ele- ments of W (m; 1). We will use (ii) and (iii) in the proof of Lemma 3.2.7. Theorem 3.2.2. [22, Theorem 2] Suppose D ∈ W (m; 1) and let r = r(D) be defined as in (3.7). Then the following are equivalent: (i) D is a regular element of W (m; 1). (ii) The kernel of D on O(m; 1) is k1 which is 1-dimensional. (iii) There exist zr+1, . . . , zm ∈ {εi+xi r+1 ≤ i ≤ m} for some εi ∈ {0, 1} and σ ∈ G i=r+1 λi(zi ∂i) for some λi ∈ k, the torus (tD)σ is spanned by such that Dσ zr+1 ∂r+1, . . . , zm ∂m, and Dσ n = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 r−1 ∂r. s = Pm (iv) All Jordan blocks of Dn have size pr. Sketch of proof. We prove the above statements are equivalent in the following or- der: (a) (i) implies (ii), (b) (ii) implies (iii), (c) (iii) implies (iv), (d) (iv) implies (ii) and finally (e) (ii) implies (i). (a) Suppose D is a regular element of W (m; 1). We prove by contradiction that Ker D = k1. Suppose dim Ker D ≥ 2. Then there exists a nonzero f ∈ Ker D ∩ m. Since p > 2, it is easy to see that if f ∈ m \ m2, then f 2 6= 0. Hence we may assume that f ∈ m2. Note that f D ∈ cW (m;1)(D) and (f D)p = f pDp = 0. It follows from Lemma 3.2.5 and Lemma 3.2.4 that f D = λDpr−1 for some λ ∈ k. n We split the proof into two cases: D /∈ W (m; 1)(0) and D ∈ W (m; 1)(0). Suppose D /∈ W (m; 1)(0). Then f D 6= 0 and we may assume that f D = Dpr−1 Dpr−1 ∈ W (m; 1)(1). As ad DW (m;1)0 = ad DnW (m;1)0 n n D D , it follows from Lemma 3.2.5 . It follows that 67 3.2. Socle involves S that there exists some y ∈ W (m; 1)0 D such that (ad Dn)pr−1(y) = D. Since p > 2, we have that D =(ad Dpr−1 n )p−1((ad Dn)pr−1−1(y)) (3.8) ∈[W (m; 1)(1), [W (m; 1)(1), W (m; 1)]] ⊆ W (m; 1)(1). This contradicts our assumption that D /∈ W (m; 1)(0). So this case cannot occur. Suppose D ∈ W (m; 1)(0). By Theorem 3.2.1, we may assume that tD ⊆ Tm, where Tm is a maximal torus in W (m; 1) with basis {x1 ∂1, . . . , xm ∂m}. Let {θ1, . . . , θm} be the corresponding dual basis in T ∗ m. Let νi denote the restriction of θi to tD. Let ad−1 denote the representation of W (m; 1)0 in gl(W (m; 1)−1) induced by the adjoint action of W (m; 1)0 to W (m; 1)−1; see (3.4) for the Z-grading on W (m; 1). Then the set of tD-weights on W (m; 1)−1 coincides with Λ := {−ν1, . . . , −νm}. By Lemma 3.2.4, we know that dim tD = m − r. Since Λ spans t∗ D, we have that Card(Λ) ≥ m − r. For ν ∈ Λ, set η(ν) := dim W (m; 1)ν η := maxν∈Λ η(ν). Then one can show that −1. Then η(ν) ≥ 1 and Pν∈Λ η(ν) = m. Set η ≤ r + 1. (3.9) Write Dn = Pi≥0 Dn,i, where Dn,i ∈ W (m; 1)i; see (3.4) for notations. Since tD ⊆ Tm, we have that Dn,i ∈ cW (m;1)(tD) for all i. In particular, each weight space W (m; 1)ν −1 is invariant under ad−1(Dn,0). As Dn is nilpotent, we get (ad−1(Dn,0))η = 0. (3.10) Since W (m; 1)(1) is a p-ideal of W (m; 1)(0), then one can show using Jacobson's formula that for all K ≥ 0. DpK n − DpK n,0 ∈ W (m; 1)(1) (3.11) Suppose r ≥ 2. Then one can show that pr−1 ≥ r + 1. By (3.9) and (3.10), n,0 ) = 0. Since ad−1 is a faithful representation of W (m; 1)0, we have that ad−1(Dpr−1 we get Dpr−1 n,0 = 0. By (3.11), we get Dpr−1 n ∈ W (m; 1)(1). Applying (3.8), we get D ∈ W (m; 1)(1). In particular, D is nilpotent. By our assumption, D is regular. This contradicts Lemma 3.2.2. Suppose r = 1. Then Card(Λ) = m−1 or m. If Card(Λ) = m, thenPν∈Λ η(ν) = m implies that η = 1. By (3.10), we have that ad−1(Dn,0) = 0. This again gives that 68 3.2. Socle involves S Dpr−1 n ∈ W (m; 1)(1). Arguing as above, we get a contradiction. If Card(Λ) = m − 1, then one can show that η = 2. Since dim tD = m − 1, we may assume that tD is spanned by elements ti = xi ∂i +ci(xm ∂m), 1 ≤ i ≤ m − 1, for some ci ∈ Fp. Note that ν1, . . . , νm−1 are linearly independent and νm = c1ν1 + · · · + cm−1νm−1. Since Card(Λ) = m − 1, we may assume that νm = νm−1, Dn,0 = λ(xm−1 ∂m) for some λ ∈ k, and Ds = m−2 Xi=1 αi(xi ∂i) + αm−1(xm−1 ∂m−1 +xm ∂m) for some αi ∈ k∗. Then cW (m;1)(tD) ⊂ W (m; 1)(0). If λ = 0, then Dn ∈ W (m; 1)(1). Applying (3.8) with r = 1, we get D ∈ W (m; 1)(1). By our assumption, D is regular. This contradicts Lemma 3.2.2. If λ 6= 0, then Dn /∈ W (m; 1)(1). Then one can show that cW (m;1)(D) ∩ W (m; 1)(1) 6= {0}. By Lemma 3.2.5, any nilpotent element of cW (m;1)(D) is a scalar multiple of Dn. Since Dn /∈ W (m; 1)(1), this is a contradiction. So this case cannot occur. If r = 0, then Dn = 0 and tD = Tm. Hence Ker D = k1, the zero weight space of Tm in O(m; 1). This shows that (i) implies (ii). (b) Suppose (ii) holds for D, i.e. Ker D = k1. We show that (ii) implies (iii). Let B denote the zero weight space of tD in O(m; 1). By Lemma 3.2.3, dim B = pr. Note that the restriction of D to B, denoted DB, is a nilpotent derivation of B. Let mB = B ∩ m be the maximal ideal of the local ring B. By our assumption, mB is not D-stable. Hence B is differentiably simple; see Definition 1.6.1. By Theorem 1.6.1, B ∼= O(r; 1) as k-algebras. Since DB is a nilpotent derivation of B and Ker DB = k1, it follows from Lemma 1.5.1 and Theorem 1.5.2 that there exist y1, . . . , yr ∈ mB whose cosets in mB/m2 B are linearly independent such that DB = ∂ ∂ y1 + yp−1 1 ∂ ∂ y2 + · · · + yp−1 1 · · · yp−1 r−1 ∂ ∂ yr . By Lemma 1.5.1(ii), Der B is a free B-module with basis DB, . . . Dpr−1 B . Hence there exists {bi,j 0 ≤ i, j ≤ r − 1} ⊂ B such that ∂ ∂ yi = (cid:18) r−1 Xj=0 bi,jDpj n (cid:19)B for 1 ≤ i ≤ r. This show that the set of partial derivatives { ∂ ∂ yi can be lifted to a system of commuting derivations of O(m; 1). 1 ≤ i ≤ r} ⊂ Der B 69 3.2. Socle involves S Since each derivation ∂ ∂ yi B = B ∩ m2. Then mB/m2 B embeds into m/m2. As a result, there exist y′ of O(m; 1) maps B ∩ m2 to mB = B ∩ m, it follows that r+1, . . . , y′ m2 m ∈ m in m/m2 are linearly independent. m such that the cosets of y1, . . . , yr, y′ r+1, . . . , y′ Since dim tD = m − r and tD acts semisimply on O(m; 1)/k1, we may assume that there exist γr+1, . . . , γm ∈ (ttor D )∗ such that y′ i + k1 ∈ (O(m; 1)/k1)γi. Since B = O(m; 1)0, then {γr+1, . . . , γm} forms a basis of the dual space t∗ D. It follows that t(y′ i) = γi(t)y′ i + γ′ i(t)1 for all t ∈ tD, where γ′ i : tD → k is a linear function. Then one can show that γ′ i is proportional to γi. So for each i ≥ r + 1, there exists εi ∈ k such that t(y′ i + εi) = γi(t)(y′ i + εi). Rescaling the y′ i's if need be, we may assume that εi ∈ {0, 1}. Set yi := y′ i + εi for r + 1 ≤ i ≤ m. By our choice of y1, . . . , ym, there is a unique σ ∈ Aut(O(m; 1)) such that σ−1(xi) = yi for 1 ≤ i ≤ r and σ−1(xi) = yi − εi for r + 1 ≤ i ≤ m. Then Dσ n = σ ◦ Dn ◦ σ−1 = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 r−1 ∂r, and Dσ s = σ ◦ Ds ◦ σ−1 = m Xi=r+1 λi(εi + xi) ∂i for some λi ∈ k. Since dim tD = m − r and each (εi + xi) ∂i is toral, it is easy to see that (tD)σ is spanned by (εr+1 + xr+1) ∂r+1, . . . , (εm + xm) ∂m. This shows that (ii) implies (iii). (c) Suppose (iii) holds for D and adapt the notations introduced in (b). Let γ = Pm i=r+1 αiγi ∈ (ttor is a free B-module of rank 1 generated by yγ = yαr+1 D )∗. Then αi ∈ Fp. By Lemma 3.2.3, the weight space O(m; 1)γ m . Moreover, (DB)pr−1 6= 0. Since Dn(yγ) = 0, then Dn acts on each O(m; 1)γ as a Jordan block of size pr. This r+1 · · · yαm shows that (iii) implies (iv). (d) Suppose (iv) holds for D. Consider the zero weight space O(m; 1)0 of tD. By Lemma 3.2.3, we see that Dn acts on O(m; 1)0 for tD as a single Jordan block of size pr. Since Ker D ⊆ O(m; 1)0, we must have that Ker D = k1. This shows that (iv) implies (ii). (e) Suppose (ii) holds for D. We show that (ii) implies (i). Since we have shown in (b) that (ii) implies (iii), we get an explicit description of D. So we may assume that D = Ds + Dn, where Ds = Pm statement (iii), and Dn = ∂1 +xp−1 is easy to see that D preserves each direct summand Lm 1 i=r+1 λizi ∂i for some λi ∈ k and zi = εi + xi as in ∂2 + · · · + xp−1 r−1 ∂r. Due to the form of D, it · · · xp−1 1 K=1 O(r; 1)zar+1 r+1 · · · zam m ∂K of 70 3.2. Socle involves S W (m; 1), where 0 ≤ ai ≤ p − 1 are fixed. Note that there are pm−r of such summands. Moreover, D acts on each summand as a direct sum of Jordan blocks of size pr with eigenvalue (ar+1λr+1 + · · · + amλm) − λK. Since the λi's are linearly independent over the prime field Fp, this eigenvalue is zero if and only if either K ≤ r and ai = 0 for all r + 1 ≤ i ≤ m or K ≥ r + 1, aK = 1 and ai = 0 for all r + 1 ≤ i ≤ m except for i = K. Due to the form of Dn, this implies that the kernel of ad D has dimension r + (m − r) = m. Hence D is regular and (ii) implies (i). This completes the sketch of proof. Now we want to state Lemma 3.2.7. But we need to define W and understand the following: Lemma 3.2.6. Let I be the ideal of O(m; 1) generated by xs+1, . . . , xm, where s ≥ 1. Define W := I ∂1 + · · · + I ∂m . Let d0 = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 s−1 ∂s be the derivation of O(s; 1). Then W is an ad d0-invariant restricted Lie subalgebra of W (m; 1) contained in W (m; 1)(0). Proof. Since I is an ideal of O(m; 1) and W (m; 1) = Pm j=1 O(m; 1) ∂j, we can describe W as IW (m; 1), i.e. the set of all f D with f ∈ I and D ∈ W (m; 1). Let f D, gE be any elements of W , where f, g ∈ I and D, E ∈ W (m; 1). By (1.4), we have that [f D, gE] = f D(g)E − gE(f )D + f g[D, E]. Since f, g ∈ I and I is an ideal of O(m; 1), it is easy to see that [f D, gE] ∈ W . Hence W is a Lie subalgebra of W (m; 1). Since O(m; 1) is a local ring, the ideal I is contained in the maximal ideal m of O(m; 1). It follows that W ⊂ W (m; 1)(0). Next we show that W is ad d0-invariant. Let f D be any element of W , where f ∈ I and D ∈ W (m; 1). Note that I is d0-invariant. Then d0(f ) ∈ I. By (1.4) again, we have that [d0, f D] = d0(f )D + f [d0, D] ∈ ID + f W (m; 1) ⊆ IW (m; 1) = W. Hence W is ad d0-invariant. It remains to show that W is restricted. Let D′ be any element of W (m; 1). Note that D′ ∈ W if and only if D′(xi) ∈ I for all 1 ≤ i ≤ m. Let D1 ∈ W . Then D1(xi) ∈ I 71 for all 1 ≤ i ≤ m and D1 preserves I. It follows that Dn 1 preserves I for all n ∈ Z>0. Hence 3.2. Socle involves S Dp 1(xi) = Dp−1 1 (D1(xi)) ∈ Dp−1 1 (I) ⊆ I for all 1 ≤ i ≤ m. Therefore, Dp 1 ∈ W and W is restricted. It follows from the above that W is an ad d0-invariant restricted Lie subalgebra of W (m; 1) contained in W (m; 1)(0). This completes the proof. Lemma 3.2.7. Let z be a nilpotent element of W (m; 1) such that z /∈ W (m; 1)(0). Then z is conjugate under G to d0 + u, where d0 = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 s−1 ∂s with 1 ≤ s ≤ m and u ∈ W ∩ W (m; 1)(p−1); see Lemma 3.2.6 for W = I ∂1 + · · · + I ∂m and (3.5) for W (m; 1)(p−1) = {Pm standard filtration of W (m; 1). j=1 fj ∂j deg fj ≥ p for all j} the component of the Proof. Let z be a nilpotent element of W (m; 1) such that z /∈ W (m; 1)(0). By Lemma 3.2.1, we may assume that z = s Xi=1 xp−1 1 · · · xp−1 i−1 (1 + xp−1 i ψi) ∂i +xp−1 1 · · · xp−1 s =d0 + s Xi=1 xp−1 1 · · · xp−1 i ψi ∂i +xp−1 1 · · · xp−1 s m Xi=s+1 ψi ∂i m Xi=s+1 ψi ∂i, where 1 ≤ s ≤ m, d0 = ∂1 +xp−1 ψi ∈ k[Xi+1, . . . , Xm]/(X p of k[Xs+1, . . . , Xm]/(X p s+1, . . . , X p i+1, . . . , X p 1 m) for 1 ≤ i ≤ s and ψi ∈ mm−s, the maximal ideal ∂2 + · · · + xp−1 1 · · · xp−1 s−1 ∂s, m), for s + 1 ≤ i ≤ m. Let I be the ideal of O(m; 1) generated by xs+1, . . . , xm. Since ψi ∈ mm−s for all s + 1 ≤ i ≤ m, it is easy to see that z preserves the ideal I. Note that the factor ring O(m; 1)/I is isomorphic to O(s; 1) = k[X1, . . . , Xs]/(X p s ), and z acts on 1 , . . . , X p O(s; 1) as a derivation. This derivation, call it y, has the same form as z except that we forget the last summand xp−1 s Pm their images ¯ψi in O(s; 1), i.e. ψi = ¯ψi + ψ′ i=s+1 ψi ∂i and replace ψi, 1 ≤ i ≤ s, by · · · xp−1 i ∈ I, and i with ψ′ 1 y = d0 + s Xi=1 xp−1 1 · · · xp−1 i ¯ψi ∂i . 72 Note also that s Xi=1 xp−1 1 · · · xp−1 i ψ′ i ∂i +xp−1 1 3.2. Socle involves S · · · xp−1 s m Xi=s+1 ψi ∂i ∈ W (m; 1)(p−1). (3.12) We show that the kernel of y on O(s; 1) is 1-dimensional, i.e. Ker y = k1. Suppose the contrary. Then there exists a nonzero f ∈ Ker y with f (0) = 0. We want to show that y(f ) 6= 0. Hence we get a contradiction. Note that f is a linear combination A = xa1 1 · · · xas of monomials x i=1 ai > 0. We want to look at the "smallest" monomial involved in f . Note that the standard degree of ¯ψi ∂i and s , where 0 ≤ ai ≤ p − 1 and Ps monomials is not a good choice. This is because y = d0 +Ps applying d0 = ∂1 +xp−1 s−1 ∂s to xA in f , the standard degree either decreases or increases; see (3.15) and (3.16) below. So it is difficult to compare d0(xA) ∂2 + · · ·+xp−1 i=1 xp−1 · · · xp−1 · · · xp−1 1 1 1 i with (cid:0)Ps i=1 xp−1 1 · · · xp−1 i This means that we cannot easily deduce that y(f ) 6= 0. Hence we are going to use ¯ψi ∂i(cid:1)(xA) and the effects of y on the other monomials in f . a monomial ordering introduced later; see Definition 3.2.42. Let x A′′ any monomial in O(s; 1). Define the p-degree of xA′′ by A′′p := s Xi=1 a′′ i pi−1 = a′′ 1 + a′′ 2p + · · · + a′′ s ps−1, = xa′′ 1 · · · xa′′ 1 s s be i.e. the p-adic expansion of nonnegative integers with digits a′′ i . It is known that for a fixed prime number p, every nonnegative integer has a unique p-adic expansion. Hence for l = 0, 1, . . . , ps − 1, there is a unique xA′′ ∈ O(s; 1) with A′′p = l. Note that if the product of two monomials is nonzero, then the p-degree of this product is given by the sum of p-degrees of the monomials; see Remark 3.2.2(iii) and (iv)(a) later. Consider d0 = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 s−1 ∂s in y. By Lemma 3.2.10 later, we know that applying d0 to any xA in f , we get d0(x A) = αx A′ , (3.13) where 1 ≤ α ≤ p − 1 and xA′ ∈ O(s; 1) with A′p = Ap − 1. in Ps Consider Ps i=1 xp−1 1 · · · xp−1 i=1 xp−1 ¯ψi ∂i in y. We first show that all nonzero summands 1 ¯ψi ∂i have positive p-degrees. Since ¯ψi ∈ O(s; 1), it is a linear · · · xp−1 i i 2Note that this monomial ordering was defined after we have proved this lemma. Due to the forms of d0 and u, we defined the p-degree of monomials as in Definition 3.2.4; see Example 3.2.1(2) and (3) later for the reasons. Here we are working with the subring O(s; 1) of O(m; 1) and the p-degree is the first part Ps i=1 a′′ i pi−1. 73 3.2. Socle involves S combination of monomials in O(s; 1). Then we can write ¯ψi = XA′′ λA′′ x A′′ , where λA′′ ∈ k and xA′′ ∈ O(s; 1) with A′′p = l ≥ 0. Hence s Xi=1 xp−1 1 · · · xp−1 i ¯ψi ∂i = s Xi=1 XA′′ λA′′xp−1 1 · · · xp−1 i x A′′ ∂i . Consider each summand xp−1 1 · · · xp−1 i xA′′ ∂i. By Example 3.2.1(1) later, we know that for 1 ≤ i ≤ s, ∂i has p-degree −pi−1. mark 3.2.2(iii) implies that xp−1 · · · xp−1 xA′′ 1 i ∂i has p-degree If xp−1 1 · · · xp−1 i xA′′ ∂i 6= 0, then Re- (cid:0)(p − 1) + (p − 1)p + · · · + (p − 1)pi−1(cid:1) + l − pi−1 =(pi − 1) + l − pi−1 =(pi − pi−1) + (l − 1). By our assumption, p ≥ 3. Since 1 ≤ i ≤ s, an easy induction on i shows that pi − pi−1 ≥ p − p0 = p − 1 ≥ 2. Since l ≥ 0, we have that l − 1 ≥ −1. It follows that (cid:0)(p − 1) + (p − 1)p + · · · + (p − 1)pi−1(cid:1) + l − pi−1 =(pi − pi−1) + (l − 1) ≥ 2 + (−1) = 1 > 0. Hence all nonzero summands xp−1 1 tive p-degrees. Then applying Ps i xA′′ · · · xp−1 i=1 xp−1 1 ∂i in Ps i=1 xp−1 1 · · · xp−1 i ¯ψi ∂i have posi- · · · xp−1 ¯ψi ∂i to any xA in f , we get i (cid:18) s Xi=1 xp−1 1 · · · xp−1 i ¯ψi ∂i(cid:19)(x A) = XA′′′ µA′′′ x A′′′ , (3.14) where µA′′′ ∈ k and xA′′′ ∈ O(s; 1) with A′′′p > Ap. By (3.13), A′p = Ap − 1. Hence A′′′p > A′p. Now we look at f and take the monomial of smallest p-degree with nonzero coefficient, say it is xA1 = xa1 1 · · · xas s , where 0 ≤ ai ≤ p − 1 and Ps i=1 ai > 0. Rescaling f if need be, we may assume that the coefficient of xA1 is 1. If a1 6= 0, then d0(x A1) = a1x A′ 1 = a1xa1−1 1 · · · xas s . (3.15) By our choice of xA1 and (3.13)-(3.14), we have that y(f ) = d0(x A1) +XA2 γA2 x A2 = a1x A′ 1 +XA2 γA2 x A2, 74 where γA2 ∈ k and xA2 ∈ O(s; 1) with A2p > A′ 1p. Since all xA2 in y(f ) have 3.2. Socle involves S 1p, they do not cancel xA′ 1. As a1xA′ A2p > A′ a1 = · · · = aK−1 = 0 and aK 6= 0, then xA1 = xaK K xaK+1 K+1 · · · xas s and 1 6= 0, we have that y(f ) 6= 0. If d0(x A1) = aK x A′ 1 = aKxp−1 1 Hence · · · xp−1 K−1xaK −1 K xaK+1 K+1 · · · xas s . (3.16) y(f ) = d0(x A1) +XA3 γA3 x A3 = aK x A′ 1 +XA3 γA3 x A3, where γA3 ∈ k and xA3 ∈ O(s; 1) with A3p > A′ 1p. Arguing similarly as above, we get y(f ) 6= 0. So f /∈ Ker y. This is a contradiction. Hence Ker y = k1. Note that since z is nilpotent so is y. Since Ker y = k1, it follows from Theo- rem 3.2.2(i) and (ii) that y is a regular element of W (s; 1). Then Theorem 3.2.2(iii) implies that y is conjugate under Aut(W (s; 1)) to d0 = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 s−1 ∂s . Since W (s; 1) is a Lie subalgebra of W (m; 1), we may identify Aut(W (s; 1)) with a subgroup of G = Aut(W (m; 1)) by letting σ(xi) = xi for all σ ∈ Aut(W (s; 1)) and s + 1 ≤ i ≤ m. Since yσ = σyσ−1 = d0 for some σ ∈ Aut(W (s; 1)), then zσ = σzσ−1 = d0 +u, where u ∈ W ∩W (m; 1)(p−1) = (I ∂1 + · · ·+I ∂m)∩W (m; 1)(p−1). Note that u ∈ W (m; 1)(p−1) follows from (3.12) and that G preserves the standard filtration of W (m; 1), in particular, G preserves the component W (m; 1)(p−1). This completes the proof. Lemma 3.2.8. Let z = d0 + u be as in Lemma 3.2.7. Then zps ∈ W (m; 1)(0). Proof. Recall that z = d0 + u, where d0 = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 s−1 ∂s and u ∈ W ∩ W (m; 1)(p−1); see Lemma 3.2.6 for W . By Lemma 1.1.1, zps = dps 0 + ups + s−1 Xl=0 vpl l , where vl is a linear combination of commutators in d0 and u. By Jacobi identity, we can rearrange each vl so that vl is in the span of [wt, [wt−1, [. . . , [w2, [w1, u] . . . ], where t = ps−l − 1 and each wα, 1 ≤ α ≤ t, is equal to d0 or to u. 75 3.2. Socle involves S We show that [wt, [wt−1, [. . . , [w2, [w1, u] . . . ] ∈ W and so vl ∈ W . By Lemma 3.2.6, W is an ad d0-invariant restricted Lie subalgebra of W (m; 1) contained in W (m; 1)(0). Since u ∈ W , it follows that [d0, u] ∈ W . It is clear that [u, u] = 0 ∈ W . This shows that [w1, u] ∈ W . Consider w2. If w2 = d0, then by the same reason, we get [d0, [d0, u]] ∈ W . If w2 = u, then W is a Lie subalgebra of W (m; 1) implies that [u, [d0, u]] ∈ W . This shows that [w2, [w1, u]] ∈ W . Continuing in this way and using the same reasons, we get [wt, [wt−1, [. . . , [w2, [w1, u] . . . ] ∈ W and so vl ∈ W for all 0 ≤ l ≤ s − 1. Since W is a restricted Lie subalgebra of W (m; 1), we have that vpl l=0 vpl l ∈ W for all 1 ≤ l ≤ s − 1. Hence Ps−1 l ∈ W ⊂ W (m; 1)(0). Similarly, since u ∈ W and W is restricted, we have that ups ∈ W ⊂ W (m; 1)(0). By Lemma 1.5.1(i), we know that dps 0 = 0. Hence zps = ups + s−1 Xl=0 vpl l ∈ W (m; 1)(0). This completes the proof. Now we consider the restricted transitive subalgebra D of W (m; 1) and show the last two results hold for any d ∈ N (D) such that d /∈ W (m; 1)(0). Lemma 3.2.9. Let d be any element of N (D) such that d /∈ W (m; 1)(0). Then d is conjugate under G to d0 + u, where d0 + u is the element in Lemma 3.2.7. Moreover, dps ∈ W (m; 1)(0). Proof. Let d be any element of N (D) such that d /∈ W (m; 1)(0). By Lemma 3.2.7, we know that d is conjugate under G to d0 + u. Here we want to point out that the use of the group action does not suppose that D or N (D) is G-stable, it is only used to bring elements of N (D) to a nice form. Since d is conjugate under G to d0 + u, there exists σ ∈ G such that d = σ(d0 + u)σ−1. Then dps = (cid:0)σ(d0 + u)σ−1(cid:1)ps = σ(d0 + u)ps σ−1. By Lemma 3.2.8, (d0 + u)ps ∈ W (m; 1)(0). Since σ ∈ G, it preserves the standard filtration of W (m; 1). Hence dps ∈ W (m; 1)(0). This completes the proof. Remark 3.2.1. Let us denote d0 +u ∈ N (D) by z. Note that zps ∈ W (m; 1)(0) implies that zps preserves the maximal ideal m of O(m; 1). This will be used in the proof of Lemma 3.2.15; see step 2. 76 3.2. Socle involves S Define M := I + z(m) to be the subspace of O(m; 1), where m is the maximal ideal of O(m; 1), I is the ideal of O(m; 1) generated by xs+1, . . . , xm with s ≥ 1, and z = d0+u as in Lemma 3.2.7. We are aiming for Lemma 3.2.11 which shows M has the property that M⊕kxp−1 · · · xp−1 s = 1 O(m; 1). To prove that result, we need to consider the image of m under z = d0 + u. In the proof of Lemma 3.2.7, we have seen that applying d0 to any monomial in m, the standard degree of monomials either decreases or increases. It is difficult to get much useful information. Therefore, we need to define a monomial ordering on O(m; 1). Let us first recall the definitions of total ordering and monomial ordering. Definition 3.2.1. [5, p. 55] A binary relation ≤ is a total ordering on a set X if the following hold for all a, b, c ∈ X: (i) Antisymmetry: if a ≤ b and b ≤ a, then a = b. (ii) Transitivity: if a ≤ b and b ≤ c, then a ≤ c. (iii) Comparability: either a ≤ b or b ≤ a. Consider the polynomial ring k[X1, . . . , Xm]. Note that there is a natural bijection between Zm ≥0 and the set of monomials in k[X1, . . . , Xm] given by (α1, . . . , αm) ←→ xα1 1 · · · xαm m , and addition in Zm ≥0 corresponds to multiplication of monomials in k[X1, . . . , Xm]. So if ≺ is an ordering on Zm (β1, . . . , βm), then xα1 1 · · · xαm m ≺ xβ1 1 · · · xβm m . Recall that ≥0, then it gives an ordering on monomials: if (α1, . . . , αm) ≺ Definition 3.2.2. [5, Definition 1, Sec. 2 and Corollary 6, Sec. 4, Chap. 2] A mono- mial ordering on k[X1, . . . , Xm] is a relation ≺ on Zm ≥0, or equivalently, a relation ≺ on the set of monomials xα1 1 · · · xαm m , (α1, . . . , αm) ∈ Zm ≥0, satisfying: (i) ≺ is a total ordering on Zm ≥0. (ii) If (α1, . . . , αm) ≺ (β1, . . . , βm) and (γ1, . . . , γm) ∈ Zm ≥0, then (α1 + γ1, . . . , αm + γm) ≺ (β1 + γ1, . . . , βm + γm). 77 3.2. Socle involves S (iii) ≺ is a well-ordering on Zm ≥0, i.e. every nonempty subset of Zm ≥0 has a smallest el- ement under ≺. This is equivalent to the condition that (0, . . . , 0) ≺ (α1, . . . , αm) for every (α1, . . . , αm) 6= (0, . . . , 0). It follows from the above definition that (0, . . . , 0) (or 1) is the smallest element in Zm ≥0 (or k[X1, . . . , Xm]) under any monomial ordering. We are interested in O(m; 1) = k[X1, . . . , Xm]/(X p 1 , . . . , X p m), where every mono- mial xa1 1 · · · xam m in O(m; 1) has 0 ≤ ai ≤ p − 1 for all 1 ≤ i ≤ m. There are many monomial orderings on O(m; 1). In particular, the following is an example. Definition 3.2.3. [5, Definition 3 and Proposition 4, Sec. 2, Chap. 2] The lexico- graphic ordering ≺lex on O(m; 1) with x1 ≺lex x2 ≺lex · · · ≺lex xm is defined as follows: for any two non-equal monomials xa1 1 · · · xam m and xa′ 1 · · · xa′ m m , 1 xa1 1 · · · xam m ≺lex xa′ 1 · · · xa′ 1 m m if and only if ai < a′ i, where i is the largest number in {1, . . . , m} for which ai 6= a′ i. Next we define the p-degree of monomials in O(m; 1). Then it induces a monomial ordering on O(m; 1). Definition 3.2.4. Let A = (a1, . . . , as, as+1, . . . , am), where 0 ≤ ai ≤ p − 1. Set x A := xa1 1 · · · xas s xas+1 s+1 · · · xam m . Define the p-degree of x A by Ap := s Xi=1 aipi−1 + ps m Xi=s+1 ai. Then we have a version of the total ordering called DegLex: if A = (a1, . . . , as, as+1, . . . , am) and A′ = (a′ i ≤ p − 1, then we say that A ≺DegLex A′ if either Ap < A′p or Ap = A′p and A m) with 0 ≤ ai, a′ s+1, . . . , a′ 1, . . . , a′ s, a′ precedes A′ in the lexicographic ordering ≺lex defined in Definition 3.2.3. Remark 3.2.2. (i) Note that 0 ≤ ai ≤ p − 1 are well defined nonnegative integers, i=1 aipi−1 = a1 + a2p + · · · + asps−1 in Ap is the p-adic expansion of nonnegative integers with digits ai. It is well known that for a fixed and the first summand Ps prime number p, every nonnegative integer has a unique p-adic expansion. 78 3.2. Socle involves S (ii) It is easy to check that DegLex is a monomial ordering on O(m; 1). Moreover, we see that DegLex first orders monomials by the p-degree, then it uses the lexicographic ordering to break ties. (iii) Let x A and x A′ be any monomials in O(m; 1). If x A A′ x = x A′′ 6= 0, i.e. all expo- nents of x1, . . . , xm are strictly less than p, then it follows from Definition 3.2.4 that A′′p = Ap + A′p. (iv) Let xA = xa1 1 · · · xas s xas+1 s+1 · · · xam m be any monomial in O(m; 1). Then Ap = a1 + a2p + · · · + asps−1 + ps m Xi=s+1 ai. Q0 = (p − 1) + (p − 1)p + · · · + (p − 1)ps−1 = ps − 1, Q = Q0 + ps((m − s)(p − 1)). Let and (a) If 0 ≤ Ap ≤ Q0, then we must have that Pm Pm i=s+1 ai ≥ 1, then Ap ≥ ps > Q0, a contradiction. Hence xA = xa1 1 · · · xas s with Ap = a1 + a2p + · · · + asps−1. By (i), we know that for a fixed prime i=s+1 ai = 0. Indeed, if p, the p-adic expansion of any nonnegative integer is unique. Hence for l = 0, 1, . . . , Q0, there is a unique monomial xA = xa1 1 · · · xas s with Ap = l. (b) If ps = Q0 + 1 ≤ Ap ≤ Q, then 1 ≤ Pm i=s+1 ai ≤ (m − s)(p − 1) and i=1 ai ≤ s(p − 1) by (a). Note that in this case, we could have distinct monomials with the same p-degree and such monomials are easy to con- 0 ≤ Ps struct. We could simply fix a1, . . . , as and choose as+1, . . . , am so that they have the same Pm i=s+1 ai. For example, the monomials xp−1 s xs+1 s xs+2 have the same p-degree which is Q0 + ps. By Defi- · · · xp−1 1 1 · · · xp−1 and xp−1 nition 3.2.3, we have that xp−1 · · · xp−1 s xs+1 ≺lex xp−1 1 · · · xp−1 s xs+2. 1 (v) Let us describe O(m; 1) using the p-degree. Keeping in mind the numbers Q0 and Q introduced in (iv). Since O(m; 1) is spanned by monomials x A = xa1 1 · · · xam m , 0 ≤ ai ≤ p − 1, 79 3.2. Socle involves S we can describe O(m; 1) as O(m; 1) = span{x A 0 ≤ Ap ≤ Q}. Let I be the ideal of O(m; 1) generated by xs+1, . . . , xm, where s ≥ 1. Let f be any polynomial in O(m; 1). Then f is a linear combination of monomials xA in O(m; 1). Since O(m; 1) = I ⊕ O(s; 1), these monomials xA are either in I or in O(s; 1). If xA ∈ I, then A = (a1, . . . , as, as+1, . . . , am), where 0 ≤ Ps s(p − 1) and 1 ≤ Pm xA /∈ I, i.e. xA ∈ O(s; 1), then A = (a1, . . . , as, 0, . . . , 0), where 0 ≤ Ps i=1 ai ≤ i=s+1 ai ≤ (m − s)(p − 1). It follows that ps ≤ Ap ≤ Q. If i=1 ai ≤ s(p − 1). It follows that 0 ≤ Ap ≤ Q0. By (iv)(a), we can write f as f = Q0 Xl=0 λAl x Al + g, where λAl ∈ k, xAl = xa1 1 · · · xas s ∈ O(s; 1) with Alp = l, and g ∈ I. We will come back to this in Lemma 3.2.14, Sec. 3.2.2. To get familiar with the p-degree defined in Definition 3.2.4, let us look at some examples. Moreover, we want to explain the reason that we define p-degree as Ap = i=1 aipi−1 + psPm Ps i=s+1 ai. Initially, we want to compute z(xA), where z = d0 + u and xA is any monomial in the maximal ideal m of O(m; 1). Due to the form of d0, the standard degree of monomials is not useful. Hence we want to define a monomial ordering so that d0 decreases the "degree" of the monomial, i.e. p-degree, whereas u increases the p-degree of the monomial; see Example 3.2.1(2) and (3) below. As a result, z(xA) is nonzero. Then we want to use these results to show the subspace M = I + z(m) has the property that M ⊕ kxp−1 s = O(m; 1). But later we · · · xp−1 1 realize there is an alternative way to prove this and we only need to do computations for d0; see Lemma 3.2.10 and Lemma 3.2.11. However, the p-degree of monomials is still useful for the proof of Lemma 3.2.14 as we may lost of control if we are using the standard degree of monomials. Example 3.2.1. (1) Let x A = xa1 1 · · · xas s xas+1 s+1 · · · xam m ∈ O(m; 1). Let A := a1 + · · · + as + as+1 + · · · + am be the standard degree of xA. Suppose A > 0. Consider the partial derivatives ∂1, . . . , ∂m. Note that for 1 ≤ j ≤ m, if aj 6= 0, then ∂j(x A) = ajxa1 1 · · · xaj −1 j · · · xam m . 80 3.2. Socle involves S Hence the degree of ∂j with respect to the standard degree of monomials is −1. Consider the p-degree. Since Ap = a1 + a2p + · · · + ajpj−1 + · · · + asps−1 + ps m Xj=s+1 aj, we see that for 1 ≤ j ≤ s, the degree of ∂j with respect to p is −pj−1, and for s + 1 ≤ j ≤ m, the degree of ∂j with respect to p is −ps. (2) Consider xA(xi ∂j), where xA ∈ O(m; 1) with A > 0, s + 1 ≤ i ≤ m and 1 ≤ j ≤ m. Suppose xAxi 6= 0. By (1), we know that the degree of xA(xi ∂j) with respect to the standard degree of monomials is A + (1 − 1) = A > 0. Consider the p-degree. By Remark 3.2.2(iii) and (1), we know that for 1 ≤ j ≤ s, the degree of xA(xi ∂j) with respect to p is Ap + (ps − pj−1). Since A > 0, we have that Ap > 0. Since 1 ≤ j ≤ s, we have that ps − pj−1 > 0. Hence Ap + (ps − pj−1) > 0. By (1) again, we know that for s + 1 ≤ j ≤ m, the degree of x A(xi ∂j) with respect to p is Ap + (ps − ps) = Ap, which is positive, too. Recall that u ∈ (Pm ideal of O(m; 1) generated by xs+1, . . . , xm with s ≥ 1, and j=1 I ∂j) ∩ W (m; 1)(p−1), where I is the W (m; 1)(p−1) = (cid:26) m Xj=1 fj ∂j deg fj ≥ p for all j(cid:27). Then u is a linear combination of xA(xi ∂j), where xAxi, s + 1 ≤ i ≤ m, is a monomial in I with p ≤ A + 1 ≤ m(p − 1), and 1 ≤ j ≤ m. By above, we know A(xi ∂j) has degree A with respect to the standard degree of monomials. that each x Hence applying xA(xi ∂j) to any xA′ = xa′ 1 1 · · · xa′ m m ∈ m with a′ j 6= 0, the standard degree increases by A ≥ p − 1, and the p-degree increases by at least Ap > 0. Since xAxi ∈ I and I is an ideal of O(m; 1), it follows that (cid:0)xA(xi ∂j)(cid:1)(xA′) ∈ I and so A′ u(x ) = XA′′ λA′′ x A′′ , 81 where λA′′ ∈ k and xA′′ ∈ I with p − 1 ≤ A = A′′ − A′ ≤ m(p − 1) − 1 and A′′p − A′p ≥ Ap > 0. 3.2. Socle involves S (3) Consider xp−1 1 · · · xp−1 r−1 ∂r, where 1 ≤ r ≤ s. By (1), the degree of xp−1 1 · · · xp−1 r−1 ∂r with respect to p is (cid:0)(p − 1) + (p − 1)p + · · · + (p − 1)pr−2(cid:1) − pr−1 = (pr−1 − 1) − pr−1 = −1. Since d0 = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 s−1 ∂s and each summand has p-degree −1, it follows that d0 has p-degree −1. Now we consider the action of z = d0 + u on the maximal ideal m of O(m; 1). Note that m = ms ⊕ I, where ms is the maximal ideal of O(s; 1) and I is the ideal of O(m; 1) generated by xs+1, . . . , xm with s ≥ 1. Since d0 = ∂1 +xp−1 ∂2 + · · · + xp−1 · · · xp−1 s−1 ∂s 1 1 is a derivation of O(s; 1), we first consider d0(ms). Lemma 3.2.10. Let d0 = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 s−1 ∂s be the derivation of O(s; 1). Let ms be the maximal ideal of O(s; 1). Let xA1 = xar r xar+1 r+1 · · · xas s be any monomial in ms, where 1 ≤ r ≤ s is the smallest index such that 1 ≤ ar ≤ p − 1, and 0 ≤ ai ≤ p − 1 for r + 1 ≤ i ≤ s. Then d0(x A1) = arx A2 = arxp−1 1 · · · xp−1 r−1xar−1 r xar+1 r+1 · · · xas s with A2p = A1p − 1. Hence d0(ms) is spanned by all monomials x xp−1 1 . In particular, dim d0(ms) = ps − 1. · · · xp−1 s A2 ∈ O(s; 1) except Proof. Let xA1 = xar r xar+1 r+1 · · · xas s be any monomial in ms, where 1 ≤ r ≤ s is the smallest index such that 1 ≤ ar ≤ p − 1, and 0 ≤ ai ≤ p − 1 for r + 1 ≤ i ≤ s. Since d0 = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 we have that · · · xp−1 r−1 ∂r +xp−1 1 · · · xp−1 r ∂r+1 + · · · + xp−1 1 · · · xp−1 s−1 ∂s, d0(x A1) = arxp−1 1 · · · xp−1 r−1xar −1 r xar+1 r+1 · · · xas s = arx A2. Since 0 ≤ ar −1 ≤ p−2 < p−1, we see that xA2 6= xp−1 1 · · · xp−1 s . By Example 3.2.1(3), we know that the degree of d0 with respect to p is −1. Hence A2p = A1p − 1. By Lemma 1.5.1(iii), we know that d0 acts on O(s; 1) as a single Jordan block of size ps 82 with zeros on the main diagonal. So dim d0(O(s; 1)) = ps − 1. Since O(s; 1) = ms ⊕ k 3.2. Socle involves S and d0(c) = 0 for all c ∈ k, we have that dim d0(ms) = ps −1. It follows from the above calculation that d0(ms) is spanned by all monomials xA2 ∈ O(s; 1) except xp−1 · · · xp−1 . 1 s Indeed, since A2 = (p − 1, . . . , p − 1, ar − 1, ar+1, . . . , as) with 0 ≤ ar − 1 ≤ p − 2 and 0 ≤ ai ≤ p − 1 for r + 1 ≤ i ≤ s, we see that as r varies from 1 to s, there are Ps r=1(p − 1)ps−r = ps − 1 different possibilities for A2 except (p − 1, . . . , p − 1) (i.e. p − 1 appears s times). So there are ps − 1 distinct monomials xA2 ∈ O(s; 1) except xp−1 1 . This completes the proof. · · · xp−1 s Lemma 3.2.11. Let z = d0 + u be as in Lemma 3.2.7. Let M = I + z(m) be the subspace of O(m; 1), where m is the maximal ideal of O(m; 1) and I is the ideal of O(m; 1) generated by xs+1, . . . , xm with s ≥ 1. Then M is spanned by all monomials xA = xa1 . In particular, dim M = pm − 1. m ∈ O(m; 1) except xp−1 1 · · · xam · · · xp−1 1 s Proof. Let M = I + z(m) be the subspace of O(m; 1), where m is the maximal ideal of O(m; 1) and I is the ideal of O(m; 1) generated by xs+1, . . . , xm with s ≥ 1. Recall that z = d0 + u, where d0 = ∂1 +xp−1 of O(s; 1) and u ∈ (Pm Since u ∈ Pm i=1 I ∂i) ∩ W (m; 1)(p−1). Note that d0 preserves the ideal I. i=1 I ∂i and I is an ideal of O(m; 1), we have that u(m) ⊆ I. Hence M = I + z(m) = I + d0(m). Consider d0(m). Note that m = ms ⊕ I, where ms is the s−1 ∂s is a derivation ∂2 + · · · + xp−1 · · · xp−1 1 1 maximal ideal of O(s; 1). Since d0(ms) ⊂ O(s; 1), d0(I) ⊆ I and O(s; 1) ∩ I = {0}, we have that d0(m) = d0(ms) ⊕ d0(I). Since d0(I) ⊆ I, we have that M = I + z(m) = I + d0(m) = I ⊕ d0(ms). (3.17) Since I is the ideal of O(m; 1) generated by xs+1, . . . , xm with s ≥ 1, then I is spanned by all monomials xa1 i=1 ai ≤ s(p − 1) and 1 · · · xas s xas+1 s+1 · · · xam m , where 0 ≤ Ps 1 ≤ Pm i=s+1 ai ≤ (m − s)(p − 1). So dim I = ps(pm−s − 1) = pm − ps. By Lemma 3.2.10, and so d0(ms) is spanned by all monomials xa1 s ∈ O(s; 1) except xp−1 · · · xp−1 1 · · · xas 1 s dim d0(ms) = ps − 1. Therefore, M = I ⊕ d0(ms) is spanned by all monomials xA = m ∈ O(m; 1) except xp−1 xa1 1 · · · xam pm − 1. This completes the proof. . In particular, dim M = (pm−ps)+(ps−1) = · · · xp−1 1 s 83 3.2. Socle involves S 3.2.2 Nilpotent elements of L In the last section we have obtained all the results required for nilpotent elements of D. Now let us consider nilpotent elements of L = (S ⊗ O(m; 1)) ⋊ D. si ⊗ fi + d be an element of L, where si ∈ S, Lemma 3.2.12. Let D′ = Pm(p−1) i=0 fi ∈ O(m; 1) with deg fi = i, and d ∈ N (D) with d ∈ W (m; 1)(0). Then D′ ∈ N (L) if and only if s0 ∈ N (S). Proof. Let D′ = Pm(p−1) i=0 N ≥ 1, si ⊗ fi + d be as in the lemma. We first show that for every (D′)pN = dpN By Lemma 1.1.1, + (spN 0 ⊗ f pN 0 ) + (other terms in S ⊗ m). (3.18) +(cid:18) m(p−1) si ⊗ fi(cid:19)pN N −1 vpl l , = dpN (D′)pN Xl=0 where vl is a linear combination of commutators in Pm(p−1) identity, we can rearrange each vl so that vl is in the span of Xi=0 i=0 + si ⊗ fi and d. By Jacobi si ⊗ fi] = 0. Suppose w1 = d. By our assump- If w1 = m(p−1) i=0 si ⊗ fi or to d. j=1 m ∂j. It is easy to see that d(O(m; 1)) ⊆ m. Hence i=0 i=0 i=0 Xi=0 si ⊗ fi] . . . ] ∈ S ⊗ m. (cid:20)wt,(cid:20)wt−1,(cid:20) . . . ,(cid:20)w2,(cid:20)w1, si ⊗ fi, then [w1,Pm(p−1) We show that [wt, [wt−1, [. . . , [w2, [w1,Pm(p−1) si ⊗ fi(cid:21) . . .(cid:21), where t = pN −l − 1 and each wα, 1 ≤ α ≤ t, is equal to Pm(p−1) Pm(p−1) tion, d ∈ W (m; 1)(0) = Pm Xi=0 This shows that [w1,Pm(p−1) If w2 = d, then [d, [w1,Pm(p−1) w2 = Pm(p−1) Xi=0 si ⊗ d(fi) ∈ S ⊗ m. si ⊗ fi(cid:21) = si ⊗ fi] ∈ S ⊗ m. Consider [w2, [w1,Pm(p−1) si ⊗ fi(cid:21)(cid:21) ∈ [S ⊗ O(m; 1), S ⊗ m] ⊆ S ⊗ m. si ⊗ fi, then Xi=0 (cid:20)d, m(p−1) m(p−1) m(p−1) m(p−1) i=0 i=0 i=0 (cid:20) si ⊗ fi,(cid:20)w1, Xi=0 This shows that [w2, [w1,Pm(p−1) i=0 that d preserves S ⊗ m and S ⊗ m is an ideal of S ⊗ O(m; 1), we eventually get si ⊗ fi]] ∈ [d, S ⊗ m] = S ⊗ d(m) ⊆ S ⊗ m. If i=0 si ⊗ fi]]. si ⊗ fi]] ∈ S ⊗ m. Continuing in this way, i.e. using 84 3.2. Socle involves S i=0 si ⊗ fi] . . . ] ∈ S ⊗ m. Since each vl is a linear combi- [wt, [wt−1, [. . . , [w2, [w1,Pm(p−1) nation of these commutators, we have that vl ∈ S ⊗ m for all 0 ≤ l ≤ N − 1. Consider vpl l , where 0 < l ≤ N − 1. Since vl is a linear combination of elements in S ⊗ m, it follows from Lemma 1.1.1 and [S ⊗ m, S ⊗ m] ⊆ S ⊗ m that vp in this way, we get vpl l ∈ S ⊗ m for all 0 < l ≤ N − 1. Therefore, l ∈ S ⊗ m. Continuing N −1 Xl=0 si ⊗ fi(cid:1)pN We consider (cid:0)Pm(p−1) i=0 m(p−1) (cid:18) Xi=0 si ⊗ fi(cid:19)pN vpl l ∈ S ⊗ m. and show that = (spN 0 ⊗ f pN 0 ) + (other terms in S ⊗ m). (3.19) (3.20) Since fi ∈ m for 0 < i ≤ m(p − 1), we have that f pN i = 0. It follows from Lemma 1.1.1 that (cid:18) m(p−1) Xi=0 si ⊗ fi(cid:19)pN = m(p−1) Xi=0 i ⊗ f pN spN i + N −1 Xl=0 upl l = (spN 0 ⊗ f pN 0 ) + N −1 Xl=0 upl l , where ul is a linear combination of commutators in si ⊗ fi, 0 ≤ i ≤ m(p − 1). By Jacobi identity, we can rearrange each ul so that ul is in the span of [ηt, [ηt−1, [. . . , [η2, [η1, s0 ⊗ f0] . . . ], where t = pN −l − 1 and each ηα, 1 ≤ α ≤ t, is equal to some si ⊗ fi, 0 ≤ i ≤ m(p − 1). Since S ⊗ m is an ideal of S ⊗ O(m; 1), one can show similarly that [ηt, [ηt−1, [. . . , [η2, [η1, s0 ⊗ f0] . . . ] ∈ S ⊗ m. Hence ul ∈ S ⊗ m for all 0 ≤ l ≤ N − 1 and upl l ∈ S ⊗ m for all 0 < l ≤ N − 1. Therefore, PN −1 l ∈ S ⊗ m. This gives (3.20) as l=0 upl desired. It follows from (3.20) and (3.19) that for every N ≥ 1, the p-th powers of D′ is given by (D′)pN = dpN = dpN si ⊗ fi(cid:19)pN + N −1 Xl=0 vpl l m(p−1) +(cid:18) Xi=0 + (spN 0 ⊗ f pN 0 ) + (other terms in S ⊗ m). By our assumption, d is a nilpotent element of D such that d ∈ W (m; 1)(0). Hence d preserves S ⊗ m. Therefore, for N ≫ 0, D′ is a nilpotent element of L if and only if s0 is a nilpotent element of S. This completes the proof. 85 Next we consider elements of the form D = Pm(p−1) i=0 3.2. Socle involves S si ⊗ fi + z, where si ∈ S, fi ∈ O(m; 1) with deg fi = i, and z = d0 + u ∈ N (D) as in Lemma 3.2.7. We first construct some automorphisms of L; see Lemma 3.2.13. Then we show that D can be reduced to an element in a nice form; see Lemma 3.2.14. Lemma 3.2.13. Let s ⊗ f be an element of S ⊗ m. Then exp(ad(s ⊗ f )) is an automorphism of L. Proof. The proof is similar to Lemma 3.1.3 in Sec. 3.1. Let s ⊗ f be an element of S ⊗ m. We first show that exp(ad(s ⊗ f )) is an automorphism of S ⊗ O(m; 1). Since f ∈ m, we have that f p = 0. Hence (ad(s ⊗ f ))p = ad(sp ⊗ f p) = 0. Moreover, it is easy to check that for any w1, w2 ∈ S ⊗ O(m; 1), X0≤i,j≤p, i+j≥p 1 i!j!(cid:2)(ad(s ⊗ f ))i(w1), (ad(s ⊗ f ))j(w2)(cid:3) = 0. Hence exp(ad(s ⊗ f )) is an automorphism of S ⊗ O(m; 1). Then it induces an auto- morphism of the derivation algebra Der(S ⊗ O(m; 1)) = (cid:0)S ⊗ O(m; 1)(cid:1) ⋊(cid:0) IdS ⊗W (m; 1)(cid:1) via conjugation. Note that L = (S ⊗ O(m; 1)) ⋊ D ⊂ Der(S ⊗ O(m; 1)). To conclude that exp(ad(s ⊗ f )) is an automorphism of L, we need to check that exp(ad(s ⊗ f )) preserves L. Explicitly, we need to show that for any d1 ∈ D, exp(ad(s ⊗ f )) ◦ d1 ◦ exp(ad(−s ⊗ f )) ∈ L. In the proof of Lemma 3.1.3 (see (3.2)), we did a similar computation. By the same computational method and the following reasons: (i) d1 is a derivation of O(m; 1), (ii) exp(ad(s ⊗ f )) exp(ad(−s ⊗ f )) = Id, (iii) f p = 0, (iv) (p − 1)! ≡ −1(mod p), and (v) S ∼= ad S via the adjoint representation and hence we may identify S ⊗ O(m; 1) with its image in gl(S ⊗ O(m; 1)) under ad, 86 3.2. Socle involves S one can show that exp(ad(s ⊗ f )) ◦ d1 ◦ exp(ad(−s ⊗ f )) = d1 − s ⊗ d1(f ) − sp ⊗ f p−1d1(f ). (3.21) It is clear that exp(ad(s ⊗ f )) ◦ d1 ◦ exp(ad(−s ⊗ f )) ∈ L. Hence exp(ad(s ⊗ f )) preserves L and it is an automorphism of L. This completes the proof. Let H denote the subgroup of Aut(L) generated by exp(ad(s ⊗ f )), where s ⊗ f ∈ S ⊗ m. Then H is a connected algebraic group with S ⊗ m ⊆ Lie(H). We show that applying suitable elements of H, the following element of L can be reduced to an element in a nice form. Lemma 3.2.14. Let D = Pm(p−1) i=0 si ⊗ fi + z be an element of L, where si ∈ S, fi ∈ O(m; 1) with deg fi = i, and z = d0 + u ∈ N (D) as in Lemma 3.2.7. Then D is conjugate under H to D1 = s′ 0 ⊗ xp−1 1 · · · xp−1 s + v′ + z, where s′ 0 ∈ S (possibly 0), v′ ∈ S⊗I, I is the ideal of O(m; 1) generated by xs+1, . . . , xm with s ≥ 1, and z = d0 + u ∈ N (D) as in Lemma 3.2.7. Strategy of the proof. Step 1 . Let D be as in the lemma. By Remark 3.2.2(v), we can rewrite D as D = Q0 Xl=0 sAl ⊗ x Al + v + z, where sAl ∈ S, xAl = xa1 1 · · · xas s ∈ O(s; 1) with 0 ≤ Alp = l ≤ Q0 = ps − 1, v ∈ S ⊗ I, and z = d0 + u ∈ N (D) as in Lemma 3.2.7. We want to show that D is conjugate some s′ under H = hexp(ad(s ⊗ f )) s ⊗ f ∈ S ⊗ mi to D1 = s′ s + v′ + z for 0 ∈ S (possibly 0) and v′ ∈ S ⊗ I. We first observe that for any exp(ad(s ⊗ f )) in H, exp(ad(s ⊗ f ))(v + u) = u + v3 for some v3 = v3(s, f ) ∈ S ⊗ I. Note that · · · xp−1 0 ⊗ xp−1 1 (S ⊗ I) ∩ (S ⊗ O(s; 1)) = {0}. Moreover, all xAl are in O(s; 1) and d0 is a derivation of O(s; 1). Hence to show that D is conjugate under H to D1, we just need to show that PQ0 l=0 sAl ⊗ xAl + d0 is conjugate under H to s′ (possibly 0). For that, we only need to apply automorphisms exp(ad(s ⊗ f )) with 0 ⊗ xp−1 1 · · · xp−1 s + d0 for some s′ 0 ∈ S s ⊗ f ∈ S ⊗ ms. Here ms denotes the maximal ideal of O(s; 1). Step 2 . Let D0 = PQ0 0 ⊗ xp−1 s′ · · · xp−1 s + d0 for some s′ l=0 sAl ⊗ xAl + d0. We show that D0 is conjugate under H to 0 ∈ S (possibly 0). If sAl = 0 for all l, then D0 is of 1 87 the desired form. If not all sAl are zero, then we look at xAl's with sAl 6= 0 and take the one with the smallest p-degree, say it is xAK with AKp = K. Then 3.2. Socle involves S D0 = Q0 Xl=K sAl ⊗ x Al + d0. , then AKp = K = Q0 and D0 is of the desired form. If x AK = If x AK = xp−1 · · · xp−1 s 1 6= xp−1 s s 1 · · · xp−1 , then 0 ≤ AKp = K < Q0 and there exist 0 ≤ ai < p − 1 xa1 1 · · · xas for some 1 ≤ i ≤ s. Let 1 ≤ r ≤ s be the smallest index such that 0 ≤ ar < p − 1. Then xAK = xp−1 s ∈ ms and s = (ar + 1)−1sAK ∈ S. It follows from Lemma 3.2.10 that applying exp(ad(s ⊗ f )) to s . Let f = x xar+1 r+1 · · · xas AK = xar +1 r+1 · · · xas . . . xp−1 r xar+1 r−1xar 1 r D0, we may assume that D0 = Q0 Xl=K+1 s′ Al ⊗ x Al + d0, where s′ Al ∈ S, xAl = xa1 1 · · · xas s ∈ O(s; 1) with Alp = l. Continue doing the above until we get D0 is conjugate under H to s′ 0 ⊗xp−1 1 · · · xp−1 s +d0 for some s′ 0 ∈ S (possibly 0). It follows that D is conjugate under H to D1. Proof. Step 1 . Let D = Pm(p−1) i=0 si ⊗ fi + z be as in the lemma. Let I be the ideal of O(m; 1) generated by xs+1, . . . , xm, where s ≥ 1. Since fi ∈ O(m; 1), it follows from Remark 3.2.2(v) that we can write fi = Q0 Xl=0 λAl,ix Al + gi, where λAl,i ∈ k, x 1 · · · xas gi ∈ I. Then we can rewrite D as Al = xa1 s ∈ O(s; 1) with 0 ≤ Alp = l ≤ Q0 = ps − 1, and D = Q0 Xl=0 sAl ⊗ x Al + v + z, (3.22) where sAl ∈ S, xAl = xa1 1 · · · xas s ∈ O(s; 1) with 0 ≤ Alp = l ≤ Q0 = ps − 1, v ∈ S ⊗ I, and z = d0 + u ∈ N (D) as in Lemma 3.2.7. We want to show that D is conjugate under H = hexp(ad(s ⊗ f )) s ⊗ f ∈ S ⊗ mi to D1 = s′ s + v′ + z for 0 ∈ S (possibly 0) and v′ ∈ S ⊗ I. We claim that this problem can be reduced · · · xp−1 some s′ 0 ⊗ xp−1 1 to show that PQ0 some s′ l=0 sAl ⊗ xAl + d0 is conjugate under H to s′ 0 ⊗ xp−1 1 · · · xp−1 s + d0 for 0 ∈ S (possibly 0). Take D as in (3.22) and let exp(ad(s ⊗ f )) be any element 88 3.2. Socle involves S Al + d0(cid:19) + exp(ad(s ⊗ f ))(v + u) sAl ⊗ x in H. By Lemma 3.2.13 (see (3.21)), exp(ad(s ⊗ f ))(D) = exp(ad(s ⊗ f ))(cid:18) Q0 Xl=0 Xl=0 Xj=1 sAl ⊗ x p−1 = Q0 Q0 1 j! Al + Xl=0 + d0 − s ⊗ d0(f ) − sp ⊗ f p−1d0(f ) + exp(ad(s ⊗ f ))(v + u). (ad s)j(sAl) ⊗ f j Al x We show that exp(ad(s ⊗ f ))(v + u) = u + v3 for some v3 = v3(s, f ) ∈ S ⊗ I. Since I is an ideal of O(m; 1), this implies that S ⊗ I is an ideal of S ⊗ O(m; 1). Hence S ⊗ I is stabilized by all exp(ad(s ⊗ f )) in H. In particular, exp(ad(s ⊗ f ))(v) = v1 for some v1 = v1(s, f ) ∈ S ⊗ I. By Lemma 3.2.13 (see (3.21)) again, exp(ad(s ⊗ f ))(u) = u − s ⊗ u(f ) − sp ⊗ f p−1u(f ). Since u ∈ (I ∂1 + · · · + I ∂m), we have that u(m) ⊆ I. In particular, u(f ) ∈ I. Hence exp(ad(s ⊗ f ))(u) = u + v2 for some v2 = v2(s, f ) ∈ S ⊗ I. It follows that exp(ad(s ⊗ f ))(v + u) = u + v3 for some v3 = v3(s, f ) = v1(s, f ) + v2(s, f ) ∈ S ⊗ I. Hence for any exp(ad(s ⊗ f )) ∈ H, exp(ad(s ⊗ f ))(D) = exp(ad(s ⊗ f ))(cid:18) Q0 Xl=0 Xl=0 Xj=1 sAl ⊗ x Al + p−1 = Q0 Q0 1 j! Xl=0 + d0 − s ⊗ d0(f ) − sp ⊗ f p−1d0(f ) + (u + v3). (ad s)j(sAl) ⊗ f j Al x sAl ⊗ x Al + d0(cid:19) + (u + v3) Applying another element exp(ad( s1 ⊗ f ′)) ∈ H to the above, we still get exp(ad( s1 ⊗ f ′))(u + v3) = u + v4 for some v4 = v4( s1, f ′) ∈ S ⊗ I. Note that (S ⊗ I) ∩ (S ⊗ O(s; 1)) = {0}. Moreover, all x Al are in O(s; 1) and d0 is a derivation of O(s; 1). Hence to show that D = PQ0 s + v′ + d0 + u, we just need to show that PQ0 · · · xp−1 0 ⊗ xp−1 l=0 sAl ⊗ xAl + v + d0 + u is H-conjugate to l=0 sAl ⊗ xAl + d0 is s + d0. For that, we only need to apply automorphisms H-conjugate to s′ 0 ⊗ xp−1 D1 = s′ · · · xp−1 1 1 exp(ad(s ⊗ f )) with s ⊗ f ∈ S ⊗ ms and do computations in O(s; 1) and W (s; 1). Here ms denotes the maximal ideal of O(s; 1). Step 2 . Let D0 = PQ0 0 ⊗ xp−1 s′ s + d0 for some s′ l=0 sAl ⊗ xAl + d0. We show that D0 is H-conjugate to 0 ∈ S (possibly 0). If sAl = 0 for all l, then D0 is of · · · xp−1 1 89 the desired form. If not all sAl are zero, then we look at xAl's with sAl 6= 0 and take the one with the smallest p-degree, say it is xAK with AKp = K. Then 3.2. Socle involves S D0 = Q0 Xl=K sAl ⊗ x Al + d0. , then AKp = K = Q0 and D0 is of the desired form. If , then 0 ≤ AKp = K < Q0 and we need to apply exp(ad(s ⊗ f )) AK = xp−1 If x xAK 6= xp−1 1 · · · xp−1 · · · xp−1 s 1 s to D0 and clear sAK ⊗ xAK . By Lemma 3.2.13 (see (3.21)), we know that for any exp(ad(s ⊗ f )) ∈ H, exp(ad(s ⊗ f ))(D0) = Q0 p−1 Q0 Al + sAl ⊗ x Xl=K + d0 − s ⊗ d0(f ) − sp ⊗ f p−1d0(f ). Xj=1 Xl=K (ad s)j(sAl) ⊗ f j 1 j! Al x (3.23) Let AK = (a1, . . . , as). Since xAK 6= xp−1 1 · · · xp−1 s , there exist 0 ≤ ai < p − 1 for some 1 ≤ i ≤ s. Note that if AKp = 0, then ai = 0 for all 1 ≤ i ≤ s. Let 1 ≤ r ≤ s be the smallest index such that 0 ≤ ar < p − 1. Then x AK = xp−1 1 . . . xp−1 r−1xar r xar+1 r+1 · · · xas s . Let f = x AK = xar+1 r xar+1 r+1 · · · xas s ∈ ms. By Lemma 3.2.10, d0(f ) = d0(x AK ) = (ar + 1)x AK with AKp = AKp − 1. Let s = (ar + 1)−1sAK ∈ S. Substituting s and f into (3.23), we get Q0 exp(ad(s ⊗ f ))(D0) = Al Q0 p−1 sAl ⊗ x Xl=K+1 Xj=1 + d0 − (ar + 1)−p+1sp Xl=K+1 1 j! + (ar + 1)−j(ad sAK )j(sAl) ⊗ (x AK )j Al x (3.24) ⊗ (x AK )p−1 x AK . AK We want to show that exp(ad(s ⊗ f ))(D0) = PQ0 Al ∈ S s ∈ O(s; 1) with Alp = l. Look at the second and the last AK ∈ ms and ms is an ideal of O(s; 1), we see that Al ⊗ xAl + d0, where s′ summands in (3.24). Since x Al = xa1 1 · · · xas l=K+1 s′ and x these two summands are in S ⊗ ms. This implies that every nonzero monomial xA in 90 these two summands has Ap ≤ Q0. Moreover, we show that every nonzero monomial 3.2. Socle involves S A′′ A′′′ = x A′ x x and xA′′ By our choice, x xA in these two summands has Ap > K + 1. We start with the second summand. By Remark 3.2.2(iii), we know that for any monomials xA′ in O(s; 1), if AK )j, where 1 ≤ j ≤ p − 1. 6= 0, then A′′′p = A′p + A′′p. Consider (x AK ∈ ms with AKp = AKp + 1 = K + 1. Hence for 2 ≤ j ≤ p − 1, if AK )j ∈ ms with p-degree j AKp ≥ 2(K + 1) > K + 1. Now look AK )j xAl in the second summand. Since l ≥ K + 1 ≥ 1, every monomial xAl is AK )j xAl is in ms and has in ms and has Alp = l ≥ K + 1. If (x p-degree j AKp + Alp ≥ AKp + Alp ≥ (K + 1) + (K + 1) > K + 1. AK )j xAl 6= 0, then (x AK )j 6= 0, then (x at (x (x AK )p−1xAK in the last summand. By above, if (x Look at (x AK )p−1 is in ms and has p-degree > K + 1. Note that x (x AK )p−1 6= 0, then AK ∈ O(s; 1) and has AK )p−1xAK is in ms and has p- AKp = K ≥ 0. So if (x degree (p − 1) AKp + AKp > (K + 1) + K ≥ K + 1. AK )p−1xAK 6= 0, then (x Hence every nonzero monomial xA in the second and the last summands has K+1 < Ap ≤ Q0. Therefore, applying exp(ad(s ⊗ f )) to D0, we may assume that D0 = Q0 Xl=K+1 s′ Al ⊗ x Al + d0 for some s′ Al ∈ S and xAl = xa1 1 · · · xas s ∈ O(s; 1) with Alp = l. Now look at D0 and repeat the above process, i.e. take xAl with the smallest p-degree for which s′ Al 6= 0. If Al = (a1, . . . , as) = (p − 1, . . . , p − 1), then D0 is of the desired form. If not, then applying a similar automorphism exp(ad(s ⊗ f )) to D0 Al ⊗ x · · · xp−1 and clear s′ 0 ⊗ xp−1 s′ H-conjugate to D1 = s′ 1 Al. Continuing in this way, we eventually get D0 is H-conjugate to s + d0 for some s′ 0 ∈ S (possibly 0). Therefore, D = D0 + v + u is 0 ⊗ xp−1 1 · · · xp−1 s + d0 + v′ + u for some s′ 0 ∈ S (possibly 0) and v′ ∈ S ⊗ I. This completes the proof. Lemma 3.2.15. Let D1 = s′ s + v′ + z be an element of L, where 0 ∈ S, v′ ∈ S ⊗ I, I is the ideal of O(m; 1) generated by xs+1, . . . , xm with s ≥ 1, and s′ z = d0 + u ∈ N (D) as in Lemma 3.2.7. Then D1 ∈ N (L) if and only if s′ · · · xp−1 0 ∈ N (S). 0 ⊗ xp−1 1 Strategy of the proof. This is a computational proof with the following key steps: Step 1 . Let D1 = s′ 0 ⊗ xp−1 1 · · · xp−1 s + v′ + z be as in the lemma. We show that Dps 1 = zps + (s′ 0 ⊗ (−1)s) + (other terms in S ⊗ m). 91 Set w = s′ 0 ⊗ xp−1 1 · · · xp−1 s + v′. By Lemma 1.1.1, Dps 1 = zps + wps + s−1 Xr=0 upr r , 3.2. Socle involves S (3.25) where ur is a linear combination of commutators in z and w. We first show that wps ∈ S ⊗ m. Then we consider Ps−1 r . By Jacobi identity, we can rearrange each ur so that ur is in the span of [wt, [wt−1, [. . . , [w2, [w1, w] . . . ], where t = ps−r − 1 and r=0 upr each wα, 1 ≤ α ≤ t, is equal to z or to w. We consider all such iterated commutators [wt, [wt−1, [. . . , [w2, [w1, w] . . . ] and show that u0 = (ad z)ps−1(w) = (s′ 0 ⊗ (−1)s) + (other terms in S ⊗ m), and for p − 1 ≤ t = ps−r − 1 ≤ ps−1 − 1, [wt, [wt−1, [. . . , [w2, [w1, w] . . . ] ∈ S ⊗ m. Hence for 1 ≤ r ≤ s − 1, ur ∈ S ⊗ m and upr Note that [z, w] = s′ 0 ⊗ z(xp−1 1 · · · xp−1 s r ∈ S ⊗ m. ) + [z, v′]. To show the above claims, we need to consider the action of z = d0 + u on xp−1 1 · · · xp−1 s ; see parts (i)-(iii). Then we consider commutators in z and w; see parts (iv)-(vi). (i) We show that for any 0 < l < ps − 1, 0 6= dl 0(xp−1 1 · · · xp−1 s ) ∈ ms, where ms is the maximal ideal of O(s; 1). (ii) We show by induction that for any 0 < l < ps − 1, zl(xp−1 1 · · · xp−1 s ) = dl 0(xp−1 1 · · · xp−1 s ) + (other terms in I) ∈ ms ⊕ I. (iii) We show that zps−1(xp−1 1 · · · xp−1 s ) = (−1)s + (other terms in I). (iv) We show that for any 0 < l < ps − 1, (ad z)l(w) ∈ (S ⊗ ms) ⊕ (S ⊗ I), and (ad z)ps−1(w) = (s′ 0 ⊗ (−1)s) + (other terms in S ⊗ m). (v) We show that for any 0 < l < ps − 1, [w, (ad z)l(w)] ∈ S ⊗ I. (vi) We show that for p − 1 ≤ t = ps−r − 1 ≤ ps−1 − 1, [wt, [wt−1, [. . . , [w2, [w1, w] . . . ] ∈ S ⊗ m, where each wα, 1 ≤ α ≤ t, is equal to z or to w. Hence for 1 ≤ r ≤ s − 1, ur ∈ S ⊗ m and upr r ∈ S ⊗ m. 92 3.2. Socle involves S It follows from the above that Dps 1 = zps + (s′ 0 ⊗ (−1)s) + (other terms in S ⊗ m). 1 = zps + (s′ Step 2 . We show that D1 ∈ N (L) if and only if s′ 0 ∈ N (S). By the last step, Dps 0 ⊗ (−1)s) + (other terms in S ⊗ m). By our assumption, z is nilpotent. By Lemma 3.2.8, zps ∈ W (m; 1)(0). Hence zps preserves S ⊗ m. Applying Lemma 3.2.12 with D′ = Dps 0 and f0 = (−1)s, we get for N ≥ 1 (see 1 , d = zps, s0 = s′ (3.18)), Dps+N 1 = zps+N + ((s′ 0)pN ⊗ (−1)spN ) + (other terms in S ⊗ m). Hence for s + N ≫ 0, D1 ∈ N (L) if and only if s′ 0 ∈ N (S). Proof. Step 1 . Let D1 = s′ 0 ⊗ xp−1 1 · · · xp−1 s + v′ + z be as in the lemma. We show that Dps 1 = zps + (s′ 0 ⊗ (−1)s) + (other terms in S ⊗ m). Set w = s′ 0 ⊗ xp−1 1 · · · xp−1 s + v′. By Lemma 1.1.1, Dps 1 = zps + wps + s−1 Xr=0 upr r , (3.26) where ur is a linear combination of commutators in z and w. We first show that wps ∈ S ⊗ m. Since xp−1 )ps = 0. By Lemma 1.1.1, s ∈ m, then (xp−1 · · · xp−1 · · · xp−1 s 1 1 wps = (v′)ps + s−1 Xr=0 pr , ηr where ηr is a linear combination of commutators in s′ 0 ⊗ xp−1 that (v′)ps ∈ S ⊗I. Since v′ ∈ S ⊗I, we can write v′ = Pn i=1 si ⊗gi for some si ∈ S and gi ∈ I. Since O(m; 1) is a local ring, the ideal I is contained in the maximal ideal m of O(m; 1). Hence gp i = 0 for all i. Moreover, [S ⊗ I, S ⊗ I] ⊆ S ⊗ I. By Lemma 1.1.1, we have that that (v′)p ∈ S ⊗ I. Continuing in this way, we get (v′)ps ∈ S ⊗ I. Next · · · xp−1 s and v′. We show 1 we show that Ps−1 0 ⊗ xp−1 s′ · · · xp−1 1 r=0 ηr s ∈ S ⊗ m and v′ ∈ S ⊗ I, we have that pr ∈ S ⊗ I. Since I ⊂ m, we have that mI ⊆ m ∩ I = I. Since [s′ 0 ⊗ xp−1 1 · · · xp−1 s , v′] ∈ [S ⊗ m, S ⊗ I] ⊆ S ⊗ mI ⊆ S ⊗ I. It is clear that [S ⊗ I, S ⊗ I] ⊆ S ⊗ I. So any iterated commutators in s′ 0 ⊗ xp−1 1 · · · xp−1 s and v′ are in S ⊗ I. Since ηr is a linear combination of these commutators, we have 93 3.2. Socle involves S that ηr ∈ S ⊗ I for all 0 ≤ r ≤ s − 1. By a similar argument as above, one can pr ∈ S ⊗ I. Therefore, show that ηr pr ∈ S ⊗ I for all 0 < r ≤ s − 1. Hence Ps−1 r=0 ηr wps = (v′)ps +Ps−1 r=0 ηr Now we consider Ps−1 pr ∈ S ⊗ I ⊂ S ⊗ m. r=0 upr r tators in z and w. By Jacobi identity, we can rearrange each ur so that ur is in the in (3.26), where ur is a linear combination of commu- span of [wt, [wt−1, [. . . , [w2, [w1, w] . . . ], where t = ps−r − 1 and each wα, 1 ≤ α ≤ t, is equal to z or to w. We consider all such iterated commutators and show that u0 = (ad z)ps−1(w) = (s′ 0 ⊗ (−1)s) + (other terms in S ⊗ m), (3.27) and for p − 1 ≤ t = ps−r − 1 ≤ ps−1 − 1, [wt, [wt−1, [. . . , [w2, [w1, w] . . . ] ∈ S ⊗ m. (3.28) Hence for 1 ≤ r ≤ s − 1, ur ∈ S ⊗ m and upr r ∈ S ⊗ m. Recall that z = d0+u, where d0 = ∂1 +xp−1 1 ∂2 + · · ·+xp−1 1 · · · xp−1 s−1 ∂s is a derivation of O(s; 1) and u ∈ (I ∂1 + · · · + I ∂m) ∩ W (m; 1)(p−1); see Lemma 3.2.7 for notations. Note that [z, w] = s′ 0 ⊗ z(xp−1 1 · · · xp−1 s ) + [z, v′]. Since d0 preserves the ideal I and so does z, we have that [z, v′] ∈ [z, S ⊗ I] = S ⊗ z(I) ⊆ S ⊗ I. Hence to show (3.27) and (3.28), we need to consider the action of z on xp−1 . We split our work into the · · · xp−1 1 s following parts (i)-(vi): (i) We show that for any 0 < l < ps − 1, 0 6= dl 0(xp−1 1 · · · xp−1 s ) ∈ ms, where ms is the maximal ideal of O(s; 1). There are two ways to prove this result: (a) using the p-degree of monomials or (b) using the standard degree of monomials and some results on d0. Let us do both ways and see the difference. (a) Let xA = xa1 1 · · · xas if Ap > 0. Consider xp−1 s be any monomial in O(s; 1). Note that xA ∈ ms if and only , where A = (p − 1, . . . , p − 1). Then Ap = ps − 1. · · · xp−1 1 s By Example 3.2.1(3), we know that the degree of d0 with respect to p is −1. Hence for any 0 < l < ps − 1, dl 0(xp−1 ps − 1 − l > 0. Therefore, 0 6= dl 1 0(xp−1 1 · · · xp−1 s ) ∈ ms. · · · xp−1 s ) is a monomial in O(s; 1) with p-degree (b) By Lemma 1.5.1(i) and (iii), we know that dpK 0 = (−1)K(∂K+1 +xp−1 K+1 ∂K+2 + · · · + xp−1 K+1 · · · xp−1 s−1 ∂s) 94 3.2. Socle involves S for all 0 ≤ K ≤ s − 1 and dps 0(xp−1 dl Here we use a similar argument given in [20, p. 151, line -8 and p. 153, line 4]. By ) 6= 0 for any 0 < l < ps − 1. We show that dl 0 = 0. Moreover, dps−1 ) = (−1)s. Hence 0(xp−1 · · · xp−1 · · · xp−1 · · · xp−1 ) ∈ ms. (xp−1 1 0 1 1 s s s (3.3), O(s; 1) is a graded W (s; 1)-module, i.e. O(s; 1) = O(s; 1)0 ⊕ O(s; 1)1 ⊕ · · · ⊕ O(s; 1)s(p−1). It is easy to see that O(s; 1)η ⊆ mη s for any η ≥ 1. Hence xp−1 1 · · · xp−1 s ∈ ms(p−1) s . Recall that W (s; 1) is a free O(s; 1)-module with basis ∂1, . . . , ∂s. Then for any D ∈ W (s; 1) and 0 < c < s(p − 1), we have that Dc(xp−1 1 · · · xp−1 s ) ∈ ms(p−1)−c s . (3.29) Since l < ps−1 = Ps−1 and Ps−1 K=0 aK < s(p − 1). Then K=0(p−1)pK, we have that l = Ps−1 K=0 aKpK, where 0 ≤ aK ≤ p−1 0(xp−1 dl 1 · · · xp−1 s ) = (cid:18) s−1 YK=0(cid:0)dpK 0 (cid:1) aK(cid:19)(xp−1 1 · · · xp−1 s ). Applying (3.29) with D = dpK 0 and c = aK for 0 ≤ K ≤ s − 1, we get 0(xp−1 dl 1 · · · xp−1 s ) = (cid:18) s−1 YK=0 (dpK 0 ) aK(cid:19)(xp−1 1 · · · xp−1 s s(p−1)−Ps−1 ) ∈ m s K=0 aK ⊆ ms. This proves (i). (ii) We show by induction that for any 0 < l < ps − 1, zl(xp−1 1 · · · xp−1 s ) = dl 0(xp−1 1 · · · xp−1 s ) + (other terms in I) ∈ ms ⊕ I. For l = 1, we have that z(xp−1 1 · · · xp−1 s ) = d0(xp−1 1 · · · xp−1 s ) + u(xp−1 1 · · · xp−1 s ). By (i), we know that d0(xp−1 · · · xp−1 ideal of O(m; 1), we have that u(xp−1 1 s 1 ) ∈ ms. Since u ∈ Pm j=1 I ∂j and I is an It is clear that ms ∩ I = {0}. · · · xp−1 ) ∈ I. s Hence the result holds for l = 1. Suppose the result holds for l = r1 < ps − 2, i.e. zr1(xp−1 ) + g for some g ∈ I. Applying z again, we get · · · xp−1 · · · xp−1 ) = dr1 0 (xp−1 1 1 s s zr1+1(xp−1 1 · · · xp−1 s ) = dr1+1 0 (xp−1 1 It is clear that u(cid:0)dr1 d0(g) ∈ I. Since r1 + 1 < ps − 1, it follows from (i) that dr1+1 · · · xp−1 0 (xp−1 1 s s 0 (xp−1 · · · xp−1 ) + d0(g) + u(cid:0)dr1 )(cid:1) + u(g). )(cid:1) + u(g) ∈ I. Since d0 preserves I, we have that · · · xp−1 1 s (xp−1 1 · · · xp−1 s ) ∈ ms. 0 95 Hence the result holds for l = r1 + 1. Therefore, for any 0 < l < ps − 1, the result holds. This proves (ii). 3.2. Socle involves S (iii) We show that zps−1(xp−1 By (ii), we know that zps−2(xp−1 1 1 · · · xp−1 s ) = (−1)s + (other terms in I). · · · xp−1 s ) = dps−2 0 (xp−1 1 · · · xp−1 s ) + g for some g ∈ I. Applying z again and using a similar argument as in (ii), we get zps−1(xp−1 1 · · · xp−1 s ) = dps−1 0 (xp−1 1 · · · xp−1 s ) + g′ for some g′ ∈ I. By Lemma 1.5.1(iii), we know that dps−1 Hence zps−1(xp−1 ) = (−1)s + g′ as desired. This proves (iii). · · · xp−1 0 1 s (xp−1 1 · · · xp−1 s ) = (−1)s. Now we consider commutators in z and w. (iv) We show that for any 0 < l < ps − 1, (ad z)l(w) ∈ (S ⊗ ms) ⊕ (S ⊗ I), and (ad z)ps−1(w) = (s′ 0 ⊗ (−1)s) + (other terms in S ⊗ m). Note that for any 0 < l < ps − 1, (ad z)l(w) = s′ 0 ⊗ zl(xp−1 1 · · · xp−1 s ) + (ad z)l(v′). Since v′ ∈ S ⊗ I and z preserves I, we have that [z, v′] ∈ [z, S ⊗ I] = S ⊗ z(I) ⊆ S ⊗ I. Then an easy induction on l shows that (ad z)l(v′) ∈ S ⊗ I. By (ii), we know that zl(xp−1 ) ∈ ms ⊕ I. Hence · · · xp−1 1 s (ad z)l(w) = s′ 0⊗zl(xp−1 1 · · · xp−1 s )+(ad z)l(v′) ∈ s′ 0⊗(ms⊕I)+S⊗I ⊆ (S⊗ms)⊕(S⊗I). Similarly, (ad z)ps−1(w) = s′ 0 ⊗ zps−1(xp−1 1 · · · xp−1 s ) + (ad z)ps−1(v′). Arguing as above, one can show that (ad z)ps−1(v′) ∈ S ⊗ I. By (iii), we know that zps−1(xp−1 ) = (−1)s + (other terms in I). Since I ⊂ m, we have that · · · xp−1 1 s (ad z)ps−1(w) = (s′ 0 ⊗ (−1)s) + (other terms in S ⊗ m). This proves (iv) and (3.27). (v) We show that for any 0 < l < ps − 1, [w, (ad z)l(w)] ∈ S ⊗ I. By (iv), we know that (ad z)l(w) ∈ (S ⊗ ms) ⊕ (S ⊗ I). Then [w, (ad z)l(w)] =[s′ 1 0 ⊗ xp−1 0 ⊗ xp−1 1 · · · xp−1 s + v′, (ad z)l(w)] · · · xp−1 s + v′, (S ⊗ ms) ⊕ (S ⊗ I)] ⊆ S ⊗ I. ∈[s′ This proves (v). 96 3.2. Socle involves S (vi) We show that for p − 1 ≤ t = ps−r − 1 ≤ ps−1 − 1, [wt, [wt−1, [. . . , [w2, [w1, w] . . . ] ∈ S ⊗ m, where each wα, 1 ≤ α ≤ t, is equal to z or to w. Hence for 1 ≤ r ≤ s − 1, ur ∈ S ⊗ m and upr r ∈ S ⊗ m. Let us consider all such iterated commutators [wt, [wt−1, [. . . , [w2, [w1, w] . . . ]. If w1 = w, then [w1, w] = 0. If w1 = · · · = wt = z, then (iv) implies that (ad z)t(w) ∈ (S ⊗ ms) ⊕ (S ⊗ I) = S ⊗ m. Hence we need to consider commutators (ad w)β(ad z)γ(w), where 1 ≤ γ < t and β > 0. By (v), we know that [w, (ad z)γ(w)] ∈ S ⊗ I. Since w ∈ S ⊗ m, we have that [w, [w, (ad z)γ(w)]] ∈ [S ⊗ m, S ⊗ I] ⊆ [S, S] ⊗ mI ⊆ S ⊗ I. Continuing in this way, we get (ad w)β(ad z)γ(w) ∈ S ⊗ I. If β + γ = t, then we are done. If β + γ < t, then we need to consider [wα, (ad w)β(ad z)γ(w)], where wα = z or w. If wα = z, then z preserves the ideal I. Hence [z, (ad w)β(ad z)γ(w)] ∈ [z, S ⊗ I] = S ⊗ z(I) ⊆ S ⊗ I. If wα = w ∈ S ⊗ m, then by the same reason as above, we get [w, (ad w)β(ad z)γ(w)] ∈ S ⊗ I. Hence [wα, (ad w)β(ad z)γ(w)] ∈ S ⊗ I. If 1 + β + γ = t, then we are done. If 1 + β + γ < t, then we need to consider [wν, [wα, (ad w)β(ad z)γ(w)]], where wν = z or w. Arguing similarly, we get [wν, [wα, (ad w)β(ad z)γ(w)]] ∈ S ⊗ I. Continuing in this way, we eventually get [wt, [wt−1, [. . . , [w2, [w1, w] . . . ] ∈ S ⊗ m for all p − 1 ≤ t = ps−r − 1 ≤ ps−1 − 1. Since ur, 1 ≤ r ≤ s − 1, is in the span of [wt, [wt−1, [. . . , [w2, [w1, w] . . . ], we have that ur ∈ S ⊗ m for all 1 ≤ r ≤ s − 1. By Lemma 1.1.1 and [S ⊗ m, S ⊗ m] ⊆ S ⊗ m, one can show that upr r ∈ S ⊗ m for all 1 ≤ r ≤ s − 1. This proves (vi) and (3.28). It follows from (3.27) and (3.28) that Dps 1 =zps + wps + s−1 Xr=0 upr r = zps + wps + (ad z)ps−1(w) + s−1 Xr=1 upr r =zps + (s′ 0 ⊗ (−1)s) + (other terms in S ⊗ m). 97 3.2. Socle involves S Step 2 . We show that D1 is a nilpotent element of L if and only if s′ 0 is a nilpotent element of S. By step 1, we know that Dps 1 = zps + (s′ 0 ⊗ (−1)s) + (other terms in S ⊗ m). By our assumption, z is nilpotent. By Lemma 3.2.8, we know that zps ∈ W (m; 1)(0). Hence zps preserves S ⊗ m. Applying Lemma 3.2.12 with D′ = Dps 1 , d = zps, s0 = s′ 0 and f0 = (−1)s, we get for N ≥ 1 (see (3.18)), Dps+N 1 = zps+N + ((s′ 0)pN ⊗ (−1)spN ) + (other terms in S ⊗ m). Hence for s + N ≫ 0, D1 is a nilpotent element of L if and only if s′ 0 is a nilpotent element of S. This completes the proof. 3.2.3 The irreducibility of N (L) We are now ready to prove that the nilpotent variety of L = (S ⊗ O(m; 1)) ⋊ D is irreducible. Recall our assumptions that D is a restricted transitive subalgebra of W (m; 1) such that N (D) is irreducible, S is a simple restricted Lie algebra such that all its derivations are inner and N (S) is irreducible. Theorem 3.2.3. The variety N (L) is irreducible. Proof. Let D be an element of L = (S ⊗ O(m; 1)) ⋊ D. Then we can write D = m(p−1) Xi=0 si ⊗ fi + d, (3.30) where si ∈ S, fi ∈ O(m; 1) with deg fi = i, and d ∈ D. Note that the surjective Lie algebra homomorphism ψ : L → D, D 7→ d induces a surjective morphism ψ : N (L) → N (D). By our assumption, N (D) is irreducible. By Theorem 1.5.1, N (L) is equidimensional. If we can prove that all fibres of ψ are irreducible and have the same dimension, then the irreducibility of N (L) follows from Lemma 2.0.1. Since D is a restricted transitive subalgebra of W (m; 1), i.e. D + W (m; 1)(0) = W (m; 1), there exist elements in D which are not in W (m; 1)(0). Hence for d ∈ N (D), 98 we have two cases to consider: either d ∈ W (m; 1)(0) or d /∈ W (m; 1)(0). Let us compute ψ−1(d) in each case. 3.2. Socle involves S Case 1 : d ∈ N (D) and d ∈ W (m; 1)(0). Let D = Pm(p−1) si ⊗fi +d be an element of L such that ψ(D) = d; see (3.30) for notations. By Lemma 3.2.12, we know that i=0 D ∈ N (L) if and only if s0 ∈ N (S). As a result, ψ−1(d) = N (S) ⊗ 1 + S ⊗ m + d ∼= N (S) ⊗ 1 + S ⊗ m. Since S ⊗ m is irreducible and N (S) is irreducible by our assumption, it follows that all fibres ψ−1(d) are irreducible. Moreover, dim ψ−1(d) = dim(N (S) ⊗ 1) + dim(S ⊗ m) = (cid:0) dim(S ⊗ 1) − MT(S)(cid:1) + dim(S ⊗ m) = dim(S ⊗ O(m; 1)) − MT(S) (by Theorem 1.5.1(iii)) = pm dim S − MT(S). Case 2 : d ∈ N (D) and d /∈ W (m; 1)(0). Step 1 . We compute ψ−1(d) for all d ∈ N (D) with d /∈ W (m; 1)(0). Then we deduce that they are irreducible. By Lemma 3.2.7 and Lemma 3.2.9, we may assume that where d = d0 + u, (3.31) d0 = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 s−1 ∂s with 1 ≤ s ≤ m, u ∈ (I ∂1 + · · · + I ∂m) ∩ W (m; 1)(p−1) and I is the ideal of O(m; 1) generated by xs+1, . . . , xm. si ⊗ fi + d be an element of L such that ψ(D) = d; see (3.30) Let D = Pm(p−1) i=0 and (3.31) for notations. Recall the subgroup H of Aut(L) which is generated by all exp(ad(s ⊗ f )), where s ⊗ f ∈ S ⊗ m; see Lemma 3.2.13. Note that H is a connected algebraic group with S ⊗ m ⊆ Lie(H). Since exp(ad(s ⊗ f )) = Pp−1 j!(ad(s ⊗ f ))j and D is of the form (3.30), it is easy to see that H stabilizes the fibres of ψ. By j=0 1 Lemma 3.2.14, we know that D is conjugate under H to D1 = s′ 0 ⊗ xp−1 1 · · · xp−1 s + v′ + d, 99 3.2. Socle involves S where s′ 0 ∈ S (possibly 0), v′ ∈ S ⊗ I and d = d0 + u as in (3.31). Then D is nilpotent if and only if D1 is nilpotent. By Lemma 3.2.15, we know that D1 ∈ N (L) if and only if s′ 0 ∈ N (S). Hence all fibres of ψ have the form H.(cid:0)N (S) ⊗ xp−1 1 · · · xp−1 s + S ⊗ I + d(cid:1) ∼= H.(cid:0)N (S) ⊗ xp−1 1 · · · xp−1 s + S ⊗ I(cid:1). Since H is connected, N (S) and S ⊗ I are irreducible, it follows that all fibres ψ−1(d) are irreducible. Step 2 . We show that all fibres of ψ have the same dimension. By case 1, we know that dim ψ−1(d) = pm dim S − MT(S) for all d ∈ N (D) with d ∈ W (m; 1)(0). In particular, dim ψ−1(0) = pm dim S − MT(S). (3.32) To finish the proof we just need to show that dim ψ−1(d) = dim ψ−1(0) for all d ∈ N (D) with d /∈ W (m; 1)(0). Step 2(i). We first show that dim ψ−1(0) ≥ dim ψ−1(d) for all d ∈ N (D) with d /∈ W (m; 1)(0). Suppose the contrary, i.e. dim ψ−1(0) < dim ψ−1(d) for some d ∈ N (D) with d /∈ W (m; 1)(0). By Theorem 1.6.7, the set W1 = (cid:8)x ∈ N (L) dim ψ−1( ψ(x)) ≥ r(cid:9) is Zariski closed in N (L) for every r ∈ N0. We now take r = dim ψ−1(0) + 1. If W1 is empty, then we are done. If W1 is nonempty, then it contains w + d ∈ N (L) with w ∈ S ⊗ m such that dim ψ−1( ψ(w + d)) ≥ r. Note that for all λ ∈ k∗, ψ(λ(w + d)) = λ ψ(w + d). Then ψ−1(cid:0) ψ(λ(w + d))(cid:1) = λ ψ−1(cid:0) ψ(w + d)(cid:1). So W1 is k∗-stable. Since W1 is Zariski closed, it contains 0. But this contradicts our choice of r. As a result, dim ψ−1(0) ≥ dim ψ−1(d) (3.33) 100 3.2. Socle involves S for all d ∈ N (D) with d /∈ W (m; 1)(0). Step 2(ii). Next we show that dim ψ−1(0) ≤ dim ψ−1(d) for all d ∈ N (D) with d /∈ W (m; 1)(0). Consider the morphism θ : H ×(cid:0)N (S) ⊗ xp−1 1 · · · xp−1 s + S ⊗ I + d(cid:1) → ψ−1(d) (h, D1) 7→ h(D1). By the work in step 1, we know that any point in the fibre ψ−1(d) has the form 0 ⊗ xp−1 1 θ(cid:0)h, s′ · · · xp−1 s + v′ + d(cid:1) for some h ∈ H, s′ 0 ∈ N (S) and v′ ∈ S ⊗ I. As H acts on the fibre it preserves smooth points. Hence we may assume that h = 1. Instead of computing dim ψ−1(d) which is difficult, we can compute the differential of θ at a smooth point of the fibre with h = 1. Since N (S) is irreducible, the set of smooth points in N (S) is nonempty. Let s′ 0 be a smooth point of N (S). Then dim Ts′ 0(N (S)) = dim N (S). Take D1 = s′ 0 ⊗ xp−1 1 · · · xp−1 s + v′ + d ∈ ψ−1(d). Then the differential of θ at (1, D1) is (dθ)(1,D1) : Lie(H) ⊕(cid:0)Ts′ 0(N (S)) ⊗ xp−1 1 · · · xp−1 s + S ⊗ I(cid:1) → TD1( ψ−1(d)) (X, Y ) 7→ [X, D1] + Y. Since D1 is a smooth point, this implies that dim TD1( ψ−1(d)) = dim ψ−1(d). In order to show that dim ψ−1(0) ≤ dim ψ−1(d), we just need to show that dim ψ−1(0) ≤ dim TD1( ψ−1(d)). It is clear that Ts′ 0(N (S)) ⊗ xp−1 1 · · · xp−1 s + S ⊗ I ⊆ Im((dθ)(1,D1)). (3.34) Moreover, [Lie(H), D1] ⊆ Im((dθ)(1,D1)). Since S ⊗ m ⊆ Lie(H), we have that [S ⊗ m, D1] = [S ⊗ m, s′ 0 ⊗ xp−1 1 · · · xp−1 s + v′ + d] ⊆ Im((dθ)(1,D1)) (3.35) Observe (3.34) and (3.35) carefully. We see that Im((dθ)(1,D1)) contains the follow- ing subspaces: 101 · · · xp−1 s which has dimension equal to dim N (S). 3.2. Socle involves S (1) Ts′ 0(N (S)) ⊗ xp−1 1 (2) S ⊗ I. (3) [S ⊗ m, s′ 0 ⊗ xp−1 1 · · · xp−1 s + v′ + d]. Look at the last two subspaces. Note that m = ms ⊕ I, where ms is the maximal ideal of O(s; 1). Moreover, mI ⊆ I. Since S ⊗ I is in Im((dθ)(1,D1)), [S ⊗ m, s′ 0 ⊗ xp−1 1 · · · xp−1 s ] = [S ⊗ (ms ⊕ I), s′ 0 ⊗ xp−1 1 · · · xp−1 s ] ⊆ S ⊗ I, and [S ⊗ m, v′] ⊆ [S ⊗ m, S ⊗ I] ⊆ S ⊗ mI ⊆ S ⊗ I, we see that Im((dθ)(1,D1)) contains S ⊗ I + [S ⊗ m, s′ 0 ⊗ xp−1 1 · · · xp−1 s + v′ + d] =S ⊗ I + [S ⊗ m, d] =S ⊗ I + S ⊗ d(m). Recall (3.31) that d = d0 + u, where d0 = ∂1 +xp−1 1 ∂2 + · · · + xp−1 1 · · · xp−1 s−1 ∂s is a derivation of O(s; 1) and u ∈ (I ∂1 + · · · + I ∂m) ∩ W (m; 1)(p−1). Since u(m) ⊆ I, we have that S ⊗ I + S ⊗ d(m) = S ⊗ I + S ⊗ d0(m). Since m = ms ⊕ I, d0(ms) ⊂ O(s; 1), d0(I) ⊆ I and O(s; 1)∩I = {0}, we have that d0(m) = d0(ms)⊕d0(I). Since d0(I) ⊆ I, we have that S ⊗I +S ⊗d(m) = S ⊗I +S ⊗d0(m) = (S ⊗I)⊕(S ⊗d0(ms)) ∼= S ⊗(I ⊕d0(ms)). (3.36) By Lemma 3.2.11 (see (3.17)), we know the subspace M = I+d(m) = I⊕d0(ms) has the property that M ⊕kxp−1 s = O(m; 1). By (3.36), we see that the sum of the last two subspaces equals S ⊗ M. This subspace of dimension (pm − 1) dim S is contained · · · xp−1 1 in Im((dθ)(1,D1)) and the complement of the first subspace Ts′ 0(N (S)) ⊗ xp−1 1 · · · xp−1 s . Therefore, dim Im((dθ)(1,D1)) ≥ dim(Ts′ 0(N (S)) ⊗ xp−1 1 · · · xp−1 s ) + dim(S ⊗ M) = dim N (S) + (pm − 1) dim S = (dim S − MT(S)) + (pm − 1) dim S = pm dim S − MT(S) = dim ψ−1(0) (see (3.32)). 102 3.2. Socle involves S Since Im((dθ)(1,D1)) ⊆ TD1( ψ−1(d)) we have that dim ψ−1(0) ≤ dim TD1( ψ−1(d)) = dim ψ−1(d). (3.37) It follows from (3.33) and (3.37) that all fibres of ψ have the same dimension. By case 1 and 2, we see that all fibres of ψ are irreducible and have the same dimension. Hence N (L) is irreducible by Lemma 2.0.1. This completes the proof. Remark 3.2.3. 1. We verified Premet's conjecture for a class of semisimple re- stricted Lie algebras L = (S ⊗ O(m; 1)) ⋊ D under the assumptions that S is a simple restricted Lie algebra over k with ad S = Der S and N (S) is irreducible, D is a restricted transitive subalgebra of W (m; 1) with N (D) is irreducible. 2. A similar argument works for Lt i=1(Si ⊗ O(mi; 1)) ⋊ (IdSi ⊗Di), where (i) each Si is a simple restricted Lie algebra such that ad Si = Der Si and N (Si) is irreducible, and (ii) each Di is a restricted transitive subalgebra of W (mi; 1) such that N (Di) is irreducible. 3. There are further cases to consider such as ad S ( Der S and other semisimple restricted Lie algebras which are not of the form given in Theorem 1.6.3. It is unclear to the author how to tackle these problems. 103 Chapter 4 The nilpotent variety of W (1; n)p is irreducible In this chapter we assume that k is an algebraically closed field of characteristic p > 3, and n ∈ N≥2. We are interested in the minimal p-envelope W (1; n)p of the Zassen- haus algebra W (1; n). This restricted Lie algebra is semisimple. Recent studies have shown that the variety N (W (1; n)) := N (W (1; n)p) ∩ W (1; n) is reducible [37, The- orem 4.8(i)]. So investigating the variety N (W (1; n)p) becomes critical for verifying Premet's conjecture. Note that this chapter contains the main result of the research paper [3] written by the author of this thesis. This chapter is organized as follows. We first recall some basic results on W (1; n) and W (1; n)p. Then we study some nilpotent elements of W (1; n)p and identify an irre- ducible component of N (W (1; n)p). Finally, we prove that N (W (1; n)p) is irreducible. The proof is similar to Premet's proof for the Jacobson-Witt algebra W (n; 1); see The- orem 1.5.2. But the (n + 1)-dimensional subspace V used in W (n; 1) has no obvious analogue for W (1; n)p. Therefore, a new V is constructed using the original definition of W (1; n) due to H. Zassenhaus. In general, constructing analogues of V for the min- imal p-envelopes of W (n; m), where m = (m1, . . . , mn) and mi > 1 for some i, would enable one to check Premet's conjecture for this class of restricted Lie algebras. 104 4.1 Preliminaries 4.1.1 W (1; n) and W (1; n)p 4.1. Preliminaries power algebra O(1; n) has a k-basis {x(a) = 1 Let k be an algebraically closed field of characteristic p > 3 and n ∈ N. The divided a! xa 0 ≤ a ≤ pn − 1}, and the product in a (cid:1)x(a+b) if 0 ≤ a + b ≤ pn − 1 and 0 otherwise. In the O(1; n) is given by x(a)x(b) = (cid:0)a+b following, we write x(1) as x. It is straightforward to see that O(1; n) is a local algebra with the unique maximal ideal m spanned by all x(a) such that a ≥ 1. A system of divided powers is defined on m, f 7→ f (r) ∈ O(1; n), where r ≥ 0; see Definition 1.4.2. A derivation D of O(1; n) is called special if D(x(a)) = x(a−1)D(x) for 1 ≤ a ≤ pn−1 and 0 otherwise; see Definition 1.4.3. The set of all special derivations of O(1; n) forms a Lie subalgebra of Der O(1; n) denoted L = W (1; n) and called the Zassenhaus algebra. When n = 1, L coincides with the Witt algebra W (1; 1) := Der O(1; 1), a simple and restricted Lie algebra. When n ≥ 2, L provides the first example of a simple, non-restricted Lie algebra; see Theorem 1.4.2. From now on, we always assume that n ≥ 2. The Zassenhaus algebra L admits an O(1; n)-module structure via (f D)(x) = f D(x) for all f ∈ O(1; n) and D ∈ L. Since each D ∈ L is uniquely determined by its effect on x, it is easy to see that L is a free O(1; n)-module of rank 1 generated by the special derivation ∂ such that ∂(x(a)) = x(a−1) if 1 ≤ a ≤ pn − 1 and 0 otherwise; see Theorem 1.4.2. Hence the Lie bracket in L is given by [x(i) ∂, x(j) ∂] = j (cid:1)(cid:1)x(i+j−1) ∂ if 1 ≤ i + j ≤ pn and 0 otherwise; see formula (1.4). i (cid:1) −(cid:0)i+j−1 (cid:0)(cid:0)i+j−1 There is a Z-grading on L, i.e. L = Lpn−2 Lpn−2 j≥i kdj for −1 ≤ i ≤ pn − 2. Then this Z-grading induces a natural filtration i=−1 kdi with di := x(i+1)∂. Put L(i) := L = L(−1) ⊃ L(0) ⊃ L(1) ⊃ · · · ⊃ L(pn−2) ⊃ 0 (4.1) on L. It is known that for i ≥ 0, dp i = di, if i = 0, dpi, if i = pt − 1 for some 1 ≤ t ≤ n − 1, (4.2) 0, otherwise.   In particular, L(0) is a restricted Lie subalgebra of Der O(1; n) and L(1) = nil (L(0)); see [37, p. 3]. We show that this implies that all nonzero tori in L(0) have dimension 1. 105 4.1. Preliminaries Consider the surjective map π : L(0) ։ L(0)/L(1). Let t be any nonzero torus in L(0). Let πt be the restriction of π to t. Then Ker πt ⊂ L(1). Since L(1) is a nilpotent p-ideal of L(0), we have that Ker πt = 0. Hence any nonzero torus t in L(0) has dimension 1; see [32, Sec. 2]. As an example, kd0 = kx ∂ is a 1-dimensional torus in L(0). Note that the Zassenhaus algebra L has another presentation. Let q = pn and let Fq ⊂ k be the set of all roots of xq − x = 0. This is a finite field with q elements. Then L has a k-basis {eα α ∈ Fq} with the Lie bracket given by [eα, eβ] = (β − α)eα+β [30, Theorem 7.6.3(1)]. We will use this presentation in Sec. 4.2.3. It is also useful to mention that there is an embedding from L into the Jacobson- Witt algebra W (n; 1); see [37, Lemma 3.1 and Proposition 4.3]. Recall that W (n; 1) is the derivation algebra of O(n; 1), where O(n; 1) = k[X1, . . . , Xn]/(X p 1 , . . . , X p n) is the truncated polynomial ring in n variables. For each 1 ≤ i ≤ n, we write xi for the image of Xi in O(n; 1). Note that W (n; 1) is a free O(n; 1)-module of rank n generated by the partial derivatives ∂1, . . . , ∂n such that ∂i(xj) = δij for all 1 ≤ i, j ≤ n; see Sec. 1.5 for details. The Zassenhaus algebra L is the set of all special derivations of the divided power algebra O(1; n). Note that O(1; n) has a k-basis {x(a) 0 ≤ a ≤ pn − 1}. For each 0 ≤ a ≤ pn − 1, let a = Pn−1 a. Define K=0 aKpK, 0 ≤ aK ≤ p − 1, be the p-adic expansion of φ : O(1; n) → O(n; 1) xai−1 i ai−1! x(a) 7→ n Yi=1 ; (4.3) see [37, (3.1.1)]. Then φ is an algebra isomorphism and it induces the following Lie algebra isomorphism: ϕ : Der O(1; n) ∼−→ W (n; 1) = Der O(n; 1) D 7→ ϕ(D), (4.4) where (ϕ(D))(u) = φ(cid:0)D(φ−1(u))(cid:1) for all u ∈ O(n; 1). Moreover, ϕ(Dp) = ϕ(D)p for all D ∈ Der O(1; n); see [37, (3.1.2)]. Since L is a Lie subalgebra of Der O(1; n), the above ϕ induces an embedding ι = ϕL : L ֒→ W (n; 1); (4.5) see [37, (4.3.2)]. More precisely, ι is induced by φ defined in (4.3). By a direct 106 4.1. Preliminaries computation, we have that ι(∂) = D1 = ∂1 + n−1 Xl=1 (−1)lxp−1 1 · · · xp−1 l ∂l+1 . Note that D1 has a similar expression to D in Lemma 1.5.1. We will use the above embedding ι in the sketch proof of Lemma 4.2.2. Let Lp = W (1; n)p denote the p-envelope of L ∼= ad L in Der L. This semisimple By Theorem 1.4.4, we have that Lp = L + Pn−1 restricted Lie algebra is referred to as the minimal p-envelope of L; see Remark 1.2.1. i=1 k∂pi. Here we identify L with as 0. Then dim Lp = pn + (n − 1). By formula (1.4), we , x(a) ∂] = x(a−pi) ∂ ad L ⊂ Der L and regard ∂pn have that for any 1 ≤ i ≤ n − 1 and 0 ≤ a ≤ pn − 1, the brackets [∂pi if a ≥ pi and 0 otherwise. Let N denote the variety of nilpotent elements in Lp. It is well known that N is Zariski closed in Lp. One should note that the maximal dimension of toral subalgebras in Lp equals n [30, Theorem 7.6.3(2)]. Moreover, Lp possesses a toral Cartan subalge- bra; see [19, p. 555]. Hence the set of all semisimple elements of Lp is Zariski dense in Lp; see [18, Theorem 2]. It follows from these facts and Theorem 1.5.1 that there exist nonzero homogeneous polynomial functions ϕ0, . . . , ϕn−1 on Lp such that N coincides with the set of all common zeros of ϕ0, . . . , ϕn−1. The variety N is equidimensional of dimension pn − 1. Furthermore, any D ∈ N satisfies Dpn = 0. (4.6) 4.1.2 The automorphism group G An automorphism Φ of O(1; n) is called admissible if Φ(f (r)) = Φ(f )(r) for all f ∈ m and r ≥ 0. By [36, Lemma 8], this is equivalent to the condition that Φ(x(pj )) = Φ(x)(pj ) for any 1 ≤ j ≤ n−1. Let G denote the group of all admissible automorphisms of O(1; n). It is well known that G is a connected algebraic group, and each Φ ∈ G is uniquely determined by its effect on x. By [36, Theorem 2], an assignment Φ(x) := Ppn−1 i=1 αix(i) with αi ∈ k such that α1 6= 0 and αpj = 0 for 1 ≤ j ≤ n − 1 extends to an admissible automorphism of O(1; n). Conversely, if Φ is an admissible automorphism of O(1; n), then Φ(x) has to be of this form. Hence dim G = pn − n. Any automorphism of the Zassenhaus algebra L is induced by a unique admissible 107 4.1. Preliminaries automorphism Φ of O(1; n) via the rule DΦ = Φ ◦ D ◦ Φ−1, where D ∈ L [26, Theo- rem 12.8]. So from now on, we shall identify G with the automorphism group of L. It is known that G respects the natural filtration of L. In [32, Sec. 1], Tyurin stated explicitly that if Φ ∈ G is such that Φ(x) = y, then Φ(g(x)∂) = (y′)−1g(y)∂ for any g(x) ∈ O(1; n). Extend this by defining Φ(∂ pi ) = Φ(∂)pi for 1 ≤ i ≤ n − 1, one gets an automorphism of Lp. It follows from the above description of G that Lie(G) ⊆ L(0). More precisely, Lemma 4.1.1. The set {di = x(i+1)∂ 0 ≤ i ≤ pn − 2 and i 6= pl − 1 for 1 ≤ l ≤ n − 1} forms a k-basis of Lie(G). Proof. Let ψ : A1 → G be the map defined by t 7→ (x 7→ x + tx(i+1)), where 0 ≤ i ≤ pn − 2 and i 6= pl − 1 for 1 ≤ l ≤ n − 1. It is easy to check that ψ is a morphism of algebraic varieties. Note that ψ(0) = Id. So the differential of ψ at 0 is the map d0ψ : k → Lie(G). Hence d0ψ(k) ⊆ Lie(G). Let us compute d0ψ(k). The morphism ψ sends A1 to the set of admissible auto- morphisms X = {Φt t ∈ A1}, where Φt(x) = x + tx(i+1). Since Φt is uniquely determined by its effect on x and "admissible" is equivalent to the condition that Φt(x(pj )) = Φt(x)(pj ) for all 1 ≤ j ≤ n − 1 (see Sec. 4.1.2), we have that Φt(x(pj )) = (x + tx(i+1))(pj ) = x(pj ) + tx(pj −1)x(i+1) + terms of higher degree in t for all 1 ≤ j ≤ n − 1; see Definition 1.4.2(iv). Now consider d0ψ : k → TId(X). Since X ⊂ G is closed in G, we have that TId(X) ⊆ Lie(G). Passing to TId(X), i.e. calculating ∂ ∂ t (Φt(x))t=0 and ∂ ∂ t(Φt(x(pj )))t=0, we get x 7→ x(i+1), x(pj ) 7→ x(pj −1)x(i+1). The above results are the same for di = x(i+1) ∂ acting on x and x(pj ), respectively. Hence di ∈ TId(X) ⊆ Lie(G). Note that the set {di = x(i+1)∂ 0 ≤ i ≤ pn − 2 and i 6= pl − 1 for 1 ≤ l ≤ n − 1} consists of pn − n linearly independent vectors. Since dim Lie(G) = dim G = pn − n, they form a basis of Lie(G). This completes the proof. 108 4.2. The variety N 4.2 The variety N 4.2.1 Some elements in N In Sec. 4.1.1, we observed that any elements of L(1) are nilpotent, but they do not tell us much information about N . The interesting nilpotent elements are contained in the i=0 αi∂pi +f (x)∂ for some f (x) ∈ m and αi ∈ k with at least one αi 6= 0. In this section, we study elements i=0 αi∂pi + f (x)∂ (not necessarily nilpotent) and we show that D is conjugate under G to an element in a nice form; see complement of L(1) in N , denoted N \L(1). They are of the formPn−1 of this form in the following way: take D = Pn−1 Lemma 4.2.1 and Lemma 4.2.4. Then we assume that D is nilpotent, i.e. Dpn = 0 by (4.6). We do some calculations to check if Dpn−1 ∈ L(0). If Dpn−1 /∈ L(0), then we show further that D is conjugate under G to an element in a nicer form; see Lemma 4.2.2, Lemma 4.2.5 and Corollary 4.2.1. In doing this, we get the results required to prove Proposition 4.2.1 in the next section, i.e. (cid:8)D ∈ N Dpn−1 /∈ L(0)(cid:9) = G.(∂ +k ∂p + · · · + k ∂pn−1 ). Let us begin with elements of the form α0 ∂ +f (x) ∂, where α0 6= 0 and f (x) ∈ m. In [37], Y.-F. Yao and B. Shu proved the following: Lemma 4.2.1. [37, Proposition 4.1] Let D = α0 ∂ +f (x) ∂ be an element of L ⊂ Lp, i=1 lix(pi−1) ∂ for where α0 6= 0 and f (x) ∈ m. Then D is conjugate under G to ∂ +Pn some li ∈ k. Sketch of proof. Take D as in the lemma. Then we can write D = Ppn−1 for some αi ∈ k with α0 6= 0. Let Φ ∈ G be such that Φ(x) = α0x. Then i=0 αix(i) ∂ Φ(D) = ∂ + pn−1 Xi=1 αi−1 0 αix(i) ∂ . So we may assume that D = ∂ + pn−1 Xi=1 βix(i) ∂ = ∂ +β1x ∂ +β2x(2) ∂ + · · · + βpn−1x(pn−1) ∂ for some βi ∈ k. Let Φ0 ∈ G be such that Φ0(x) = x + β1x(2). Then one can check that Φ0(D) = Φ0(cid:0) ∂ + pn−1 Xi=1 2x(2) ∂ +β′ 3x(3) ∂ + · · · + β′ pn−1x(pn−1) ∂ βix(i) ∂(cid:1) = ∂ +β′ 109 4.2. The variety N for some β′ 2, . . . , β′ pn−1 ∈ k. In general, if Φi−1 · · · Φ0(D) = ∂ +µi+1x(i+1) ∂ +µi+2x(i+2) ∂ + · · · + µpn−1x(pn−1) ∂ for 1 ≤ i ≤ p − 3, then take Φi ∈ G with Φi(x) = x + µi+1x(i+2). Applying Φi to the above, we get ΦiΦi−1 · · · Φ0(D) = ∂ +µ′ i+2x(i+2) ∂ +µ′ i+3x(i+3) ∂ + · · · + µ′ pn−1x(pn−1) ∂ for some µ′ i+2, . . . , µ′ pn−1 ∈ k. Consequently, we get Φp−3Φp−4 · · · Φ0(D) = ∂ +νp−1x(p−1) ∂ +νpx(p) ∂ + · · · + νpn−1x(pn−1) ∂ for some νp−1, . . . νpn−1 ∈ k. Set Φp−2 = Id and take Φp−1 ∈ G with Φp−1(x) = x + νpx(p+1). Then one can check that Φp−1Φp−2Φp−3 · · · Φ0(D) = ∂ +νp−1x(p−1) ∂ +ν′ p+1x(p+1) ∂ + · · · + ν′ pn−1x(pn−1) ∂ for some νp−1, ν′ p+1, . . . , ν′ pn−1 ∈ k. Continuing in this way, we finally get Φpn−3Φpn−4 · · · Φ0(D) = ∂ + n Xi=1 lix(pi−1) ∂ for some li ∈ k. This completes the sketch of proof. Then Y.-F Yao and B. Shu assumed that D is nilpotent and they proved that Lemma 4.2.2. [37, Proposition 4.3] Let D = α0 ∂ +f (x) ∂ be a nilpotent element of L ⊂ Lp, where α0 6= 0 and f (x) ∈ m. Then (i) Dpn−1 /∈ L(0). (ii) D is conjugate under G to ∂. Sketch of proof. (i) Take D as in the lemma. Then we can write D = Ppn−1 for some αi ∈ k with α0 6= 0. By Jacobson's formula, one can show that i=0 αix(i) ∂ Dpn−1 = αpn−1 0 ∂pn−1 + n−2 Xj=0 j ∂pj α′ +w for some α′ j ∈ k and w ∈ L(0). Since α0 6= 0, this implies that Dpn−1 6∈ L(0). This proves (i). 110 ∂ +Pn (ii) The proof splits into three steps: Step 1 . Take D as in the lemma. By Lemma 4.2.1, D is conjugate under G to 4.2. The variety N i=1 lix(pi−1) ∂ for some li ∈ k. Since D is nilpotent, it follows from (4.6) that n lix(pi−1) ∂(cid:1)pn Xi=1 We show that this implies that ∂ +Pn i=1 lix(pi−1) ∂ is conjugate under G to ∂. Here we will embed the Zassenhaus algebra L into the Jacobson-Witt algebra W (n; 1) and use (cid:0) ∂ + = 0. A. Premet's results on N (W (n; 1)). In Sec. 4.1.1, we have described the construction process of this embedding. Recall the algebra isomorphism φ : O(1; n) → O(n; 1) = k[X1, . . . , Xn]/(X p 1 , . . . , X p n) and the induced Lie algebra isomorphism ϕ : Der O(1; n) ∼−→ W (n; 1) = Der O(n; 1); see (4.3) and (4.4) for their definitions. Then ϕ gives rise to the following embedding: ι = ϕL : L ֒→ W (n; 1); see (4.5). It follows from the definition of φ that n−1 ι(∂) = D1 = ∂1 + (−1)lxp−1 1 Xl=1 (−1)ilixp−1 1 · · · xp−1 i ∂1 . · · · xp−1 l ∂l+1, and (4.7) n n ι(cid:0) ∂ + Xi=1 lix(pi−1) ∂(cid:1) = D1 + Xi=1 Note that D1 has a similar expression to D in Lemma 1.5.1. Since ϕ(Dp) = ϕ(D)p for all D ∈ Der O(1; n), we have that n n Since (cid:0) ∂ +Pn Xi=1 ϕ(cid:0)(∂ + i=1 lix(pi−1) ∂(cid:1)pn (cid:0)D1 + n Xi=1 · · · xp−1 i i.e. D1 +Pn i=1(−1)ilixp−1 1 lix(pi−1) ∂)pn(cid:1) = (cid:0)D1 + (−1)ilixp−1 1 Xi=1 · · · xp−1 i . ∂1(cid:1)pn = 0, the above implies that (−1)ilixp−1 1 · · · xp−1 i ∂1(cid:1)pn = 0, ∂1 is a nilpotent element of W (n; 1). 111 4.2. The variety N Step 2 . Set D(n) = D1 + n Xi=1 (−1)ilixp−1 1 · · · xp−1 i ∂1 . We show that D(n)pn = 0 implies that D(n) is conjugate under Aut(W (n; 1)) to D1. For n = 1, D1 = ∂1 and D(1) = ∂1 −l1xp−1 show that D(1)p = l1(1 − l1xp−1 ∂1. By considering D(1)p(x1), it is easy to ) ∂1. By our assumption, D(1)p = 0. Since 1 − l1xp−1 1 1 1 is invertible in O(n; 1), this implies that l1 = 0. Hence D(1) = ∂1 = Id(D1). Suppose now n ≥ 2. Then one can check that D(n)pn−1 (xn) =D(n)pn−1−1((−1)n−1xp−1 1 · · · xp−1 n−1) ≡(∂1 −xp−1 1 ∂2 + · · · + (−1)n−2xp−1 1 · · · xp−1 n−2 ∂n−1)pn−1−1 · ((−1)n−1xp−1 1 · · · xp−1 n−1) (mod M) ≡ ± 1 (mod M) (by Lemma 1.5.1(iii)). Here M denotes the unique maximal ideal of O(n; 1) generated by x1, . . . , xn. It follows i=1 fi ∂i fi ∈ M for all i}. Since D(n)pn = 0 and that D(n)pn−1 /∈ W (n; 1)(0) = {Pn D(n)pn−1 /∈ W (n; 1)(0), it follows from Lemma 1.5.3 that D(n) ∈ Aut(W (n; 1)).D1. Step 3 . Let u1, u2 ∈ L. Then [37, Lemma A.3] states that u1, u2 are in the same G-orbit if and only if ι(u1), ι(u2) are in the same Aut(W (n; 1))-orbit. Set u1 = i=1 lix(pi−1) ∂. By (4.7), ι(u1) = D(n). Set u2 = ∂. By (4.7), ι(u2) = D1. Since i=1 lix(pi−1) ∂ ∈ G. ∂. ∂ +Pn D(n) ∈ Aut(W (n; 1)).D1, the above result implies that ∂ +Pn This proves (ii). Now we consider the other elements of N \ L(1), i.e. elements of the form +g(x) ∂, where 1 ≤ t ≤ n − 1, βi ∈ k and g(x) ∈ m. We want to ∂pt i=0 βi ∂pi +Pt−1 show that they are conjugate under G to elements in nice forms. For that, we need a result proved by S. Tyurin which tells us how admissible automorphisms of O(1; n) with identical linear part act on these elements. Lemma 4.2.3. [32, Theorem 1] Let Φ ∈ G be such that Φ(x) = y = x +Ppn−1 where νj ∈ k with νpi = 0 for 1 ≤ i ≤ n − 1. Then j=2 νjx(j), Φ(∂) = (y′)−1 ∂ ≡ ∂ (mod L(0)), and Φ(∂pi ) = ∂pi −(y′)−1(∂pi y) ∂ ≡ ∂pi (mod L(0)) for 1 ≤ i ≤ n − 1. 112 4.2. The variety N Hence for any D = ∂pt + Pt−1 g(x) ∈ m, i=0 βi∂pi + g(x)∂ ∈ Lp with 1 ≤ t ≤ n − 1, βi ∈ k and Φ(D) = ∂pt + t−1 Xi=1 βi∂pi + (y′)−1(cid:18)β0 + Φ(g(x)) − t−1 Xi=1 βi∂pi y − ∂pt y(cid:19)∂. Sketch of proof. Recall from Sec. 4.1.2 that if Φ is any admissible automorphism of O(1; n) with Φ(x) = y (not necessarily with identical linear part), then Φ(q(x) ∂) = (y′)−1Φ(q(x))∂ for any q(x) ∈ O(1; n), and Φ(∂pi ) = Φ(∂)pi for 1 ≤ i ≤ n − 1. (4.8) (4.9) In this lemma, we only consider admissible automorphisms of O(1; n) with identical linear part, i.e. Φ ∈ G with Φ(x) = y = x + Ppn−1 j=2 νjx(j), where νj ∈ k with νpi = 0 for 1 ≤ i ≤ n − 1. By (4.8), we have that Φ(∂) = (y′)−1 ∂. We show that Φ(∂) ≡ ∂ (mod L(0)). This is equivalent to show that (y′)−1 ∂ − ∂ ∈ L(0). (4.10) Note that y′ = 1 +Ppn−1 j=2 νjx(j−1) which is invertible in O(1; n). Since L(0) is invariant under multiplication of invertible elements of O(1; n), we can multiply both sides of (4.10) by y′ and show that (1 − y′) ∂ ∈ L(0). This is clearly true. Hence Φ(∂) = (y′)−1 ∂ ≡ ∂ (mod L(0)). It remains to compute Φ(∂pi ) for 1 ≤ i ≤ n−1. Since Φ(∂) ≡ ∂ (mod L(0)), we may write Φ(∂) = ∂ +φ0(x) ∂ for some φ0(x) ∈ m. By (4.9), Φ(∂ p) = Φ(∂)p. By Jacobson's formula (Definition 1.1.1(3)) and the fact that L(0) is a restricted Lie subalgebra of Der O(1; n), we get Φ(∂p) = ∂p +φ1(x) ∂ for some φ1(x) ∈ O(1; n). Since ∂p x = 0, it follows that 0 = Φ(∂p x) = Φ(∂p)y = (∂p +φ1(x) ∂)y = ∂p y + φ1(x)y′. Hence φ1(x) = −(y′)−1(∂p y) and so Φ(∂ p) = ∂p −(y′)−1(∂p y) ∂. We show that Φ(∂p) ≡ ∂p (mod L(0)). This is equivalent to show that −(y′)−1(∂p y) ∂ ∈ L(0). By the same reason as above, we can multiply both sides by −y′ and show that (∂p y) ∂ ∈ L(0). Due to the form of y, i.e. νp = 0, this is clearly true. Hence Φ(∂p) = ∂p −(y′)−1(∂p y) ∂ ≡ ∂p (mod L(0)). 113 In a similar way, one can show that Φ(∂ pi ) = ∂pi −(y′)−1(∂pi y) ∂ ≡ ∂pi (mod L(0)) for 2 ≤ i ≤ n − 1. 4.2. The variety N Let D = ∂pt +Pt−1 and g(x) ∈ m. It follows from (4.8) and the above that i=0 βi∂pi + g(x)∂ be an element of Lp, where 1 ≤ t ≤ n − 1, βi ∈ k Φ(D) = ∂pt + t−1 Xi=1 βi∂pi + (y′)−1(cid:18)β0 + Φ(g(x)) − t−1 Xi=1 βi∂pi y − ∂pt y(cid:19)∂. This completes the sketch of proof. Lemma 4.2.4. [32, Theorem 1] Let D = ∂pt + Pt−1 Lp, where 1 ≤ t ≤ n − 1, βi ∈ k and g(x) ∈ m. Then D is conjugate under G to i=0 βi∂pi + g(x)∂ be an element of ∂pt + t−1 Xi=0 βi∂pi + x(pn−pt)h(x)∂ for some h(x) = Ppt−1 η=0 µηx(η) with µη ∈ k. Strategy of the proof. The proof splits into two parts. The first part (a) was proved by S. Tyurin; see [32, Theorem 1, p. 68, line -8]. The second part (b) follows from (a) and Corollary 1.4.1. Explicitly, we prove the following: (a) Take D as in the lemma. We show that for 1 ≤ j ≤ pn − pt − 1, if g(x) ∂ ≡ γjx(j) ∂ (mod L(j)) for some γj ∈ k and Φj ∈ G is such that Φj(x) = yj = x + γjx(pt+j), then Φj(D) ≡ ∂pt + t−1 Xi=0 βi ∂pi (mod L(j)). We start with j = 1 and continue checking the above equivalence until we get D is conjugate under G to ∂pt + t−1 Xi=0 βi ∂pi (mod L(pn−pt−1)). (b) By part (a), we may assume that D = ∂pt + t−1 Xi=0 βi∂pi + pt−1 Xη=0 µηx(pn−pt+η) ∂ for some µη ∈ k. Note that for any 0 ≤ η ≤ pt − 1, x(pn−pt)x(η) = (cid:18)pn − pt + η pn − pt (cid:19)x(pn−pt+η). 114 4.2. The variety N The result then follows from Corollary 1.4.1 which states that for any 0 ≤ η ≤ pt − 1, the following congruence holds: (cid:18)pn − pt + η pn − pt (cid:19) ≡ 1 (mod p). Proof. (a) Take D = ∂pt +Pt−1 i=0 βi∂pi + g(x)∂ as in the lemma. By Lemma 4.2.3, we know that if Φ(x) = y is any admissible automorphism of O(1; n) with identical linear part, then Φ(D) = ∂pt + t−1 Xi=1 βi∂pi + (y′)−1(cid:18)β0 + Φ(g(x)) − t−1 Xi=1 βi∂pi y − ∂pt y(cid:19)∂. If g(x)∂ ≡ γ1x∂ (mod L(1)) for some γ1 ∈ k and Φ1 ∈ G is such that Φ1(x) = y1 = x + γ1x(pt+1), then we show that Φ1(D) ≡ ∂pt + t−1 Xi=0 βi∂pi (mod L(1)). If γ1 = 0, then g(x)∂ ∈ L(1) and the result is clear. If γ1 6= 0, then we show this congruence by proving that Φ1(D) − ∂pt −Pt−1 i=0 βi∂pi ∈ L(1), i.e. (y′ 1)−1(cid:18)β0 + Φ1(g(x)) − t−1 βi∂pi y1 − ∂pt Xi=1 y1(cid:19) ∂ −β0 ∂ ∈ L(1). (4.11) Note that y′ 1 = 1 + γ1x(pt) which is invertible in O(1; n). Since L(1) is invariant under multiplication of invertible elements of O(1; n), we can multiply both sides of (4.11) by y′ 1 and show that (cid:18)β0 + Φ1(g(x)) − t−1 Xi=1 βi∂pi y1 − ∂pt y1(cid:19) ∂ −β0y′ 1 ∂ ∈ L(1). Since g(x)∂ ≡ γ1x∂ (mod L(1)) and Φ1 preserves the natural filtration of L, in partic- ular, it preserves L(1), hence y1 − ∂pt y1(cid:19) ∂ −β0y′ 1 ∂ t−1 Xi=1 βiγ1x(pt−pi+1) − γ1x(cid:19) ∂ t−1 βi∂pi Xi=1 (cid:18)β0 + Φ1(g(x)) − ≡(cid:18)β0 + γ1(x + γ1x(pt+1)) − − β0(1 + γ1x(pt)) ∂ ≡0 (mod L(1)). 115 4.2. The variety N Here we used our assumption that p > 3, i.e. pt −pi ≥ pt −pt−1 ≥ p−1 > 2. Therefore, D is conjugate to ∂pt i=0 βi∂pi (mod L(1)). +Pt−1 In general, if l < pn − pt − 1 and D ≡ ∂pt + Pt−1 i=0 βi∂pi + γlx(l) ∂ (mod L(l)) for i=0 βi∂pi + gl(x) ∂ with gl(x) ∂ ≡ γlx(l) ∂ (mod L(l)), then some γl ∈ k, i.e. D = ∂pt +Pt−1 applying Φl ∈ G with Φl(x) = yl = x + γlx(pt+l) to D, we get Φl(D) = ∂pt + t−1 Xi=1 βi∂pi + (y′ l)−1(cid:18)β0 + Φl(gl(x)) − t−1 Xi=1 βi∂pi yl − ∂pt yl(cid:19)∂. We show that Φl(D) ≡ ∂pt + t−1 Xi=0 βi∂pi (mod L(l)). If γl = 0, then gl(x) ∂ ∈ L(l) and the result is clear. If γl 6= 0, then we show this congruence by proving that (y′ l)−1(cid:18)β0 + Φl(gl(x)) − t−1 Xi=1 βi∂pi yl − ∂pt yl(cid:19)∂ − β0 ∂ ∈ L(l). (4.12) By the same reason as before, we can multiply both sides of (4.12) by y′ l and show that (cid:18)β0 + Φl(gl(x)) − t−1 Xi=1 βi∂pi yl − ∂pt yl(cid:19) ∂ −β0y′ l ∂ ∈ L(l). Indeed, since Φl preserves L(l), we have that t−1 βi∂pi Xi=1 (cid:18)β0 + Φl(gl(x)) − ≡(cid:18)β0 + γl(x + γlx(pt+l))(l) − − β0(1 + γlx(pt+l−1)) ∂ ≡0 (mod L(l)). yl − ∂pt yl(cid:19) ∂ −β0y′ l ∂ t−1 Xi=1 βiγlx(pt−pi+l) − γlx(l)(cid:19) ∂ Hence D is conjugate to ∂pt +Pt−1 Xi=0 D ≡ ∂pt t−1 + i=0 βi∂pi (mod L(l)). Then βi∂pi + γl+1x(l+1)∂ (mod L(l+1)) for some γl+1 ∈ k. x+ γl+1x(pt+l+1) to D we can show that D is conjugate to ∂pt +Pt−1 If γl+1 6= 0, then applying Φl+1 ∈ G with Φl+1(x) = yl+1 = i=0 βi∂pi (mod L(l+1)). 116 4.2. The variety N Continue doing this until we get D is conjugate to ∂pt +Pt−1 i=0 βi∂pi (mod L(pn−pt−1)). Then D ≡ ∂pt + t−1 Xi=0 βi∂pi + γpn−ptx(pn−pt) ∂ (mod L(pn−pt)) 6= 0, then we were supposed to apply Φpn−pt ∈ G for some γpn−pt ∈ k. with Φpn−pt(x) = x + γpn−ptx(pn) to D. But since x(j) = 0 for j ≥ pn in O(1; n), the If γpn−pt automorphism Φpn−pt is the identity automorphism and we stop here. Therefore, D is conjugate under G to ∂pt + t−1 Xi=0 βi∂pi (mod L(pn−pt−1)). (b) By part (a), we may assume that D = ∂pt + t−1 Xi=0 βi∂pi + pt−1 Xη=0 µηx(pn−pt+η) ∂ for some µη ∈ k. Note that for 0 ≤ η ≤ pt − 1, x(pn−pt)x(η) = (cid:18)pn − pt + η pn − pt (cid:19)x(pn−pt+η) ≡ x(pn−pt+η) (mod p) by Corollary 1.4.1. As a result, D ≡ ∂pt + = ∂pt + t−1 Xi=0 t−1 Xi=0 βi∂pi + pt−1 Xη=0 µηx(pn−pt)x(η) ∂ (mod p) βi∂pi + x(pn−pt) pt−1 Xη=0 µηx(η) ∂ . Set h(x) = Ppt−1 η=0 µηx(η), we get the desired result. This completes the proof. Next we assume that D = ∂pt + Pt−1 i=0 βi∂pi + x(pn−pt)Ppt−1 We check that if Dpn−1 ∈ L(0). Let us first consider the case t = n − 1. η=0 µηx(η) ∂ is nilpotent. Lemma 4.2.5. Let D = ∂pn−1 +Pn−2 i=0 βi∂pi + x(pn−pn−1)Ppn−1−1 η=0 element of Lp. µηx(η)∂ be a nilpotent (i) If βi = 0 for all i, then µ0 = µ1 = 0 and Dpn−1 ∈ L(1). (ii) (a) Let j ≥ 0 be the smallest index such that βj 6= 0. Then µ0 = 0 and Dpn−1−j is conjugate under G to pn−1−1 Xη=2 for some νη ∈ k. Hence Dpn−1 ∈ L(1) if j ≥ 1. +x(pn−pn−1) ∂pn−1 νηx(η) ∂ 117 4.2. The variety N (b) In particular, if β0 6= 0, then Dpn−1 is conjugate under G to ∂pn−1 . Hence D = ∂pn−1 i=0 γi ∂pi +Pn−2 for some γi ∈ k with γ0 6= 0. The proof is long as it involves many calculations. Let us first explain the strategy of the proof. Strategy of the proof. This is a computational proof and the following steps are crucial: Step 1 . Take D = ∂pn−1 +Pn−2 µηx(η)∂ as in the lemma. Since D is nilpotent, then Dpn = 0 by (4.6). We first calculate Dp which will be used i=0 βi∂pi. By i=0 βi∂pi +x(pn−pn−1)Ppn−1−1 in step 2. Set D1 = x(pn−pn−1)Ppn−1−1 µηx(η)∂ and D2 = ∂pn−1 + Pn−2 η=0 η=0 Jacobson's formula, Dp = n−2 Xi=0 i ∂pi+1 βp + (ad D2)p−1(D1) + µ(1)x(pn−1)∂ for some µ(1) ∈ k; see (4.14) in the proof. Step 2 . We consider the scalars βi in the following two cases and prove statements (i) and (ii) in the lemma. Step 2(i). Suppose βi = 0 for all i. Then D = ∂pn−1 + x(pn−pn−1)Ppn−1−1 µηx(η)∂. We show that µ0 = µ1 = 0 and Dpn−1 ∈ L(1). By the calculation in step 1, we have µηx(η) ∂ +µ(1)x(pn−1)∂; see (4.15). We show that if µ0 6= 0, then Dpn 6= 0, a contradiction. Hence µ0 = 0. Similarly, we show that µ1 = 0. As a result, Dp ∈ L(1). Since L(1) is restricted, we get Dpn−1 ∈ L(1) as desired. that Dp = Ppn−1−1 η=0 η=0 Step 2(ii)(a). Let j ≥ 0 be the smallest index such that βj 6= 0, and let l be the largest index such that βl 6= 0, i.e. 0 ≤ j ≤ l ≤ n − 2 and l pn−1−1 D = ∂pn−1 Xη=0 We show that µ0 = 0 and Dpn−1−j is conjugate under G to + x(pn−pn−1) Xi=j βi∂pi + µηx(η)∂. ∂pn−1 +x(pn−pn−1) pn−1−1 Xη=2 νηx(η) ∂ for some νη ∈ k. Let us start with the special case j = l. We prove by induction that for any 1 ≤ r ≤ n − 1 − j, Dpr is conjugate under G to ∂pj+r + βpr−1 0,(1) ∂pr−1 +x(pn−pj+r) pj+r−1 Xη=0 µη,(r)x(η) ∂ 118 4.2. The variety N for some β0,(1) ∈ k∗µ0 and µη,(r) ∈ k. Here we will use Jacobson's formula and Lemma 4.2.4. In particular, Dpn−1−j is conjugate under G to ∂pn−1 + βpn−2−j 0,(1) ∂pn−2−j +x(pn−pn−1) pn−1−1 Xη=0 µη,(n−1−j)x(η) ∂; see (4.16). Then we calculate Dpn and use Dpn = 0 to show that β0,(1) = µ0 = 0 and µ0,(n−1−j) = µ1,(n−1−j) = 0. This gives the desired result in this case. For the general case, j < l, one can show similarly that Dpn−1−j is conjugate under G to ∂pn−1 + λ ∂pn−2−j + n−3−j Xi=0 λi ∂pi +x(pn−pn−1) pn−1−1 Xη=0 νηx(η) ∂ for some λ ∈ k∗µ0 and λi, νη ∈ k; see (4.18). Then we show similarly that Dpn = 0 implies that λ = µ0 = 0, λi = 0 for 0 ≤ i ≤ n − 3 − j and ν0 = ν1 = 0. This gives the desired result. Suppose now j ≥ 1. We show that (∂pn−1 (4.19). Since Dpn−1−j is conjugate under G to ∂pn−1 νηx(η) ∂ and G preserves L(1), we have that Dpn−j ∈ L(1). As L(1) is restricted, we get Dpn−1 ∈ L(1) +x(pn−pn−1)Ppn−1−1 η=2 +x(pn−pn−1)Ppn−1−1 η=2 νηx(η) ∂)p ∈ L(1); see as desired. Step 2(ii)(b). Suppose β0 6= 0. By step 2(ii)(a), Dpn−1 is conjugate under G to νηx(η) ∂ for some νη ∈ k. We show that Dpn = 0 implies that ∂pn−1 νη = 0 for all η. Hence Dpn−1 is conjugate to ∂pn−1 +x(pn−pn−1)Ppn−1−1 η=2 . Then we find an expression for D. Let n−2 + k ∂pi Xi=1 S := (cid:26)D ∈ (cid:0) ∂pn−1 +L(cid:1) ∩ N Dpn−1 Note that S is a subset of the centralizer cLp(∂pn−1 cLp(∂pn−1 ) and show that any D in S has the form D = ∂pn−1 is conjugate under G to ∂pn−1(cid:27). ). Then we consider elements of i=0 γi ∂pi +Pn−2 for some γi ∈ k with γ0 6= 0. Proof. Step 1 . Let D = ∂pn−1 +Pn−2 µηx(η)∂ be a nilpo- tent element of Lp. Then Dpn = 0 by (4.6). Let us first calculate Dp. Recall Jacobson's i=0 βi∂pi + x(pn−pn−1)Ppn−1−1 η=0 formula, (D1 + D2)p = Dp 1 + Dp 2 + p−1 Xi=1 si(D1, D2) (4.13) 119 4.2. The variety N for all D1, D2 ∈ Lp, and si(D1, D2) can be computed by the formula ad(tD1 + D2)p−1(D1) = p−1 Xi=1 isi(D1, D2)ti−1, where t is a variable. Set and D1 = x(pn−pn−1) pn−1−1 Xη=0 µηx(η)∂ D2 = ∂pn−1 + n−2 Xi=0 βi∂pi . 1 = 0. By Corollary 1.4.1, x(pn−pn−1)x(η) ≡ x(pn−pn−1+η) (mod p). µηx(pn−pn−1+η)∂. By Lemma 1.1.1, We first show that Dp Then D1 = Ppn−1−1 η=0 pn−1−1 Dp 1 = Xη=0 (cid:0)µηx(pn−pn−1+η)∂(cid:1)p + w, where w is a linear combination of commutators in µηx(pn−pn−1+η)∂, 0 ≤ η ≤ pn−1 − 1. By Jacobi identity, we can rearrange w so that w is in the span of commutators [wp−1, [wp−2, [. . . , [w2, [w1, w0] . . . ], where each wν, 0 ≤ ν ≤ p − 1, is equal to some µηx(pn−pn−1+η)∂, 0 ≤ η ≤ pn−1 − 1. We show that such iterated commutator equals 0 and so w = 0. Recall (4.1) the natural filtration {L(α)}α≥−1 of L, where L(α) = 0 for all α > pn − 2. Since 0 ≤ η ≤ pn−1 − 1, we have that µηx(pn−pn−1+η)∂ ∈ L(pn−pn−1−1) for all η. Then [wp−1, [wp−2, [. . . , [w2, [w1, w0] . . . ] ∈ L(p(pn−pn−1−1)) by Definition 1.3.3(ii). We show that L(p(pn−pn−1−1)) = 0, i.e. p(pn − pn−1 − 1) > pn − 2. Note that p(pn − pn−1 − 1) > pn − 2 ⇐⇒ pn+1 − pn − p > pn − 2 ⇐⇒ pn+1 − p + 2 > 2pn. We consider the p-adic expansions of these two numbers. By our assumption, p > 3. Then p − 1 > 2. So 2pn + 0pn−1 + · · · + 0p + 0 is the p-adic expansion of 2pn. By (1.2) in Corollary 1.4.1, the p-adic expansion of pn+1 − p + 2 is pn+1 − p + 2 = n Xj ′=1 (p − 1)pj ′ + 2 = (p − 1)pn + (p − 1)pn−1 + · · · + (p − 1)p + 2. Since p − 1 > 2, it is clear that pn+1 − p + 2 > 2pn. Hence L(p(pn−pn−1−1)) = 0 and so [wp−1, [wp−2, [. . . , [w2, [w1, w0] . . . ] = 0. As a result, w = 0. By (4.2), we have that (cid:0)µηx(pn−pn−1+η)∂(cid:1)p = 0 for all 0 ≤ η ≤ pn−1 − 1. Therefore, Dp 1 = 0. 120 4.2. The variety N By Lemma 1.1.1 again, we get Dp i ∂pi+1. By (4.1) the natural filtration of L, we have that for any 1 ≤ s ≤ p − 2, 2 = Pn−2 i=0 βp [D1, (ad D2)s(D1)] ∈ [L(pn−pn−1−1), L(pn−(s+1)pn−1−1)] ⊆ [L(pn−pn−1−1), L(pn−1−1)] ⊆ L(pn−2) = span{x(pn−1)∂}. This last term will appear if and only if s = p − 2. So Dp = n−2 Xi=0 for some µ(1) ∈ k. i ∂pi+1 βp + (ad D2)p−1(D1) + µ(1)x(pn−1)∂ (4.14) Step 2 . We consider the scalars βi in the following two cases. Step 2(i). If βi = 0 for all i, then D2 = ∂pn−1 . By (4.14), Dp = (ad ∂pn−1 )p−1(D1) + µ(1)x(pn−1)∂. Since ∂pn−1 is a derivation of L and ∂pn−1(Ppn−1−1 η=0 pn−1−1 µηx(η)) = 0, we have that Dp = Xη=0 µηx(η) ∂ +µ(1)x(pn−1)∂. (4.15) If µ0 6= 0, then Dp ≡ µ0∂ (mod L(0)). By Lemma 1.1.1, Dpn ≡ µpn−1 0 ∂pn−1 + n−2 Xi=0 i ∂pi µ′ (mod L(0)) for some µ′ i ∈ k. As µ0 6= 0, this implies that Dpn 6≡ 0 (mod L(0)) and so it is not equal to 0. This contradicts that D is nilpotent. Hence µ0 = 0. Similarly, if µ1 6= 0 then Dp ≡ µ1x∂ (mod L(1)). But Dpn ≡ µpn−1 µ1 = 0. Therefore, Dp is an element of L(1). Since L(1) is restricted we have that Dpn−1 ∈ L(1). This proves (i). x∂ 6≡ 0 (mod L(1)), a contradiction. Thus 1 Step 2(ii)(a). Let j ≥ 0 be the smallest index such that βj 6= 0, and let l be the largest index such that βl 6= 0, i.e. 0 ≤ j ≤ l ≤ n − 2 and D = ∂pn−1 + l Xi=j βi∂pi + x(pn−pn−1) pn−1−1 Xη=0 µηx(η)∂. We first consider the special case j = l, i.e. D = ∂pn−1 + βj∂pj + x(pn−pn−1) pn−1−1 Xη=0 µηx(η)∂. 121 4.2. The variety N We prove by induction that for any 1 ≤ r ≤ n − 1 − j, Dpr is conjugate under G to ∂pj+r + βpr−1 0,(1) ∂pr−1 +x(pn−pj+r) pj+r−1 Xη=0 µη,(r)x(η) ∂ for some β0,(1) ∈ k∗µ0 and µη,(r) ∈ k. For r = 1, the previous calculation (4.14) gives Dp = βp j ∂pj+1 + ad(cid:0)∂pn−1 + βj∂pj(cid:1)p−1(cid:18)x(pn−pn−1) pn−1−1 Xη=0 µηx(η)∂(cid:19) + µ(1)x(pn−1)∂. Note that ad(cid:18)∂pn−1 = ad(cid:18) p−1 Xm=0 (−1)mβp−1−m j + βj∂pj(cid:19)p−1(cid:18)x(pn−pn−1) pn−1−1 Xη=0 µηx(η)∂(cid:19) ∂mpn−1+(p−1−m)pj(cid:19)(cid:18) pn−1−1 Xη=0 µηx(pn−pn−1+η)∂(cid:19) pn−1−1 = Xη=0 µηx(η)∂ − βj pn−1−1 Xη=0 µηx(pn−1−pj+η)∂ + . . . + βp−1 j pn−1−1 Xη=0 µηx(pn−pn−1−(p−1)pj+η)∂. The above result can be rewritten as µ0 ∂ +g(x) ∂ for some g(x) ∈ m. Hence Dp = βp j ∂pj+1 + µ0 ∂ +g(x) ∂ +µ(1)x(pn−1)∂. Then the automorphism Φ(x) = αx with αpj+1 = βp j reduces Dp to the form Dp = ∂pj+1 + β0,(1) ∂ +f1(x) ∂, where β0,(1) ∈ k∗µ0 and f1(x) ∈ m. Then Lemma 4.2.4 implies that Dp is conjugate under G to ∂pj+1 + β0,(1) ∂ +x(pn−pj+1) pj+1−1 Xη=0 µη,(1)x(η) ∂ for some µη,(1) ∈ k. Thus, the result is true for r = 1. Suppose the result is true for r = K − 1 < n − 1 − j, i.e. DpK−1 is conjugate under G to ∂pj+K−1 + βpK−2 0,(1) ∂pK−2 +x(pn−pj+K−1) pj+K−1−1 Xη=0 µη,(K−1)x(η) ∂ 122 4.2. The variety N for some β0,(1) ∈ k∗µ0 and µη,(K−1) ∈ k. Let us calculate DpK . Set and D1 = x(pn−pj+K−1) pj+K−1−1 Xη=0 µη,(K−1)x(η) ∂ D2 = ∂pj+K−1 + βpK−2 0,(1) ∂pK−2 in the Jacobson's formula (4.13). Then Dp 1 ∈ L(1) and Dp 2 = ∂pj+K + βpK−1 0,(1) ∂pK−1 . By (4.1) the natural filtration of L, we have that (ad D2)p−1(D1) ∈ L(pn−pj+K −1) ⊆ L(1). Similarly, for any 1 ≤ s ≤ p − 2, [D1, (ad D2)s(D1)] ∈ [L(pn−pj+K−1−1), L(pn−(s+1)pj+K−1−1)] ⊆ [L(1), L(pn−(s+1)pj+K−1−1)] ⊆ L(pn−(s+1)pj+K−1), ⊆ L(pn−(p−1)pj+K−1), ⊆ L(pn−(p−1)pn−2) (since j + K − 1 ≤ n − 2) ⊆ L(1). Hence DpK = ∂pj+K + βpK−1 By Lemma 4.2.4, DpK is conjugate under G to 0,(1) ∂pK−1 +fK(x) ∂ for some fK(x) ∈ m with fK(x) ∂ ∈ L(1). ∂pj+K + βpK−1 0,(1) ∂pK−1 +x(pn−pj+K ) pj+K −1 Xη=0 µη,(K)x(η) ∂ for some µη,(K) ∈ k, i.e. the result is true for r = K. Therefore, we proved by induction that for any 1 ≤ r ≤ n − 1 − j, Dpr is conjugate under G to ∂pj+r + βpr−1 0,(1) ∂pr−1 +x(pn−pj+r) pj+r−1 Xη=0 µη,(r)x(η) ∂ for some β0,(1) ∈ k∗µ0 and µη,(r) ∈ k. In particular, Dpn−1−j is conjugate under G to ∂pn−1 + βpn−2−j 0,(1) ∂pn−2−j +x(pn−pn−1) pn−1−1 Xη=0 µη,(n−1−j)x(η) ∂ . (4.16) 123 4.2. The variety N By Jacobson's formula, Dpn−j = βpn−1−j 0,(1) ∂pn−1−j + pn−1−1 Xη=0 µη,(n−1−j)x(η) ∂ +fn−j(x) ∂ +µ(n−j)x(pn−1) ∂ (4.17) for some fn−j(x) ∂ ∈ L(1) and µ(n−j) ∈ k. Then Dpn ≡ βpn−1 0,(1) ∂pn−1 +µpj 0,(n−1−j) ∂pj + j−1 Xi=0 i ∂pi µ′′ (mod L(0)) for some µ′′ i ∈ k. But Dpn = 0, this implies that β0,(1) = 0. Since β0,(1) ∈ k∗µ0, we have i = 0 for all i. Substituting that µ0 = 0. We must also have that µ0,(n−1−j) = 0 and µ′′ these into (4.17), we get Dpn−j = pn−1−1 Xη=1 µη,(n−1−j)x(η) ∂ +fn−j(x) ∂ +µ(n−j)x(pn−1) ∂ ≡ µ1,(n−1−j)x ∂ (mod L(1)). Then one can show similarly that µ1,(n−1−j) = 0. Hence Dpn−1−j (4.16) is conjugate under G to ∂pn−1 + x(pn−pn−1) pn−1−1 Xη=2 µη,(n−1−j)x(η) ∂ . If j < l, i.e. D = ∂pn−1 + Pl show similarly that Dpn−1−j is conjugate under G to i=j βi∂pi + x(pn−pn−1)Ppn−1−1 η=0 µηx(η)∂, then one can ∂pn−1 + λ ∂pn−2−j + n−3−j Xi=0 λi ∂pi +x(pn−pn−1) pn−1−1 Xη=0 νηx(η) ∂ (4.18) for some λ ∈ k∗µ0 and λi, νη ∈ k. Then by the same arguments as above, one can show that λ = µ0 = 0, λi = 0 for 0 ≤ i ≤ n − 3 − j and ν0 = ν1 = 0. As a result, Dpn−1−j is conjugate under G to ∂pn−1 + x(pn−pn−1) pn−1−1 Xη=2 νηx(η)∂. Suppose now j ≥ 1. By a similar calculation as in (4.14) and (4.15), we get + x(pn−pn−1) (cid:0)∂pn−1 pn−1−1 pn−1−1 Xη=2 νηx(η)∂(cid:1)p = Xη=2 νηx(η)∂ + µ(n−j)x(pn−1) ∂ (4.19) 124 4.2. The variety N for some µ(n−j) ∈ k. This is an element of L(1). Since Dpn−1−j is conjugate under G νηx(η)∂ and G preserves L(1), we have that Dpn−j ∈ L(1). to ∂pn−1 + x(pn−pn−1)Ppn−1−1 η=2 As L(1) is restricted, we have that Dpn−1 ∈ L(1). This proves (ii)(a). Step 2(ii)(b). If β0 6= 0, then (ii)(a) implies that Dpn−1 is conjugate under G to ∂pn−1 + x(pn−pn−1) pn−1−1 Xη=2 νηx(η) ∂ for some νη ∈ k. If q is the smallest index such that νq 6= 0, then Dpn = pn−1−1 Xη=q νηx(η) ∂ +µ(n)x(pn−1) ∂ for some µ(n) ∈ k. As νq 6= 0, this implies that Dpn 6= 0, a contradiction. Hence νη = 0 for all η. Therefore, we are interested in the set S := (cid:26)D ∈ (cid:0) ∂pn−1 + n−2 Xi=1 k ∂pi +L(cid:1) ∩ N Dpn−1 is conjugate under G to ∂pn−1(cid:27). Since [D, Dpn−1] = 0, the above set S is a subset of the centralizer cLp(∂pn−1 to verify that cLp(∂pn−1 ) is spanned by ∂pn−1 and W (1, n − 1)p. Since W (1, n − 1)p is a ). It is easy restricted Lie subalgebra of Lp, we may regard the automorphism group of W (1, n−1)p as a subgroup of G. Let D = ∂pn−1 ), where γi ∈ k and v ∈ W (1, n − 1). If v = 0, then Dpn−1 = 0 which is not conjugate to ∂pn−1 +v be an element of cLp(∂pn−1 +Pn−2 i=1 γi ∂pi . So v 6= 0. If v 6∈ W (1, n − 1)(0), then v = γ0 ∂ for some γ0 6= 0. It is easy to see that Dpn−1 is conjugate under G to ∂pn−1 λix(i) ∂ . If v ∈ W (1, n − 1)(0), then v = Ppn−1−1 i=1 with λi 6= 0 for some i. It follows from Lemma 4.2.4 that D is conjugate under G to ∂pn−1 + n−2 Xi=1 γi ∂pi +x(pn−pn−1) pn−1−1 Xη=0 λ′ ηx(η) ∂ η ∈ k. If γi = 0 for all i, then (i) of this lemma implies that Dpn−1 ∈ L(1) for some λ′ which is not conjugate to ∂pn−1 . Similarly, if j ≥ 1 is the smallest index such that γj 6= 0, then (ii)(a) of this lemma implies that Dpn−1 ∈ L(1) which is again not conjugate i=0 γi ∂pi to ∂pn−1 . Therefore, the set S consists of elements of the form D = ∂pn−1 , +Pn−2 where γi ∈ k with γ0 6= 0. This proves (ii)(b). We see that the last proof involves calculations using Jacobson's formula and ap- plications of Lemma 4.2.4. The only property of D that we used is Dpn = 0. Hence 125 using the same arguments, we can prove a very similar result for nilpotent elements 4.2. The variety N i=0 αi∂pi + x(pn−pm)Ppm−1 ∂pm +Pm−1 Corollary 4.2.1. Let E = ∂pm +Pm−1 n − 2 be a nilpotent element of Lp. η=0 µηx(η) ∂, where 1 ≤ m ≤ n − 2. i=0 αi∂pi + x(pn−pm)Ppm−1 η=0 µηx(η) ∂ with 1 ≤ m ≤ (i) If αi = 0 for all i, then Epn−1 ∈ L(1). (ii) (a) Let q ≥ 0 be the smallest index such that αq 6= 0. Then Epn−1−q is conjugate under G to pn−1−1 Xη=2 for some νη ∈ k. Hence Epn−1 ∈ L(1) if q ≥ 1. +x(pn−pn−1) ∂pn−1 νηx(η) ∂ (b) In particular, if α0 6= 0, then Epn−1 is conjugate under G to ∂pn−1 . Hence E = ∂pm i=0 γi ∂pi +Pm−1 for some γi ∈ k with γ0 6= 0. Strategy of the proof. By a similar argument as in step 2(ii)(a) of Lemma 4.2.5, one can show that Epn−1−m is conjugate under G to D1 = ∂pn−1 + m−1 Xi=0 αpn−1−m i ∂pi+n−1−m +x(pn−pn−1) pn−1−1 Xη=0 µη,(n−1−m)x(η) ∂ for some µη,(n−1−m) ∈ k. Note that D1 has a similar expression to D in Lemma 4.2.5. Applying that lemma to D1, we get most of the desired results. The other results such as showing that Epn−1 ∈ L(1) in (i) and finding an expression for E in (ii)(b) follow from the same arguments as in the proof of that lemma. Proof. Take E = ∂pm +Pm−1 η=0 µηx(η) ∂ as in the corollary. By a similar argument as in step 2(ii)(a) of Lemma 4.2.5, i.e. using induction on r, one i=0 αi∂pi + x(pn−pm)Ppm−1 can show that for any 1 ≤ r ≤ n − 1 − m, Epr is conjugate under G to ∂pm+r + m−1 Xi=0 αpr i ∂pi+r +x(pn−pm+r) pm+r−1 Xη=0 µη,(r)x(η) ∂ for some µη,(r) ∈ k. In particular, Epn−1−m is conjugate under G to D1 = ∂pn−1 + m−1 Xi=0 αpn−1−m i ∂pi+n−1−m +x(pn−pn−1) pn−1−1 Xη=0 µη,(n−1−m)x(η) ∂ . 126 4.2. The variety N Note that D1 has a similar expression to D in Lemma 4.2.5. Since E is nilpotent, we have that Epn = 0 by (4.6). As Epn−1−m is conjugate under G to D1, this implies that Dpm+1 = 0. 1 (i) Suppose αi = 0 for all i. Applying Lemma 4.2.5(i) to D1, we get µ0,(n−1−m) = µ1,(n−1−m) = 0. Then (4.15) in step 2(i) of Lemma 4.2.5 gives pn−1−1 Dp 1 = µη,(n−1−m)x(η) ∂ +µ(1)x(pn−1)∂ Xη=2 1 ∈ L(1). Since G preserves L(1), we have that Epn−m ∈ L(1). for some µ(1) ∈ k. Hence Dp As L(1) is restricted, we have that Epn−1 ∈ L(1). This proves (i). (ii)(a) Let q ≥ 0 be the smallest index such that αq 6= 0. Then Epn−1−m is conjugate under G to D1 = ∂pn−1 + m−1 Xi=q αpn−1−m i ∂pi+n−1−m +x(pn−pn−1) pn−1−1 Xη=0 µη,(n−1−m)x(η) ∂ . Applying Lemma 4.2.5(ii)(a) to D1, we get µ0,(n−1−m) = 0 and Dpm−q 1 is conjugate under G to ∂pn−1 +x(pn−pn−1) pn−1−1 Xη=2 νηx(η) ∂ for some νη ∈ k. Hence (Epn−1−m)pm−q = Epn−1−q is conjugate under G to the above element. It follows from Lemma 4.2.5(ii)(a) that Epn−1 ∈ L(1) if q ≥ 1. This proves (ii)(a). (b) If α0 6= 0, then it follows from the above and Lemma 4.2.5(ii)(b) that Epn−1 is . Here we used that Epn = 0. By the same arguments as conjugate under G to ∂pn−1 in step 2(ii)(b) of Lemma 4.2.5, one can show that E = ∂pm i=0 γi ∂pi +Pm−1 for some γi ∈ k with γ0 6= 0. This proves (ii)(b). 4.2.2 An irreducible component of N The results in the last section enable us to prove the following: Proposition 4.2.1. Define Nreg := (cid:8)D ∈ N Dpn−1 /∈ L(0)(cid:9). Then Nreg = G.(∂ +k ∂p + · · · + k ∂pn−1 ). 127 4.2. The variety N )pn−1 = ∂pn−1 i=1 αi ∂pi , this shows are contained ) ⊆ Nreg. To show that Nreg is a subset of ), we observe that if D1 ∈ L(1), then D1 is nilpotent and i=1 αi ∂pi )pn = 0 and (∂ +Pn−1 Proof. Since (∂ +Pn−1 that any elements which are conjugate under G to ∂ +Pn−1 in Nreg. So G.(∂ +k ∂p + · · · + k ∂pn−1 G.(∂ +k ∂p + · · · + k ∂pn−1 Dpn−1 ∈ L(1) ⊂ L(0). Hence D1 6∈ Nreg. As a result, Nreg ⊆ N \ L(1). i=1 αi ∂pi 1 Let D ∈ Nreg. Note that elements of N \ L(1) have the form n−1 Xi=0 αi∂pi + f (x)∂, where f (x) ∈ m and αi ∈ k with at least one αi 6= 0. If α0 6= 0 and αi = 0 for all i ≥ 1, then (α0 ∂ +f (x) ∂)pn−1 6∈ L(0) by Lemma 4.2.2(i). Hence α0 ∂ +f (x) ∂ ∈ Nreg. Take D to be such an element. It follows from Lemma 4.2.2(ii) that D is conjugate under G to ∂. Thus D ∈ G.(∂ +k ∂p + · · · + k ∂pn−1 ) in this case. For the other elements of N \ L(1), let 1 ≤ t ≤ n − 1 be the largest index such that i=0 αi∂pi + f (x)∂. Let Φ ∈ G be such αt 6= 0, i.e. we consider elements of the form Pt that Φ(x) = αx, where αpt = αt. Then Φ reduces Pt t−1 i=0 αi∂pi + f (x)∂ to ∂pt + Xi=0 βi∂pi + g(x) ∂ for some βi ∈ k∗αi and g(x) ∈ m. By Lemma 4.2.4, ∂pt conjugate under G to i=0 βi∂pi + g(x) ∂ is +Pt−1 ∂pt + t−1 Xi=0 βi∂pi + x(pn−pt) pt−1 Xη=0 µηx(η) ∂ for some µη ∈ k. If βi = 0 for all i, then Lemma 4.2.5 and Corollary 4.2.1(i) imply that +x(pn−pt) (cid:0) ∂pt pt−1 Xη=0 µηx(η) ∂(cid:1)pn−1 ∈ L(1) ⊂ L(0), and so elements of this form are not in Nreg. Now let j ≥ 1 be the smallest index such that βj 6= 0. Then Lemma 4.2.5 and Corollary 4.2.1(ii)(a) imply that + (cid:0) ∂pt t−1 Xi=j βi∂pi + x(pn−pt) pt−1 Xη=0 µηx(η) ∂(cid:1)pn−1 ∈ L(1) ⊂ L(0), and so elements of this form are not in Nreg. But if β0 6= 0, then it is easy to see that + (cid:0) ∂pt t−1 Xi=0 βi∂pi + x(pn−pt) pt−1 Xη=0 µηx(η) ∂(cid:1)pn−1 6∈ L(0), 128 4.2. The variety N and so elements of this form are in Nreg. Take D to be such an element. It follows from Lemma 4.2.5 and Corollary 4.2.1(ii)(b) that D = ∂pt for some γi ∈ k with i=0 γi ∂pi γ0 6= 0. Let Φ1 ∈ G be such that Φ1(x) = γ0x. Applying Φ1 to D, we get Φ1(D) = i ∈ k∗γi. Hence D ∈ G.(∂ +k ∂p + · · · + k ∂pn−1 +(γ0)−pt ∂pt for some γ′ i ∂pi ) +Pt−1 ∂ +Pt−1 i=1 γ′ in this case. Since we have exhausted all elements of Nreg, this completes the proof. Before we proceed to show that the Zariski closure of Nreg is an irreducible com- ponent of N , we need the following results: Lemma 4.2.6. Let D = ∂ +Pn−1 i=1 λi ∂pi cLp(D)) the centralizer of D in L (respectively Lp). Then with λi ∈ k and denote by cL(D) (respectively (i) cL(D) = span{∂}. (ii) cLp(D) = span{∂, ∂p, . . . , ∂pn−1 }. (iii) cL(D) ∩ Lie(G) = {0}. Proof. (i) Clearly, span{∂} ⊆ cL(D). Since (ad D)pn−1 6= 0 and (ad D)pn = 0, the theory of canonical Jordan normal form says that there exists a basis B of L such that the matrix of ad D with respect to B is a single Jordan block of size pn with zeros on the main diagonal. Hence the matrix of ad D has rank pn − 1. This implies that Ker(ad D) has dimension 1. By definition, Ker(ad D) = cL(D). Hence cL(D) = span{∂}. (ii) It is clear that span{∂, ∂p, . . . , ∂pn−1 } ⊆ cLp(D). Suppose v ∈ cLp(D). Then i=0 αi ∂pi we can write v = Pn−1 must have that v1 ∈ cLp(D). By (i), the centralizer of D in L is k ∂ which is not in L(0). Hence v1 = 0 and cLp(D) = span{∂, ∂p, . . . , ∂pn−1 +v1 for some v1 ∈ L(0). Since Pn−1 ∈ cLp(D), we i=0 αi ∂pi }. (iii) It follows from (i) and Lemma 4.1.1. This completes the proof. Lemma 4.2.7. The Zariski closure of Nreg is an irreducible component of N . Proof. By Proposition 4.2.1, it suffices to show that the Zariski closure of G.(∂ +k ∂p + · · · + k ∂pn−1 ) is an irreducible component of N . Set X = ∂ +X0, where X0 = k ∂p + · · · + k ∂pn−1 . Note that X ∼= X0 ∼= An−1. So X is irreducible. Moreover, G is a connected algebraic group and so G.X is an irreducible variety 129 4.2. The variety N contained in N . Hence dim G.X ≤ dim N . If dim G.X ≥ dim N , then we get the desired result. Define Ψ to be the morphism Ψ : G × X → G.X (g, D) 7→ g.D Since G.X is dense in G.X, it contains smooth points of G.X. As the set of smooth points is G-invariant, there exists D ∈ X such that Ψ(1, D) = D is a smooth point in G.X. We may assume that D = ∂ +Pn−1 i=1 λi ∂pi and the differential of Ψ at the smooth point (1, D) is the map for some λi ∈ k. Then TD(X) = X0 (dΨ)(1,D) : Lie(G) ⊕ X0 → TD(G.X) (Y, Z) 7→ [Y, D] + Z. Since dim TD(G.X) = dim G.X, it is enough to show that dim TD(G.X) ≥ dim N = pn − 1. This can be done by showing that (dΨ)(1,D) is injective. By Lemma 4.1.1, j=1 µjx(j) ∂, where µj ∈ k with we know that any Y ∈ Lie(G) has the form Y = Ppn−1 µpl = 0 for all 1 ≤ l ≤ n − 1. By (1.4), [Y, D] = (cid:20) pn−1 Xj=1 Xj=1 = − pn−1 µjx(j) ∂, ∂ + µjx(j−1) ∂ − n−1 Xi=1 Xi=1 n−1 λi ∂pi(cid:21) pn−1 Xj=pi+1 λiµjx(j−pi) ∂ . Hence [Y, D] ∈ L for all Y ∈ Lie(G). As L ∩ X0 = {0}, we have that Ker((dΨ)(1,D)) ∼= cLie(G)(D). By Lemma 4.2.6(iii), cLie(G)(D) = {0}. Hence Ker((dΨ)(1,D)) = 0 and (dΨ)(1,D) is injective. As a result, dim TD(G.X) ≥ dim Im((dΨ)(1,D)) = dim Lie(G) + dim X0 = (pn − n) + (n − 1) = pn − 1 = dim N . This completes the proof. 130 4.2. The variety N 4.2.3 The irreducibility of N Our goal is to prove the irreducibility of the variety N . To achieve this, we need the following result: Proposition 4.2.2. Define Nsing := N \ Nreg = (cid:8)D ∈ N Dpn−1 ∈ L(0)(cid:9). Then dim Nsing < dim N . Clearly, Nsing is Zariski closed in N . To prove this proposition, we need to construct an (n + 1)-dimensional subspace V in Lp such that V ∩ Nsing = {0}. Then the result follows from Theorem 1.6.8; see Lemma 1.5.5 for a similar proof. The way V is constructed relies on the original definition of L due to H. Zassenhaus and the following lemmas. Recall that L has a k-basis {eα α ∈ Fq} with the Lie bracket given by [eα, eβ] = (β − α)eα+β. Here Fq ⊂ k is a finite field with q = pn elements. The multiplicative group F∗ q of Fq is cyclic of order pn − 1 with generator ξ; see Sec. 4.1.1 for detail. Since L is simple, it follows from Theorem 1.2.2(iii) that all its minimal p- envelopes are isomorphic as restricted Lie algebras. Moreover, L ∼= ad L via the adjoint representation. It follows from Remark 1.2.1 that the minimal p-envelope Lp = (ad L)p is the p-subalgebra of Der L generated by ad L. We identify L with ad L. Since L is a subalgebra of the restricted Lie algebra Der O(1; n) and {eα α ∈ Fq} is a basis of L, it follows from Lemma 1.1.3 that Lp = Xα∈Fq,i≥0 kepi α , (4.20) i.e. Lp consists of all iterated p-th powers of eα and L. Lemma 4.2.8. (i) Let σ ∈ GL(L) be such that σ(eα) := ξ−1eξα for any α ∈ Fq. Then σ is a diagonalizable automorphism of L. (ii) Let σ be as in (i). Define σ(epi α ) := ξ−piepi ξα for any i ≥ 1 and α ∈ Fq. Then σ extends to an automorphism of Lp. Proof. (i) By definition, [σ(eα), σ(eβ)] = [ξ−1eξα, ξ−1eξβ] = ξ−2(ξβ − ξα)eξα+ξβ = ξ−1(β − α)eξ(α+β) = σ([eα, eβ]) 131 for any α, β ∈ Fq. So the endomorphism σ is an automorphism of L. Since ξpn−1 = 1, we have that σpn−1 = Id. As k is an algebraically closed field, the automorphism σ is 4.2. The variety N diagonalizable. This proves (i). α ) := ξ−piepi (ii) Define σ(epi ξα for any i ≥ 1 and α ∈ Fq. We show that σ extends to an automorphism of Lp. By definition, [σ(epi α ), σ(epj β )] = [ξ−pi epi ξα, ξ−pj epj ξβ] for any i, j ≥ 1 and α, β ∈ Fq. On the other hand, it follows from Definition 1.1.1(1) that α , epj [epi β ] = −(ad eα)pi−1 ◦ (ad eβ)pj (eα). Since σ is an automorphism of L such that σ(eα) = ξ−1eξα for any α ∈ Fq, we have that σ[epi α , epj β ] = σ(cid:0) − (ad eα)pi−1 ◦ (ad eβ)pj (eα)(cid:1) (ad eξα)pi−1 ◦ (ad eξβ)pj = −ξ−pi ξ−pj (eξα) = [ξ−pi = [σ(epi epj ξβ] epi ξα, ξ−pj α ), σ(epj β )]. Hence σ extends to an automorphism of Lp. This proves (ii). Note that [e0, eβ] = βeβ for any β ∈ Fq. So ad e0 is a semisimple endomorphism of L. Since ad L ∼= L, then e0 generates a torus under the [p]-th power map. Lemma 4.2.9. Let T denote the p-envelope (ke0)p in Lp. Then T is an n-dimensional torus of Lp such that T ∩ L = ke0. Proof. Let T denote the p-envelope (ke0)p in Lp. Then T = Pi≥0 kepi subalgebra of Lp generated by ke0; see Definition 1.1.3 and Lemma 1.1.3. We first is the p- 0 show that T is a torus of Lp. By Definition 1.1.1(1), 0 , epj [epi 0 ] = −(ad e0)pi−1 ◦ (ad e0)pj (e0) = 0 for any i, j ≥ 0. Hence T is an abelian p-subalgebra of Lp. Then we show that T consists of semisimple elements. By Lemma 1.1.4(iii), it is enough to show that e0 is semisimple. By Lemma 4.2.8, we know that σ acts on T as σ(epi 0 ) = ξ−pi epi 0 132 (4.21) 4.2. The variety N for all i ≥ 0. Since ξpn−1 = 1, we see that ξ−1, ξ−p, . . . , ξ−pn−1 are the eigenvalues 0 . So e0 and epn of σ on T . Moreover, σ(e0) = ξ−1e0 and σ(epn 0 correspond to the same eigenvalue ξ−1 and they are in the same eigenspace. We show that e0 and epn 0 . Recall that for any α, β ∈ Fq, the Lie bracket of L is given by [eα, eβ] = (β − α)eα+β. Then for any α ∈ Fq and i > 0, 0 are linearly dependent with e0 = epn 0 ) = ξ−pnepn 0 = ξ−1epn we have that [e0 − epn 0 , epi α ] = [e0, epi α ] − [epn 0 , epi α ] (e0) + (ad e0)pn−1 ◦ (ad eα)pi (e0) = −(ad eα)pi = 0 + 0 = 0. Similarly, for any α ∈ Fq and i = 0, we have that [e0 − epn 0 , eα] = [e0, eα] − (ad e0)pn (eα) = αeα − αpn eα = αeα − αeα = 0. We show the above calculations imply that e0 − epn p-envelope Lp of L is given by Lp = Pα∈Fq,i≥0 kepi powers of eα and L. It follows from the above calculations that e0 − epn 0 ∈ z(Lp). By (4.20), the minimal α , i.e. Lp consists of all iterated p-th 0 ∈ z(Lp). Since L is simple, it follows from Theorem 1.2.2(iii) that Lp is semisimple. Since z(Lp) is an abelian p-ideal of Lp, we must have that z(Lp) = 0. As a result, e0 = epn implies that e0 is semisimple. By Lemma 1.1.4(iii), epi 0 is semisimple for every i ≥ 1 0 . This and t is semisimple for every t ∈ (ke0)p = T . Hence T is an abelian p-subalgebra of Lp consisting of semisimple elements, i.e. T is a torus of Lp. Next we show that dimk T = n. Since e0 = epn 0 , this implies that the [p]-th 0 0 = epn+i for all i ≥ 0. Hence any t ∈ T has power map on T is periodic, i.e. epi the form t = Pn−1 i=0 µiepi for some µi ∈ k. Thus {e0, ep over, it follows from (4.21) that the eigenvectors e0, ep tinct eigenvalues ξ−1, ξ−p, . . . , ξ−pn−1. Hence they are linearly independent. Therefore, {e0, ep } is a basis for T and dimk T = n. 0, . . . , epn−1 0, . . . , epn−1 } spans T . More- correspond to dis- 0 0, . . . , epn−1 0 0 0 It remains to show that T ∩ L = ke0. Note that e0 /∈ L(0). Indeed, if e0 ∈ L(0), then T is contained in L(0) as L(0) is restricted. But this contradicts that any nonzero torus of L(0) has dimension 1; see (4.2) in Sec. 4.1.1. Therefore, e0 /∈ L(0). Since e0 is semisimple, we may assume that e0 = ∂ +z0 for some 0 6= z0 ∈ L(0). Let us compute epi 0 for 1 ≤ i ≤ n − 1. By Jacobson's formula, ep 0 = ∂p +α1,0 ∂ +z1 for some α1,0 ∈ k 133 4.2. The variety N and 0 6= z1 ∈ L(0) (otherwise ep 0 is nilpotent). Continue doing this, one can show that for 1 ≤ i ≤ n − 1, epi 0 = ∂pi + i−1 Xj=0 αi,j ∂pj +zi for some αi,j ∈ k and 0 6= zi ∈ L(0) (otherwise epi 0 is nilpotent). It is clear that epi 0 /∈ L for all 1 ≤ i ≤ n − 1. Hence T ∩ L = ke0. This completes the proof. Remark 4.2.1. It follows from Lemma 4.2.8 and the last proof that epi 0 with i ≥ 0 satisfies the following: (i) σ(epi 0 ) = ξ−piepi 0 for all i ≥ 0. (ii) epi 0 is semisimple with epi 0 = epn+i 0 for all i ≥ 0. (iii) e0 is an element of L = L(−1) such that e0 /∈ L(0). Hence for any 0 ≤ i ≤ n − 1, we may assume that epi 0 = ∂pi + i−1 Xj=0 αi,j ∂pj +zi for some αi,j ∈ k and 0 6= zi ∈ L(0). Note that epi n − 1. Moreover, {e0, ep 0, . . . , epn−1 0 } forms a basis for the n-dimensional torus 0 /∈ L for all 1 ≤ i ≤ T = (ke0)p. Since we have proved in Lemma 4.2.8 that σ is an automorphism of L, then it preserves the natural filtration {L(i)}i≥−1 of L. Moreover, we can prove the following: Lemma 4.2.10. For −1 ≤ i ≤ pn −2, the automorphism σ acts on each 1-dimensional vector space L(i)/L(i+1) as ξi Id. Proof. We prove this result by induction on i. For i = −1, consider the surjective map π−1 : L(−1) ։ L(−1)/L(0). By Remark 4.2.1(iii), we know that e0 is an element of L(−1) such that e0 /∈ L(0). Then the vector space L(−1)/L(0) is spanned by π(e0). This implies that σ acts on L(−1)/L(0) as ξ−1 Id. Hence the result holds for i = −1. Suppose the result holds for −1 ≤ i = j ≤ pn − 3, i.e. σ acts on L(j)/L(j+1) as ξj Id. We want to show the result holds for i = j + 1 ≤ pn − 2, i.e. σ acts on L(j+1)/L(j+2) as ξj+1 Id. Since L = L(0) + ke0 and [L, L(i+1)] = L(i), we have that [e0, L(i+1)] * L(i+1). So we can find for each i ∈ {0, . . . , pn − 2} a ui ∈ L(i) such that [e0, ui] /∈ L(i). By writing ui as a sum of σ-eigenvectors corresponding to the same eigenvalue, we see 134 4.2. The variety N that we may assume that each ui is a σ-eigenvector corresponding to an eigenvalue, say λi. Then σ[e0, ui] = [σ(e0), σ(ui)] = [ξ−1e0, λiui] = ξ−1λi[e0, ui]. So ξ−1λi is the eigenvalue of σ on L(i−1)/L(i). Since i ∈ {0, . . . , pn − 2}, then i − 1 ∈ {−1, . . . , pn − 3}. By the induction hypothesis, σ acts on L(i−1)/L(i) as ξi−1 Id. So we must have that ξ−1λi = ξi−1 and hence λi = ξi. Now consider the surjective map πi : L(i) ։ L(i)/L(i+1). Since ui ∈ L(i) is such that [e0, ui] /∈ L(i), i.e. ui /∈ L(i+1), then the vector space L(i)/L(i+1) is spanned by πi(ui). It follows that σ acts on L(i)/L(i+1) as ξi Id. Therefore, we proved by induction that for −1 ≤ i ≤ pn − 2, σ acts on each 1-dimensional vector space L(i)/L(i+1) as ξi Id. This completes the proof. Lemma 4.2.11. Let u denote the element u0 in the last proof. Then ku is a 1- dimensional torus in L(0). Proof. Recall from the last proof that u ∈ L(0) is such that [e0, u] /∈ L(0). Moreover, u is a σ-eigenvector corresponding to the eigenvalue ξ0 = 1, i.e. σ(u) = u. We want to show that ku is a 1-dimensional torus in L(0). We first show that u is not nilpotent. Note that u /∈ L(1). Indeed, if u ∈ L(1), then Remark 4.2.1(iii) implies that [e0, u] ∈ [L(−1), L(1)] ⊆ L(0). But this contradicts that [e0, u] /∈ L(0). Hence u ∈ L(0) is such that u /∈ L(1). If u is nilpotent, then L(0) would be p-nilpotent. But this contradicts (4.2). Hence u is not nilpotent. Next we show that up is a multiple of u. Since u ∈ L(0) and L(0) is restricted, we have that up ∈ L(0). Since σ is an automorphism of the restricted Lie algebra L(0), we have that σ(up) = σ(u)p = up. So up is σ-fixed. By Lemma 4.2.10, the eigenvalues of σ on L(0) are ξ0 = 1, ξ, . . . , ξpn−3 and ξpn−2 = ξ−1, and they all have multiplicity 1. Since both u and up are σ-fixed, it follows that up is a multiple of u. Therefore, ku is a 1-dimensional torus in L(0). This completes the proof. Let us summarize what we know so far. Remark 4.2.2. It follows from Lemma 4.2.10 that (i) the eigenvalues of σ on L = span{∂, x ∂, x(2) ∂, . . . , x(pn−2) ∂, x(pn−1) ∂} are ξ−1, ξ0 = 1, ξ, . . . , ξpn−3 and ξpn−2 = ξ−1. All have multiplicity 1 except ξ−1 which has multiplicity 2; 135 4.2. The variety N (ii) the eigenvalues of σ on L(0) = span{x ∂, x(2) ∂, . . . , x(pn−2) ∂, x(pn−1) ∂} are ξ0 = 1, ξ, . . . , ξpn−3 and ξpn−2 = ξ−1. All have multiplicity 1; (iii) the eigenspace L[i] := {D ∈ L σ(D) = ξiD} corresponding to the eigenvalue ξi, where 0 ≤ i ≤ pn − 3, has dimension 1. In particular, the eigenspace L[0] = ku is a 1-dimensional torus in L(0); see Lemma 4.2.11. Since any torus in a restricted Lie algebra has a basis consisting of toral elements (see Lemma 1.1.5), we may assume that u is toral, i.e. up = u; (iv) the eigenspace L[−1] = span{e0, v v ∈ L(pn−2)} and it has dimension 2; see Re- mark 4.2.1(i) and (iii). We are now ready to prove Proposition 4.2.2 which states that Nsing := N \ Nreg = (cid:8)D ∈ N Dpn−1 ∈ L(0)(cid:9) has dim Nsing < dim N . Proof of Proposition 4.2.2. Recall the n-dimensional torus T = Pn−1 and the 1-dimensional torus ku in L(0); see Lemma 4.2.9 and Lemma 4.2.11. Put i=0 kepi 0 in Lp V := T ⊕ ku. We want to show that V ∩ Nsing = {0}. Then the result follows from Theorem 1.6.8. Suppose for contradiction that V ∩ Nsing 6= {0}. Then take a nonzero element y in V ∩ Nsing, we can write y = n−1 Xi=0 λiepi 0 + µu for some λi, µ ∈ k with at least one λi 6= 0. Case 1 . Suppose λ0 6= 0. We want to show this implies that ypn−1 /∈ L(0). But y ∈ Nsing by our assumption, this contradicts the definition of Nsing. By Lemma 1.1.1 and the facts that epn 0 = e0 and up = u (see Remark 4.2.1(ii) and Remark 4.2.2(iii)), we have that ypn−1 =(cid:18) n−1 Xi=0 =λpn−1 λiepi 0 + µu(cid:19)pn−1 n−1 epn−2 0 + λpn−1 epn−1 0 0 + λpn−1 n−2 epn−3 0 + · · · + λpn−1 2 0 + λpn−1 ep 1 e0 (4.22) + µpn−1 u + n−2 Xr=0 upr r , where ur is a linear combination of commutators in u and epi 0 , 0 ≤ i ≤ n − 1. We show that ur ∈ L for all 0 ≤ r ≤ n − 2. By Lemma 4.2.11, u ∈ L(0) ⊂ L. By Re- mark 4.2.1(iii), for any 0 ≤ i ≤ n−1, we may assume that epi +zi 0 = ∂pi j=0 αi,j ∂pj +Pi−1 136 4.2. The variety N for some αi,j ∈ k and 0 6= zi ∈ L(0). So e0 ∈ L is such that e0 /∈ L(0) and for 1 ≤ i ≤ n − 1, epi 0 /∈ L. Since L is an ideal in Lp and the ∂pi 0 ≤ i ≤ n − 1, commute amongst each other, it follows immediately that ur ∈ L for 0 ∈ Lp is such that epi , all 0 ≤ r ≤ n − 2. Next we show that Pn−2 r=0 upr r ∈ k ∂pn−2 +k ∂pn−3 + · · ·+ k ∂p +L. By Theorem 1.4.2, Der O(1; n). By Definition 1.1.3(ii), Lpr L = span{x(γ) ∂ 0 ≤ γ ≤ pn − 1} is a subalgebra of the restricted Lie algebra := {Dpr D ∈ L}. Since ur ∈ L for all r=0 Lpr. By Lemma 1.1.3 0 ≤ r ≤ n − 2, we have that upr r=0 upr r ∈ Lpr . Hence Pn−2 r ∈ Pn−2 and the fact that L(0) ⊂ L is a restricted subalgebra of Der O(1; n), we get Lpr = k ∂pn−2 +k ∂pn−3 + · · · + k ∂p +L. +k ∂pn−3 + · · · + k ∂p +L as desired. n−2 Xr=0 r ∈ k ∂pn−2 Hence Pn−2 r=0 upr Now look at µpn−1u in (4.22). Since u ∈ L(0), it follows from the above that we can write µpn−1 u + n−2 Xr=0 upr r = n−2 Xi=0 i ∂pi λ′ +z for some λ′ i ∈ k and z ∈ L(0). Substituting this and epi 0 = ∂pi (0 ≤ i ≤ n − 1) into (4.22), we get ypn−1 = λpn−1 0 ∂pn−1 + n−2 Xi=0 i ∂pi λ′′ +z j=0 αi,j ∂pj +Pi−1 +zi (4.23) for some λ′′ i ∈ k and z ∈ L(0). Since λ0 6= 0, this shows that ypn−1 /∈ L(0). But y ∈ Nsing = (cid:8)D ∈ N Dpn−1 ∈ L(0)(cid:9), this contradicts the definition of Nsing. Case 2 . Suppose now λ0 = 0 and let 1 ≤ s ≤ n − 1 be the largest index such that λs 6= 0. Then y = s Xi=1 λiepi 0 + µu. We want to show this implies that yp2n−s−1 6= 0. But y ∈ Nsing, in particular, y is nilpotent with ypn = 0. As 2n − s − 1 ≥ n, this contradicts that y is nilpotent. By Lemma 1.1.1 and the facts that epn 0 = e0 and up = u (see Remark 4.2.1(ii) and Remark 4.2.2(iii)), we have that ypn−s = λpn−s s e0 + λpn−s s−1 epn−1 0 + λpn−s s−2 epn−2 0 + · · · + λpn−s 1 epn−s+1 0 + µpn−s u + n−s−1 Xl=0 vpl l , 137 4.2. The variety N where vl is a linear combination of commutators in u and epi 0 , 1 ≤ i ≤ s. By Jacobi identity, we can rearrange each vl so that vl is in the span of [wt, [wt−1, [. . . , [w1, u] . . . ], where t = pn−s−l − 1 and each wν, 1 ≤ ν ≤ t, is equal to u or to some epi 0 , 1 ≤ i ≤ s. Arguing similarly as in case 1, one can show that [wt, [wt−1, [. . . , [w1, u] . . . ] ∈ L. More precisely, we show that [wt, [wt−1, [. . . , [w1, u] . . . ] ∈ L(0). Since σ(u) = u and σ(epi for 1 ≤ i ≤ s (see Remark 4.2.1(i) and Remark 4.2.2(iii)), the σ- 0 ) = ξ−piepi 0 eigenvalue of each such iterated commutator is ξ−a, where 1 < p ≤ a ≤ tps = pn−l − ps ≤ pn − p < pn − 2. So this eigenvalue is not equal to ξ−1. Then [wt, [wt−1, [. . . , [w1, u] . . . ] ∈ L(0)\L(pn−2) ⊂ L(0); see Remark 4.2.2. Hence vl ∈ L(0). Since L(0) is restricted, we have that vpl and so Pn−s−1 vpl l ∈ L(0). Therefore, l ∈ L(0) l=0 ypn−s = λpn−s s e0 + γn−1epn−1 0 + γn−2epn−2 0 + · · · + γn−s+1epn−s+1 0 + f1 for some γi ∈ k, λs 6= 0 and f1 ∈ L(0). Note that ypn−s has a similar expression to y in case 1. By Lemma 1.1.1 and a similar argument as in case 1, one can show that (ypn−s )pn−1 = yp2n−s−1 = λp2n−s−1 s ∂pn−1 + n−2 Xi=0 i ∂pi γ′ + f i ∈ k and f ∈ L(0); see (4.23). Since λs 6= 0, this shows that yp2n−s−1 6= 0. for some γ′ But y ∈ Nsing, in particular, y is nilpotent and ypn = 0 by (4.6). Since 2n − s − 1 ≥ n, the above contradicts that y is nilpotent. Therefore, we proved by contradiction that V ∩ Nsing = {0}. Then the result follows from Theorem 1.6.8. This completes the proof. Theorem 4.2.1. The variety N coincides with the Zariski closure of Nreg = G.(∂ +k ∂p + · · · + k ∂pn−1 ) and hence is irreducible. Proof. By Theorem 1.5.1, we know that the variety N is equidimensional of dimension pn − 1. The ideal defining N is homogeneous, hence any irreducible component of N contains 0. It follows from Lemma 4.2.7 that the Zariski closure of Nreg is an irreducible component of N . Let Z1, . . . , Zt be pairwise distinct irreducible components of N , and 138 4.2. The variety N set Z1 = Nreg. Suppose t ≥ 2. Then Z2\Z1 is contained in Nsing, which is Zariski closed in N with dim Nsing < dim N by Proposition 4.2.2. Since Z2 \ Z1 = Z2 \ (Z1 ∩ Z2), this set is Zariski dense in Z2. Then its closure Z2 is also contained in Nsing, i.e. dim Z2 = dim N ≤ dim Nsing. This is a contradiction. Hence t = 1 and the variety N is irreducible. This completes the proof. 139 Bibliography [1] G. Benkart, T. Gregory and A. Premet, The recognition theorem for graded Lie algebras in prime characteristic, Mem. Amer. Math. Soc. 197 (920), 2009. [2] R. E. Block, Determination of the differentiably simple rings with a minimal ideal, Ann. of Math., Vol. 90, No. 3 (1969) 433 -- 459. [3] C. Chen, The nilpotent variety of W (1; n)p is irreducible, Jour- nal of Algebra and Its Applications, Vol. 18, No. 3 (2019), http://dx.doi.org/10.1142/S0219498819500567, arXiv preprint (2017), arXiv:1707.02881v3 [math.RA]. [4] C. Chevalley, Sur certains groupes simples, Tohoku Math. J. 7 (1955) 14 -- 66 (French). [5] D. A. Cox, J. Little and D. Oshea, Ideals, Varieties, and Algorithms, An In- troduction to Computational Algebraic Geometry and Commutative Algebra, 4th edn., Undergraduate Texts in Mathematics, (Springer International Publishing Switzerland, 2015). [6] V. I. Danilov and V. V. Shokurov, Algebraic Geometry I: Algebraic Curves, Al- gebraic Manifolds and Schemes, ed. I. Shafarevich, translated from Russian by D. Coray, V. N. Shokurov, Encyclopaedia of Mathematical Sciences, Vol. 23, (Springer-Verlag Berlin Heidelberg, 1994). [7] S. P. Demushkin, Cartan subalgebras of simple Lie p-algebras Wn and Sn, Sibirsk. Mat. Zh. 11 (1970) 310 -- 325 (Russian); Siberian Math. J. 11 (1970) (English translation). 140 BIBLIOGRAPHY [8] M. Geck, An Introduction to Algebraic Geometry and Algebraic Groups, Oxford Graduate Texts in Mathematics, Vol. 10 (OUP Oxford, 2013). [9] R. Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics, Vol. 52 (Springer-Verlag New York, 1977). [10] J. E. Humphreys, Linear Algebraic Groups, Graduate Texts in Mathematics, Vol. 21 (Springer-Verlag New York, 1975). [11] N. Jacobson, Classes of restricted Lie algebras of characteristic p. II., Duke Math. J. 10 (1943) 107 -- 121. [12] N. Jacobson, Lie Algebras, (Dover Publications, Inc., New York, 1979). [13] J. Jantzen, Nilpotent Orbits in Representation Theory, in Lie Theory: Lie Alge- bras and Representations, eds. J.-P. Anker, B. Orsted, Progress in Mathematics, Vol. 228 (Birkhauser Boston, Boston, MA, 2004), pp. 1 -- 211. [14] A. I. Kostrikin and I. R. Shafarevich, Cartan pseudogroups and Lie p-algebras, Dokl. Akad. Nauk SSSR 168 (1966) 740 -- 742 (Russian); Soviet Math. Dokl. 7 (1966) 715 -- 718 (English translation). [15] A. I. Kostrikin and I. R. Shafarevich, Graded Lie algebras of finite characteristics, Izv. Akad. Nauk SSSR Ser. Mat. 33 (1969) 251 -- 322 (Russian); Math. USSR Izv. 3(1) (1969) 237 -- 304 (English translation). [16] G. M. Melikian, On simple Lie algebras of characteristic 5, Uspekhi Mat. Nauk 35 (1980) 203 -- 204 (Russian). [17] A. Premet, Lie algebras without strong degeneration, Mat. Sb. 129(1) (1986) 140 -- 153 (Russian); Math. USSR Sb. 57(1) (1987) 151 -- 164 (English translation). [18] A. Premet, On Cartan subalgebras of Lie p-algebras, Izv. Akad. Nauk SSSR Ser. Mat. 50(4) (1986) 788 -- 800 (Russian); Math. USSR Izv. 29(1) (1987) 145 -- 157 (English translation). [19] A. Premet, Regular Cartan subalgebras and nilpotent elements in restricted Lie al- gebras, Mat. Sb. 180(4) (1989) 542 -- 557 (Russian); Math. USSR Sb. 66(2) (1990) 555 -- 570 (English translation). 141 BIBLIOGRAPHY [20] A. Premet, The theorem on restriction of invariants, and nilpotent elements in Wn, Mat. Sb. 182(5) (1991) 746 -- 773 (Russian); Math. USSR Sb. 73(1) (1992) 135 -- 159 (English translation). [21] A. Premet, Nilpotent commuting varieties of reductive Lie algebras, Invent. Math. 154(3) (2003) 653 -- 683. [22] A. Premet, Regular derivations of truncated polynomial rings, Contemp. Math. 652 (2015) 123 -- 139, Amer. Math. Soc., Providence, RI. [23] A. Premet and D. I. Stewart, Classification of the maximal subalgebras of ex- ceptional Lie algebras over fields of good characteristic, arXiv preprint (2017), arXiv:1711.06988v2 [math.RA]. [24] A. Premet and H. Strade, Classification of finite dimensional simple Lie algebras in prime characteristics, Contemp. Math. 413 (2006) 185 -- 214, Amer. Math. Soc., Providence, RI. [25] A. Premet and H. Strade, Simple Lie algebras of small characteristic VI. Com- pletion of the classification, J. Algebra 320 (2008) 3559 -- 3604. [26] R. Ree, On generalized Witt algebras, Trans. Amer. Math. Soc. 83 (1956) 510 -- 546. [27] G. B. Seligman, Modular Lie Algebras, (Springer-Verlag Berlin Heidelberg, 1967). [28] I. R. Shafarevich, Basic Algebraic Geometry 1: Varieties in Projective Space, 3rd edn., translated from the 2007 Russian edition and with notes by M. Reid (Springer-Verlag Berlin Heidelberg, 2013). [29] S. Skryabin, Invariant polynomial functions on the Poisson algebra in character- istic p, J. Algebra 256 (2002) 146 -- 179. [30] H. Strade, Simple Lie Algebras over Fields of Positive Characteristic I: Structure Theory, 2nd edn., De Gruyter Expositions in Mathematics, Vol. 38 (Walter de Gruyter & Co., Berlin, 2017). 142 BIBLIOGRAPHY [31] H. Strade and R. Farnsteiner, Modular Lie Algebras and their Representations, Monographs and Textbooks in Pure and Applied Mathematics, Vol. 116 (Marcel Dekker, New York, 1988). [32] S. A. Tyurin, Classification of tori in the Zassenhaus algebra, Izv. Vyssh. Uchebn. Zaved. Mat. 2 (1998) 69 -- 76 (Russian); Russian Math. (Iz. VUZ) 42(2) (1998) 66-73 (English translation). [33] J. Wei, The nilpotent variety and invariant polynomial functions in the Hamilto- nian algebra, arXiv preprint (2014), arXiv:1401.6532v1 [math.RT]. [34] J. Wei, H. Chang and X. Lu, The variety of nilpotent elements and invariant polynomial functions on the special algebra Sn, Forum Math. 27(3) (2015) 1689- 1715. [35] B. Ju. Weisfeiler, Infinite dimensional filtered Lie algebras and their connection with graded Lie algebras, Funktsional. Anal. i Prilozen. 2(1) (1968) 94-95 (Rus- sian). [36] R. Wilson, Classification of generalized Witt algebras over algebraically closed fields, Trans. Amer. Math. Soc. 153 (1971) 191-210. [37] Y.-F. Yao and B. Shu, Nilpotent orbits of certain simple Lie algebras over trun- cated polynomial rings, J. Algebra 458 (2016) 1-20. 143
1706.01279	2	1706	2017-11-09T16:17:49	On the Dixmier-Moeglin equivalence for Poisson-Hopf algebra	[ "math.RA" ]	We prove that the Poisson version of the Dixmier-Moeglin equivalence holds for cocommutative affine Poisson-Hopf algebras. This is a first step towards understanding the symplectic foliation and the representation theory of (cocommutative) affine Poisson-Hopf algebras. Our proof makes substantial use of the model theory of fields equipped with finitely many possibly noncommuting derivations. As an application, we show that the symmetric algebra of a finite dimensional Lie algebra, equipped with its natural Poisson structure, satisfies the Poisson Dixmier-Moeglin equivalence.	math.RA	math	ON THE DIXMIER-MOEGLIN EQUIVALENCE FOR POISSON-HOPF ALGEBRAS ST´EPHANE LAUNOIS AND OMAR LE ´ON S ´ANCHEZ Abstract. We prove that the Poisson version of the Dixmier-Moeglin equiva- lence holds for cocommutative affine Poisson-Hopf algebras. This is a first step towards understanding the symplectic foliation and the representation theory of (cocommutative) affine Poisson-Hopf algebras. Our proof makes substantial use of the model theory of fields equipped with finitely many possibly noncom- muting derivations. As an application, we show that the symmetric algebra of a finite dimensional Lie algebra, equipped with its natural Poisson structure, satisfies the Poisson Dixmier-Moeglin equivalence. Contents Introduction 1. 2. Some preliminaries 2.1. Poisson-Hopf algebras 2.2. The Poisson Dixmier-Moeglin equivalence 2.3. Differential algebras 3. On affine D-varieties and isotriviality 4. The D-DME for commutative affine D-groups over constants 5. Main results on Poisson-Hopf algebras References 1 3 3 4 5 6 13 15 18 1. Introduction We fix a field k of characteristic zero, and recall that by an affine k-algebra one means a finitely generated one that is a domain. The aim of this note is to study the representation theory and the geometry of affine Poisson-Hopf k-algebras via methods from model theory of differential fields. For the reader unfamiliar with Poisson-Hopf algebras, let us mention at this point that these include symmetric algebras of finite-dimensional Lie algebras (endowed with their natural Poisson bracket, see Section 2.1). For further examples, we refer the reader to [10] and [14]. As it is often the case, classifying simple representations or symplectic leaves of Poisson(-Hopf) algebras is too wide a problem, and so we are approaching it by studying the so-called Poisson-primitive ideals. There are several equivalent Date: November 5, 2018. 2010 Mathematics Subject Classification. Primary 17B63; Secondary 03C98, 12H05, 16T05. Key words and phrases. Dixmier-Moeglin equivalence, Poisson-Hopf algebras, model theory of differential fields. The work of SL was supported by EPSRC grant EP/N034449/1. 1 2 ST´EPHANE LAUNOIS AND OMAR LE ´ON S ´ANCHEZ ways to define these (prime Poisson) ideals. Given our representation theoretic and geometric motivation, we just give two equivalent definitions at this stage. Let A = O(V ) be an affine Poisson k-algebra (so that V is a Poisson variety over k). The Poisson-primitive ideals of A are the defining ideals of the Zariski closure of the symplectic leaves of V . Equivalently, they are precisely the annihilators of the simple Poisson A-modules. Thus, classifying Poisson-primitive ideals of a Poisson(- Hopf) algebra is a first step towards understanding both its symplectic foliation and its representation theory. The goal of this paper is to provide a topological criterion to characterise Poisson- primitive ideals among prime Poisson ideals of a Poisson-Hopf algebra. More pre- cisely, we prove that Poisson-primitive ideals of a cocommutative affine Poisson- Hopf algebra are exactly those prime ideals that are Poisson-locally closed. In [1], the authors together with Bell and Moosa proved that Poisson-primitive ideals co- incide with the so-called Poisson-rational ideals. So, combining this with our results here, we obtain that for affine cocommutative Poisson-Hopf algebras, the notions of Poisson-rational, Poisson-primitive and Poisson-locally closed coincide. The co- incidence of these three notions is often referred to as the Poisson Dixmier-Moeglin equivalence (see Section 2.2), so that we can state our main result as follows. Theorem 1. Any cocommutative affine Poisson-Hopf k-algebra satisfies the Pois- son Dixmier-Moeglin equivalence. We note that the Poisson Dixmier-Moeglin equivalence does not always hold for a Poisson algebra. By [17, 1.7(i), 1.10], in an affine Poisson k-algebra A, we have that Poisson-locally closed implies Poisson-primitive, and Poisson-primitive implies Poisson-rational (this also follows from Proposition 3.4 below). It was shown in [1], that when A has Krull-dimension at most three, then Poisson-rational implies Poisson-locally closed, and so the Poisson-DME holds in this case. However, in that same paper, counterexamples of Krull-dimension d were build, for any d ≥ 4. Hence, the main point of this note is to show that those (counter-)examples cannot admit a cocommutative Poisson-Hopf algebra structure; and that in fact in this case Poisson-rational does imply Poisson-locally closed. While in general one cannot remove the Hopf algebra assumption in Theorem 1, one natural question to ask at this point is: can we remove the cocommutative assumption? That is, does the Poisson Dixmier-Moeglin equivalence hold in any affine Poisson-Hopf algebra? While we currently do not have an answer, in Re- mark 5.6(2) below we suggest how one could address this question. We note that in the differential-Hopf algebra context in a single derivation the cocommutative assumption can indeed be removed, see [2, Theorem 2.19], and also that Bell and Leung have asked a similar question in the noncommutative setting, see [3, Con- jecture 1.3]. As symmetric algebras of finite dimensional Lie algebras are examples of cocom- mutative affine Poisson-Hopf algebras, Theorem 1 applies to this family of Poisson k-algebras. Thus, we obtain (with very different methods) a Poisson analogue of the foundational result of Dixmier and Moeglin, later generalized by Irving and Small, that asserts that primitive ideals in the enveloping algebra U (g) of a finite- dimensional Lie k-algebra g are precisely the locally closed prime ideals of U (g) [8]. ON THE DME FOR POISSON-HOPF ALGEBRAS 3 Theorem 2. If A is the symmetric algebra of a finite dimensional Lie k-algebra g, equipped with its natural Poisson bracket, then A satisfies the Poisson Dixmier- Moeglin equivalence. We expect that our results on the representation theory of Poisson-Hopf algebras will help us better understand representations of Hopf algebras in general. This is justified by the fact that connected Hopf algebras arise (in a rough sense) as deformations of Poisson-Hopf algebras [19]. It is important to note that a significant part of the proof of Theorem 1 makes substantial use of the model theory of differential fields, via the theory of alge- braic D-varieties and D-groups (see Sections 3 and 4). The novelty of this paper, compared to [1] or [2] where the model theory of ordinary differential fields was used, is that we work in the context of several possibly noncommuting derivations. While the model theory of differential fields with commuting derivations has fruit- fully been applied in other areas of mathematics (for instance, in differential Galois theory, see [13]), to the authors knowledge this paper contains the first application of the model-theoretic properties of the theory of differential fields where no com- mutativity assumption is made among the derivations. We expect (and hope) that the ideas presented here will motivate the further use of these tools in new areas of algebra, and perhaps initiate the study of the model theory of Poisson rings. 2. Some preliminaries Recall that for us k denotes a field of characteristic zero, and that by an affine algebra we mean a finitely generated one that is a domain. 2.1. Poisson-Hopf algebras. Recall that a Poisson k-algebra is a commutative k-algebra A equipped with a Lie bracket {−, −} such that {a, bc} = {a, b}c + b{a, c}, for all a, b, c ∈ A. In other words, for each a ∈ A, the map {a, −} : A → A is a derivation. Given a Poisson k-algebra (A, {−, −}) the tensor algebra A ⊗ A can be naturally equipped with a Poisson k-algebra structure as follows; define {a ⊗ b, a′ ⊗ b′} = {a, a′} ⊗ b b′ + a a′ ⊗ {b, b′} for a, b, a′a, b′ ∈ A and extend to all of A ⊗ A by k-linearity. Definition 2.1. A Poisson-Hopf k-algebra A is a Poisson k-algebra with the addi- tional structure of a Hopf algebra such that the Poisson bracket {−, −} commutes with coproduct ∆; that is, for all a, b ∈ A. ∆({a, b}) = {∆(a), ∆(b)} Remark 2.2. Let (A, {−, −}) be a Poisson algebra with a Hopf algebra structure. In order to prove that the Poisson bracket commutes with coproduct it suffices to check that if G is a set of generators of A (as a k-algebra) then ∆({a, b}) = {∆(a), ∆(b)} for all a ∈ G and b ∈ G. Indeed, suppose a ∈ G and b ∈ A, then b = f (g1, . . . , gs) for some polynomial f over k and gi ∈ G. Since {a, −} and {∆(a), −} are k-linear 4 ST´EPHANE LAUNOIS AND OMAR LE ´ON S ´ANCHEZ derivations, we get ∆({a, b}) =∆ s Xi=1 Xi=1 ∂f ∂ti ∂f ∂ti (g1, . . . , gs) · {a, gi}! (∆(g1), . . . , ∆(gs)) · {∆(a), ∆(gi)} s = ={∆(a), f (∆(g1, . . . , ∆(gs))} ={∆(a), ∆(b)}. The general case, when a, b ∈ A, follows from this in similar fashion; now writing a as h(g1, . . . , gs) and using the fact that {−, b} and {−, ∆(b)} are k-linear derivations. We now describe a general example of cocommutative Poisson-Hopf algebra. Namely, let A be the symmetric algebra of a finite dimensional Lie k-algebra g. Such a (symmetric) algebra A is of the form k[x1, . . . , xs] where the xi's form a k-basis for the Lie algebra g. It is well known that A becomes a Poisson algebra with Poisson bracket defined by: {f, g} :=Xi<j [xi, xj ](cid:18) ∂f ∂xi ∂g ∂xj − ∂f ∂xj ∂g ∂xi(cid:19) . Moreover, we can equip A with a Hopf algebra structure where the xi's are all primitive (so that A is cocommutative). One can check that then the Poisson bracket commutes with the coproduct. Indeed, it follows from Remark 2.2 that we only need to prove that ∆({xi, xj}) = {∆(xi), ∆(xj )} for all i, j. Using the fact that every element of g is primitive and that ∆ is an algebra homomorphism, we get {∆(xi), ∆(xj )} ={1 ⊗ xi + xi ⊗ 1, 1 ⊗ xj + xj ⊗ 1} =1 ⊗ {xi, xj} + {xi, xj} ⊗ 1 =1 ⊗ [xi, xj] + [xi, xj] ⊗ 1 =∆([xi, xj]) =∆({xi, xj }) for all i, j. Thus A is indeed a cocommutative affine Poisson-Hopf k-algebra. Further examples of (affine) Poisson-Hopf algebras are given by the coordinate ring of Poisson affine algebraic groups [10]. These examples are not, however, generally cocommutative, unless of course the algebraic group is abelian. 2.2. The Poisson Dixmier-Moeglin equivalence. Let (A, {−, −}) be a Poisson k-algebra . An ideal I of A is a Poisson ideal if {A, I} ⊆ I. The Poisson-center of A is defined as Zp(A) = {a ∈ A : {A, a} = 0}. Recall that when A is a domain, there is a natural Poisson structure on Frac A (induced by the quotient rule of derivations). ON THE DME FOR POISSON-HOPF ALGEBRAS 5 We denote by SpecP A the subspace of SpecA consisting of Poisson ideals. The Poisson core of an ideal I of A is defined as the largest Poisson ideal contained in I. A prime Poisson ideal P of A is said to be • Poisson-locally closed if \P (Q∈SpecP A Q 6= P • Poisson-primitive if P is the Poisson core of a maximal ideal of A. • Poisson-rational if ZP (Frac A/P ) is an algebraic extension of k. We say that a A satisfies the Poisson Dixmier-Moeglin equivalence (or Poisson- DME) if a prime Poisson ideal of A is Poisson-locally closed iff it is Poisson-primitve iff it is Poisson-rational. Let us remark that, by [5, Lemma 3.5], the above defini- tion of Poisson-primitive does correspond to the equivalent definitions given in the introduction. 2.3. Differential algebras. We now briefly recall some facts about differential k- algebras that will be useful in subsequent sections. For any k-algebra A, we denote by Derk(A) the A-vector space of k-linear derivations on A. Remark 2.3. Let (A, {−, −}) be a Poisson k-algebra and G a set of generators of A. Recall that a Hamiltonian of A is an element of Derk(A) of the form {a, −} for some a ∈ A. An easy computation shows that the A-vector subspace of Derk(A) spanned by the Hamiltonians of A is equal to spanA({a, −} : a ∈ G). Due to the above remark, to check that an ideal in a Poisson algebra is a Poisson ideal one only needs to check that it is invariant under the Hamiltonians of a set of generators. More generally, we have Lemma 2.4. Let (A, {−, −}) be a Poisson k-algebra. If D ⊆ Derk(A) is such that spanA D = spanA({a, −} : a ∈ A), then (1) an ideal I of A is Poisson iff it is a D-ideal (i.e., invariant under D), and (2) the Poisson-center ZP (A) equals the D-constants of A (i.e., ∩δ∈D ker δ). Furthermore, if A is a domain, ZP (Frac A) = D-constants(Frac A). Proof. This is an easy exercise. We leave the details to the reader. (cid:3) Suppose A is a commutative k-algebra equipped with a family of k-linear deriva- tions D. We denote by SpecDA the subspace of SpecA consisting of D-ideals. The D-core of an ideal I of A is defined as the largest D-ideal contained in I. A prime D-ideal P of A is said to be • D-locally closed if \P (Q∈SpecD A Q 6= P • D-primitive if P is the D-core of a maximal ideal of A. • D-rational if the D-constants of Frac A/P is an algebraic extension of k. 6 ST´EPHANE LAUNOIS AND OMAR LE ´ON S ´ANCHEZ We say that a A satisfies the D-Dixmier-Moeglin equivalence (or D-DME) if a prime D-ideal of A is D-locally closed iff it is D-primitve iff it is D-rational. The following is an easy consequence of Lemma 2.4. Corollary 2.5. Let (A, {−, −}) be a Poisson k-algebra. If D ⊆ Derk(A) is such that spanA D = spanA({a, −} : a ∈ A), then a prime ideal P of A is (1) Poisson-locally closed iff it is D-locally closed, (2) Poisson-primitive iff it is D-primitive, and (3) Poisson -rational iff it is D-rational. Consequently, (A, {−, −}) satisfies the Poisson-DME iff it satisfies the D-DME. Given a commutative Hopf k-algebra A equipped with a family of k-linear deriva- tions D, we say that A is a differential-Hopf algebra if each derivation commutes with coproduct; that is, δ(∆(a)) = ∆(δa) for all a ∈ A and δ ∈ D. Here recall that the derivations D naturally lift to A ⊗ A as follows; for all a, b ∈ A and extend by k-linearity to all of A ⊗ A. δ(a ⊗ b) = δa ⊗ b + a ⊗ δb Remark 2.6. (1) As we did in Remark 2.2, one can check that, in a commutative Hopf k- algebra A equipped with k-linear derivations D, the derivations D commute with coproduct if and only if δ(∆(a)) = ∆(δa) for all δ ∈ D and a varying in a set of generators of A. (2) Suppose (A, {−, −}) is an affine Poisson-Hopf k-algebra, we do not know if there is D ⊆ Derk(A) with spanA D = spanA({a, −} : a ∈ A) and such that (A, D) is a differential-Hopf algebra. Nonetheless, in Propo- sition 5.5, we prove that it is possible to find such D in the case when k is algebraically closed and A is cocommutative. 3. On affine D-varieties and isotriviality In this section we present the basics of the theory of affine algebraic D-varieties in the context of finitely many (possibly noncommuting) derivations, together with the notions of isotriviality and compound isotriviality. It is worth noting that the theory of D-varieties in the context of commuting derivations appears in [6]. We make, somewhat freely (specially compared to [1, 2]), use of basic model- theoretic terminology, for which [15] should suffice. The appropriate model-theoretic context here is that of fields equipped with finitely many (possibly noncommuting) derivations. Fix a positive integer m. We work in the first-order language of differ- ential rings equipped with m derivations Ldif f = Lrings ∪ {δ1, . . . , δm}. One can easily axiomatize the class of differential fields (K, δ1, . . . , δm) of charac- teristic zero (where no commutativity assumption is made among the derivations). ON THE DME FOR POISSON-HOPF ALGEBRAS 7 Such a differential field is called existentially closed if any quantifier-free Ldif f - formula with a realisation in a differential field extension of K already has a real- isation in K. Note that such differential fields exist (by a standard Zorn's lemma and chain construction argument). It turns out that there is a first-order axiomatization of the class of existentially closed differential fields of characteristic zero. This is a consequence of the general results in [16], and following their notation we denote this theory by D -CF0. In that same paper, the authors established that this is a complete stable theory with quantifier elimination and elimination of imaginaries. Moreover, they showed that given (U, δ1, . . . , δm) = D -CF0, if K is a differential subfield then K = dclU (K), and if additionally K is algebraically closed then K = aclU (K). We now fix a sufficiently large saturated model (U, D = {δ1, . . . , δm}) = D -CF0. This means that given a small (i.e., K < U) differential subfield K of U, and a (possibly infinite) collection Σ of Ldif f -formulas with parameters from K, if every finite subcollection of Σ is satisfiable in U then so is all of Σ. One of the most important definable subsets of U is its subfield of constants, which is defined as CU = m \i=1 ker δi. A subset of U n that is an arbitrary (possibly infinite) intersection of definable sets will be called a type-definable set. Remark 3.1. The field CU is algebraically closed and it is purely stably embedded. This means that any subset of Cn U that is definable in the Ldif f -structure U, over some differential subfield K, is actually definable in the Lrings-structure CU over CK. In particular, by ω-stability of the theory of algebraically closed fields, if G ⊆ Cn U is a type-definable group over K in the differential structure (U, δ1, . . . , δm), then G is definable over CK in the pure-field structure CU . We now discuss affine algebraic D-varieties. Fix a (small) differential subfield K < U. We say that a Zariski closed set V ⊆ U n defined over K is a D-variety if its coordinate ring K[V ] is equipped with a family of m derivations ¯∂ = {∂1, . . . , ∂m} such that ∂i extends δiK. We now wish to give a more algebro-geometric characterization of affine D- varieties. Given a Zariski closed V ⊆ U n over K and δ ∈ D, the δ-prolongation of V is the Zariski-closed set τδ ⊆ U 2n defined by the equations f (¯x) = 0 and ∂f ∂xi n Xi=1 (¯x) · yi + f δ(¯x) = 0 for all f ∈ I(V /K) := {f ∈ K[¯x] : f (V ) = 0}, where ¯x = (x1, . . . , xn) and f δ ∈ K[¯x] is obtained by applying δ to the coefficient of f . It is easy to check that it suffices to vary f in a family of generators of the ideal I(V /K). Consequently, if V is defined over a field k of constants (i.e., k < CU ), then τδV is nothing more that the tangent bundle T V of V . More generally, the D-prolongation of V , denoted by τDV ⊂ U n(m+1), is defined as the fibred-product τDV = τδ1 V ×V · · · ×V τδm V. 8 ST´EPHANE LAUNOIS AND OMAR LE ´ON S ´ANCHEZ Note that τDV comes equipped with a canonical projection map τDV → V . The D-prolongation has the characteristic property that for any a ∈ V we have that (a, δ1a, . . . , δma) ∈ τDV ; in other words, the map ¯x 7→ (¯x, δ1 ¯x, . . . , δm ¯x) defines a differential regular section of τDV → V . If V is defined over a field k of constants, then τDV equals the m-fold fibred- product of T V and hence it comes with a canonical (algebraic) section, the zero section s0 : V → τDV . On the other hand, for arbitrary V defined over K, the existence of an algebraic regular section s : V → τDV over K turns out to be equiv- alent to a D-variety structure on V . Indeed, if ¯∂ = {∂1, . . . , ∂m} are derivations on the coordinate ring K[V ] extending those on K, and we let ¯z = (z1, . . . , zn) be its coordinate functions and set s(¯z) = (¯z, ∂1(¯z), . . . , ∂m(¯z)) where ∂i(¯z) = (∂i(z1), . . . , ∂i(zn)), then it is not hard to check (using the fact that the ∂i's are derivations) that this s yields a section of τDV → V which is regular and over K. On the other hand, any such section s = (Id, s1, . . . , sm) corresponds to the derivations on K[V ] induced by setting ∂i(¯z) = si(¯z). From now on, we will usually refer to a D-variety as a pair (V, s) where V is an affine algebraic variety (viewed as a Zariski closed subset of U n) and s is a section of τDV → V . A point a ∈ V will be called a D-point of V if s(a) = (a, δ1(a), . . . , δm(a)). We will denote the set of D-points of V by (V, s)#. Note that this is an example of a definable set in the structure (U, δ1, . . . , δm) and in fact this will be the main source of such examples for us. Also, note that if V is defined over a field of constants and s = s0 (the zero section), then (V, s)# equals V (CU ), the CU -points of V . A Zariski closed subset W of a D-variety (V, s) is said to be a D-subvariety if s(W ) ⊆ τDW ; of course, in this case (W, sW ) will be a D-variety. Also, a regular map f : V → W between D-varieties (V, s) and (W, t) is said to be a D-morphism if f maps D-points to D-points. Remark 3.2. It is easy to check that W ⊂ V is a D-subvariety iff the ideal I(W/K) ⊂ K[V ] is a ¯∂-ideal. Also, a regular map between D-varieties f : V → W is a D-morphism iff the pull-back f ∗ : K[W ] → K[V ] is a ¯∂-ring homomorphism. Lemma 3.3. Let (V, s) and (W, t) be affine algebraic D-varieties over K and f a D-morphism between them. Then, (1) each K-irreducible component of V is a D-subvariety, (2) the set of D-points of V is Zariski-dense in V , (3) if X is a D-subvariety of W , then f −1(X) is a D-subvariety of V , and (4) if Y is a D-subvariety of V , then the Zariski closure of f (Y ) is a D- subvariety of W . Proof. (1) Let W be a K-irreducible component of V and a a Zariski K-generic point of W . It suffices to show that s(a) ∈ τDW . As a is not contained in any of the other K-irreducible components of V , from the nature of the equations defining ON THE DME FOR POISSON-HOPF ALGEBRAS 9 τDW , we get that τDWa (the fibre above a) coincides with τDVa. The claim now follows. (2) By (1), we may assume that V is K-irreducible. Let W be a proper Zariski closed subset of V and a a Zariski K-generic point of V (hence a /∈ W ). Let b = (b1, . . . , bm) ∈ U nm be such that (a, b) = s(a), since b ∈ τDVa, the equations of τDV yield that there are derivations ∂i : K(a, b) → K(a, b), for i = 1, . . . , m, such that ∂iK = δiK and ∂i(a) = bi (see for instance [12, Chapter 7, §5]). By saturation of U, there is a differential field embedding (K(a, b), ¯∂) → (U, D) fixing K. So there is point (a′, b′) in U n(m+1), namely the image of (a, b), such that a′ ∈ V \ W and s(a′) = (a′, b′) = (a, δ1a′, . . . , δma′). That is, a′ is a D-point of V not in W . Thus, the set of D-points is dense in V . (3) To prove that f −1(X) is a D-subvariety of V it suffices to show that I(f −1(X)/K) ⊆ K[V ] is a D-ideal. Recall that this ideal is given as the radical ideal generated by f ∗(I(X/K)). As radical ideals of D-ideals are again D-ideals (see [9, Lemma 1.8]), it suffices to show that the ideal generated by f ∗(I(X/K)) in K[V ] is a D-ideal. Let δ ∈ D, then for an expression of the form Pi gif ∗(hi), with gi ∈ K[V ] and hi ∈ I(X/K), we have δ Xi gif ∗(hi)! =Xi δ(gi)f ∗(hi) + gif ∗(δhi), where we have used that f is a D-morphism (so f ∗ commute with δ). Since X is a D-subvariety of W , δhi ∈ I(X/K), and so the above term is in the ideal generated by f ∗(I(X/K)), as desired. (4) To prove that Z, the Zariski closure of f (Y ), is a D-subvariety of W , it suffices to show that each K-irreducible component of Z is such. Thus, we assume that Z is K-irreducible. Let a be a Zariski K-generic D-point of an K-irreducible component of Y that maps dominantly onto Z. Then, b = f (a) is a Zariski K- generic of Z, and since f is a D-morphism we have that b is a D-point of W . Thus, s(b) = (b, δ1b, . . . , δmb) ∈ τDZ, and, by Zariski genericity of b, we must have s(Z) ⊆ τDZ, as desired. (cid:3) From the above lemma we see that if (V, s) is K-irreducible (meaning that V is K-irreducible), then it contains a Zariski K-generic D-point. Indeed, if this were not the case, by saturation of U there would be a finite collection of proper Zariski closed subsets of V defined over K that contain all D-points of V . But as V is K-irreducible this finite collection does not cover all of V and hence we contradict part (2) of the lemma. Note that if a is a Zariski K-generic D-point of V (assuming V is K-irreducible), then the function field K(V ) ∼= K(a) equipped with the derivations ¯∂ is a differential subfield of (U, D). Thus, from now on, we will assume that (K[V ], ¯∂) ≤ (K(V ), ¯∂) < (U, D), and, moreover, we identify ¯∂ with D. 10 ST´EPHANE LAUNOIS AND OMAR LE ´ON S ´ANCHEZ Proposition 3.4. Let k < CU and (V, s) be a k-irreducible affine algebraic D- variety. Then, for any prime D-ideal P of k[V ] we have D-locally closed =⇒ D-primitive =⇒ D-rational. Furthermore, suppose V is geometrically irreducible (i.e., kalg-irreducible), if in kalg[V ] a prime D-ideal is D-rational only if it is D-locally closed, then the same holds in k[V ]. Proof. By passing to the quotient k[V ]/P we may assume that P = (0). Now assume (0) is D-locally closed. This means that \Q∈SpecD k[V ]\(0) Q 6= (0). At the level of D-subvarieties of V (recall that each such Q corresponds to a proper D-subvariety of V ), this is equivalent to V having a proper D-subvariety W over k that contains all such. Take a point a ∈ V \ W (kalg). Then I(a/k) ⊂ k[V ] is a maximal ideal and any D-ideal inside it must be zero (as a /∈ W ). This shows that (0) is the D-core of a maximal ideal of k[V ]; in other words, (0) is D-primitive. On the other hand, assume (0) is D-primitive. That is, there is a maximal ideal m with D-core (0). Let a ∈ V (kalg) be such that m = I(a/k). Now let f be a D-constant of k(V ). We must show that f algebraic over k. We first show that a is not in the singular locus of f . Towards a contradiction suppose it is. Then, for every representation f = p q we have q(a) = 0; that is, q ∈ m. Since f is a D-constant, we have that for each δ ∈ D 0 = δf = δp · q − p · δq q2 δq = p and so, either δq = 0, or δp q = f in which case δq(a) = 0. In any case, δq ∈ m. Repeating this process we obtain that δ′δq ∈ m for any δ′ ∈ D, and so on, hence we get that the D-ideal generated by q is contained in m. This contradicts the fact that the D-core of m is (0), and so f is defined at a. Write f = p q where q(a) 6= 0. Now let h ∈ k[t] be the minimal polynomial of f (a) ∈ kalg. There is a sufficiently large integer s such that if we set r = qs · (h ◦ f ) then r ∈ k[V ]; and, since r(a) = qs(a)h(f (a)) = 0, we also have that r ∈ m. Let δ ∈ D, since δ(h ◦ f ) = (h′ ◦ f ) · δf = 0, we get that δr = δ(qs) · (h ◦ f ); and so δr(a) = 0, implying that δr ∈ m. Repeating this process we obtain that δ′δr ∈ m for any δ′ ∈ D, and so on, hence we get that the D-ideal generated by r is contained in m. By the choice of m, r must be zero. This implies that h(f ) = 0 and so f ∈ kalg. This shows D-rationality of (0). For the 'furthermore' clause, suppose V is geometrically irreducible and that a prime D-ideal of kalg[V ] is D-rational only if it is D-locally closed. Let P be a prime D-rational ideal of k[V ]. We must show that P is D-locally closed. Let W be the k-irreducible D-subvariety of V that corresponds to P . Also, let Y be one of the kalg-irreducible components of W . By Lemma 3.3(1), Y is a D-subvariety of V (over kalg). Let b be a Zariski kalg-generic D-point of Y ; then b is a Zariski k-generic point D-point of W . Since kalg(Y ) = kalg(b) ⊆ k(b)alg ON THE DME FOR POISSON-HOPF ALGEBRAS 11 and the D-constants of k(b) are algebraic over k (by D-rationality of P ), we get that the D-constants of kalg(Y ) are precisely kalg. In other words, the D-ideal I(Y /kalg) of kalg[V ] is D-rational; by our assumption, this ideal is D-locally closed. That is, Y contains a proper D-subvariety Y ′ over kalg that contains all such. Letting W ′ be the Zariski k-closure of Y ′, we obtain a proper D-subvariety of W over k containing all such; equivalently, P is D-locally closed. (cid:3) We will need one more piece of model-theoretic terminology. We denote by AutD(U/K) the differential automorphisms of U fixing K. Given a tuple a ∈ U n, we define the (complete) type of a over K as the orbit of a under the action of AutD(U/K) on U n; in other words, tp(a/K) := {b ∈ U n : σ(a) = b for some σ ∈ AutD(U/K)}. In model-theoretic parlance, we are identifying a type with its set of realizations in U. A type tp(a/K) is always given as an (infinite) intersection of Ldif f -definable sets in U; namely, the intersections of all definable sets over K containing a. We will call the set defined by tp(a/K), in U n, the set of its realisations. We say that tp(a/K) is isolated if its set of realizations is a definable set (it will necessarily be definable over K). Remark 3.5. If (V, s) is a K-irreducible D-variety and a is a Zariski K-generic D- point, then the type tp(a/K) is precisely the set of all Zariski K-generic D-points of V . Indeed, if b is another Zariski K-generic D-point, then, by saturation of U, there is a field automorphism σ of U such that b = σ(a). But since both, a and b, are D-points, σ is in fact a differential homomorphism, and so σ ∈ AutD(U/K). The other implication is obvious. Proposition 3.6. Let k be a subfield of CU . Let (V, s) be a k-irreducible D-variety, and a a Zariski k-generic D-point of V . Then tp(a/k) is isolated if and only if (0) is a D-locally closed D-ideal of k[V ]. Proof. Suppose tp(a/k) is isolated. Then its set of realizations is definable over k. By quantifier elimination and the fact that all such realizations are D-points of V , this definable set must the form (V \ W ) ∩ (V, s)# where W is a proper Zariski closed subset of V defined over k. Since, by Remark 3.5, the realisations of tp(a/K) is precisely the set of all Zariski k-generic D-points of V , all D-subvarieties of V defined over k must be contained in W . At the level of D-ideals of k[V ], this is equivalent to saying that all the nonzero D-ideals contain I(W/k). This shows that \Q∈SpecD k[V ]\(0) Q 6= (0), and so (0) is D-locally closed. On the other hand, assume (0) is D-locally closed. Let X be the proper D- subvariety of V corresponding to the D-ideal ∩Q∈SpecD k[V ]\(0)Q. Note that X contains all proper D-subvarieties of V defined over k. If a ∈ (V \ X) is a D-point, then it must a Zariski k-generic of V (otherwise, its Zariski k-locus would yield a proper D-subvariety over k not contained in X). This shows that the set (3.1) (V \ X) ∩ (V, s)# coincides with the set of Zariski k-generic D-points of V . By Remark 3.5, this latter set is precisely tp(a/K), so this type is given by the formula (3.1) and therefore it is isolated. (cid:3) 12 ST´EPHANE LAUNOIS AND OMAR LE ´ON S ´ANCHEZ Definition 3.7 (c.f. [2]). Let (V, s) be a K-irreducible affine algebraic D-variety. (1) We say that (V, s) is isotrivial if there is a differential field F < U extension of K and an injective D-morphism over F from (V, s) to (W, s0), where W is defined over CF and s0 is the zero section. (2) We say that (V, s) is compound isotrivial in ℓ-steps if there is an sequence of K-irreducible D-varieties (Vi, si), i = 1, . . . , ℓ, and dominant D-morphisms over K V = Vℓ fℓ / Vℓ−1 fℓ−1 / · · · f2 / V1 f1 / V0 = 0 such that for each i = 0, 1, . . . , ℓ − 1, if a is a Zariski K-generic D-point of Vi, then f −1 i+1(a) is a K(a)-irreducible D-subvariety of Vi+1 (by Lemma 3.3(3)). i+1(a) is isotrivial. Here note that f −1 We now prove one of the key results of the paper. Theorem 3.8. Let V be a K-irreducible affine compound isotrivial D-variety. If CK(V ), the D-constants of K(V ), is algebraic over CK, then the type of a Zariski K-generic D-point of V is isolated. Proof. We let a be a Zariski K-generic D-point of V . We must show that tp(a/K) is isolated. Suppose V is compound isotrivial in ℓ-steps. We proceed by induction on ℓ. For the base case, ℓ = 1, V must be isotrivial. By definition, there is a definable (with possibly additional parameters) injective map from the D-points of V to CU . In model theoretic terms this means that the type tp(a/K) is internal to CU . This in turn implies that tp(a/K alg) is internal to CU as well. On the other hand, the condition that CK(a) = CK(V ) ⊂ Calg K translates in model-theoretic terms to the type tp(a/K alg) being weakly orthogonal to CU . Indeed, to see this, one must show that K alg(a) ∩ K alg(CU ) is contained in K alg. Taking d in the intersection we get d = h(c) for some polynomial h over K alg and tuple c from CU . If X is the set of tuples x from CU such that d = h(x), then, by Remark 3.1, X is Lrings-definable in CU over CK alg(a). Hence, there is a tuple c′ from Calg K that satisfies X; and so d = h(c′) ∈ K alg, as desired. K(a) ⊂ Calg It is a well known model-theoretic fact on binding groups of automorphisms (for a proof see [7, Appendix B]) that the above two conditions (internality and weak orthogonality) imply the existence of a type-definable group G ⊆ Cn U over K alg that acts definably (over K alg) and transitively on tp(a/K alg). By Remark 3.1, G must be definable (over CK alg ), and so the realisations of tp(a/K alg) form a definable set; in other words, tp(a/K alg) is isolated. Since the type of any tuple of K alg over K isolated, we must have that tp(a/K) is isolated as well, as desired. Now assume ℓ > 1, and suppose we have V = Vℓ fℓ / Vℓ−1 fℓ−1 / · · · f2 / V1 f1 / V0 = 0 as in Definition 3.7(2). Let b = fℓ(a), then b is a Zariski K-generic D-point of Vℓ−1. Since CK(b) ⊆ CK(a), we have that CK(b) is algebraic over CK; and so, by induction, tp(b/K) is isolated. We now claim that tp(a/K(b)) is also isolated. Indeed, a is a Zariski K(b)-generic D-point of W := f −1(b). By definition of compound isotriviality, W is isotrivial, and so by the base case (ℓ = 1) the type of a over K(b) is isolated. The result now follows from the fact that both types, tp(b/K) and / / / / / / / / ON THE DME FOR POISSON-HOPF ALGEBRAS 13 tp(a/K(b)), are isolated (this is an easy exercise but a proof appears in [15, Lemma 4.2.21]). (cid:3) Corollary 3.9. Let k be a subfield of CU and let (V, s) be a k-irreducible affine compound isotrivial D-variety. If (0) is a D-rational ideal of k[V ], then it is also D-locally closed. Proof. The fact that (0) is D-rational translates to Ck(V ) being algebraic over k = Ck (this equality holds since k < CU ). Now, by Theorem 3.8, the type of a Zariski k- generic D-point of V is isolated; which, by Proposition 3.6, implies that (0) is a D-locally closed ideal of k[V ], as desired. (cid:3) 4. The D-DME for commutative affine D-groups over constants In this section we discuss affine algebraic D-groups, and show that the connected commutative ones defined over the constants are compound isotrivial in 2-steps. We carry on the notation from the previous section; in particular, (U, D = {δ1, . . . , δm}) is a sufficiently large saturated model of D -CF0 and all base differential fields, k or K, of parameters are assumed to be small (i.e., of cardinality less than that of U). Given an affine algebraic group G over K, just as the tangent bundle T G of G has the structure of an algebraic group, the D-prolongation of G also has a canonical structure of an algebraic group (over K). Definition 4.1. An affine algebraic D-group (G, s) over K is an affine algebraic group with the additional structure of a D-variety s : G → τDG, both over K, such that s is a group homomorphism. Remark 4.2. At the level of the coordinate ring K[G], a section s : G → τDG is a group homomorphism if and only if the derivations D on K[G] commute with coproduct ∆. Indeed, the section s = (Id, s1, . . . , sm) : G → τDG = τδ1 G ×G · · · ×G τδm G is a group homomorphism iff each section (Id, si) : G → τδi G, for i = 1, . . . , m, is a group homomorphism. But each such section is a group homomorphism iff δi commutes with ∆ (this is well known but a proof appears in [2, Lemma 2.18]). If G is defined over a field of constants k, then the zero section s0 : G → τDG is a group homomorphism. Thus, in this case (G, s0) is a D-group. Our main focus here is on the (compound) isotriviality of connected commutative D-groups over a field of constants. It turns out that to establish compound isotriviality of such D-groups one essentially only needs to understand the commutative unipotent case. Lemma 4.3. Suppose (G, s) is an algebraic D-group over K. If G = Gn n, then (G, s) is isotrivial. a for some Proof. By the comments in Remark 4.2, each section (Id, si) : Gn a → G2n a is a group homomorphism. Thus, si : Gn Ai ∈ M atn(K), i = 1, . . . , m. We can find b ∈ Gn a → Gn a is of the form si(¯x) = Ai ¯x for some a = Gn a (U) such that δib = Aib 14 ST´EPHANE LAUNOIS AND OMAR LE ´ON S ´ANCHEZ for i = 1, . . . , m. Set f : Gn computation now to check that for every D-point a of Gn a be the map f (¯x) = ¯x − b. a we have a → Gn It is an easy δi(f (a)) = 0 Hence, f is an injective D-morphism between (Gn that s0 is the zero section). The result follows. a , s) and (Gn a , s0) (where recall (cid:3) Proposition 4.4. Suppose (G, s) is a connected algebraic D-group over a field of constants k. If G is commutative, then every k-irreducible D-subvariety is com- pound isotrivial in 2-steps. Proof. As G is over a field of constants, namely k, it comes equipped with the zero section s0 as well. Set f (¯x) = s(¯x) · s0(¯x)−1 where the product and inverse occur in τDG. Then f is a regular (algebraic) map from G to the m-th power of the Lie algebra L(G) of G (here we use again that k < CU ). Moreover, as G is commutative, f is group homomorphism. Let H be the image of f ; then H = Gn a for some n. The section s induces, via f , a D-group structure t on H. Thus, f becomes a surjective group D-morphism between (G, s) and (H, t). We claim that G f / H / 0 witnesses the compound isotriviality of (G, s). Indeed, (H, t) is isotrivial by Lemma 4.3, so it suffices to show that if a is a Zariski k-generic D-point of H then f −1(a) is isotrivial. Let g be a D-point of f −1(a) and N := ker f . Then g induces an injective D-morphism from f −1(a) onto N (as D-subvarieties of G) given by h 7→ g−1 · h. As N is defined over k and sN is the zero section, f −1(a) is indeed isotrivial. Note that this argument actually shows that f −1(a) is isotrivial for any D-point a of H (not necessarily Zariski generic). Now let V be an arbitrary k-irreducible D-subvariety of G. Letting W be the D-subvariety of H given by the Zariski closure of f (V ) (see Lemma 3.3(4)) and g := f V , we get that V g / W / 0 witnesses the compound isotriviality of V . Indeed, (W, tW ) is isotrivial because it is a D-subvariety of the isotrivial (H, t). Also, the argument in the above paragraph shows that if b is a Zariski k-generic D-point of W then f −1(b) is isotrivial, but then g−1(b) = f −1(b) ∩ V is isotrivial as well. The result follows. (cid:3) Remark 4.5. (1) We note that in Proposition 4.4 one cannot obtain in general compound isotriviality in 1-step (in other words, isotriviality). For example, in the single derivative case D = {δ}, consider G = Ga × Gm with D-group structure s : G → T G given by s(x, y) = (x, y, 0, xy). Then G is not isotrivial, see [18, §2] for details. (2) It follows from [11, Fact 2.7(iii)], that the center Z(G) of an affine algebraic D-group (G, s) is a (normal) D-subgroup. In the case of a single derivation D = {δ}, it was shown in [11, Theorem 2.10] that G/Z(G) with its induced D-group structure is isotrivial. While this result extends to the case of several commuting derivations, it is not yet known if it holds in the general / / / / ON THE DME FOR POISSON-HOPF ALGEBRAS 15 situation of possibly noncommuting derivations. It is worth noting that if such a result does hold, then one can extend the argument of Proposi- tion 4.4 to show that any connected algebraic D-group over the constants is compound isotrivial in 3-steps (this would yield an interesting extension of [2, Proposition 2.15]). Putting Proposition 4.4 together with the results of Section 3, we obtain Corollary 4.6. Suppose (G, s) is a connected algebraic D-group over a field of constants k. If G is commutative, then (k[G], D) satisfies the D-DME. Proof. By Proposition 4.4, every k-irreducible D-subvariety of G is compound isotrivial. Hence, by Corollary 3.9, a prime D-ideal of k[G] is D-rational only if it is D-locally closed. The result now follows from Proposition 3.4. (cid:3) 5. Main results on Poisson-Hopf algebras Recall that k denotes a field of characteristic zero. In this section we prove the main result of the paper; namely, Theorem 5.1. Any cocommutative affine Poisson-Hopf k-algebra satisfies the Pois- son Dixmier-Moeglin equivalence. Remark 5.2. By [17, 1.7(i), 1.10], in any affine Poisson algebra (A, {−, −}) we have that Poisson-locally closed implies Poisson-primitive, and Poisson-primitive implies Poisson-rational. We note that this also follows from our results in Section 3. Indeed, if we let D denote the (finite) family of Hamiltonians of a collection of generators of A, then, by Proposition 3.4, in the differential k-algebra (A, D) we have that D-locally closed implies D-primitive, and D-primitive implies D-rational. The result now follows from Corollary 2.5. We will make use of the following consequence of Proposition 3.4. Lemma 5.3. Let (A, {−, −}) be an affine Poisson k-algebra such that A ⊗ kalg is a domain. If in A ⊗ kalg a prime ideal is Poisson-rational only if it is Poisson-locally closed, then the same holds in A. Proof. The assumptions imply that A is of the form k[V ] for some geometrically irreducible affine algebraic variety V over k. Letting D be the (finite) family of Hamiltonians of a collection of generators of A, we get that in kalg[V ] a prime ideal is D-rational only if it is D-locally closed (by Corollary 2.5). By the 'furthermore' clause of Proposition 3.4, we get the same implication holds in k[V ]. Again by Corollary 2.5, we get that in A = k[V ] a prime ideal is Poisson-rational only if it is Poisson-locally closed, as desired. (cid:3) A commutative and cocommutative affine Hopf k-algebra is nothing more than the coordinate ring k[G] of a connected commutative affine algebraic group G over k. The following well known theorem characterizes such groups over kalg (see [4] for instance). Theorem 5.4. Let G be a connected commutative affine algebraic group over k. Then G is isomorphic over kalg to Gs m for some s and t. a × Gt 16 ST´EPHANE LAUNOIS AND OMAR LE ´ON S ´ANCHEZ A consequence of this result is that for any affine Hopf k-algebra A, that is commutative and cocommutative, we have that (5.1) A ⊗ kalg = kalg[x1, . . . , xs, y± 1 , . . . , y± t ] where the xi's are primitive and the yi's are group-like. We use this fact to prove: Proposition 5.5. Let (A, {−, −}) be an affine Poisson k-algebra. Suppose further that k is algebraically closed and A is a cocommutative Hopf algebra. Then, there is D ⊆ Derk(A) with (5.2) spanA D = spanA({a, −} : a ∈ A), and such that (A, {−, −}) is a Poisson-Hopf algebra if and only if (A, D) is a differential-Hopf algebra. Proof. Write A = k[G] where G is a connected commutative affine algebraic group over k. By Theorem 5.4 (or (5.1) rather), we may assume that A = k[x1, . . . , xs, y± 1 , . . . , y± t ] where the xi's are primitive and the yi's are group-like. Consider the Hamiltonians δxi := {xi, −} : A → A, for i = 1, . . . , s, and the normalized Hamiltonians δyi := y−1 i {yi, −} : A → A, for i = 1, . . . , t. We claim that D := {δx1, . . . , δxs.δy1, . . . , δyt } is the desired set of k-linear derivations. Clearly (5.2) is satisfied (see Remark 2.3). Now suppose that (A, {−, −}) is a Poisson-Hopf algebra. Let 1 ≤ i ≤ s and set x := xi. Recall that ∆x = x ⊗ 1 + 1 ⊗ x. Now let a ∈ A, using sumless Sweedler notation we write ∆a = a(1) ⊗ a(2). We then have δx(∆(a)) = δx(a(1) ⊗ a(2)) = δxa(1) ⊗ a(2) + a(1) ⊗ δxa(2) = {x, a(1)} ⊗ a(2) + a(1) ⊗ {x, a(2)} = {x ⊗ 1, a(1) ⊗ a(2)} + {1 ⊗ x, a(1) ⊗ a(2)} = {x ⊗ 1 + 1 ⊗ x, a(1) ⊗ a(2)} = {∆x, ∆a} = ∆({x, a}) = ∆(δxa) (since A is a Poisson-Hopf algebra) ON THE DME FOR POISSON-HOPF ALGEBRAS 17 Now let 1 ≤ j ≤ t and set y = yj. Recall that ∆y = y ⊗ y. Now let a ∈ A. We then have δy(∆(a)) = δy(a(1) ⊗ a(2)) = δya(1) ⊗ a(2) + a(1) ⊗ δya(2) = y−1{y, a(1)} ⊗ a(2) + a(1) ⊗ y−1{y, a(2)} =(cid:0)y−1 ⊗ y−1(cid:1)(cid:0){y, a(1)} ⊗ ya(2) + ya(1) ⊗ {y, a(2)}(cid:1) = ∆(y−1) {y ⊗ y, a(1) ⊗ a(2)} = ∆(y−1) {∆y, ∆a} = ∆(y−1) ∆({y, a}) = ∆(y−1{y, a}) = ∆(δya) (since A is a Poisson-Hopf algebra) We have shown that all these derivations commute with coproduct; in other words, that (A, D) is a differential-Hopf algebra. The other implication (i.e., that if D commutes with ∆ then {−, −} commutes (cid:3) with ∆) follows from a similar series of equalities and applying Remark 2.2. Remark 5.6. (1) In terms of Poisson groups over an algebraically closed field k, in the sense of [10, §1.3], the above proposition shows that given a commutative affine algebraic group G over k equipped with a Poisson variety structure, one can find D ⊆ Derk(k[G]) such that (5.2) holds, with k[G] in place of A, and with the property that G is a Poisson algebraic group if and only if it is an algebraic D-group (with respect to D). (2) We do not know at this point whether or not the cocommutativity as- sumption can be removed from Proposition 5.5. Nonetheless, we note that if this were the case and if algebraic D-groups over constants were com- pound isotrivial (see Remark 4.5(2)), then the proof below of Theorem 5.1 would work for any (not necessarily cocommutative) affine Poisson-Hopf k-algebra. We can now prove Theorem 5.1. Proof of Theorem 5.1. By Remark 5.2, it suffices to show that D-rational implies D- locally closed. By Lemma 5.3, we may assume that k is algebraically closed. Write A as k[G] where G = Gs m, and let D be the family of k-linear derivations of A obtained in Proposition 5.5. Then (A, D) is a differential-Hopf algebra. a × Gt By Remark 4.2, the induced section s : G → τDG is a group homomorphism; in other words, (G, s) is a D-group. By Corollary 4.6, (A, D) satisfies the D-DME; in particular, a prime D-ideal of A is D-rational only if it is D-locally closed. By Corollary 2.5, this in turn implies that a prime Poisson ideal of A if Poisson-rational only if it is Poisson-locally closed, as desired. (cid:3) We conclude with the following application: 18 ST´EPHANE LAUNOIS AND OMAR LE ´ON S ´ANCHEZ Theorem 5.7. If A is the symmetric algebra of a finite dimensional Lie algebra g over k, equipped with its natural Poisson bracket, then A satisfies the Poisson Dixmier-Moeglin equivalence. Proof. We know from Section 2.1 that A is a cocommutative affine Poisson-Hopf k-algebra, and so the result follows from Theorem 5.1. (cid:3) References [1] Jason Bell, St´ephane Launois, Omar Le´on S´anchez and Rahim Moosa. Poisson algebras via model theory and differential algebraic geometry. Journal of the European Mathematical Society, 19:2019 -- 2049, 2017. [2] Jason Bell, Omar Le´on S´anchez and Rahim Moosa. D-groups and the Dixmier-Moeglin equiv- alence. Preprint 2017. arXiv:1612.00069 [3] Jason Bell and Wing Hong Leung. The Dixmier-Moeglin equivalence for cocommutative Hopf algebras of finite Gelfand-Kirillov dimension. Algebra and Representation Theory 17(6): 1843 -- 1852, 2014. [4] Armand Borel. Linear algebraic groups. Graduate Texts in Mathematics. Vol. 126. Springer, 1991. [5] Kenneth Brown and Iain Gordon. Poisson orders, symplectic reflection algebras and repre- sentation theory. J. Reine Angew. Math., 559:193 -- 216, 2003. [6] Alexandru Buium. Differential algebraic groups of finite dimension. Lecture Notes in Math- ematics. Vol. 1506. Springer-Verlag, 1992. [7] Ehud Hrushovski. Computing the galois group of a linear differential equation. In Differential Galois theory, Vol. 58, pp. 97 -- 138. Banach Center Publications, Polish Academy of Sciences, 2002. [8] Ronald Irving and Lance Small. On the characterization of primitive ideals in enveloping algebras. Math. Z., 173(3):217 -- 221, 1980. [9] Irving Kaplansky. An introduction to differential algebra. Publications de l'institut de math´ematique de l'universit´e de Nancago. Hemann, 1957, [10] Leonid Korogodski and Yan Soibelman. Algebras of functions on quantum groups: Part I. Mathematical Surveys and Monographs, Vol. 56, American Mathematical Society, Provi- dence, RI, 1998. [11] Piotr Kowalski and Anand Pillay. Quantifier elimination for algebraic D-groups. Transactions of the American Mathematical Society, 358(1):167 -- 181, 2006. [12] Serge Lang. Algebra. Springer-Verlag. Third Edition, 2002. [13] Omar Le´on S´anchez and Anand Pillay. Some definable Galois theory and examples. To appear in the Bulletin of Symbolic Logic. [14] Jiafeng Lu, Xingting Wang and Guangbin Zhuang. Universal enveloping algebras of Poisson Hopf algebras. Journal of Algebra, 426:92 -- 136, 2015. [15] David Marker. Model Theory: An Introduction. Graduate Texts in Mathematics. Vol. 217. Springer, 2002. [16] Rahim Moosa and Tom Scanlon. Model theory of fields with free operators in characteristic zero. Journal of Mathematical Logic, 14(2), 2014. [17] Sei-Qwon Oh. Symplectic ideals of Poisson algebras and the Poisson structure associated to quantum matrices. Communications in Algebra, 27(5):2163 -- 2180, 1999. [18] Anand Pillay. Finite-dimensional differential algebraic groups and the Picard-Vessiot the- ory. In Differential Galois theory, Vol. 58, pp. 97 -- 138. Banach Center Publications, Polish Academy of Sciences, 2002. [19] Guangbin Zhuang. Properties of pointed and connected Hopf algebras of finite Gelfand- Kirillov dimension. Journal of the London Mathematical Society, 87(3):877 -- 898, 2013. St´ephane Launois, University of Kent, School of Mathematics, Statistics, and Ac- tuarial Science, Canterbury, Kent CT2 7NZ, UK E-mail address: [email protected] Omar Le´on S´anchez, University of Manchester, School of Mathematics, Oxford Road, Manchester, M13 9PL. E-mail address: [email protected]
1712.07751	1	1712	2017-12-21T00:20:15	q-generalized (anti -) flexible algebras and bialgebras	[ "math.RA" ]	In this work, we provide a q-generalization of flexible algebras and related bialgebraic structures, including center-symmetric (also called antiflexible) algebras, and their bialgebras. Their basic properties are derived and discussed. Their connection with known algebraic structures, previously developed in the literature, is established. A q-generalization of Myung theorem is given. Main properties related to bimodules, matched pairs and dual bimodules as well as their algebraic consequences are investigated and analyzed. Finally, the equivalence between q-generalized flexible algebras, their Manin triple and bialgebras is established.	math.RA	math	q-generalized (anti -) flexible algebras and bialgebras Mahouton Norbert Hounkonnou∗ and Mafoya Landry Dassoundo† Abstract. In this work, we provide a q-generalization of flexible algebras and related bialge- braic structures, including center-symmetric (also called antiflexible) algebras, and their bialge- bras. Their basic properties are derived and discussed. Their connection with known algebraic structures, previously develped in the literature, is established. A q-generalization of Myung theorem is given. Main properties related to bimodules, matched pairs and dual bimodules as well as their algebraic consequences are investigated and analyzed. Finally, the equivalence between q-generalized flexible algebras, their Manin triple and bialgebras is established. Keywords. Lie algebra, Lie-admissible algebra, flexible algebra, antiflexible algebra, center- symmetric algebra, matched pair, Manin triple, bialgebra. Mathematics Subject Classification (2010). Primary 16Yxx,17Axx, 16-XX, 16T10, 17A30, 17D25, 16D20, 17D99; Secondary 16S80, 16T25. October 21, 2018 Contents Introduction 1. 2. Basic properties of a q-generalized flexible algebra 3. Bimodules and matched pairs of q-generalized flexible algebras 4. Basic properties of the q-generalized flexible algebras 5. Manin triple of q-generalized flexible algebras and bialgebras 6. Application to octonion algebra 7. Concluding remarks References 1 2 5 8 13 15 17 17 1. Introduction Alternative algebras were introduced by Zorn [29] who established their fundamental iden- tities, studied their nucleus by modifying the characteristic of the field, and investigated their Lie admissibility using the corresponding Jacoby identity. Furthermore, Zorn derived their power associativity conditions. Later, Schafer [26] gave a new formulation of these algebras in terms of left and right multiplication operators, and in terms of division algebras of degree two. He also provided the isotopes of these algebras. Santilli [27] introduced Lie admissible algebras and gave their basic properties. He extended his study to mutation algebras, examined their relation to associative algebras, Lie algebras, Jordan and special Jordan algebras, and established the pas- sage from one type of algebra to another by using a hexahedron with oriented edges. Radicals of flexible Lie admissible algebras were introduced, and some of their properties were established and discussed in [6]. Classes of flexible Lie admissible algebras were also investigated and discussed in [18]. Albert in [1] elaborated fundamental concepts, and studied the isotropy of nonassociative 1 2 MAHOUTON NORBERT HOUNKONNOU∗ AND MAFOYA LANDRY DASSOUNDO† algebras. Simple and semi-simple algebras, and their characterization from nonassociative alge- braic structures were developed and discussed in [2]. For more details, see a self-contained book by Shafer [25] addressing a nice compilation of basic properties of nonassociative algebras. Contrarily to Lie algebras, and except for some classification based on the characteristics of closed fields (see [7] and references therein), a full classification of nonassociative algebras still remains a tremendous task. Some interesting properties and algebraic identities of antiflexible structures were investigated and discussed in [8]. The properties of simple, semisimple and nearly semisimple antiflexible algebras were also derived and analyzed in [22 -- 24]. Among the nonassociative algebras, the alternative algebras, with an associator preserving certain symmetry by exchanging its elements, play a central role in both mathematics and physics as they possess interesting properties. A nice repertory of their applications in physics, including gauge theory and Yang-Mills gauge theory formulation from nonassociative algebras can be found in [20], (and references therein). A theory of nuclear boson-expansion for odd-fermion system in the context of nonassociative algebras was also examined in [21]. A study on quark structure and octonion algebras was performed in [11]. Further, a generalization of the classical Hamiltonian dynamics to a three-dimensional phase space, generating equations of motion with two Hamilto- nians and three canonical variables, was performed with an analog of Poisson bracket realized by means of the associator of nonassociative algebras [19]. Besides, flexible algebras were also investigated in terms of degree of algebras [14]. Other characterizations and applications of nonassociative algebras can be found in [17], [15] and [12] (and references therein). Similarly to algebraic properties of quantum groups developed by Drinfeld [10], some nonas- sociative algebras possess interesting identities with applications in physics, and generate the so-called associative or classical Yang-Baxter equations [3, 9, 13], (and references therein). Fur- thermore, the bialgebras constructed from Jordan algebras [28] are related to the Lie bialgebras. The left-symmetric algebras, also called pre-Lie algebras [5], are known as Lie admissible alge- bras, and admit the left multiplication operator as a representation. They can also be used to produce symplectic Lie algebras, while their coboundary bialgebras lead to the identity known as S-equation, and generate para-Kahler Lie algebras. The case of associative algebras also furnished remarkable properties investigated by Aguiar [3] and Bai [4]. The center-symmetric algebras stud- ied in [12] are also Lie admissible algebras. The present work addresses a q-generalization of flexible algebras and related bialgebraic struc- tures, including center-symmetric (also called antiflexible) algebras, and their bialgebras. Their basic properties are derived and discussed. Their connection with known algebraic structures existing in the literature is established. A q-generalization of Myung theorem is given. Main properties related to bimodules, matched pairs and dual bimodules, and their algebraic conse- quences are investigated and analyzed. Finally, the Manin triple of q-generalized flexible algebras, and its link to q-generalized flexible bialgebras are buit together with the equivalence with the matched pair of q-generalized flexible algebras. 2. Basic properties of a q-generalized flexible algebra In this section, a q-generalization of algebras encompassing flexible, anti-flexible and asso- ciative algebras is provided. New classes of algebras are induced. Their relevant properties and link with known algebras are derived. Jordan identity and Lie admissibility condition are also established. Definition 2.1. Let A(q), where q ∈ K, be a finite dimensional vector space. The couple (A(q), ·) is called a q-generalized flexible algebra if, for all x, y, z ∈ A(q), the following relation is satisfied: (x, y, z) = q(z, y, x), or, equivalently, µ ◦ (µ ⊗ id) + q(µ ◦ τ ) ◦ ((µ ◦ τ ) ⊗ id) = µ ◦ (id ⊗µ) + q(µ ◦ τ ) ◦ (id ⊗(µ ◦ τ )) q-GENERALIZED (ANTI -) FLEXIBLE ALGEBRAS AND BIALGEBRAS 3 where (x, y, z) := (x · y) · z − x · (y · z) is the associator of the bilinear product "·" on A(q); µ is defined by µ(x, y) = x · y; id is the identity map on A(q); and τ stands for the exchange map on A(q) given by τ (x ⊗ y) = y ⊗ x. This can be described by the following commutative diagram: A(q) ⊗ A(q) ⊗ A(q) µ⊗id − id ⊗µ A(q) ⊗ A(q) (µ◦τ )⊗id − id ⊗(µ◦τ ) A(q) qµ◦τ µ / A(q) Remark 2.2. We have: • For q = 0, the algebra A(q) is reduced to an associative algebra; • For q = −1, A(q) becomes a flexible algebra; • For q = 1, A(q) turns to be a center-symmetric algebra [12], (also called nonflexible algebra [15]). In the sequel, (A(q), ·) denotes a q-generalized flexible algebra over K. Besides, for notation simplification, we write xy instead of x · y, for x, y ∈ A(q), i.e. the product "·" is omitted when there is no confusion. Definition 2.3. Suppose L and R be left and right multiplication operators defined on A(q) as: A(q) −→ gl(A(q)) L : x 7−→ Lx : A(q) −→ A(q) y 7−→ Lx(y) := x · y A(q) → gl(A(q)) R : x 7→ Rx : A(q) → A(q) y 7→ Rx(y) := y · x. Then, their associated dual maps are defined as follows: A(q) → gl(A(q)∗) A(q)∗ → A(q)∗ L∗ : x 7→ L∗ x : a A(q) → K 7→ L∗ x(a) : x 7→ < L∗ x(a), y >= ± < a, Lx(y) > (2.1) (2.2) (2.3) A(q) → gl(A(q)∗) A(q)∗ → A(q)∗ R∗ : x 7→ R∗ x : a A(q) → K (2.4) 7→ R∗ x(a) : x 7→ < R∗ x(a), y >= ± < a, Rx(y) > Proposition 2.4. Let L and R be the above defined left and right multiplication operators. The following relations are satisfied for all x, y ∈ A(q) : Lxy − LxLy = q(RxRy − Ryx), [Rx, Ly] = q[Ry, Lx], RxRy − Ryx = q(Lxy − LxLy). (2.5) (2.6) (2.7) Proof: Let A(q) be a q-generalized flexible algebra over the field K. For all x, y, z ∈ A(q), the proof follows from the equivalences: (x, y, z) = q(z, y, x) ⇐⇒ (xy)z − x(yz) = q(zy)x − qz(yx) / / / 4 . MAHOUTON NORBERT HOUNKONNOU∗ AND MAFOYA LANDRY DASSOUNDO† (x, y, z) = q(z, y, x) ⇐⇒ (Lxy − LxLy)(z) = q(RxRx − Ryx)(z) (x, y, z) = q(z, y, x) ⇐⇒ (RzLx − LxRz)(y) = q(RxLz − LzRx)(y) (x, y, z) = q(z, y, x) ⇐⇒ [Rz, Lx](y) = q[Rx, Lz](y) (x, y, z) = q(z, y, x) ⇐⇒ (RzRy − Ryz)(x) = q(Lzy − LzLy)(x) (cid:3) Proposition 2.5. Provided the sub-adjacent algebra G(A(q)) := (A(q), [, ]), where the bilinear product [, ] is the commutator associated to the product on A(q), we have, for all x, y, z ∈ A(q): J(x, y, z) := [x, [y, z]] + [y, [z, x]] + [z, [x, y]] = (q − 1){(x, y, z) + (y, z, x) + (z, x, y)}. Proof: It stems from a straightforward computation. Lemma 2.6. (1) From the Proposition2.5, for all x, y, z ∈ A(q), the relation S(x, y, z) := (x, y, z) + (y, z, x) + (z, x, y) = 0 (2.8) (cid:3) (2.9) is a sufficient condition for A(q) to become a Lie admissible algebra, i.e. for (A(q), [, ]) to be a Lie algebra. In the particular case where q = 1, we get a center symmetric algebra which is Lie admissible as developed in [12]. (2) The q-generalized algebra A(q) is Lie admissible if and only if, for all x, y, z ∈ A(q), we have: (q − 1)S(x, y, z) = 0. In particular, any q-generalized flexible algebra defined on a field Kq−1 of characteristic q − 1 is Lie admissible. Proof: Let (A(q), ·) be a q-generalized flexible algebra. (1) From the relation(2.8), we have J(x, y, z) = (q − 1)S(x, y, z). Indeed, A(q) is Lie admis- sible if and only if J(x, y, z) = 0 yielding S(x, y, z) = 0. (2) The Lie admissibility condition J(x, y, z) = 0 implies that the relation (q − 1)S(x, y, z) = 0 giving S(x, y, z) = 0, or q − 1 = 0, or, in other words, the field K on which A(q) defines a vector space is of characteristic q − 1. (cid:3) Proposition 2.7. The following relation is satisfied for all x, y, z ∈ A(q) : (Lxy − LxLy + RxLy − LyRx + RyRx − Rxy) = q(RxRy − Ryx + RyLx −LxRy + Lyx − LyLx), (2.10) where L and R are representations of left and right multiplication operators, respectively. Proof: Let us write the associator with the operators L and R. For all x, y, z ∈ A(q), (x, y, z) = (xy)z − x(yz) = (Lxy − LxLy)(z) = (RzLx − LxRz)(y) = (RzRy − Ryz)(x). It follows that: (x, y, z) + (y, z, x) + (z, x, y) = (Lxy − LxLy + RxLy − LyRx + RyRx −Rxy)(z) = q(z, y, x) + q(x, z, y) + q(y, x, z) = q(RxRy − Ryx + RyLx − LxRy + Lyx −LyLx)(z). Therefore, for all x, y, z ∈ A(q): Lxy − LxLy + RxLy − LyRx + RyRx − Rxy = q(RxRy − Ryx + RyLx − LxRy + Lyx − LyLx). (cid:3) Remark 2.8. The result (2.10) can also be derived from the Proposition 2.4 by summing the relations (2.5), (2.6) and (2.7). q-GENERALIZED (ANTI -) FLEXIBLE ALGEBRAS AND BIALGEBRAS 5 Theorem 2.9. The following relation is satisfied: ∀x, y, z ∈ A(q), [xy − qyx, z] + [yz − qzy, x] + [zx − qxz, y] = 0, (2.11) where the bilinear product [, ] is the commutator associated to the product "·" defined on A(q). Proof: By a direct computation. (cid:3) Remark 2.10. By setting the parameter q = 1, we get the Jacoby identity from the relation (2.11) indicating that the underlying algebra is Lie admissible as shown in [12]. We are therefore in right to set the following: Definition 2.11. Setting G(A(q)) := (A(q), [, ]), where A(q) is the underlying vector space associated to a q-generalized flexible algebra (A(q), ·), the equation (2.11) defines a q-generalized Jacobi identity. Theorem 2.12. For a q-generalized flexible algebra A(q), the following propositions are equiv- alent: For all x, y, z ∈ A(q), (1) (2) where {x, y}q := xy + qyx; [z, xy] = {z, x}qy − x{y, z}q, [z, x ∗q y] = [z, x] ∗q y + x ∗q [z, y], where x ∗q y = 1 2 (xy − qyx); (3) A(q) is a Lie-admissible q-generalized flexible algebra, i.e. Remark 2.13. From Theorem 2.12, we observe that: [[x, y], z] + [[y, z], x] + [[z, x], y] = 0. (2.12) (2.13) (2.14) (1) For q = −1, the equation (2.12) turns out to be the derivation for the commutator of a flexible algebra as postulated by the well known Myung Theorem [18], [20] (and references therein). (2) For q = 0, the equation (2.12) becomes an evidence by using the associativity. (3) For q = 1, the equation (2.12) leads to the relation (x, y, z) + (y, z, x) + (z, x, y) = 0, which characterizes an anti-flexible algebra structure, see [12] (and references therein) (4) For q = 1, the equation (2.13) is equivalent to the Jacoby identity in a field of a charac- teristic 0, what is the case for a center-symmetric (also called anti-flexible) algebra; (5) For q = 0, the equation (2.13) describes the derivation property of the commutator (or Lie bracket) of a Lie algebra induced by an associative algebra; (6) For q = −1, the flexibility condition (2.13) defines the derivation property of the Jordan product given as x · y = 1 2 (xy + yx), see [20]. 3. Bimodules and matched pairs of q-generalized flexible algebras Definition 3.1. The triple (l, r, V ), where V is a finite dimensional vector space and l, r : A(q) → gl(V ) are two linear maps satisfying the following relations for all x, y ∈ A(q) : lxy − lxly = q(rxry − ryx), [rx, ly] = q[ry, lx], rxry − ryx = q(lxy − lxly), (3.1) (3.2) (3.3) is called a bimodule of A(q), also simply denoted by (l, r). Proposition 3.2. Let l, r : A(q) → gl(V ) be two linear maps. The couple (l, r) is a bimodule of the q-generalized flexible algebra A(q) if and only if there exists a q-generalized flexible algebra structure "∗" on the semi-direct vector space A(q) ⊕ V given by (x + u) ∗ (y + v) := x · y + lxv + ryu, ∀x, y ∈ A(q), and ∀u, v ∈ V. 6 MAHOUTON NORBERT HOUNKONNOU∗ AND MAFOYA LANDRY DASSOUNDO† We denote such a q-generalized flexible algebra structure "∗" on the semi-direct vector space A(q) ⊕ V by (A(q) ⊕ V, ∗) or simply A(q) ⋉ V. Proof: Let x, y, z ∈ A(q), where A(q) is a generalized flexible algebra, and u, v, w ∈ V , where V is a finite dimensional vector space. Using the bilinear product defined, for all x, y ∈ A(q), and all u, v ∈ V by: (x + u) ∗ (y + v) := x · y + lxv + ryu, where l, r : A → gl(V ) are linear maps, the associator of the bilinear product ∗ can be written as: (x + u, y + v, z + w) = ((x + u) ∗ (y + v)) ∗ (z + w) −(x + u) ∗ ((y + v) ∗ (z + w)) = (x · y + lxv + ryu) ∗ (z + w) −(x + u) ∗ (y · z + lyw + rzv) = (x · y) · z + lx·yw + rz(lxv) + rz(ryu) − x · (y · z) −lx(lyw) + lx(rzv) − ry·zu (x + u, y + v, z + w) = (x, y, z) + (lx·y − lxly)w + (zzlx − lxrz)v +(rzry − ry·z)u. Then, we have, ∀x, y, z ∈ A(q) and ∀u, v, w ∈ V : (x + u, y + v, z + w) = (x, y, z) + (lx·y − lxly)w + (zzlx − lxrz)v +(rzry − ry·z)u. Besides, q(z + w, y + v, x + u) = q(z, y, x) + q(lz·y − lzly)u + q(zxlz − lzrx)v +q(rxry − ry·x)w. (3.4) (3.5) By setting (x + u, y + v, z + w) = q(z + w, y + v, x + u), ∀x, y, z ∈ A(q) and ∀u, v, w ∈ V , i.e. (A(q) ⊕ V, ∗) is a q-generalized flexible algebra, we get from the right hand side of the equations (3.4) and (3.5): (x, y, z) = q(z, y, x), (lx·y − lxly) = q(rxry − ry·x), (zzlx − lxrz) = q(zxlz − lzrx), (rzry − ry·z) = q(lz·y − lzly), which are equivalent to the relations (3.1), (3.2), (3.3) defining the bimodule of a q-generalized flexible algebra. (cid:3) Example 3.3. According to the Proposition2.4, (L, R), where R and L are the representations of the right and left multiplication operators, respectively, is a bimodule of a q-generalized flexible algebra A(q). Indeed, L and R satisfy the equations (3.1), (3.2) and (3.3). Theorem 3.4. Let (A(q), ·) and (B(q), ∗) be two q-generalized flexible algebras. Suppose there exist linear maps lA, rA : A(q) → gl(B(q)) and lB, rB : B(q) → gl(A(q)) satisfying the following relations: (lB(a)x) · y + lB(rA(x)a)y − lB(a)(x · y) = q(rB(a)(y · x) − y · (rB(a)x) −rB(lA(x)a)y), rB(a)(x · y) − x · (rB(a)y) − rB(lA(y)a)x = q((lB(a)y) · x + lB(rA(y)a)x −lB(a)(y · x)), (rB(a)x) · y + lB(lA(x)a)y − x · (lB(a)y) − rB(rA(y)a)x = q(((rB(a)y) · x +lB(lA(y)a)x − y · (lB(a)x) − rB(rA(x)a)y), (lA(x)a) ∗ b + lA(rB(a)x)b − lA(x)(a ∗ b) = q(rA(x)(b ∗ a) − b ∗ (rA(x)a) −rA(lB(a)x)b)), (3.6) (3.7) (3.8) (3.9) q-GENERALIZED (ANTI -) FLEXIBLE ALGEBRAS AND BIALGEBRAS 7 rA(x)(a ∗ b) − a ∗ (rA(x)b) − rA(lB(b)x)a = q((lA(x)b) ∗ a + lA(rB(b)x)a −lA(x)(b ∗ a)), (3.10) (rA(x)a) ∗ b + lA(lB(a)x)b − a ∗ (lA(x)b) − rA(rB(b)x)a = q((rA(x)b) ∗ a +lA(lB(b)x)a − b ∗ (lA(x)a) − rA(rB(a)x)b)), (3.11) ∀x, y ∈ A(q) and ∀a, b ∈ B(q). It follows that there is a q-generalized flexible algebra structure "⋆" on the direct sum of vector spaces A(q) ⊕ B(q) given as: (x + a) ⋆ (y + b) = (x · y + lB(a)y + rB(b)x) + (a ∗ b + lA(x)b + rA(y)a). Proof: Let (A(q), ·) and (B(q), ∗) be two q-generalized flexible algebras, lA, rA : A(q) → gl(B(q)) and lB, rB : B(q) → gl(A(q)) be four linear maps satisfying the relations (3.6), (3.7), (3.8), (3.9), (3.10) and (3.11). Consider the bilinear product "⋆" defined on the vector space A(q) ⊕ B(q) as: (x + a) ⋆ (y + b) = (x · y + lB(a)y + rB(b)x) + (a ∗ b + lA(x)b + rA(y)a), ∀x, y ∈ A(q); a, b ∈ B(q). We have: (x + a, y + b, z + c) = {(x + a) ⋆ (y + b)} ⋆ (z + c) −(x + a) ⋆ {(y + b) ⋆ (z + b)} = (x, y, z) + (a, b, c) + {rB(c)(x · y) + lA(x · y)c −x · (rB(c)y) − lA(x)(lA(y)c) − rB(lA(y)c)x} +{rB(c)(rB(b)x) + lA(rB(b)x)c − rB(b ∗ c)x +(lA(x)b) ∗ c − lA(x)(b ◦ c)} +{(rB(b)x) · zlB(lA(x)b)z + rA(z)(lA(x)b) −x · (lB(b)z) − rB(rA(z)b)x − lA(x)(rA(z)b)} +{(lB(a)y) · z + lB(rA(y)a)z + rA(z)(rA(y)a) −lB(a)(y · z) − rA(y · z)a} + {rB(c)(lB(a)y) +(rA(y)a) ∗ c + lA(lB(a)y)c − +lB(a)(rB(c)y) −a ∗ (lA(y)c) − rA(rB(c)y)a} + {lB(a ∗ b)z +rA(z)(a ∗ b) − +lB(a)(lB(b)z) − a ∗ (rA(z)b) −rA(lB(b)z)a} = (x, y, z) + (x, y, c) + (x, b, z) + (x, b, c) + (a, y, z) +(a, b, z) + (a, b, c) Then, the q-generalized flexibility condition of the bilinear product ⋆ is given as: (x + a, y + b, z + c) = q(z + c, y + b, x + a) ⇐⇒ (x, y, z) = q(z, y, x) (a, b, c) = q(c, b, a) (x, y, c) = q(c, y, x) (x, b, z) = q(z, b, x) (x, b, c) = q(c, b, x) (a, y, z) = q(z, y, a) (a, y, c) = q(c, y, a) (a, b, z) = q(z, b, a)   8 MAHOUTON NORBERT HOUNKONNOU∗ AND MAFOYA LANDRY DASSOUNDO† ⇐⇒ (x + a, y + b, z + c) = q(z + c, y + b, x + a) ⇐⇒ (lA, rA, B(q)), (lB, rB, A(q)) are bimodules of A(q), B(q), respectively, (x, y, a) = q(a, y, x) (x, a, y) = q(y, a, x) (x, a, b) = q(b, a, x) (a, x, y) = q(y, x, a) (a, x, b) = q(b, x, a) (a, b, x) = q(x, b, a) (lA, rA, B(q)), (lB, rB, A(q)) are bimodules of A(q), B(q), respectively, and the following equations: (3.6), (3.7), (3.8), (3.9), (3.10), (3.11) are satisfied.     Therefore, we obtain a q-generalized flexible algebra structure given, for all x, y ∈ A(q) and all a, b ∈ B(q), by: (x + a) ⋆ (y + b) = (x · y + lB(a)y + rB(b)x) + (a ∗ b + lA(x)b + rA(y)a) on the direct sum of the underlying vector spaces A(q) and B(q) of the bimodules (lA, rA, B(q)) and (lB, rB, A(q)) of the associated q-generalized flexible algebras A(q) and B(q). (cid:3) In this case, the obtained q-generalized flexible algebra (A(q)⊕B(q), ⋆) is denoted by A(q) ⊲⊳lA,rA lB,rB B(q), or simply A(q) ⊲⊳ B(q). Definition 3.5. The sixtuple (A(q), B(q), lA, rA, lB, rB) satisfying the conditions of Theo- rem 3.4 is called matched pair of the generalized flexible algebras A(q) and B(q). Remark 3.6. Theorem 3.4 is a q-generalization of main theorems, well known in the literature. Indeed, • For q = 0, Theorem 3.4 is exactly the fundamental theorem for the matched pair of associative algebras. See [4] and references therein. • For q = 1, Theorem 3.4 is reduced to the fundamental theorem for the matched pair of center-symmetric algebras formulated in [12]. • For q = −1, Theorem 3.4 becomes the fundamental theorem for the matched pair of flexible algebras. 4. Basic properties of the q-generalized flexible algebras In this section, we construct and discuss the basic definitions and main properties of the q-generalized flexible algebras. Definition 4.1. Let l, r : A(q) → gl(V ) be the two above mentioned linear maps. Their dual maps are defined as: A(q) → gl(V ∗) V ∗ → V ∗ l∗ : x 7→ l∗ x : V → K (4.1) v∗ 7→ l∗ x(v∗) : u 7→ < l∗ x(v∗), u >= < v∗, lx(u) >, q-GENERALIZED (ANTI -) FLEXIBLE ALGEBRAS AND BIALGEBRAS 9 A(q) → gl(V ∗) V ∗ → V ∗ r∗ : x 7→ r∗ x : V → K (4.2) v∗ 7→ r∗ x(v∗) : u 7→ < r∗ x(v∗), u >= < v∗, rx(u) >, where V ∗ = Hom(V, K) and gl(V ∗) is the linear group of V ∗. Theorem 4.2. For any finite dimensional vector space V , suppose l, r : A(q) → gl(V ) be two linear maps such that l∗ and r∗ are dual maps of l and r, respectively. The following propositions are equivalent: (1) (l, r, V ) is a bimodule of the q-generalized flexible algebra A(q), (2) (r∗, l∗, V ∗) is a bimodule of the q-generalized flexible algebra A(q). Remark 4.3. It is worth noticing that the dual bimodule does not depend on the parameter q. Clearly, it is the dual bimodule obtained for the center-symmetric algebra in [12]. Therefore, the dual bimodule of center-symmetric algebras is the same as the dual bimodule of flexible algebras. Proposition 4.4. The quadruple (R∗ · , L∗ · , A(q)∗), where A(q)∗ is the dual space of A(q) and given by A(q)∗ = Hom(A(q), K), is a bimodule of A(q). Proof: By considering Definition 2.3, Proposition 2.4 and Proposition 4.2, we deduce that (cid:3) · , A(q)∗) is a bimodule of A(q). · , L∗ (R∗ Theorem 4.5. Let (A(q), ·) be a q-generalized flexible algebra. Suppose that there is a q- generalized flexible algebra structure "◦" on its dual space A(q)∗ = Hom(A(q), K). The sixtuple ◦) is a matched pair of the q-generalized flexible algebras A(q) and A(q)∗ (A(q), A(q)∗, R∗ if and only if the linear maps R∗ ◦ satisfy the following relations for all x, y ∈ A(q) and all a, b ∈ A(q)∗: · , R∗ · , R∗ · , L∗ ◦, L∗ · , L∗ ◦, L∗ (R∗ −y · (L∗ ◦(a)x) · y + R∗ ◦(a)x) − L∗ ◦(L∗ ◦(R∗ · (x)a)y), · (x)a)y − R∗ ◦(a)(x · y) = q(L∗ ◦(a)(y · x) L∗ +R∗ ◦(a)(x · y) − x · (L∗ · (y)a)x − R∗ ◦(L∗ ◦(a)y) − L∗ ◦(a)(y · x)), ◦(R∗ · (y)a)x = q((R∗ ◦(a)y) · x (L∗ q((L∗ ◦(a)x) · y + R∗ ◦(R∗ · (x)a)y − x · (R∗ ◦(a)y) − L∗ ◦(L∗ ◦(a)y) · x + R∗ ◦(R∗ · (y)a)x − y · (R∗ ◦(a)x) − L∗ · (y)a)x = · (x)a)y). ◦(L∗ (4.3) (4.4) (4.5) Proof: Consider a q-generalized flexible algebra (A(q), ·) and assume that there is a q-generalized flexible algebra structure "◦" on its dual space (A(q))∗. Using Definition 2.3, Proposition 2.4 and Proposition 4.2, we deduct that (R∗ ◦, A(q)) is a bimodule of A(q)∗. It remains to show that the quadruple ((R∗ ◦) satisfies the relations (3.6), (3.7), (3.8), (3.9), (3.10) and (3.11). By taking the correspondences lA → R∗ · , rA → L∗ · , lB → R∗ · , we straightforwardly obtain the relations (3.6), (3.7) and (3.8). To establish now the relations (3.9), (3.10) and (3.11), let us compute, ∀x, y ∈ A(q) and ∀a, b ∈ B(q) : · , A(q)∗) is a bimodule of A(q) and (R∗ · and rA → L∗ · , R∗ · , L∗ ◦, L∗ · , L∗ ◦, L∗ < (R∗ −y · (L∗ −R∗ =< y, L∗ · (L∗ −q(R∗ ◦(a)x) · y + R∗ ◦(a)x) − L∗ ◦(L∗ ◦(R∗ ◦(a)(L·(x)y) − q(L∗ · (x)a)y − R∗ · (x)a)y), b >=< L·(R∗ ◦(a)(R·(x)y) − R·(L∗ ◦(a)(x · y) − q(L∗ ◦(a)x)y + R∗ ◦(a)x)y − L∗ · (x)(b ◦ a) − q((R∗ ◦(a)(y · x) ◦(L∗ ◦(R∗ · (x)(a ◦ b)) · (x)a)y · (x)a)y), b > ◦(a)x)b + (L∗ · (R∗ ◦(a)x)b) + q((R∗ · (x)a) ◦ b − L∗ · (x)a) ◦ b) > . Then we get for all x, y ∈ A(q) and all a, b ∈ B(q) : · (x)a)y − R∗ · (x)a)y), b >=< y, L∗ · (x)(a ◦ b) − R∗ ◦(L∗ ◦(R∗ · (x)(b ◦ a) − q((R∗ ◦(a)x) · y + R∗ ◦(a)x) − L∗ < (R∗ −y · (L∗ −L∗ · (L∗ ◦(a)(x · y) − q(L∗ ◦(a)(y · x) ◦(a)x)b + (L∗ · (R∗ · (x)a) ◦ b ◦(a)x)b − (R∗ · (x)a) ◦ b) > . 10 MAHOUTON NORBERT HOUNKONNOU∗ AND MAFOYA LANDRY DASSOUNDO† The following holds: (R∗ ◦(a)x) · y + R∗ ◦(L∗ · (x)a)y − R∗ ◦(a)(x · y) = q(L∗ −L∗ ◦(a)(y · x) − y · (L∗ ◦(R∗ · (x)a)y) ◦(a)x) or, equivalently, · (R∗ L∗ ◦(a)x)b + b ◦ (L∗ · (x)a) − L∗ · (x)(b ◦ a) = q((R∗ · (x)(a ◦ b) − R∗ −(R∗ · (L∗ ◦(a)x)b · (x)a) ◦ b) which is exactly the equation (3.10) by taking the correspondences: lA → R∗ rA → L∗ · , a → b, and we have · , rA → L∗ · , lB → R∗ · , L∗ +R∗ · (x)(a ◦ b) − a ◦ (L∗ · (L∗ ◦(b)x)a − R∗ · (x)b) − L∗ · (x)(b ◦ a)). · (R∗ ◦(b)x)a = q((R∗ · (x)b) ◦ a (4.6) This shows that the relation (4.3) is equivalent to (4.6). By the same way, we have: ◦(a)(x · y) − x · (L∗ < L∗ +R∗ ◦(L∗ −L∗ ◦(R∗ =< x, R∗ +b ◦ (L∗ · (y)a)x − R∗ · (y)a)x − q(L·(R∗ · (y)(a ◦ b) − R∗ · (y)a) − L∗ ◦(a)y) − L∗ ◦(R∗ ◦(a)(y · x)), b >=< L∗ ◦(L∗ ◦(a)y)b − (R∗ ◦(a)y)x + R∗ · (L∗ · (y)a)x − q((R∗ ◦(a)y) · x ◦(a)(R·(y)x) − R·(L∗ · (y)a)x − R∗ · (y)a) ◦ b − q[L∗ · (R∗ ◦(a)y)b ◦(a)(L·(y)x)), b > ◦(a)y)x · (y)(b ◦ a)] > . Then, we have that is equivalent to L∗ +R∗ ◦(a)(x · y) − x · (L∗ · (y)a)x − R∗ ◦(L∗ ◦(a)y) − L∗ ◦(a)(y · x)) ◦(R∗ · (y)a)x = q((R∗ ◦(a)y) · x R∗ +b ◦ (L∗ · (y)(a ◦ b) − R∗ · (y)a) − L∗ ◦(a)y)b − (R∗ · (L∗ · (y)(b ◦ a)). · (y)a) ◦ b = q(L∗ · (R∗ ◦(a)y)b This exactly gives the equation (3.9) by setting the correspondence lA → R∗ rA → L∗ · , y → x, and then we get: · , rA → L∗ · , lB → R∗ · , (R∗ −b ◦ (L∗ · (x)a) ◦ b + R∗ · (x)a) − L∗ · (L∗ · (R∗ ◦(a)x)b). ◦(a)x)b − R∗ · (x)(a ◦ b) = q(L∗ · (x)(b ◦ a) This shows that the relation (4.4) is equivalent to (4.7): · (x)a)y − x · (R∗ · (y)a)x − y · (R∗ ◦(a)x) · y + R∗ ◦(a)y) · x + R∗ · (L◦(a)x)b + b ◦ (R∗ ◦(R∗ ◦(R∗ < (L∗ −q((L∗ < y, L∗ −q(a ◦ (R∗ · (x)b) + L∗ · (L∗ · (x)a) − (L∗ ◦(b)x)a) − R∗ · (y)a)x ◦(a)y) − L∗ ◦(a)x) − L∗ · (x)b) ◦ a − R∗ · (R∗ ◦(L∗ ◦(L∗ · (R∗ ◦(a)x)b − (L∗ · (x)a)y), b >= ◦(b)x)a · (x)a) ◦ b) > . Therefore, the following relation (L∗ ◦(a)x) · y + R∗ ◦(R∗ · (x)a)y − x · (R∗ ◦(R∗ +R∗ ◦(a)y) − L∗ · (y)a)x − y · (R∗ · (y)a)x = q((L∗ ◦(L∗ ◦(a)x) − L∗ ◦(L∗ ◦(a)y) · x · (x)a)y) is equivalent to L∗ · (L◦(a)x)b + b ◦ (R∗ · (x)a) − (L∗ +L∗ · (x)b) ◦ a − R∗ ◦(b)x)a) − R∗ · (L∗ · (R∗ · (R∗ ◦(b)x)a − q(a ◦ (R∗ ◦(a)x)b − (L∗ · (x)b) · (x)a) ◦ b) which is exactly the equation (3.11) by setting the correspondences lA → R∗ · , a → b, and we have · , rA → L∗ · , rA → L∗ · , lB → R∗ (L∗ q((L∗ · (x)a) ◦ b + R∗ · (R∗ ◦(a)x)b − a ◦ (R∗ · (x)b) ◦ a + R∗ · (R∗ ◦(b)x)a) − b ◦ (R∗ · (x)b) − L∗ · (L∗ · (x)a) − L∗ ◦(b)x)a = · (L∗ ◦(a)x)b)). This shows that the relation (4.5) is equivalent to (4.8). (4.7) (4.8) q-GENERALIZED (ANTI -) FLEXIBLE ALGEBRAS AND BIALGEBRAS 11 Therefore, the sixtuple (A(q), (A(q)∗, R∗ · , L∗ · , R∗ ◦, L∗ flexible algebras A(q) and A(q)∗ if and only if the linear maps R∗ (4.3), (4.4) and (4.5). ◦) is a matched pair of the q-generalized ◦ satisfy the equations (cid:3) · , R∗ · , L∗ ◦, L∗ Remark 4.6. Theorem 4.5 encompasses particular results known in the literature, namely: • For q = 0, Theorem4.5 is exactly reduced to the result obtained by Bai in [4], (see also references therein) giving the construction of the dual matched pair for the associative algebras. • For q = 1, we recover the theorem relating the dual matched pair of center-symmetric algebras with the dual matched pair of Lie algebras, investigated in [12]. • For q = −1, Theorem4.5 gives the dual matched pair of flexible algebras. This is a new result, given in this work for the first time, to our best knowledge of the literature. Proposition 4.7. Assume that there is a q-generalized flexible algebra structure "◦" on the dual space A(q)∗. There is a q-generalized flexible algebra structure "⋆" on the vector space A(q)⊕ A(q)∗ given, for all x, y ∈ A(q) and all a, b ∈ A(q)∗, by: (x + a) ⋆ (y + b) = (x · y + R∗ +(a ◦ b + R∗ ◦(a)y + L∗ · (x)b + L∗ ◦(b)x) · (y)a), (4.9) if and only if the sixtuple (A(q), A(q)∗, R∗ algebras A(q) and A(q)∗. · , L∗ · , R∗ ◦, L∗ ◦) is a matched pair of the q-generalized flexible Proof: It is well known that A(q) ⊕ A(q)∗ is a vector space as a direct sum of vector spaces. The product "⋆" is also a bilinear product by definition. Then, it only remains to show that the q-generalized flexibility identity for the product "⋆" is equivalent to the fact that the sixtuple (A(q), A(q)∗, R∗ ◦) is a matched pair of A(q) and A(q)∗. For all x, y, z ∈ A(q), and all a, b, c ∈ A(q)∗, the left and right hand sides of the associator of the bilinear product ⋆ are given, respectively, by: · , R∗ · , L∗ ◦, L∗ {(x + a) ⋆ (y + b)} ⋆ (z + c) = {x · y + R∗ · (x)b)z} + (x · y) · z ◦(a)y) ◦(c)(R∗ ◦(b)x) · z + R∗ ◦(R∗ +{(L∗ +R∗ +L∗ +R∗ +R∗ +L∗ +L∗ ◦(c)(x · y) + L∗ · (x)b) ◦ c · (y)a) ◦ c ◦(b)x) + {(R∗ ◦(b)x)c} + {(L∗ ◦(a)y)c} + (a ◦ b) ◦ c + R∗ ◦(a ◦ b)z + L∗ ◦(c)(L∗ · (L∗ · (R∗ · (z)(a ◦ b) + L∗ · (z)(L∗ · (z)(R∗ · (y)a) · (x)b) · (x · y)c ◦(a)y + L∗ · (y)a} ⋆ (z + c) = ((x · y + R∗ ◦(b)x) · z + R∗ ◦(c)(x · y + R∗ ◦(b)x + a ◦ b + R∗ ◦(a)y · (x)b + L∗ ◦(b)x) ◦(a ◦ b + R∗ ◦(a)y + L∗ · (x)b · (y)a)z +L∗ +L∗ +L∗ +(a ◦ b + R∗ +R∗ +L∗ · (x)b + L∗ · (x · y + R∗ · (z)(a ◦ b + R∗ ◦(a)y + L∗ · (x)b + L∗ · (y)a) · (y)a) ◦ c ◦(b)x)c = (x · y) · z + (R∗ ◦(a)y) · z + (L∗ · (x)b)z + R∗ ◦(b)x) · z ◦(L∗ · (y)a)z ◦(R∗ ◦(c)(R∗ ◦(a)y) ◦(a ◦ b)z + R∗ ◦(c)(x · y) + L∗ ◦(c)(L∗ +R∗ +L∗ +L∗ +(a ◦ b) ◦ c + (R∗ +(L∗ +R∗ +L∗ {(x + a) ⋆ (y + b)} ⋆ (z + c) = {(R∗ · (y)a) ◦ c + R∗ · (L∗ ◦(b)x)c + L∗ · (z)(R∗ ◦(a)y) · z + R∗ ◦(b)x) · (x)b) + L∗ ◦(L∗ · (x)b) ◦ c · (x · y)c + R∗ · (z)(a ◦ b) · (z)(L∗ · (y)a)z} · (y)a) · (R∗ ◦(a)y)c MAHOUTON NORBERT HOUNKONNOU∗ AND MAFOYA LANDRY DASSOUNDO† 12 and, (x + a) ⋆ {(y + b) ⋆ (z + c)} = (x + a) ⋆ {y · z + R∗ · (y)c + L∗ +b ◦ c + R∗ ◦(b)z + L∗ · (z)b} ◦(c)y = x · (y · z + R∗ ◦(b)z + L∗ ◦(b ◦ c + R∗ ◦(a)(y · z + R∗ +R∗ L∗ +a ◦ (b ◦ c + R∗ +R∗ +L∗ · (x)(b ◦ c + R∗ · (y · z + R∗ ◦(c)y) ◦(b)z + L∗ · (z)b)x · (y)c + L∗ ◦(c)y) · (y)c + L∗ · (z)b) · (y)c + L∗ · (z)b) ◦(b)z + L∗ ◦(c)y)a ◦(b)z) + x · (L∗ ◦(a)(R∗ ◦(b)z) ◦(c)y) = x · (y · z) + x · (R∗ ◦(a)(y · z) + R∗ +R∗ +R∗ ◦(a)(L∗ +L∗ ◦(L∗ +a ◦ (L∗ +R∗ +L∗ ◦(c)y) + L∗ ◦(b ◦ c)x + L∗ ◦(R∗ · (y)c)x · (z)b)x + a ◦ (b ◦ c) + a ◦ (R∗ · (z)b) + R∗ · (y)c) · (z)b) + L∗ · (y · z)a · (x)(R∗ · (R∗ (x + a) ⋆ {(y + b) ⋆ (z + c)} = {x · (R∗ · (y)c) + R∗ · (L∗ ◦(b)z)a + L∗ ◦(b)z) + L∗ ◦(L∗ · (x)(b ◦ c) · (x)(L∗ ◦(c)y)a. · (z)b)x} · (y)c)x} ◦(b)z) ◦(R∗ ◦(a)(R∗ ◦(c)y) + L∗ +{x · (L∗ +R∗ ◦(a)(y · z) + R∗ +R∗ ◦(a)(L∗ +{a ◦ (R∗ +{a ◦ (L∗ +x · (y · z) + a ◦ (b ◦ c) + R∗ +R∗ · (x)(L∗ ◦(c)y) + L∗ · (y)c) + L∗ · (z)b) + L∗ ◦(b ◦ c)x · (L∗ · (R∗ · (y)c) + R∗ · (x)(R∗ ◦(c)y)a} ◦(b)z)a} · (x)(b ◦ c) · (z)b) + L∗ · (y · z)a Therefore, the associator can be rewritten as: ((x + a), (y + b, (z + c)) = {(x + a) ⋆ (y + b)} ⋆ (z + c) −(x + a) ⋆ {(y + b) ⋆ (z + c)} ((x + a), (y + b, (z + c)) = {(L∗ −L∗ −L∗ −R∗ −R∗ +{(L∗ · (L∗ −L∗ · (R∗ −L∗ + {R∗ · (x · y)c − R∗ ◦(b)x) · z + R∗ ◦(R∗ · (z)b)x} + {L∗ ◦(L∗ ◦(R∗ · (y)c)x} + {(R∗ ◦(a)(y · z)} + {(R∗ · (x)(b ◦ c)} ◦(b)z) · (x)b)z − x · (R∗ ◦(c)(x · y) − x · (L∗ ◦(a)y) · z + R∗ ◦(L∗ · (x)b) ◦ c + R∗ · (L∗ ◦(c)y) · (y)a)z ◦(b)x)c · (y)c) · (z)b) ◦(b)z)} · (y)a) ◦ c + R∗ ◦(c)y)a} + {L∗ ◦(b)z)a} + {R∗ · (x)(R∗ · (x)b) − R∗ · (y)a) − L∗ ◦(a)y)c − a ◦ (R∗ · (R∗ · (z)(a ◦ b) − a ◦ (L∗ ◦(a ◦ b)z − R∗ ◦(a)(R∗ · (y)c)} · (x)(L∗ · (y · z)a} ◦(c)(R∗ · (z)(R∗ · (z)(L∗ +{L∗ +{L∗ +(a, b, c) + (x, y, z) + {L∗ −R∗ ◦(c)y)} + {L∗ ◦(b)x) − L∗ ◦(a)(L∗ ◦(c)(L∗ · (z)b)} ◦(a)y) ◦(b ◦ c)x}. Similarily, we have: q (z + c, y + b, x + a) = q{(L∗ · (y)c)x ◦(a)y) · (z)b)x − z · (R∗ ◦(c)y) · x + R∗ ◦(R∗ · (x)b)z} + q{(R∗ ◦(b)x) ◦(L∗ ◦(a)(z · y) − z · (L∗ · (x)(c ◦ b) − c ◦ (L∗ · (R∗ · (y)c) ◦ a + R∗ ◦(a)y)c} + {(R∗ · (z)b) ◦ a ◦(b)z) · x + R∗ ◦(L∗ −L∗ ◦(c)(y · x)} + q{L∗ −R∗ ◦(R∗ · (y)a)z} + q{L∗ −L∗ ◦(b)x)c} + q{(L∗ −L∗ · (R∗ · (L∗ · (y)a) − L∗ −c ◦ (R∗ ◦(b)z)a − R∗ +R∗ · (L∗ · (z)(b ◦ a)} + q(c, b, a) + q(z, y, x) +q{R∗ · (z · y)a − R∗ · (z)(R∗ · (x)(R∗ +q{L∗ +q{L∗ ◦(a)(L∗ · (z)b) − R∗ ◦(b)z) − L∗ · (z)(L∗ ◦(b ◦ a)z} · (x)b) ◦(c)y)a · (y)a)} · (x)b)} q-GENERALIZED (ANTI -) FLEXIBLE ALGEBRAS AND BIALGEBRAS 13 +q{L∗ −R∗ ◦(c)(R∗ ◦(a)(R∗ ◦(c)y) − R∗ ◦(b)x)} + q{L∗ ◦(c)(L∗ · (x)(L∗ ◦(a)y)} + q{R∗ · (y)c) − L∗ ◦(c ◦ b)x · (y · x)c}. Hence, the equality: (x + a, y + b, z + c) = q(z + c, y + b, x + a) is equivalent to the following: · , L∗ (R∗ respectively, (L∗ · , A(q)∗) and (R∗ ◦, L∗ ◦(b)x) · z + R∗ ◦, A(q)) are bimodules of A(q) and A(q)∗, ◦(R∗ · (x)b)z − x · (R∗ ◦(b)z) − L∗ · (z)b)x = ◦(L∗ q((L∗ ◦(b)z) · x + R∗ ◦(R∗ ◦(c)(x · y) − x · (L∗ ◦(L∗ ◦(c)y) · x + R∗ ◦(b)x) − L∗ · (z)b)x − z · (R∗ ◦(c)y) − L∗ ◦(R∗ · (y)c)x − R∗ · (x)b)z), L∗ q((R∗ (R∗ q(L∗ (R∗ q(L∗ L∗ q((R∗ ◦(L∗ ◦(a)y) · z + R∗ ◦(a)(z · y) − z · (L∗ · (x)b) ◦ c + R∗ · (L∗ · (x)(c ◦ b) − c ◦ (L∗ · (z)(a ◦ b) − a ◦ (L∗ · (L∗ · (y)a)z − R∗ ◦(a)y) − L∗ ◦(b)x)c − R∗ · (x)b) − L∗ · (z)b) − L∗ ◦(b)z)a − R∗ · (z)b) ◦ a + R∗ ◦(L∗ · (y)c)x = ◦(c)(y · x)), ◦(a)(y · z) = ◦(R∗ · (y)a)z), · (x)(b ◦ c) = · (R∗ ◦(b)x)c), ◦(b)z)a = · (z)(b ◦ a)), · (R∗ (L∗ · (y)a) ◦ c + R∗ · (R∗ q((L∗ −c ◦ (R∗ ◦(a)y)c − a ◦ (R∗ · (R∗ · (y)c) ◦ a + R∗ · (y)a) − L∗ · (L∗ · (y)c) − L∗ ◦(c)y)a ◦(a)y)c). · (L∗ ◦(c)y)a =   Therefore, by Theorem 4.5, the bilinear product "⋆" defines a q-generalized flexible algebra struc- ture on the vector space A(q) ⊕ A(q)∗ if and only if (A(q), A(q)∗, R∗ ◦) is a matched pair of the q-generalized flexible algebras A(q) and A(q)∗. (cid:3) · , R∗ · , L∗ ◦, L∗ Remark 4.8. From Proposition 4.7, we conclude that both the flexible and antiflexible algebras have the same matched pairs given on A(q) and A(q)∗ for q = ±1. The same result extends to associative algebras obtained for the parameter q = 0. 5. Manin triple of q-generalized flexible algebras and bialgebras We start with the following definitions, consistent with analogous formulation for Lie algebras [16]: Definition 5.1. Let (A(q), ·) be a q-generalized flexible algebra. Suppose that there is a q-generalized flexible algebra structure "◦" on its dual space A(q)∗. A Manin triple of the q- generalized flexible algebras A(q) and A(q)∗ associated to a symmetric, non-degenerate, invariant bilinear form B defined on the vector space A(q) ⊕ A(q)∗ by: B(x + a, y + b) =< x, b > + < y, a >, (5.1) for all x, y ∈ A(q) and all a, b ∈ A(q)∗, where the bilinear product <, > is the natural pairing between the vector spaces A(q) and A(q)∗, is a triple (A(q) ⊕ A(q)∗, A(q), A(q)∗) such that the bilinear product "⋆" defined for all x, y ∈ A(q) and all a, b ∈ A(q)∗ by: (x + a) ⋆ (y + b) = (x · y + R∗ ◦(a)y + L∗ ◦(b)x) + (a ◦ b + R∗ · (x)b + L∗ · (y)a) realizes a q-generalized flexible algebra structure on A(q) ⊕ A(q)∗. Definition 5.2. The triple (A(q), A1(q), A2(q)), where: • A(q) is a q-generalized flexible algebra together with a nondegenerate, invariant and symmetric bilinear form, and • A1(q) and A2(q) are two Lagrangian sub-q-generalized flexible algebras of A(q) such that A(q) = A1(q) ⊕ A2(q). is a q-generalized flexible bialgebra. 14 MAHOUTON NORBERT HOUNKONNOU∗ AND MAFOYA LANDRY DASSOUNDO† Theorem 5.3. Suppose there is a q-generalized flexible algebra structure "◦" on the dual space A(q)∗. Then, the sixtuple (A(q), A(q)∗, R∗ ◦) is a matched pair of the generalized flexible algebras (A(q), ·) and (A(q)∗, ◦) if and only if (A(q) ⊕ A(q)∗, A(q), A(q)∗) is a Manin triple of the q-generalized flexible algebras (A(q), ·) and (A(q)∗, ◦). · , R∗ · , L∗ ◦, L∗ Proof: Let (A(q), ·) be a q-generalized flexible algebra. Assume that there is a q-generalized flexible algebra structure "◦" on its dual vector space A(q)∗. From Theorem 4.7, the vector space A(q) ⊕ A(q)∗ has a q-generalized flexible algebra structure here denoted by "⋆" given, for all x, y ∈ A(q) and all a, b ∈ A(q)∗, by: (x + a) ⋆ (y + b) = (x · y + R∗ ◦(b)x) + (a ◦ b + R∗ · (x)b + L∗ · (y)a). Then, the sixtuple (A(q), A(q)∗, R∗ ◦) is a matched pair of the q-generalized flexible algebras (A(q), ·) and (A(q)∗, ◦). From Definition 5.1, it remains to show that the bilinear form B defined by using the natural pairing between A(q) and its dual in (5.1) satisfies the relation: · , L∗ ◦(a)y + L∗ · , R∗ ◦, L∗ B((x + a) ⋆ (y + b), (z + c)) = B((x + a), (y + b) ⋆ (z + c)). (5.2) The left hand side of the equation (5.2) is given, for all x, y, z ∈ A(q) and all a, b, c ∈ A(q)∗, by: B((x + a) ⋆ (y + b), (z + c)) = B(x · y + R∗ ◦(a)y + L∗ ◦(b)x + a ◦ b + R∗ · (x)b +L∗ · (y)a, z + c) = < x · y + R∗ ◦(a)y + L∗ ◦(b)x, c > + < z, a ◦ b + R∗ · (x)b + L∗ · (y)a > = < x · y, c > + < R∗ + < L∗ + < z, R∗ ◦(a)y, c > ◦(b)x, c > + < z, a ◦ b > · (x)b > + < z, L∗ · (y)a >= < x · y, c > + < y, R◦(a)c > + < x, L◦(b)c > + < z, a ◦ b > + < R·(x)z, b > + < L·(y)z, a > B((x + a) ⋆ (y + b), (z + c)) = < x · y, c > + < y, c ◦ a > + < x, b ◦ c > + < z, a ◦ b > + < z · x, b > + < y · z, a > . Then, we have: B((x + a) ⋆ (y + b), (z + c)) = < x · y, c > + < y, c ◦ a > + < x, b ◦ c > + < z, a ◦ b > + < z · x, b > (5.3) + < y · z, a > . Besides, the right hand side of the equation (5.2) can be developed as follows: B((x + a), (y + b) ⋆ (z + c)) = B((x + a), y · z + R∗ ◦(b)z + L∗ ◦(c)y + b ◦ c +R∗ · (y)c + L∗ · (z)b) = < x, b ◦ c > + < x, R∗ + < x, L∗ + < R∗ · (y)c > · (z)b > + < y · z, a > ◦(b)z, a > + < L∗ ◦(c)y, a > = < x, b ◦ c > + < R·(y)x, c > + < L·(z)x, b > + < y · z, a > + < z, R◦(b)a > + < y, L◦(c)a > B((x + a), (y + b) ⋆ (z + c)) = < x, b ◦ c > + < x · y, c > + < z · x, b > + < y · z, a > + < z, a ◦ b > + < y, c ◦ a > Hence, B((x + a), (y + b) ⋆ (z + c)) = < x, b ◦ c > + < x · y, c > + < z · x, b > + < y · z, a > + < z, a ◦ b > + < y, c ◦ a > . Therefore, from the relations (5.3) and (5.4), we have the required result. (5.4) (cid:3) q-GENERALIZED (ANTI -) FLEXIBLE ALGEBRAS AND BIALGEBRAS 15 Theorem 5.4. Suppose that there is a q-generalized flexible algebra structure "◦" on the dual space A(q)∗. The following propositions are equivalent: (1) (A(q)⊕A(q)∗, A(q), A(q)∗, ω) is a Manin triple of the q-generalized flexible algebras A(q) and A(q)∗ with the nondegenerate symmetric bilinear form ω difined on A(q) ⊕ A(q)∗, for all x, y ∈ A(q) and all a, b ∈ A(q)∗ by: ω(x + a, y + b) :=< x, b > + < y, a >, where <, > is the natural pairing between A(q) and A(q)∗. (2) The sixtuple (A(q), A(q)∗, R∗ · , L∗ algebras (A(q), ·) and (A(q)∗, ◦). · , R∗ ◦, L∗ ◦) is a matched pair of the q-generalized flexible (3) (A(q), A(q)∗) is a q-generalized flexible bialgebra. Proof: By considering the Theorem 5.3, we deduct (1) ⇐⇒ (2). By the definition, we also (cid:3) have the equivalence (1) ⇐⇒ (3). 6. Application to octonion algebra Definition 6.1. An octonion algebra O is an eight dimensional vector space spanned by elements {e0, e1, · · · e7} satisfying the following relations: ∀i, j, k = 1, · · · 7, e2 0 = e0, eie0 = ei = e0ei, eiej = −δije0 + cijkek, (6.1) where the fully antisymmetric structure constants cijk are taken to be 1 for the combination of indexes: (ijk) ∈ {(124), (137), (156), (235), (267), (346), (457)}, with the bilinear product given in Table 1. e6 e6 e4 e3 e1 y e0 e2 e3 e4 e0 e0 e1 e2 e7 −e2 e1 −e0 e1 e4 e2 −e4 −e0 e2 e5 e3 −e7 −e5 −e0 e3 e4 e4 e5 −e6 e5 e6 e6 e7 e7 e7 e5 e5 e7 e6 −e5 −e3 e7 −e6 e1 e3 −e5 e1 e4 e4 −e3 −e1 −e0 e2 e5 −e4 −e2 −e0 e1 −e3 e6 e2 −e1 −e6 −e0 e2 −e4 e7 e3 −e2 −e7 −e0 e5 −e7 e3 e6 −e1 Table 6.1. Multiplication table of octonion algebra. The associator of the octonion algebra O = Span{e0, e1, · · · e7} defined as: eijk := (ei, ej, ek) = (eiej)ek − ei(ejek), ∀i, j, k ∈ {0, 1, 2, · · · 7} obeys the following relations: ∀i, j, k ∈ {1, 2, · · · , 7}, (e0, ei, ej) = (ei, e0, ej) = (ei, ej, e0) = (ei, ei, ej = (ei, ej, ei) = (ei, ej, ei) = 0 (6.2) (6.3) 7 (ei, ej, ek) = (cijmδmk − cjkmδim)e0 + X m=1 7 X n=1 7 (cijmcmkn − cjkmcimn)en, (6.4) X m=1 where the associator is written as (ei, ej, ek) := (eiej)ek − ei(ejek) := eijek. Proposition 6.2. Let O be an octonion algebra with basis {e0, e1, · · · , e7}. We have: (1) The 4 dimensional sub-algebras, spanned by the elements {e0, ei, ej, ek} where the index (ijk) ∈ {(124), (137), (156), (235), (267)(346), (457)}, are associative, i.e their associator vanishes. So far, the associator (ei, ej, ek) such that indexes are repeated, or contain zero, also vanishes, and the vector space {e0, ei, ej, ek} does not have a sub-algebra property. 16 MAHOUTON NORBERT HOUNKONNOU∗ AND MAFOYA LANDRY DASSOUNDO† e6 e4 e1 e7 e3 e2 e5 Figure 1. Realization of octonion algebra e2 0 0 0 0 0 0 0 −2e6 −2e7 e3 0 0 0 0 0 0 0 0 e4 0 0 0 0 0 0 0 0 0 0 e5 0 0 0 2e7 0 0 0 e6 0 0 0 2e3 −2e5 0 2e1 0 0 0 y e0 0 e00 0 e01 0 e11 0 e12 0 e22 0 −2e6 e23 0 e33 0 e34 0 e44 0 e45 0 e55 0 e56 0 e66 0 −2e4 e67 0 e77 Table 6.2. Table of composition of associator of octonion. 0 2e5 −2e7 0 2e3 0 0 0 0 2e6 −2e1 0 2e4 0 0 0 0 2e7 −2e2 0 2e5 0 0 2e1 −2e3 0 0 e7 0 0 0 0 2e4 0 2e2 0 0 0 0 0 0 −2e2 0 0 0 0 0 −2e1 0 0 0 0 0 e1 0 0 0 0 0 0 −2e3 (2) Other associators (ei, ej, ek), where the indexes (ijk) are fully skew-symmetric for the combinations : (ijk) ∈ {(123), (125), (126), (127), (234), (236), (237), (341), (342), (345), (347), (451), (452), (453), (456), (562), (563), (564), (567), (671), (673), (674), (675)}, do not vanish, and are anti-left symmetric, anti-right symmetric and anti-center symmetric. Definition 6.3. Let O be an octonion algebra. Consider the triple (l, r, V ), where V is a finite dimensional vector space, and l, r : O → gl(V ) are two linear maps. Then, the following relations are satisfied: for all ei ∈ O, i = 0, 1, · · · , 7, le0 = id = re0 , lei = −rei , [rei , lej ] = [rej , lei ], δij + lei lej = cijklek , (6.5) (6.6) (6.7) q-GENERALIZED (ANTI -) FLEXIBLE ALGEBRAS AND BIALGEBRAS 17 where the structure constants cijk, given in the equation (6.1), are well defined. Proposition 6.4. Let O be an octonion algebra, and l, r : O → gl(V ) be two linear maps. The couple (l, r) is a bimodule of the octonion algebra O if and only if there exists an octonion algebra structure "∗" on the semi-direct vector space O ⊕ V given by (ei + u) ∗ (ej + v) := eiej + lei v + rej u, ∀ei, ej ∈ O, ∀u, v ∈ V, i, j = 0, 1, · · · 7. Theorem 6.5. For an octonion algebra O spanned by {e0, e1, · · · , e7}, the following relations are equivalent: (1) (2) [ek, eiej] = [ek, ei]ej + ei[ek, ej], cijmckml = ckimcmjl + ckjmciml, (6.8) (6.9) where the reals Cijk are defined in the equation(6.1), for all i, j, k ∈ {0, 1, · · · 7}. Theorem 6.5 is known as Myung Theorem. For more details, see [18, 20]. Proposition 6.6. Let O be an octonion algebra with basis {e0, e1, · · · e7}. The following relation is satisfied: or, equivalently, 2δij + cijkrek + 2rej rei = 0, 2δij − cijklek + 2lej lei = 0, (6.10) (6.11) ∀i, j, k = 0, 1, · · · , 7, where cijk are defined in the equation(6.1), and lei , rei are linear operators satisfying the relations (6.5), (6.6) and (6.7). 7. Concluding remarks In this work, we have provided a q-generalization of flexible algebras, including center-symmetric, (also called antiflexible), algebras, and their bialgebras. Basic properties have been derived and discussed, as well as their connection to known algebraic structures investigated in the litera- ture. A q-generalization of Myung theorem has been given. Main properties related to bimodules, matched pairs, dual bimodules, and their algebraic consequences have been derived and discussed. Finally, the equivalence between q-generalized flexible algebras, their Manin triple and bialgebras has been elucidated. References [1] A.A. Albert, Nonassociative algebras: I. Fundamental concepts and isotropies, Annals of Maths., second series, Vol 43, N◦4 (1942) pp. 685-707. [2] A.A. Albert, Nonassociative algebras: II. New simple algebras, Annals of Maths., second series, Vol 43, N◦4 (1942) pp. 708-723. [3] M. Aguiar, On the associative analog of Lie bialgebras, Journal of Algebra 244,(2001), pp. 492-532. [4] C. Bai, Double constructions of Frobenius algebras, Connes cocycles and their duality, J. Noncommut. Geom. 4 (2010), pp. 475-530. [5] C. Bai, Left-symmetric bialgebras and an analogue of the classical Yang-Baxter equation, Comm. in Contemp. Math. Vol.10, No. 2 (2008), pp. 221-260. [6] G.M. Benkart and L.M. Osborn, Flexible Lie-admissible algebra, J. of Algebra, Vol.71 (1981), pp. 11-31. [7] M.C. Bhandari, On the Classification of simple antiflexible algebras, Trans. of the Amer. Math. Soc., Vol.173 (1972), pp. 159-181. [8] H.A. C¸ elik, On Primitive and prime antiflexible ring, Journal of Algebra, Vol.20 (1972), pp.428-440. [9] V.G. Drinfeld, Hamiltonian structure on the Lie groups, Lie bialgebras and the geometric sense of the classical Yang-Baxter equations, Soviet Math. Dokl. 27 (1983) pp. 68-71. [10] V.G. Drinfeld, Quantum groups, Proc. Internat. Congress Math.(Berkeley, 1986), Vol.1, Amer. Math. Soc., povidence, (1987), pp. 798-820. [11] M. Gunaydin and F. Gursey, Quark structure and octonions, J. Math. Phys., Vol.14, N◦11, (1973), pp. 1651-1667. 18 MAHOUTON NORBERT HOUNKONNOU∗ AND MAFOYA LANDRY DASSOUNDO† [12] M.N. Hounkonnou and M.L. Dassoundo, Center-symmetric algebras and bialgenras: relevant properties and consequences, Geometric Methods in Physics, XXXIV Workshop (2015), Trends in Maths., pp. 261-273. [13] M. Jimbo, Introduction to the Yang-Baxter equation, Int. J. of Mod. Phys. A, Vol.4 N◦15 (1989), pp. 3759- 3777. [14] E. Kleinfeld and L.A. Kokoris, Flexible algebras of degree one, Proc. of Amer. Math. Soc., Vol.13, N◦6 (1962), pp. 891-893. [15] F. Kosier, On a class of non-flexible algebras, Trans. of Amer. Math. Soc., Vol.102, N◦2 (1962), pp. 299-318. [16] Y. Kosmann-Schwarzbach, Integrability of nonlinear systems, 2nd ed., Lecture notes in physics, Pringer-Verlag, (2004), pp. 107-173. [17] J.H. Mayne, Flexible algebras of degree two, Trans. of Amer. Math. Soc. Vol.172, (1972), pp.69-81. [18] H.C. Myung, Some classes of flexible Lie-admissible algebras, Trans. of the Amer. Math. Soc., Vol.167, (1972), pp. 79-88. [19] Y. Nambu, Generalized Hamiltonian dynamics, Physical Review D, Vol.7, N◦8, (1973), pp. 2405-2412. [20] S. Okubo, Introduction to octonion and other nonassociative algebras in physics, Cambridge University Press, 1995. [21] S. Okubo, Nonassociative algebra in nuclear boson-expansion theory, Physical Review C, Vol.10, N◦5, (1974), pp. 2045-2047. [22] D.J. Rodabaugh, A theorem of semisimple antiflexible algebras, Communications in Algebra Vol.6 N◦11, (1978) pp. 1081-1090. [23] D.J. Rodabaugh, On semisimple antiflexible algebras, Portugaliae mathematica, vol.6, fasc. 3, (1967), pp. 261-271. [24] L.W. Davis and D.J. Rodabaugh, Simple nearly antiflexible algebras have unity elements, Soc. vol.2,1 (1969), pp. 69-72. [25] R.D. Schafer, An introduction to nonassociative algebras, New York Academic Press,1966. [26] R.D. Schafer, Alternative algebras over an arbitrary field, Bull. of the Amer. Math. Soc., vol.49 (1943), pp. 549-555. [27] R.M. Santilli, Supplemento al nuovo cimento, Series I. vol.6 (1968), pp. 1225-1249. [28] V.N. Zhelyabin, Jordan bialgebras and their relation to Lie bialgebras, Algebra and Logic, Vol.36, N◦1, (1997), pp. 1-15. [29] M. Zorn, Theorie der Alternativen Ringe, Abh. Math. Sem. Hamburgischen Univ. vol.8 (1930), pp. 123-147. (∗) University of Abomey-Calavi, International Chair in Mathematical Physics and Applications, ICMPA-UNESCO Chair, 072 BP 50, Cotonou, Rep. of Benin E-mail address: [email protected], with copy to [email protected] (†) University of Abomey-Calavi, International Chair in Mathematical Physics and Applications, ICMPA-UNESCO Chair, 072 BP 50, Cotonou, Rep. of Benin E-mail address: [email protected]
0912.2534	2	0912	2011-10-20T17:50:44	CSR expansions of matrix powers in max algebra	[ "math.RA" ]	We study the behavior of max-algebraic powers of a reducible nonnegative n by n matrix A. We show that for t>3n^2, the powers A^t can be expanded in max-algebraic powers of the form CS^tR, where C and R are extracted from columns and rows of certain Kleene stars and S is diadonally similar to a Boolean matrix. We study the properties of individual terms and show that all terms, for a given t>3n^2, can be found in O(n^4 log n) operations. We show that the powers have a well-defined ultimate behavior, where certain terms are totally or partially suppressed, thus leading to ultimate CS^tR terms and the corresponding ultimate expansion. We apply this expansion to the question whether {A^ty, t>0} is ultimately linear periodic for each starting vector y, showing that this question can be also answered in O(n^4 log n) time. We give examples illustrating our main results.	math.RA	math	CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA SERGEI SERGEEV AND HANS SCHNEIDER Abstract. We study the behavior of max-algebraic powers of a reducible nonnegative matrix A ∈ Rn×n + . We show that for t ≥ 3n2, the powers At can be expanded in max-algebraic sums of terms of the form CS tR, where C and R are extracted from columns and rows of certain Kleene stars, and S is diagonally similar to a Boolean matrix. We study the properties of individual terms and show that all terms, for a given t ≥ 3n2, can be found in O(n4 log n) operations. We show that the powers have a well-defined ultimate behavior, where certain terms are totally or partially suppressed, thus leading to ulti- mate CS tR terms and the corresponding ultimate expansion. We apply this expansion to the question whether {Aty, t ≥ 0} is ultimately linear periodic for each starting vector y, showing that this question can be also answered in O(n4 log n) time. We give examples illustrating our main results. 1. Introduction By max algebra we understand the analogue of linear algebra developed over the max-times semiring Rmax,× which is the set of nonnegative numbers R+ equipped with the operations of "addition" a⊕b := max(a, b) and the ordinary multiplication a ⊗ b := a × b. Zero and unity of this semiring coincide with the usual 0 and 1. The operations of the semiring are extended to the nonnegative matrices and vectors in the same way as in conventional linear algebra. That is if A = (aij), B = (bij) and C = (cij) are matrices of compatible sizes with entries from R+, we write C = A⊕B all i, j. if cij = aij ⊕ bij for all i, j and C = A ⊗ B if cij = Lk aikbkj = maxk(aikbkj) for If A is a square matrix over R+ then the iterated product A⊗ A⊗ ...⊗ A in which the symbol A appears k times will be denoted by Ak. These are the max-algebraic powers of nonnegative matrices, the main object of our study. The max-plus semiring Rmax,+ = (R ∪ {−∞}, ⊕ = max, ⊗ = +), developed over the set of real numbers R with adjoined element −∞ and the ordinary addition playing the role of multiplication, is another isomorphic "realization" of max alge- bra. In particular, x 7→ exp(x) yields an isomorphism between Rmax,+ and Rmax,×. In the max-plus setting, the zero element is −∞ and the unity is 0. The main results of this paper are formulated in the max-times setting, since some important connections with nonnegative and Boolean matrices are more trans- parent there. However, the max-plus setting is left for the examples in the last section, in order to appeal to the readers who work with max-plus algebra and applications in scheduling problems and discrete event systems [1, 5, 8, 17]. The Cyclicity Theorem is a classical result of max algebra, in the max-times set- ting it means that the max-algebraic powers of any irreducible nonnegative matrix 2010 Mathematics Subject Classification. Primary:15A80, 15A23, 15A21. This work was supported by EPSRC grant RRAH12809 and RFBR grant 08-01-00601. 1 2 SERGEI SERGEEV AND HANS SCHNEIDER are ultimately periodic (up to a scalar multiple), with the period equal to cyclicity of the critical graph. This theorem can be found in Heidergott et al. [17, Theorem 3.9], see also Baccelli et al. [1, Theorem 3.109] and Cuninghame-Green [8, Theorem 27-6] (all stated in the max-plus setting). However, the length of the pre-periodic part can be arbitrarily large and the result does not have an evident extension to reducible matrices. The behavior of max-algebraic powers of reducible matrices relies on connec- tions between their strongly connected components and the hierarchy of their max- algebraic eigenvalues. This has been well described in a monograph of Gavalec [16], see also [9, 15, 20]. The relation of the Cyclicity Theorem to the periodicity of Boolean matrices is understood but not commonly and explicitly used in max algebra. For instance the construction of cyclic classes [2, 3] appears in the proof of Lemma 3.3 in [17] (without references to literature on Boolean matrices). This relation becomes particularly apparent after application of a certain D−1AD similarity scaling called visualization in [25, 26], see also [10, 11]. Semanc´ıkov´a [22, 23] realized that the cyclic classes of Boolean matrices are helpful in treating computational complexity of the periodicity problems in max-min algebra, with analogous max-algebraic applications in mind [24]. A result by Nachtigall [21] states that, though the length of the preperiodic part cannot be polynomially bounded, the behavior of matrix powers after O(n2) can be represented as max-algebraic sums of matrices from certain periodic sequences. Moln´arov´a [19] studies this Nachtigall expansion further, showing that for a given matrix power after O(n2) the representing Nachtigall matrices can be computed in O(n5) time. In the general reducible case, the sequences of entries {at ij, t ≥ 0} of At are ultimately generalized periodic [9, 20], meaning that for t ≥ T where T is sufficiently large, they may consist of several ultimately periodic subsequences which grow with different rates. Gavalec [15, 16] showed that deciding whether {at ij, t ≥ 0} is ultimately periodic is in general NP-hard. In a related study, Butkovic et al. [5, 6] considered robust matrices, such that for any given x the sequence {Atx} is ultimately periodic with period 1. The conditions formulated in [6] can be verified in polynomial time, which suggests that such "global" periodicity questions must be tractable. The main goal of this paper is to find a common ground for the above pieces of knowledge on matrix periodicity in max algebra. We introduce the concept of CSR expansion, in which a matrix power At is represented as the max-algebraic sum of terms of the form CStR, called CSR products. Here C and R have been extracted from columns and rows of certain Kleene star (the same for both), and S can be made Boolean by a similarity scaling D−1SD. The matrix CR appeared previously as spectral projector [1, 7], and S typically arises as the incidence matrix of a critical graph. After giving necessary preliminaries in Section 2, we start in Section 3 by study- ing the CSR products. We show that they form a cyclic group and describe the action of this group on the underlying Kleene star. We also emphasize the path sense of these operators, see Theorem 3.3, thus providing connection to the ap- proach of [9, 15, 16, 20, 22, 23]. In Section 4 we establish the algebraic form of CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 3 the Nachtigall expansion which controls the powers At after t ∼ O(n2). See The- orem 4.2. In Section 5 we show that at large t certain Nachtigall terms become totally or partially suppressed by heavier ones, leading to the ultimate expansion of matrix powers. This result, see Theorem 5.6, can be understood as a generalization of the Cyclicity Theorem to reducible case. In Section 6 we treat the computational complexity of computing the terms of CSR expansions for a given matrix power, showing that in general this can be done in no more than O(n4 log n) operations. In Section 7, which extends the results of Butkovic et al. [5, 6] on robust matrices, we describe orbit periodic matrices A, i.e., such that the orbit Aty is ultimately linear periodic for all initial vectors y, see Theorem 7.6. We use the ultimate expansion to show that the conditions for orbit periodicity can be verified in no more than O(n4 log n) operations, see Theorem 7.8 and its Corollary. We conclude by Section 8 which contains some examples given in the max-plus setting. 2. Preliminaries In this section we recall some important notions of max algebra. These are the maximum cycle geometric mean, the critical graph and the Kleene star. We close the section with nonnegative similarity scalings and cyclic classes of the critical graph. Let A = (aij ) ∈ Rn×n + . The weighted digraph D(A) = (N (A), E(A)), with the set of nodes N (A) = {1, . . . , n} and the set of edges E(A) = {(i, j) aij 6= 0} with weights w(i, j) = aij , is called the digraph associated with A. Suppose that P = (i1, ..., ip) is a path in DA, then the weight of P is defined to be w(P ) = ai1i2 ai2i3 . . . aip−1ip if p > 1, and 1 if p = 1. If i1 = ip then P is called a cycle. The length of P , denoted by l(P ), is the number of edges in P (it equals p − 1 here). When any two nodes in D(A) can be connected to each other by paths,the matrix A is called irreducible. Otherwise, it is called reducible. In the reducible case, there are some (maximal) strongly connected components of D(A), and a number of nodes that do not belong to any cycle. We will refer to such nodes as to trivial components of D(A). The maximum cycle geometric mean of A, further denoted by λ(A), is defined by the formula (2.1) λ(A) = max Pc (w(Pc))1/k, where the maximization is taken over all cycles Pc = (i1, . . . , ik), for k = 1, . . . , n, in the digraph D(A). The Cyclicity Theorem ([17, Theorem 3.9], see also [1, 5, 8]) states that if A is irreducible then after a certain time T (A), there exists a number γ such that At+γ = λγ(A)At for all t ≥ T (A). Thus λ is the ultimate growth rate of matrix powers in this case. If λ = 1 then At+γ = At for t ≥ T (A), in which case we say that {At, t ≥ 0} is ultimately periodic. Remarkably λ(A) is also the largest max-algebraic eigenvalue of A, meaning the + such that A ⊗ x = λx, largest number λ for which there exists a nonzero x ∈ Rn see [1, 4, 5, 8, 17] and references therein. The operation of taking the maximal cycle geometric mean (m.c.g.m. for short) is homogeneous: λ(αA) = αλ(A). Hence any matrix, which has λ(A) 6= 0 meaning that D(A) is not acyclic, can be scaled so that λ(A/λ(A)) = 1. Following [4], matrix A ∈ Rn×n + with λ(A) = 1 will be called definite. 4 SERGEI SERGEEV AND HANS SCHNEIDER A cycle Pc = (i1, . . . , ik) in D(A) is called critical, if (w(Pc))1/k = λ(A). Every node and edge that belongs to a critical cycle is called critical. The set of critical nodes is denoted by Nc(A), the set of critical edges is denoted by Ec(A). The critical digraph of A, further denoted by C(A) = (Nc(A), Ec(A)), is the digraph which consists of all critical nodes and critical edges of D(A). The cyclicity of an irreducible graph is defined as the g.c.d. (greatest common divisor) of the lengths of all its simple cycles. The critical graph defined above may have several strongly connected components, and in this case the cyclicity is the l.c.m. (least common multiple) of their cyclicities. This gives the number γ which appears in the Cyclicity Theorem [1, 17], and it can be shown that the ultimate period cannot be less than γ (in particular, this follows from the approach of the present paper). There is no obvious subtraction in max algebra, however we have an analogue of (I − A)−1 defined by (2.2) A∗ := I ⊕ A ⊕ A2 ⊕ . . . , where I is the identity matrix. This series converges to a finite matrix if and only if λ(A) ≤ 1 [1, 4, 8, 17], and then this matrix A∗ = (a∗ ij ) is called the Kleene star of A. This matrix has properties (A∗)2 = A∗ and, clearly, A∗ ≥ I. It is important that the entries of max-algebraic powers Ak = (ak ij) express the maximal weights of certain paths: ak ij is equal to the greatest weight of paths P that connect i to j and have length k. The entry a∗ ij for i 6= j is equal to the greatest weight of paths that connect i to j with no restriction on their lengths. As in the nonnegative linear algebra, we have only few invertible matrices, in the sense of the existence of (nonnegative) A−1 such that A−1 ⊗ A = A ⊗ A−1 = I. More precisely, such matrices can be diagonal matrices (2.3) X = diag(x) :=  x1 ... 0 . . . 0 ... . . . . . . xn   for a positive x = (x1, . . . , xn), or monomial matrices obtained from the diagonal matrices by permutations of rows or columns. Nevertheless, such matrices give rise to very convenient diagonal similarity scalings A 7→ X −1AX. Such transfor- mations do not change λ(A) and C(A) [12]. They commute with max-algebraic multiplication of matrices and hence with the operation of taking the Kleene star. Geometrically, they correspond to automorphisms of Rn +, both in the case of max algebra and in the case of nonnegative linear algebra. The importance of such scalings in max algebra was emphasized already in [8], Ch. 28. By an observation of Fiedler and Pt´ak [13], for any definite matrix A there is a scaling X such that X −1AX is visualized meaning that all critical entries of the matrix equal 1 and all the rest are less than or equal to 1. It is also possible to make X −1AX strictly visualised [26], meaning that only critical entries are equal to 1. If a matrix is visualised, or strictly visualised, the same is true for all powers of this matrix, meaning that the critical graph can be seen as a Boolean matrix that "lives by itself". Thus there is a clear connection to the powers of Boolean matrices. CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 5 The periodicity of powers of Boolean matrices is ruled by cyclic classes [2] also known as imprimitivity sets [3], which we explain below. We note that this notion appeared already in a work of Frobenius [14]. Proposition 2.1 (e.g. Brualdi-Ryser [3]). Let G = (N, E) be a strongly connected digraph with cyclicity γG.Then the lengths of any two paths connecting i ∈ N to j ∈ N (with i, j fixed) are congruent modulo γG. Proposition 2.1 implies that the following equivalence relation can be defined: i ∼ j if there exists a path P from i to j such that l(P ) ≡ 0(mod γG). The equivalence classes of G with respect to this relation are called cyclic classes [2, 22, 23]. The cyclic class of i will be denoted by [i]. Consider the following access relations between cyclic classes: [i] →t [j] if there exists a path P from a node in [i] to a node in [j] such that l(P ) ≡ t(mod γG). In this case, a path P with l(P ) ≡ t(mod γG) exists between any node in [i] and any node in [j]. Further, by Proposition 2.1 the length of any path between a node in [i] and a node in [j] is congruent to t, so the relation [i] →t [j] is well-defined. Cyclic classes can be computed in O(E) time by Balcer-Veinott digraph con- densation [2], where E denotes the number of edges in G. At each step of this algorithm, we look for all edges which issue from a certain node i, and condense all end nodes of these edges into a single node. Another efficient algorithm is described in [3]. Let S = (sij ) be the incidence matrix of G, meaning that sij = 1 if (i, j) ∈ E and sij = 0 otherwise. The ultimate periodicity of such Boolean matrices has been well studied. If γG = 1 then the periodicity of St starts latest after the Wielandt number W (n) := (n − 1)2 + 1 [3, 18]. This bound is sharp and is due to Wielandt [28]. If γ > 1 then there are even better sharp bounds due to Schwartz [27]. Assume w.l.o.g. that S is irreducible, and let n = αγ + t. If α > 1 then the periodicity starts at most after W (α)γ + t which does not exceed n2 γ + γ. If α = 1 then it starts almost "straightaway", after at most max(1, t). 3. CSR products In this section, given a nonnegative matrix A ∈ Rn×n + , we consider max-algebraic products of the form CStR, where S is associated with some subdigraph of the critical graph C(A), and matrices C and R are extracted from a certain Kleene star related to A. We show that these products form a cyclic group and study their periodic properties. We also show that CStR are related to a distinguished set of paths which we call C-heavy. We start with a remark that the concept of cyclic classes discussed in Section 2 can be generalized to completely reducible digraphs, which consist of (possibly sev- eral) strongly connected components, not connected with each other. Importantly, the critical digraph of any A ∈ Rn×n + with λ(A) > 0 is completely reducible. Let A ∈ Rn×n + have λ(A) = 1. Consider any completely reducible subdigraph C = (Nc, Ec) of C(A). In particular, C may consist of several disjoint cycles of C(A), or we can take a component of C(A), or we may just have C = C(A). Denote by γ the cyclicity of C and take B := (Aγ )∗. Define the matrices C = (cij ) ∈ Rn×n + , 6 SERGEI SERGEEV AND HANS SCHNEIDER C CR (Aγ)∗ Figure 1. The scheme of C and R defined in (3.1) R = (rij ) ∈ Rn×n + and S = (sij) ∈ Rn×n + by (3.1) 0, cij =(bij, sij =(aij, 0, if j ∈ Nc otherwise, rij =(bij, 0, if i ∈ Nc otherwise, if (i, j) ∈ Ec otherwise. The nonzero entries of C, respectively R, can only be in the submatrix of B = (Aγ)∗ extracted from columns, respectively rows, in Nc. All nonzero entries of S are in the principal submatrix SNcNc extracted from rows and columns in Nc. See Figure 1 for a schematic display. We show in the next proposition that S can be assumed to be 0 − 1. It can be also deduced from the results in [12]. Proposition 3.1. Let A ∈ Rn×n have λ(A) = 1 and let S be defined as above. + + such that D−1SD for D := diag(z) is a 0 − 1 There exists a positive z ∈ Rn matrix. Proof. As λ(S) = 1, we can take z :=Ln ·j. This vector is positive, and observe that Sz ≤ z since SS ∗ ≤ S ∗. From this and D(S) = C(S) = C, it can be deduced by multiplying z−1 i sijzj = 1 for all (i, j) ∈ D(S), while z−1 (cid:3) i sijzj ≤ 1 along cycles that z−1 i sijzj = 0 for all (i, j) /∈ D(S). j=1 S ∗ As S can be scaled to 0 − 1 matrix, we conclude that {St, t ≥ 0} becomes periodic at most after the Wielandt number W (nc) = (nc − 1)2 + 1, where nc is the number of nodes in Nc. Matrix A will be called S-visualized, if S defined in (3.1) is Boolean. In the S-visualized case, the asymptotic form of St for t ≥ T is determined by the cyclic classes of C [3]: (3.2) st ij =(1, if [i] →t [j], 0, otherwise. Now we study the CSR products (3.3) assuming that A ∈ Rn×n + P (t) := CStR, has λ(A) = 1. CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 7 We start by the observation that if T is the number after which {St, t ≥ 0} becomes periodic, then SlγR = R, CSlγ = C, (3.4) Indeed, (3.2) implies that all diagonal entries of Slγ with indices in Nc are 1 if lγ ≥ T , which implies SlγR ≥ R and CSlγ ≥ C. On the other hand, S ≤ A and so Slγ ≤ (Aγ)∗. Further, (Aγ)∗R ≤ R and C(Aγ )∗ ≤ C since (Aγ)∗(Aγ )∗ = (Aγ)∗, and so SlγR ≤ (Aγ)∗R ≤ R and CSlγ ≤ C(Aγ )∗ ≤ C. ∀lγ ≥ T. As {St, t ≥ T } is periodic, so is {P (t), t ≥ T }. Moreover, we conclude from (3.4) that this periodicity starts from the very beginning. Proposition 3.2 (Periodicity). P (t+γ) = P (t) for all t ≥ 0. Proof. It follows from Eqn. (3.4) that P (t+lγ) = P (t) and P (t+(l+1)γ) = P (t+γ) for lγ ≥ T . But P (t+lγ) = P (t+(l+1)γ), which implies that also P (t+γ) = P (t). (cid:3) It is also useful to understand the meaning of P (t) in terms of paths. Given a set of paths Π, we denote by w(Π) the greatest weight of paths in Π, assuming w(Π) = 0 if Π is empty. A path will be called C-heavy if it goes through a node in Nc. The set of all C-heavy paths on D(A) that connect i to j and have length t will be denoted by Πh ij,t. We also denote by τ the maximal cyclicity of the components of C(A). Theorem 3.3 (C-Heavy Paths). Let A ∈ Rn×n such that {St, t ≥ T } is periodic. + have λ(A) = 1 and let T ≥ 0 be 1. For t ≥ 0, 2. For t ≥ T + 2τ (n − 1), (3.5) (3.6) w(Πh ij,t) ≤ P (t) ij . w(Πh ij,t) ≥ P (t) ij . Proof. By Proposition 3.1 there exists a diagonal matrix D such that D−1AD is S-visualized. As both sides of (3.5) and (3.6) are stable under similarity scaling of A, we will assume that A is already S-visualized. 1.: Let P ∈ Πh ij,t and w(P ) 6= 0. We need to show that (3.7) w(P ) ≤ P (t) ij Path P can be decomposed as P = Pbeg ◦ Pend, where Pbeg connects i to a node m ∈ Nc and Pend connects m to j. Adjoining to P any sufficiently large number of cycles of C that go through m and whose total length is a multiple of γ, we obtain beg ◦ P ′ a path P ′ that we can decompose as P ′ = P ′ beg connects i to m1 ∈ C and l(P ′ beg) is a multiple of γ, P ′ int connects m1 to m2 ∈ Nc, has length t and belongs entirely to C, and P ′ end) is a multiple of int) = st γ. We conclude that w(P ′ m1m2 = 1. We obtain w(P ) = w(P ′) ≤ cim1 st m1m2rm2j, which implies (3.7), and hence (3.5). m1m2 rm2j. This is the weight of a path P decomposed as P = Pbeg ◦ Pint ◦ Pend, where Pbeg connects i to m1, Pint connects m1 to m2 and Pend connects m2 to j. Here Pint has length t and belongs to the component of C which we denote by T and whose cyclicity we denote by π. The lengths of Pbeg and Pend are respectively l(Pbeg) = l1π and 2.: There exist indices m1 and m2 such that P (t) end connects m2 to j and l(P ′ end, where P ′ beg) ≤ cim1 , w(P ′ end) ≤ rm2j and w(P ′ int ◦ P ′ ij = cim1 st 8 SERGEI SERGEEV AND HANS SCHNEIDER l(Pend) = l2π for some l1, l2, since l(Pbeg) and l(Pend) are multiples of γ and γ is itself a multiple of π. Paths Pbeg and Pend correspond to certain paths on D(Aπ) with lengths l1 and l2. If l1 ≥ n or l2 ≥ n, then we can perform cycle deletion (w.r.t. D(Aπ)) and obtain paths P 2 end with lengths k1π and k2π where k1 < n and k2 < n. For the resulting path P 2 := P 2 end we will have w(P 2) ≥ w(P ) since λ = 1. Now we have l(P 2 end) ≤ 2τ (n − 1). If t ≥ T + 2τ (n − 1), then the principal submatrix of St corresponding to the component T coincides with that of St−(k1+k2)π, which implies that Pint can be int with length t − (k1 + k2)π, so that w(P 3) = w(P 2) ≥ P (t) replaced by a path P 3 ij beg ◦ P 3 where P 3 := P 2 int ◦ P 2 (cid:3) beg ◦ Pint ◦ P 2 beg) + l(P 2 beg and P 2 end ∈ Πh ij,t. This "path sense" of P (t) simplifies the proof of the following important law. Theorem 3.4 (Group law). P (t1+t2) = P (t1)P (t2) for all t1, t2 ≥ 0 Proof. As (Aγ)∗(Aγ )∗ = (Aγ)∗, we have (RC)ii = (Aγ)∗ hence RCS ≥ S. We use this to obtain that P (t1)P (t2) ≥ P (t1+t2): ii = 1 for all i ∈ Nc, and (3.8) P (t1)P (t2) = CSt1 RCSt2R ≥ CSt1+t2 R = P (t1+t2). But Theorem 3.3 implies that P (t1)P (t2) ≤ P (t1+t2) for all large enough t1, t2, since the concatenation of two C-heavy paths is again a C-heavy path. As P (t) are periodic, it follows that P (t1)P (t2) ≤ P (t1+t2) for all t1, t2, and we obtain the claim combining this with the reverse inequality. (cid:3) Formulas (3.2) and (3.4) imply that if A is S-visualized then all rows of R or columns of C with indices in the same cyclic class of C coincide. Hence, when working with P (t)we can assume without loss of generality that all cyclic classes have just 1 element and consequently, that SNcNc is a permutation matrix. This captures the structure of P (t), which form a cyclic group of order γ. As usual ei denotes the vector which has all coordinates equal to 0 except for the ith which equals 1. For the rows P (t) ·j of P (t) we have: (3.9) i· and columns P (t) i· , P (t)ej = P (t) ·j . i P (t) = P (t) eT Next we study the periodicity of P (t) in more detail. It turns out that the columns and rows of P (t+1) with indices in Nc can be obtained from those of P (t) by means of a permutation on cyclic classes, while the rest of the columns (or rows) are max-linear combinations of the critical ones. We start with the following observation on the spectral projector P (0) := CR, which can be found in [1, 7]. Lemma 3.5. P (0) Proof. As (Aγ)∗ We see that i· = Ri· and P (0) ii = 1 for all i ∈ Nc, we obtain eT ·i = C·i for all i ∈ Nc. i C = Ci· ≥ eT i and Rei = R·i ≥ ei. (3.10) P (0) P (0) i CR ≥ eT i· = eT ·i = CRei ≥ Cei = C·i. i R = Ri·, But CR ≤ ((Aγ )∗)2 = (Aγ )∗, which implies the reverse inequalities P (0) P (0) i· ≤ Ri· and (cid:3) ·i ≤ C·i. CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 9 Theorem 3.6. Let A ∈ Rn×n then + have λ(A) = 1 and be S-visualized. If [i] →t [j], (3.11) for all s, t ≥ 0. P (t+s) i· = P (s) j· , P (s) ·i = P (t+s) ·j Proof. We prove the first equality of (3.11). Using the group law we assume that i St = ej. Using s = 0. We also assume that SNcNc is a permutation matrix, then eT this and Lemma 3.5 we obtain P (0) j· = eT i StR ≤ j P (0) = eT ≤ eT j R = eT i CStR = eT i P (t) = P t i·. (3.12) Analogously we have P (0) the group law and periodicity, we obtain that P (t) with (3.12) we obtain the desired property. . Multiplying this inequality by P (t) and using . Combining this (cid:3) i· ≤ P (γ−t) j· = P (0) i· ≤ P (γ) j· j· Corollary 3.7. Let A ∈ Rn×n (CSt)·i for all i ∈ Nc. + have λ(A) = 1. Then P (t) i· = (StR)i· and P (t) ·i = Proof. We assume that A is S-visualized, and we also assume that SNcNc is a permutation matrix. For [i] →t [j], Theorem 3.6 and Lemma 3.5 imply that i P (t) = eT P (t) i· = eT P (t) ·j = P (t)ej = P (0)ei = Cei = CStej = (CSt)·j. i StR = (StR)i·, j P (0) = eT j R = eT The claim is proved. (cid:3) Corollary 3.8. Let A ∈ Rn×n αik and βki, where k ∈ Nc, such that + have λ(A) = 1. For each k = 1, . . . , n there exist (3.13) P (t) ·k = Mi∈Nc αikP (t) ·i , P (t) k· = Mi∈Nc βkiP (t) i· . Proof. By Corollary 3.7 we have P (t) i· = (StR)i· for all i ∈ Nc. Eqn. (3.13) follows directly from P (t) = CStR, the coefficients αik (resp. βki) being taken from the kth column of R (resp. the kth row of C). (cid:3) ·i = (CSt)·i and P (t) 4. Nachtigall expansions In this section we show that the powers At of A ∈ Rn×n can be expanded for t ≥ 3n2 as sum of CSR products. This establishes a more general algebraic form of the Nachtigall expansion studied in [19, 21]. + Let A = (aij ) ∈ Rn×n + . Define λ1 = λ(A), and let C1 = (N1, E1) be a completely reducible subdigraph of C(A). Set A1 := A and K1 := N , where N = {1, . . . , n}. The elements of a Nachtigall expansion will be now defined inductively for µ ≥ 2. Namely, we define Kµ := N \ ∪µ−1 i=1 Ni and Aµ = (aµ ij ) ∈ Rn×n + by (4.1) aµ ij =(aij, 0, if i, j ∈ Kµ, otherwise. 10 SERGEI SERGEEV AND HANS SCHNEIDER R1 ∗ ∗ ∗ 0 0 0 ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ 0 0 0 S1 ∗ ∗ 0 0 0 C1 0 ∗ ∗ 0 0 0 0 0 0 0 0 0 0 S2 ∗ 0 0 0 C2 R2 0 ∗ ∗ ∗ ∗ ∗ 0 ∗ ∗ ∗ ∗ ∗ 0 ∗ ∗ ∗ ∗ ∗ Figure 2. Formation of first two terms in a Nachtigall expansion: a schematic example. Note that the structure of strongly connected components of D(A) is of no use here. Further define λµ := λ(Aµ). If λµ = 0 then stop, otherwise select a completely reducible subdigraph Cµ = (Nµ, Eµ) of the critical digraph C(Aµ), and proceed as above with µ := µ + 1. By the above procedure we define Kµ, Aµ, λµ and Cµ = (Nµ, Eµ) for µ = 1, . . . , m, where m ≤ n is the last number µ such that λµ > 0. Denote L :=Sm i=1 Nµ and L = N \L. For each µ = 1, . . . , m, let γµ be the cyclicity of Cµ. Since λ((Aµ/λµ)γµ ) = λ(Aµ/λµ) = 1, the Kleene star Bµ := ((Aµ/λµ)γµ)∗ is finite. Define the matrices Cµ = (cµ and Sµ = (sµ + , Rµ = (rµ ij ) ∈ Rn×n ij ) ∈ Rn×n + ij ) ∈ Rn×n + by (4.2) if j ∈ Nµ, otherwise, rµ ij =(bµ ij, 0, if i ∈ Nµ, otherwise, cµ ij, 0, ij =(bµ ij =(aµ sµ ij/λµ, 0, if (i, j) ∈ Eµ otherwise. A schematic example of Nachtigall expansion is given in Figure 2. It follows from Proposition 3.1 that each Sµ can be scaled to a 0 − 1 matrix using a certain vector denoted here by zµ. Note that the sets Nµ are pairwise disjoint. Defining z ∈ Rn + by (4.3) zi =(zµ i , 1, if i ∈ Nµ, if i ∈ L, and letting D := diag(z), we obtain that the matrix A := D−1AD is totally S- visualized, meaning that all corresponding matrices Sµ are Boolean. As Sµ can be scaled to be Boolean, the sequences of their max-algebraic powers {St µ t ≥ 0}, being powers of Boolean matrices when scaled, are ultimately periodic with periods γµ. This periodicity starts at most after the corresponding Wielandt numbers (kµ − 1)2 + 1 where kµ is the number of elements in Nµ We proceed with some notation. Denote µ(i) = µ if i ∈ Nµ, and µ(i) = +∞ if i ∈ L. Denote by Πij,t the set of paths on D(A) which connect i to j and have length t. Denote by Πµ ij,t the set of paths P ∈ Πij,t such that mini∈NP µ(i) = µ, CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 11 where NP is the set of nodes visited by P . (Note that greater values of µ correspond to smaller λµ.) The paths in Πµ ij,t will be called µ-heavy, since they are Cµ-heavy (see Sect. 3) in Aµ. Any path P ∈ Πij,t with t ≥ n has at least one cycle. Hence there are no paths with length t ≥ n that visit only the nodes in L, for otherwise the subdigraph of D(A) induced by L would contain a cycle and the number of components would be more than m. We can express the entries of At = (at ij ) for t ≥ n as follows: ij) and At µ = (aµ,t at ij = w(Πij,t) = w(Πµ ij,t), t ≥ n m Mµ=1 (4.4) Denote (4.5) aµ,t ij = m Mν=µ w(Πν ij,t), t ≥ n. N (t) µ := CµSt µRµ. These CSR products are defined from Aµ and Cµ in the same way as P (t), see Sect. 3, were defined from A and C, and it follows that the sequence N (t) µ is periodic with period γµ. Further denote by τµ the greatest cyclicity of a component in Cµ and by nµ the number of nodes in Kµ. The following is a version of Theorem 3.3 for N (t) µ . Theorem 4.1 (µ-Heavy Paths). Let Tµ be such that {St µ, t ≥ Tµ} is periodic. 1. For t ≥ 0, (4.6) (4.7) w(Πµ ij,t) ≤ λt µ(N (t) µ )ij 2. For t ≥ Tµ + 2τµ(nµ − 1) , w(Πµ ij,t) ≥ λt µ(N (t) µ )ij Proof. We can w.l.o.g. assume that λµ = 1, since both (4.6) and (4.7) are stable under scalar multiplication of A. After this, the claim follows from Theorem 3.3. (cid:3) In the theorem above, we can choose Tµ equal to each other and of the order O(n2) for all µ. The main result of this section immediately follows now from Theorem 4.1 and Eqn. (4.4), noting that Tµ + 2τµ(nµ − 1) ≤ 3n2 for all µ. Theorem 4.2 (Nachtigall expansion). Let A ∈ Rn×n + . Then for all t ≥ 3n2 (4.8) In particular, (4.9) At µ = At = m Mν=µ m Mν=1 ν N (t) λt ν . νN (t) λt ν . 12 SERGEI SERGEEV AND HANS SCHNEIDER ∗ ∗ ∗ 0 0 0 ∗ ∗ ∗ 0 0 0 ∗ ∗ ∗ 0 0 0 ∗ D1 ∗ 0 0 0 S1 ∗ ∗ 0 0 0 C1 S ◦ 1 ∗ ∗ 0 0 0 C ◦ 1 ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ R1 R◦ 1 ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ 0 0 0 0 0 0 0 0 0 0 0 0 0 S2 ∗ 0 0 0 C2 0 0 0 0 0 0 0 ∗ ∗ 0 0 0 0 0 0 0 0 0 R2 R◦ 2 0 ∗ ∗ ∗ ∗ ∗ 0 0 0 ∗ ∗ ∗ 0 ∗ ∗ ∗ ∗ ∗ 0 0 0 ∗ D2 ∗ 0 ∗ ∗ ∗ ∗ ∗ 0 0 0 S ◦ 2 ∗ ∗ C ◦ 2 Figure 3. Formation of first two terms in a Nachtigall expansion (upper part) and ultimate expansion (lower part): a schematic example. The associated digraph consists of two components, de- noted D1 and D2. 5. Ultimate expansion In this section we construct a different expansion of At which we call the ultimate expansion, in order to describe the ultimate behavior of At. This expansion is related to the Nachtigall expansion of Section 4 with the selection rule Cµ := C(Aµ). The latter expansion will be called the canonical Nachtigall expansion. The elements of the ultimate expansion will be labeled by ◦, since we need to distinguish them from those of the canonical Nachtigall expansion. For instance, we will write A◦ µ versus Aµ, C ◦ µ versus Cµ and λ◦ 1 = λ(A), let C ◦ µ versus λµ, etc. 1 = (N ◦ 1 , E◦ + . Define λ◦ 1 ) be the critical graph 1 the set of nodes in all components of D(A) that contain Let A = (aij ) ∈ Rn×n of A and denote by M ◦ the components of C ◦ 1 . Set A◦ 1 := A and K ◦ 1 := N . µ := N \ ∪µ−1 By induction for µ ≥ 2, define K ◦ i=1 M ◦ in the case of a Nachtigall expansion), and define A◦ i (instead of Kµ = N \ ∪µ−1 µ = (aµ◦ ij ) ∈ Rn×n by + i=1 Ni (5.1) aµ◦ ij =(aij, 0, if i, j ∈ K ◦ µ, otherwise. CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 13 Define λ◦ µ := λ(A◦ µ = (N ◦ µ, E◦ By the above procedure we define K ◦ µ). If λ◦ = 0 then stop, otherwise let C ◦ critical graph of Aµ, let M ◦ contain the components of C ◦ µ) be the µ be the set of all nodes in the components of D(A) which µ, and proceed with the above definition for µ := µ + 1. µ for µ 6= 0. Note that µ, µ = 1, . . . , m◦} is the set of m.c.g.m. of all nontrivial components of D(A), µ consists of the critical digraphs of (possibly several) such components. µ be the cyclicity of C ◦ µ be defined from µ in full analogy with Bµ, Cµ, Sµ and Rµ in Sect. 4. The matrices S ◦ µ µ = (N ◦ µ = 1, . . . , m◦, where m◦ ≤ n is the last number µ such that λ◦ {λ◦ and each C ◦ (A◦ can be again simultaneously scaled to 0 − 1 form. Let γ ◦ µ)γ ◦ µ/λ◦ µ, and let B◦ µ, E◦ µ) and M ◦ µ, C ◦ µ, S ◦ µ and R◦ µ, A◦ µ, λ◦ µ, C ◦ Essentially in the new construction we contract by the components of D(A) in- stead of the components of C(Aµ) in the case of the canonical Nachtigall expansion. See Figure 3 for a visual comparison. Denote by λ(i) the m.c.g.m. of the component of D(A) to which i belongs, and let λ(i) = 0 if {i} is a trivial component of D(A). For a path P define λ(P ) := maxi∈NP λ(i) where NP is the set of nodes visited by P . Recall that Πij,t denotes the set of paths on D(A) which connect i to j and have length t. Denote by Πµ◦ ij,t the set of paths P ∈ Πij,t such that P contains a node in N ◦ µ and λ(P ) = λµ. Such paths will be called µ-hard. Note that they are C(A◦ µ)-heavy with respect to A◦ µ. Denote (5.2) U (t) µ := C ◦ µ(S ◦ µ)tR◦ µ. These CSR products are defined from A◦ µ and C ◦ defined in Sect. 3 from A and C, and it follows that the sequence {U (t) µ the greatest cyclicity of a component in C ◦ µ. Denote by τ ◦ the period γ ◦ the number of nodes in K ◦ µ. The next result follows from Theorem 3.3. µ in the same way as P (t) were µ , t ≥ 0} has µ and by n◦ µ Theorem 5.1 (µ-Hard paths). Let T ◦ µ be such that {(S ◦ µ)t, t ≥ T ◦ µ } is periodic. 1. For t ≥ 0, w(Πµ◦ ij,t) ≤ (λ◦ µ)t(U t µ)ij . 2. For t ≥ T ◦ µ + 2τ ◦ µ(n◦ µ − 1) , w(Πµ◦ ij,t) ≥ (λ◦ µ)t(U t µ)ij . (5.3) (5.4) Comparing the constructions above with those of the canonical Nachtigall ex- pansion (see Sect. 4 assuming that Cµ := C(Aµ)), we see that C1 is the same as C ◦ 1 and λ1 is the same as λ◦ 1, however, other components and values may be not the same. We next describe relation between them. Proposition 5.2. Each ν = 1, . . . , m◦ corresponds to a unique µ = 1, . . . , m such that λ◦ ν = λµ and all components of C ◦ ν are also components of Cµ. Proof. Consider the canonical Nachtigall expansion. Note that λµ strictly decrease, as at each step of the definition we remove the whole critical digraph. Now pick ar- bitrary λ◦ ν , which is the m.c.g.m. of some component of D(A). There is a reduction step when Cµ for the first time intersects with a component of D(A) whose m.c.g.m. is λ◦ ν , precisely as they are. This proves the claim. (cid:3) ν , and Cµ has to contain all components of C ◦ ν . Then λµ = λ◦ 14 SERGEI SERGEEV AND HANS SCHNEIDER Further we renumber λ◦ ν = λµ, meaning that the numbering ν is adjusted to that of λµ. This defines a subset Σ of {1, . . . , m}, such that ν so that ν = µ if λ◦ of λ◦ λµ = λ◦ µ if and only if µ ∈ Σ. Corollary 5.3. γµ is a multiple of γ ◦ µ for each µ ∈ Σ. Proof. With the new numbering, all components of C ◦ by Proposition 5.2. µ are also components of Cµ (cid:3) Unlike µ-heavy paths, µ-hard paths do not cover the whole path sets Πij,t in general. However evidently µ)t (5.5) (A◦ ij = w(Πµ◦ ij,t), if i ∈ N ◦ µ or j ∈ N ◦ µ. From this and Theorem 5.1 we deduce the following. Proposition 5.4. Let A ∈ Rn×n ij = λt + . For all t ≥ 3n2 if i ∈ N ◦ µU (t) µ , (5.6) (A◦ µ)t µ or j ∈ N ◦ µ. For the sequel we need to establish some relation between connectivity on D(A) and nonzero entries of U (t) µ . We denote by γ ◦ the l.c.m. of all cyclicities γ ◦ µ. Recall that we denote by NP the set of nodes visited by a path P and by λ(P ) the greatest m.c.g.m. of a component visited by P . µ of all components C ◦ Proposition 5.5. Let i, j ∈ N, l ≥ 0 and µ ∈ Σ. The following are equivalent. µ )ij 6= 0; 1. (U (l) 2. For all t ≡ l(mod γ ◦) such that t ≥ 3n2, there is a µ-hard path of length t connecting i to j; 3. For some t ≡ l(mod γ ◦) there exists a µ-hard path of length t connecting i to j; 4. For some t ≡ l(mod γ ◦) there exists a path P of length t such that λ(P ) = λµ. Proof. Implications 1.⇔2. and 3⇔1. follow from Theorem 5.1 and the periodicity of U (t) µ . Implications 2.⇒3.⇒4. are evident. It remains to prove 4.⇒3. Let k be a node in NP which belongs to M ◦ µ (that is, a critical node) in the same component of D(A) as k. There exists a cycle containing both k and l. Adjoining γ ◦ copies of this cycle to P we obtain a µ-hard path, whose length is congruent to l(mod γ ◦). (cid:3) µ, and let l be a node in N ◦ Now we establish the ultimate expansion of matrix powers, as an ultimate form of the canonical Nachtigall expansion. We will write a(t) T= b(t) if a(t) = b(t) for all t ≥ t′ where t′ is an unknown integer, and analogously for inequalities. Theorem 5.6 (Ultimate expansion). Let A ∈ Rn×n + . For all µ ∈ Σ (5.7) In particular, (5.8) ν U (t) λt ν . (A◦ µ)t T= Mν∈Σ : ν≥µ At T=Mν∈Σ νU (t) λt ν . CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 15 µ ≤ N (t) µN (t) µ for all µ ∈ Σ, since any µ by the canonical Nachtigall Proof. It suffices to prove (5.8). First note that U (t) expansion, it suffices to prove that µ-hard path is a µ-heavy path. As At T= Lµ λt ≤ Mµ∈Σ Mµ=1 λt µN (t) µ (5.9) m T For all µ, i and j such that (N (t) (U (t) µ )ij = (N (t) Assume that either µ /∈ Σ, or µ ∈ Σ but (N (t) λt µU (t) µ µ )ij 6= 0, we will show that either µ ∈ Σ and µ )ij , or there exists ν ∈ Σ such that λν > λµ and (U (t) ν )ij 6= 0. µ )ij > (U (t) µ(N (t) µ )ij . Theorem 4.1 implies that for all l such that l ≥ 3n2 and l ≡ t(mod γµ) there exist paths P ∈ Πµ ij,l such that w(P ) = λl µ )ij. We are going to show that these paths are not µ-hard. If µ /∈ Σ then this is immediate. If µ ∈ Σ then by Corollary 5.3 γµ is a multiple of γ ◦ µ, and hence l ≡ t(mod γ ◦ µ )ij by Theorem 5.1, which implies (U (t) µ )ij contradicting our assumptions. Hence P are not µ- hard, meaning that for any such path there exists ν ∈ Σ such that λν = λ(P ) > λµ. Applying Proposition 5.5, we obtain that (U (t) ν )ij 6= 0 with λν > λµ. The claim is proved. (cid:3) µ). If P is µ-hard, then w(P ) ≤ λl µ )ij ≥ (N (t) µ(U (t) If A is irreducible, then the ultimate expansion has only one term, which cor- responds to its critical graph C(A). In general, it has several terms (up to n) corresponding to the critical graphs of the components of D(A) (or possibly clus- ters of critical graphs of the components with the same m.c.g.m.) Thus, Theorem 5.6 can be regarded as a generalization of the Cyclicity Theorem, see [17] Theorem 3.9 or [1] Theorem 3.108, which it implies as a special irreducible case. 6. Computational complexity Given A ∈ Rn×n + , we investigate the computational complexity of the following of a Nachtigall expansion with a pre- problems. (P1) For given t, reconstruct all terms N (t) µ scribed selection rule for Cµ. (P2) For given t: 0 ≤ t < γ, reconstruct all terms U (t) µ of the ultimate expansion. In (P1) we assume that selecting the subdigraph Cµ of C(Aµ) does not take more than O(n3) operations. This holds in particular if Cµ is an arbitrary cycle of C(Aµ) as in [19, 21]. Problem (P1) is close to the problem considered by Moln´arov´a [19], and Problem (P2) is extension of a problem regarded by Sergeev [25]. An O(n4 log n) solution of these problems is given below. It is based on visualisation, square multiplication and permutation of cyclic classes. See Semanc´ıkov´a [22, 23] for closely related studies in max-min algebra. Theorem 6.1. For any A ∈ Rn×n O(n4 log n) operations. + , problems (P1) and (P2) can be solved in Proof. (P1): First we need to compute λµ, Aµ and Cµ for all µ. At each step the computation requires no more than O(n3) operations, based on Karp and Floyd- Warshall methods applied to each component of D(Aµ). The total complexity is no more than O(n4). After this, we find all cyclic classes in each Cµ, which has total 16 SERGEI SERGEEV AND HANS SCHNEIDER complexity O(n2), and hence the cyclicities γσ of all components of the graphs Cµ. At this stage we can also find a scaling which leads to a total S-visualization of A (and hence all Aµ). This relies on Floyd-Warshall method applied to each Sµ and takes no more than O(n3) operations in total. ν µ = CµSt By Theorem 4.2, At µ admit Nachtigall expansion for all µ and all t ≥ 3n2. The µ with indices in Nµ are determined at t ≥ 3n2 only by rows and columns of At N t µRµ, since by construction these rows and columns are zero in all terms N (t) for ν > µ. This means in particular that these rows and columns become periodic after 3n2 time. By repeated squaring Aµ, A2 µ, . . . , we reach a power µ with r ≥ 3n2, which requires no more than O(n3 log n) operations. Now we Ar can use Corollary 3.7 identifying CµSr µRµ as submatrices extracted from columns, resp. rows, of At µRµ from Sr µ by the permutation on cyclic classes determined by the remainders r(mod γσ) and (t − r)(mod γσ), for each cyclicity γσ of a component of Cµ. This takes O(n2) overrides. Finally we compute N t µRµ (O(n3) operations). We conclude that the total complexity for all µ does not exceed O(n4 log n) operations. µ with indices in Nµ. By Theorem 3.6 we can obtain St µRµ and Cµ from CµSr µ = CµSt µ and Sr µ, A4 (P2): It is clear that the computation of all prerequisites for the ultimate ex- pansion is done like in the first para of the proof of (P1), and takes no more than O(n4) operations. After that we use Proposition 5.4 which means that the critical rows and columns in each (A◦ µ . Hence the factors of each U (t) µ, followed by µ a permutation on cyclic classes and matrix multiplication, which overall takes no more than O(n3 log n) operations. We conclude that the total complexity for all µ does not exceed O(n4 log n) operations. (cid:3) µ)r for r ≥ 3n2 are determined only by U (r) can be computed by matrix squaring of A◦ 7. Orbit periodic matrices Being motivated by the results of Butkovic et al. [6] on robust matrices, we are going to derive necessary and sufficient conditions for orbit periodicity and to show that they can be verified in polynomial time. A matrix A ∈ Rn×n is called orbit periodic if for each y ∈ Rn + there exists λ(y) ∈ Rn Aty for all sufficiently large t, where (as above)γ ◦ is the joint cyclicity (l.c.m.) of the critical graphs of all components of D(A). + such that At+γ ◦ y = (λ(y))γ ◦ + A sequence {Aty, t ≥ 0} with the above property will be called ultimately linear periodic and λ(y) will be called its ultimate growth rate. The same wording will be used for the sequences {at , s ≥ 0} has ultimate growth rate λ, if there exists αij 6= 0 such that al+γ ◦s = αijλl+γ ◦s for all s starting from a sufficiently large number. ij, t ≥ 0. We say that a subsequence {al+γ ◦s ij ij It may seem more general if in the above definition of linear periodicity we replace γ ◦ by γ(y). But using the ultimate expansion (5.8) we conclude that {Aty, t ≥ 0} is ultimately linear periodic if and only if there exists µ ∈ Σ such that Aty T= λt µ y. As U (t+γ ◦) = U (t) µ for all µ and t, we conclude that the exact period of {Aty, t ≥ 0} has to divide γ ◦. µU (t) µ The ultimate expansion leads to the following properties of the sequences {at ij, t ≥ 0}, already known in max algebra [9, 16, 20]. As above, λ(P ) is the largest m.c.g.m. CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 17 of the components of D(A) visited by P , and Πij,t denotes the set of paths of length t connecting i to j. Lemma 7.1. For each l : 0 ≤ l < γ ◦, the subsequence {al+γ ◦s , s ≥ 0} is ultimately zero or has an ultimate growth rate λij (l). In the latter case, for each such l and each sufficiently large t ≡ l(mod γ ◦) there exists P ∈ Πij,t such that λ(P ) = λij (l). ij ij µ (U (l) = λl+γ ◦s Proof. The ultimate expansion (5.8) implies that for each l there exists µ ∈ Σ such that al+γ ◦s µ )ij at sufficiently large s, so the subsequence has growth rate λij (l) := λµ. The second part of the statement follows from Proposition 5.5. (cid:3) Lemma 7.2. If P ∈ Πij,l then {al+γ ◦s λ(P ). , s ≥ 0} has ultimate growth rate at least ij Proof. Using Proposition 5.5 cond. 4, we obtain that (U (l) λµ = λ(P ), and then al+γ ◦s ≥ (λ(P ))l+γ ◦s(U (l) µ )ij 6= 0 for µ such that (cid:3) ij µ )ij at sufficiently large s. Denote by L◦ the set of nodes in the nontrivial components of D(A). The next statement follows from the Cyclicity Theorem [1, 17]. For the sake of completeness we deduce it from the ultimate expansion. Lemma 7.3. If i, j ∈ L◦ belong to the same component of D(A), then {at ij, t ≥ 0} is ultimately linear periodic and its growth rate is the m.c.g.m. of that component. Further, for each i ∈ L◦ and all t there exist k and l in the same component of D(A) such that at ik 6= 0 and at li 6= 0. Proof. All paths which connect i to j are µ-hard, with λµ = λ(i) = λ(j). Using Theorem 5.1 we obtain that only (U (t) µ )ij is nonzero and hence the ultimate expan- sion reads at µ )ij . For the second part note that i belongs to a cycle with ij nonzero weight. (cid:3) µ(U (t) T= λt If i is connected to j by a path, we denote this by i → j. Observe that if i → j then also k → l for each k in the same component of D(A) with i and for each l in the same component of D(A) with j. We also denote i ↔ j if both i → j and j → i (i.e., if i and j are in the same component of D(A)). In the next theorem we describe, in terms of such relations, when the sequences of columns {Atei, t ≥ 0} are ultimately linear periodic. Proposition 7.4. Let A ∈ Rn×n ultimately linear periodic if and only if for all i ∈ L◦, i → j implies λ(i) ≤ λ(j). and j ∈ L◦. The sequence {Atej, t ≥ 0} is + kj 6= 0, and the sequence {at+sγ ◦ Proof. The "only if" part: Lemma 7.3 implies that for each t there exists k such that at , s ≥ 0} has ultimate growth rate λ(j). If the condition does not hold, there exists a path P leading from i to j such that λ(P ) = λ(i) > λ(j), and by Lemma 7.2 there is a subsequence of {al ij, l ≥ 0} with ultimate growth rate at least λ(i). kj The "if" part: If the sequence {Atej, t ≥ 0} is not ultimately linear periodic, then some of its entries by Lemma 7.3 have ultimate growth rate λ(j) and there is a subsequence of {at kj, t ≥ 0}, for some k ∈ N , which has a different ultimate growth rate. Lemma 7.1 implies that this growth rate has to be greater than λ(j), and must be the m.c.g.m. of a component which has access to j. (cid:3) 18 SERGEI SERGEEV AND HANS SCHNEIDER We denote i ⇒ j and say that i strongly accesses j, if i can be connected to j by a path of any length starting from a certain number Tij. For example, i strongly accesses j if i → j and the cyclicities of the components of D(A) containing i and j are coprime. Observe that if i ⇒ j then also k ⇒ l for each k ↔ i and l ↔ j. Now we show that strong access relations are essential for the ultimate linear periodicity of all sequences {At(ei ⊕ ej), t ≥ 0}. Proposition 7.5. Let A ∈ Rn×n and i, j ∈ L◦. Suppose that {Atei, t ≥ 0} and {Atej, t ≥ 0} are ultimately linear periodic. Then {At(ek ⊕ el), t ≥ 0} is also ultimately linear periodic for all k ↔ i and l ↔ j, if and only if either i ⇒ j, or j ⇒ i, or both are false but λ(i) = λ(j). + Proof. If λ(i) = λ(j) then both {Atek, t ≥ 0} and {Atel, t ≥ 0} for all k ↔ i and l ↔ j have this growth rate and {At(ek ⊕ el), t ≥ 0} is ultimately linear periodic with this growth rate. So it remains to consider the case λ(i) < λ(j). In this case Proposition 7.4 implies j 6→ i, and therefore we have to show that {At(ek ⊕ el), t ≥ 0} are ultimately linear periodic for all k ↔ i and l ↔ j if and only if i ⇒ j. mk The "if" part: For each t : 0 ≤ t < γ ◦, if there exists m ∈ N and s1 ≥ 0 such 6= 0, then there exists a path P ∈ Πmk,t+s1γ ◦ . As k ⇒ l, this path can 6= 0. 6= 0 for all sufficiently large s and it dominates that at+s1γ ◦ be joined with a path from k to l of length s2γ ◦, and we get that at+(s1+s2)γ ◦ Using Lemma 7.2 we obtain at+sγ ◦ over at+sγ ◦ since it has larger growth rate. ml ml mk The "only if" part: The ultimate linear periodicity of {At(ek ⊕ ej), t ≥ 0}, for any k ↔ i, implies that supp(Atek) ⊆ supp(Atej) for all large enough t. By Lemma 7.3 there is k ↔ i such that (At)ik 6= 0, hence also (At)ij 6= 0. As we reasoned for any t, it follows that i ⇒ j. (cid:3) Theorem 7.6 (Orbit periodicity). A ∈ Rn×n following conditions hold for all i, j ∈ L◦: + is orbit periodic if and only if the 1. i → j implies λ(i) ≤ λ(j), 2. if neither j ⇒ i nor i ⇒ j then λ(i) = λ(j), or equivalently if {At(ei ⊕ ej), t ≥ 0} are ultimately linear periodic for all i, j ∈ L◦. Proof. We need only prove that 1. and 2. are sufficient for orbit periodicity, the rest relies on Propositions 7.4 and 7.5. Let D(A, y) be the subgraph induced by the set of nodes that have access to supp(y) := {i : yi 6= 0}. Nontrivial components Dσ of D(A, y) are ordered by relation Dσ1 (cid:22) Dσ2 if i ⇒ j for some (and hence all) i ∈ Dσ1 and j ∈ Dσ2 . Consider the maximal components with respect to this relation, by conditions 1. and 2. they must have the same m.c.g.m. and it must be the greatest one. We denote this m.c.g.m. by λ and show that it is the ultimate growth rate of Aty. By Lemma 7.1 for each l : 0 ≤ l < γ, the subsequence {al+sγ ◦ , s ≥ 0} has a certain ultimate growth rate if it is not ultimately zero. We have to show that λ is the maximal growth rate of {al+sγ ◦ , s ≥ 0} over k ∈ supp(y), for every fixed i ∈ N and l : 0 ≤ l < γ ◦. Then it follows that At+γ ◦ y T= λγ ◦ Aty. ik ij To avoid trivialities we assume that there exists k ∈ supp(y) and such that {al+sγ ◦ , s ≥ 0} is not ultimately zero. Then for some k ∈ supp(y) there exists a path P ∈ Πik,t where t ≡ l(mod γ ◦) which visits a nontrivial component Dσ1 ik CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 19 ik If the m.c.g.m. of Dσ1 is λ, then by Lemma 7.2 the growth rate of of D(A, y). {al+sγ ◦ , s ≥ 0} is not less than λ, hence it must be λ and we are done. If the m.c.g.m. of Dσ1 is less than λ, then Dσ1 strongly accesses some component Dσ2 with m.c.g.m. λ, and Dσ2 accesses a node k′ ∈ supp(y). Due to the strong access we can adjust the length of the path from Dσ1 to Dσ2 if necessary, and we obtain a path P ′ ∈ Πik′,t′ where t′ ≡ l(mod γ ◦). By Lemma 7.2 we obtain that the growth rate of {al+sγ (cid:3) , s ≥ 0} is not less than λ, hence it must be λ. ik′ To assess the computational feasibility of condition 2. in Theorem 7.6 we need the following observation which uses the ultimate expansion, see Theorem 5.6 and Section 5. Theorem 7.7. Let A ∈ Rn×n be such that the sequences {Atei, t ≥ 0} are ulti- mately linear periodic for all i ∈ L◦. Then A is orbit periodic if and only if the following holds for all µ, ν ∈ Σ: + (7.1) λµ < λν ⇒ [i∈N ◦ µ supp(U (1) µ ei) ⊆ \j∈N ◦ ν supp(U (1) ν ej). T= Proof. If {Atei, t ≥ 0} is ultimately linear periodic and i ∈ L◦, then Atei µU (t) µ ei for λµ = λ(i), since λµ must be the growth rate of Atei and all other terms λt of the ultimate expansion have different growth rates. Then also At(ei ⊕ ej) T= µU (t) λt µ and ν ). If λµ < λν then {At(ei ⊕ ej), t ≥ 0} is ultimately linear periodic if and j ∈ M ◦ only if At(ei ⊕ ej) T= λt (7.2) ν ej where λµ = λ(i) and λν = λ(j) (equivalently, i ∈ M ◦ ν ej. This happens if and only if λµ < λν ⇒ supp(U (t) µ ei) ⊆ supp(U (t) ν ej) ∀i ∈ M ◦ µ, j ∈ M ◦ ν , µ ei ⊕ λt ν U (t) ν U (t) holds for all µ, ν ∈ Σ. Corollary 3.8 implies that any column of U (t) µ , resp. U (t) is a max-linear combi- ν ν , so that the support of that column ν . Hence ν . Further, ν ) with indices in N ◦ µ ν ) just permute as t changes, so the inclusions need be verified only for nation of columns with indices in N ◦ is the union of supports of certain columns with indices in N ◦ we need to check the support inclusions only for i ∈ N ◦ Corollary 3.7 implies that the columns of U (t) µ (or resp. N ◦ t = 1. This shows that (7.2) is equivalent to µ and j ∈ N ◦ (or resp. U (t) µ, resp. N ◦ µ, resp. N ◦ (7.3) µ ei) ⊆ supp(U (1) which is equivalent to (7.1). The claim is proved. λµ < λν ⇒ supp(U (1) ν ej) ∀i ∈ N ◦ µ, j ∈ N ◦ ν , (cid:3) Now we give a polynomial bound on the computational complexity of verifying the orbit periodicity of a reducible matrix. Theorem 7.8. Let A ∈ Rn×n + . Suppose that all components of D(A) and access relations between them are known, and λµ and U (1) are computed for all µ ∈ Σ. µ Then the orbit periodicity of A can be verified in no more than O(n3) operations. Proof. For all pairs µ, ν, we must verify condition 1. of Theorem 7.6 and condition (7.1). The first of these conditions is verified for the pairs of components of D(A) and it takes no more than O(n2), if all access relations between them are known. µ ei) and To verify condition (7.1), we need to compute all unions Si∈N ◦ supp(U (1) µ 20 SERGEI SERGEEV AND HANS SCHNEIDER intersections Ti∈N ◦ µ ei), which requires O(n2) operations, and then make O(n2) comparisons of Boolean vectors, which requires no more than O(n3) opera- tions. (cid:3) µ supp(U (1) We combine the results of Theorems 6.1 and 7.8. Corollary 7.9. Given A ∈ Rn×n verify whether it is orbit periodic or not. + , it takes no more than O(n4 log n) operations to 8. Examples All examples in this section will be in the max-plus setting Rmax,+ := (R ∪ {−∞}, ⊕ = max, ⊗ = +). Example 1. We construct the canonical Nachtigall expansion of At for We start with A1 = A. The maximal cycle mean is λ1 = 0, and the component C1 has two nodes 1, 2 and two edges (1, 2) and (2, 1). The cyclicity is γ1 := 2. We proceed by setting the entries in first two rows and columns to −∞, thus 0 −7 −6 −1 0 −1 −5 −4 −7 −5 −1 −3 −6 −4 −3 −2 A =  .   A2 =  A3 =  −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −1 −3 −∞ −∞ −3 −2 −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −2 . .     (8.1) obtaining (8.2) (8.3) C3: (8.4) The maximum cycle mean is λ2 = −1, and the component C2 has one node 3 and one edge (3, 3). The cyclicity is γ2 = 1. Now we set everything to −∞ except for the entry (4, 4): The maximum cycle mean is λ3 = −2, and the component C3 has one node 4 and one edge (4, 4). The cyclicity is γ3 = 1. We obtain matrices S1, S2 and S3 which correspond, respectively, to C1, C2 and , S2 =  −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ 0 −∞ −∞ −∞ −∞ −∞ ,   S1 =  S3 =  0 −∞ 0 −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ 0     . These are Boolean matrices in the max-plus setting, with entries ∞, 0 instead of 0, 1. CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 21 Further we need to compute Kleene stars (Aγ1 )∗, ((A2 − λ2)γ2)∗ = (1 + A2)∗ and ((A3 − λ3)γ3 )∗ = (2 + A3)∗, and construct matrices C1, R1, C2, R2, C3 and R3. The critical parts of the Kleene stars are shown below, the rest of the elements being denoted by · as we do not need them: (8.5) (2 ⊗ A3)∗ =  · · · · · · · · · −∞ −∞ −∞ 0 −∞ −∞ −∞ .   ,   0 0 −1 −5 −4 −1 −6 −5 −5 −6 −10 −9 −4 −5 −9 −8 ,   0 −1 −6 −5 0 −1 −5 −4 −6 −5 −11 −10 −5 −4 −10 −9 −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ 0 −2 −∞ −∞ −2 −4 −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ 0 , .     = N (0) 1 ⊕ (−2) ⊗ N (0) 2 ⊕ (−4) ⊗ N (0) 3 = N (1) 1 ⊕ (−3) ⊗ N (0) 2 ⊕ (−6) ⊗ N (0) 3 , . The Nachtigall expansion starts to work already at t = 2. Indeed, Further we compute the Nachtigall matrices (A2)∗ =  (1 ⊗ A2)∗ =  0 −1 −5 −4 0 −6 −5 −1 · −5 −6 −4 −5 · · · ,   , · · · −∞ −∞ 0 −2 · · · −∞ · −∞ · −2 ·   1 = C1 ⊗ R1 =  N (0) N (1) 1 = C1 ⊗ S1 ⊗ R1 =  (8.6) (8.7) (8.8) N (0) N (0) 2 = C2 ⊗ R2 =  3 = C3 ⊗ R3 =      −1 0 −6 −5 0 −1 −5 −4 −6 −5 −3 −5 −5 −4 −5 −6 0 −1 −5 −4 −1 0 −6 −5 −5 −6 −2 −4 −4 −5 −4 −4 A2 =  A3 =  A4 =  Starting from t = 4 the third term can be forgotten: 0 −1 −5 −4 −1 0 −6 −5 −5 −6 −4 −6 −4 −5 −6 −8   = N (0) 1 ⊕ (−4) ⊗ N (0) 2 . 22 SERGEI SERGEEV AND HANS SCHNEIDER The ultimate periodic behavior starts after T (A) = 10: (8.9) = N (0) 1 = U (0) 1 . A10 =  0 0 −1 −5 −4 −1 −6 −5 −5 −6 −10 −9 −4 −5 −9 −8   Example 2. The following example will illustrate the ultimate expansion: (8.10) A = −2 0 −3 −7 −∞ −∞ −∞ 0 −2 −5 −7 −∞ −∞ −∞ −9 −7 −9 −8 −∞ −∞ −∞ −9 −6 −4 −4 −∞ −∞ −∞ −8 −5 −5 −4 −1 −7 −5 −7 −8 −5 −6 −3 −6 −8 −6 −4 −9 −3 −5 −5 −5   .   We compute the elements of the ultimate expansion. Firstly, A◦ 1 = A, λ1 = 0, and the component C ◦ 1 = C(A) consists of two nodes 1, 2 and two edges (1, 2) and (2, 1). On the next step we set all entries in the first four rows and columns of A to −∞, 2. Its essential submatrix with finite entries is extracted thus obtaining matrix A◦ from the remaining rows and columns 5 to 7: (8.11) 5 to 7 × 5 to 7. A◦ 2 ess =  −1 −7 −5 −3 −6 −8 −5 −5 −5  , We compute λ2 = −1, and the component C2 = C(A2) consists of the loop (5, 5). The components C1 and C2 determine Boolean (i.e., 0, −∞) matrices S1 and S2. We compute (A2)∗ and (1 ⊗ A◦ 2)∗: (8.12) (A2)∗ =   0 −2 −5 −7 −∞ −∞ −∞ 0 −3 −7 −∞ −∞ −∞ −2 −7 −9 0 −12 −∞ −∞ −∞ −∞ −∞ −∞ −6 −8 −8 0 −8 −6 −5 −6 −6 −5 0 −8 −8 −7 −8 −7 −4 0 −4 −6 −7 −7 −6 −10 0 ,   2 ess)∗ =  0 −6 −4 −2 0 −6 −4 −4 1 and C ◦ 0   , 2 , R◦ 1 , R◦ (8.13) (1 ⊗ A◦ 5 to 7 × 5 to 7. Next we build matrices C ◦ 2 whose essential parts are shown below: (8.14) (8.15) C ◦ R◦ −2 0 −9 −8 −6 −7 −6(cid:19)T 1 ess =(cid:18) 0 −2 −7 −6 −5 −8 −4 1 ess =(cid:18) 0 −2 −5 −7 0 −3 −7(cid:19) , 1 to 2× 1 to 4. −2 C ◦ R◦ 2 ess =(cid:0)0 −2 −4(cid:1)T 2 ess =(cid:0)0 −6 −4(cid:1) , , 5 to 7× 5 5 × 5 to 7. , 1 to 7× 1 to 2, CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 23 Using (8.14) and (8.15) we compute the ultimate terms U (0) C ◦ 1 ⊗ S ◦ meaning 1 = 2. The ultimate expansion starts to work at t = 9, 1 and U (0) 1 = C ◦ 2 = C ◦ 1, U (1) 1 ⊗ R◦ 2 ⊗ R◦ 1 ⊗ R◦ (8.16) A9 = U (1) A11 = U (1) 1 ⊕ (−9) ⊗ U (0) 1 ⊕ (−11) ⊗ U (0) 2 , A10 = U (0) 2 , A12 = U (0) 1 ⊕ (−10) ⊗ U (0) 2 , 1 ⊕ (−12) ⊗ U (0) 2 , . . . In this case, the canonical Nachtigall expansion starts to work already at t = 3, and after t = 4 only the first two terms are essential. In this expansion, matrix A2 results from setting only the first two (instead of four) columns and rows to −∞. The second Nachtigall term N2 is equal to C2 ⊗ R2, where C2 = C ◦ 2 and R2 6= R◦ 2 has essential part (8.17) Example 3. We illustrate the total periodicity. Let R2 ess =(cid:0)−4 −3 0 −6 −4(cid:1) , 5 × 3 to 7. (8.18) A = (8.19) B =     −∞ 1 −∞ −∞ 1 −∞ −∞ −∞ 1 −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ 1 −∞ −2 −∞ −∞ −∞ 0 −∞ −2 −∞ −∞ 0 −∞ −∞ 1 −∞ −∞ 1 −∞ −∞ −∞ 1 −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ 1 −∞ −2 −∞ −∞ −∞ 0 −∞ −∞ −2 −∞ 0 −∞     Note that A and B are almost the same, except for the entries (6, 2) and (6, 3). In both cases the ultimate expansion coincides with the canonical Nachtigall expan- sion, and we have two critical components C1 = (N1, E1) with N1 = {1, 2, 3, 4}, E1 = {(1, 2), (2, 3), (3, 4), (4, 1)} and C2 = (N2, E2) with N2 = {5, 6} and E2 = {(5, 6), (6, 5)}. The eigenvalues are λ1 = 1 and λ2 = 0, and the cyclicities are γ1 = 4 and γ2 = 2, their l.c.m. is γ = 4. Notr that M ◦ 1 = N1 and M ◦ 2 = N2. In both cases the component with λ1 does not have access to the component with λ2 which is smaller, hence condition 1. of Theorem 7.6 is true meaning that all columns of At and Bt are ultimately periodic. Condition 2. of Theorem 7.6 holds for A but it does not hold for B. In particular, there are only paths of odd length connecting node 6 to node 3. Hence A is orbit periodic and B is not. Consider also condition (7.1). We need the terms of the ultimate expansion for 1 and U A 2 , t(mod γ ◦) = 1. In each case there are two terms, which we denote by U A resp. U B 2 , for the case of A, resp. B. We have 1 and U B (8.20) U A 1 = −∞ 0 −∞ −∞ 0 −∞ −∞ −∞ 0 −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ 0 −4 −3 −6 −5 −∞ −∞ −4 −3 −6 −5 −∞ −∞     24 SERGEI SERGEEV AND HANS SCHNEIDER (8.21) U B 2 =   −∞ 0 −∞ −∞ 0 −∞ −∞ −∞ 0 −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −∞ −3 −∞ −5 −∞ −∞ −4 −∞ −3 −∞ −∞ −∞ 0   (8.22) U A 2 ess = U B 2 ess =(cid:18)−∞ 0 −∞(cid:19) , 0 5 to 6 × 5 to 6. Observe that supp(U A {1, 2, 3, 4}, but this condition does not hold for B. 2 ei) ⊆ supp(U A 1 ej) for all i ∈ M ◦ 2 = {5, 6} and j ∈ M ◦ 1 = To see the difference between Atx and Btx, take x = [0 − ∞ − ∞ − ∞ − ∞ 0]T . The sequence {Atx} is ultimately periodic starting from t = 4 with period 4 and In particular the last component of {Atx} yields the following growth rate 1. number sequence for t ≥ 4: (8.23) (Atx)6 = {1 1 1 1 5 5 5 5 9 9 9 9 . . .}, t ≥ 4. The sequence {Btx} is not ultimately periodic. In particular, the last component of {Btx} yields the following number sequence for t ≥ 2: (8.24) (Btx)6 = {0 0 0 1 0 4 0 5 0 8 0 9 . . .}, t ≥ 2, which can be expressed as (8.25) (Btx)6 =  0, t − 3, t − 5, if t is even and t ≥ 2, if t = 4k + 3 and k ≥ 0, if t = 4k + 5 and k ≥ 0. 9. Acknowledgement We thank Peter Butkovic for his encouraging support, numerous discussions and careful reading of the paper. We acknowledge the work of the anonimous reviewer, who helped us to eliminate some mistakes and typos. We are also grateful to Abdulhadi Aminu, Trivikram Dokka and Glenn Merlet for their comments, help, and advice. References [1] F.L. Baccelli, G. Cohen, G.-J. Olsder, and J.-P. Quadrat, Synchronization and linearity: an algebra for discrete event systems, Wiley, 1992. [2] Y. Balcer and A.F. Veinott, Computing a graph's period quadratically by node condensation, Discrete Mathematics 4 (1973), 295 -- 303. [3] R.A. Brualdi and H.J. Ryser, Combinatorial matrix theory, Cambridge Univ. Press, 1991. [4] P. Butkovic, Max-algebra: the linear algebra of combinatorics?, Linear Algebra Appl. 367 (2003), 313 -- 335. , Max-linear systems: theory and algorithms, Springer, 2010. [5] [6] P. Butkovic, R. A. Cuninghame-Green, and S. Gaubert, Reducible spectral theory with appli- cations to the robustness of matrices in max-algebra, To be published in SIAM J. on Matrix Anal. and Appl., 2009. [7] J. Cochet-Terrasson, S. Gaubert, and J. Gunawardena, A constructive fixed-point theorem for min-max functions, Dynamics and Stability of Systems 14 (1999), no. 4, 407 -- 433. [8] R. A. Cuninghame-Green, Minimax algebra, Lecture Notes in Economics and Mathematical Systems, vol. 166, Springer, Berlin, 1979. CSR EXPANSIONS OF MATRIX POWERS IN MAX ALGEBRA 25 [9] B. de Schutter, On the ultimate behavior of the sequence of consecutive powers of a matrix in the max-plus algebra, Linear Algebra Appl. 307 (2000), 103 -- 117. [10] L. Elsner and P. van den Driessche, On the power method in max algebra, Linear Algebra Appl. 302-303 (1999), 17 -- 32. [11] , Modifying the power method in max algebra, Linear Algebra Appl. 332-334 (2001), 3 -- 13. [12] G.M. Engel and H. Schneider, Cyclic and diagonal products on a matrix, Linear Algebra Appl. 7 (1973), 301 -- 335. [13] M. Fiedler and V. Pt´ak, Diagonally dominant matrices, Czechoslovak Math. J. 17 (1967), no. 92, 420 -- 433. [14] G. Frobenius, Uber matrizen aus nicht negativen elementen, Sitzber. Preuss. Akad. Wiss. (1912), 456 -- 477. [15] M. Gavalec, Linear matrix period in max-plus algebra, Linear Algebra Appl. 307 (2000), 167 -- 182. [16] [17] B. Heidergott, G.-J. Olsder, and J. van der Woude, Max-plus at work, Princeton Univ. Press, , Periodicity in extremal algebra, Gaudeamus, Hradec Kr´alov´e, 2004. 2005. [18] K.H. Kim, Boolean matrix theory and applications, Marcel Dekker, New York, 1982. [19] M. Moln´arov´a, Computational complexity of nachtigall's representation, Optimization 52 (2003), 93 -- 104. [20] , Generalized matrix period in max-plus algebra, Linear Algebra Appl. 404 (2005), 345 -- 366. [21] K. Nachtigall, Powers of matrices over an extremal algebra with applications to periodic graphs, Mathematical Methods of Operations Research 46 (1997), 87 -- 102. [22] B. Semanc´ıkov´a, Orbits in max-min algebra, Linear Algebra Appl. 414 (2006), 38 -- 63. [23] , Orbits and critical components of matrices in max-min algebra, Linear Algebra Appl. 426 (2007), 415 -- 447. [24] B. Semanc´ıkov´a, Private communication, 2009. [25] S. Sergeev, Max algebraic powers of irreducible matrices in the periodic regime: An applica- tion of cyclic classes, Linear Algebra Appl. 431 (2009), 1325 -- 1339. [26] S. Sergeev, H. Schneider, and P. Butkovic, On visualization scaling, subeigenvectors and Kleene stars in max algebra, Linear Algebra Appl., 2008, accepted. E-print arXiv:0808.1992. [27] S. Schwartz, On a sharp estimation in the theory of binary relations on a finite set, Czechoslo- vak Math. J. 20 (1970), 703 -- 714. [28] H. Wielandt, Unzerlegbare nichtnegative matrizen, Math. Z. 52 (1950), 642 -- 645. University of Birmingham, School of Mathematics, Watson Building, Edgbaston B15 2TT, UK. E-mail address: [email protected] Department of Mathematics, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA. E-mail address: [email protected]
1208.5278	2	1208	2012-09-20T07:19:24	Low dimensional cohomology of Hom-Lie algebras and q-deformed W(2,2) algebra	[ "math.RA" ]	This paper aims to study the low dimensional cohomology of Hom-Lie algebras and q-deformed W(2,2) algebra. We show that the q-deformed W(2,2) algebra is a Hom-Lie algebra. Also, we establish a one-to-one correspondence between the equivalence classes of one dimensional central extensions of a Hom-Lie algebra and its second cohomology group, leading us to determine the second cohomology group of the q-deformed W(2,2) algebra. In addition, we generalize some results of derivations of finitely generated Lie algebras with values in graded modules to Hom-Lie algebras. As application we compute all $\a^k$-derivations and in particular the first cohomology group of the q-deformed W(2,2) algebra.	math.RA	math	Low dimensional cohomology of Hom-Lie algebras and q-deformed W (2, 2) algebra1 Lamei Yuan ‡, Hong You ‡ † ‡Academy of Fundamental and Interdisciplinary Sciences, Harbin Institute of Technology, Harbin 150080, China †Department of Mathematics, Suzhou University, Suzhou 200092, China E-mail: [email protected], [email protected] Abstract. This paper aims to study the low dimensional cohomology of Hom-Lie algebras and q-deformed W (2, 2) algebra. We show that the q-deformed W (2, 2) algebra is a Hom-Lie algebra. Also, we establish a one-to-one correspondence between the equivalence classes of one dimensional central extensions of a Hom-Lie algebra and its second cohomology group, leading us to determine the second cohomology group of the q-deformed W (2, 2) algebra. In addition, we generalize some results of derivations of finitely generated Lie algebras with values in graded modules to Hom-Lie algebras. As application we compute all αk-derivations and in particular the first cohomology group of the q-deformed W (2, 2) algebra. Key words: Hom-Lie algebras, q-deformed W (2, 2) algebra, derivation, second cohomology group, first cohomology group. Mathematics Subject Classification (2000): 17A30, 17A60, 17B68, 17B70. 1. Introduction The notion of Hom-Lie algebras was initially introduced in [3] motivated by examples of deformed Lie algebras coming from twisted discretizations of vector fields. In this paper we will follow the slightly more general definition of Hom-Lie algebras given by Makhlouf and Silvestrov in [6]. Precisely, a Hom-Lie algebra is a triple (L, [·, ·], α) consisting of a vector space L, a bilinear map [·, ·] : L × L → L and a linear map α : L → L such that [x, y] = −[y, x], (skew-symmetry) (cid:9)x,y,z [[x, y], α(z)] = 0, (Hom-Jacobi identity) for all x, y, z ∈ L, and where the symble (cid:9)x,y,z denotes summation over the cyclic permu- tation on x, y, z. One sees that the classical Lie algebras recover from Hom-Lie algebras if the twisting map α is the identity map. The Hom-Lie algebras were discussed intensively in [7, 8, 9, 10] while the graded cases were considered in [1, 5, 11]. But the cohomology with values in graded Hom-modules is not very clear. Therefore, one of the aims of the present 1Corresponding author: [email protected] 1 paper is to fill this gap. The W (2, 2) Lie algebra was introduced in [13] for the study of classification of vertex operator algebras generated by vectors of weight 2. It is an extension of the Virasoro algebra. In the following we denote by W the centerless W (2, 2) Lie algebra, which is an infinite dimensional Lie algebra generated by Ln and Mn (n ∈ Z) satisfying the following Lie brackets [Lm, Ln] = (n − m)Lm+n, [Lm, Mn] = (n − m)Mm+n, [Mm, Mn] = 0, for m, n ∈ Z. In [12] we presented a realization of the centerless W (2, 2) Lie algebra W by using bosonic and fermionic oscillators. The bosonic oscillator a and its hermitian conjugate a+ obey the commutation relations: [a, a+] = aa+ − a+a = 1, [1, a+] = [1, a] = 0. (1.1) It follows by induction on n that [a, (a+)n] = n(a+)n−1, for all n ∈ Z. The fermionic oscillators b and b+ satisfy the anticommutators {b, b+} = bb+ + b+b = 1, b2 = (b+)2 = 0. (1.2) Moreover, we set [a, b] = [a, b+] = [a+, b] = [a+, b+] = 0. Lemma 1.1 ([12]) With notations above. The generators of the form Ln ≡ (a+)n+1a, Mn ≡ (a+)n+1b+a, for all n ∈ Z, (1.3) realize the centerless W (2, 2) Lie algebra W under the commutator [A, B] = AB − BA, for all A, B ∈ W. Now fix a nonzero q ∈ C such that q is not a root of unity. We introduce the following notation and the q-number [A, B](α,β) = αAB − βBA, [n]q = qn − q−n q − q−1 . 2 It is clear to see that [−n]q = −[n]q. Furthermore, one can deduce that qn[m]q − qm[n]q = [m − n]q, q−n[m]q + qm[n]q = [m + n]q. (1.4) Then the generators Ln and Mn (n ∈ Z) satisfy the following q-brackets: [Ln, Lm](qn−m, qm−n) = [m − n]qLm+n, [Ln, Mm](qn−m, qm−n) = [m − n]qMm+n, [Mn, Mm](qn−m,qm−n) = 0, for all m, n ∈ Z. We call this algebra the q-deformed W (2, 2) algebra, which is the second object considered in this paper. In the following we will denote q-deformed W (2, 2) algebra by Wq and simply write the q-bracket as [·, ·]q. In [12] we determined quantum groups and one dimensional central extensions of Wq. In this paper, we will study its low dimensional cohomology theory. That is the second aim of this paper. Throughout this paper, C denotes the field of complex number and Z denotes the set of all integers. All vector spaces and algebras are assumed to be over C. 2. Second cohomology group In this section, we first recall some basic definitions and in particular central extension of Hom-Lie algebras. Then we establish a one-to-one correspondence between the equivalence classes of one dimensional central extensions of a Hom-Lie algebra and its second cohomology group with coefficients in C. As application we determine the second cohomology group of the q-deformed W (2, 2) algebra which is considered as a Hom-Lie algebra. In the sequel we will often simply write a Hom-Lie algebra (L, [·, ·], α) as (L, α). A Hom- Lie algebra (L, α) is said to be multiplicative if the twisting map α is an endomorphism. Let G be an abelian group. A Hom-Lie algebra (L, α) is said to be G-graded, if its underlying vector space is G-graded (i.e., L = ⊕g∈GLg) satisfying [Lg, Lh] ⊆ Lg+h, and if α is an even map, i.e., α(Lg) ⊆ Lg, for all g, h ∈ G. The theory of central extensions of Hom-Lie algebras was studied in [3, 4]. An extension of a Hom-Lie algebra (L, ζ) by an abelian Hom-Lie algebra (a, ζa) is a commutative diagram with exact rows 0 −−−→ a ι−−−→ L pr −−−→ L −−−→ 0 ζa  y 0 −−−→ a ζ  y ι−−−→ L 3 ζ  y pr −−−→ L −−−→ 0 where ( L, ζ) is a Hom-Lie algebra. The extension is central if ι(a) ⊆ Z( L) = {x ∈ L [x, L] L = 0}. In the following we focus on the central extension of (L, α) by a one-dimensional center Cc, where c = ι(1). Note that the center Cc can be considered as the one-dimensional trivial Hom-Lie algebra with the identity map. Definition 2.1 Let (L, α) be a Hom-Lie algebra. A bilinear map ψ : L × L → C is called a 2-cocycle on L if the following conditions are satisfied ψ(x, y) = −ψ(y, x), ψ(α(x), [y, z]) + ψ(α(y), [z, x]) + ψ(α(z), [x, y]) = 0, (2.1) (2.2) for all x, y, z ∈ L. Now we have the following theorem: Theorem 2.2 Let (L, α) be a Hom-Lie algebra and ψ : L × L → C be a bilinear map. Define on the vector space L = L ⊕ C the following bracket and linear map by [x + c, y + b] L = [x, y]L + ψ(x, y), α(x + c) = α(x) + c, (2.3) (2.4) for all x, y ∈ L and c, b ∈ C. Then ( L, [·, ·] L, α) is a Hom-Lie algebra one dimensional central extension of (L, α) if and only if ψ is a 2-cocycle on (L, α). If, in addition, (L, α) is multiplicative and ψ satisfies ψ(α(x), α(y)) = ψ(x, y), for all x, y ∈ L, then the Hom-Lie algebra ( L, α) is also multiplicative. Proof. Since [·, ·]L is skew-symmetric, the new bracket [·, ·] L is skew-symmetric if and only if the map ψ is skew-symmetric. For any x, y, z ∈ L and a, b, c ∈ C, we have [ α(x + a), [y + b, z + c] L] L = [α(x) + a, [y, z]L + ψ(y, z)] L = [α(x), [y, z]L]L + ψ(α(x), [y, z]L). Consequently, [ α(x + a), [y + b, z + c] L] L + [ α(y + b), [z + c, x + a] L] L + [ α(z + c), [x + a, y + b] L] L = 0 4 if and only if ψ(α(x), [y, z]L) + ψ(α(y), [z, x]L) + ψ(α(z), [x, y]L) = 0, which proves ( L, [·, ·] L, α) is a Hom-Lie algebra if and only if ψ is a 2-cocycle on (L, α). If (L, α) is multiplicative, then we have α([x + a, y + b] L) = α([x, y]L + ψ(x, y)) = α([x, y]L) + ψ(x, y) = [α(x), α(y)]L + ψ(x, y). On the other hand, we have [ α(x + a), α(y + b)] L = [α(x) + a, α(y) + b] L = [α(x), α(y)]L + ψ(α(x), α(y)). According to the hypothesis that ψ(α(x), α(y)) = ψ(x, y) for all x, y ∈ L, we have α([x + a, y + b] L) = [ α(x + a), α(y + b)] L, for all x, y ∈ L, a, b ∈ C, which shows that ( L, α) is multiplicative. Finally, we define pr and ι as the natural projection and inclusion respectively by pr : L → L, pr(x + a) = x; ι : C → L, ι(a) = 0 + a. Then it is easy to show that ( L, α) is a one-dimensional central extension of (L, α). (cid:3) Denote by Z 2(L, C) the vector space of all 2-cocycles on a Hom-Lie algebra (L, α). For any linear map f : L → C, we can define a 2-cocycle ψf by ψf (x, y) = f ([x, y]), for all x, y ∈ L. (2.5) Such a 2-cocycle is called a 2-coboundary or a trivial 2-cocycle on L. Let B2(L, C) denote the vector space of all 2-coboundaries on L. The quotient space H 2(L, C) = Z 2(L, C)/B2(L, C) is called the second cohomology group of L with trivial coefficients C. A 2-cocycle ψ is said to be equivalent to another 2-cocycle φ if ψ − φ is trivial. For a 2-cocycle ψ, let [ψ] be the equivalent class of ψ. Then we have the following corollary: 5 Corollary 2.3 For any Hom-Lie algebra (L, α), there exists a one-to-one correspondence between the equivalence classes of one dimensional central extensions of (L, α) and its second cohomology group H 2(L, C). In the following, we consider the q-deformed W (2, 2) algebra Wq. Note that Wq is not a Lie algebra, because the classical Jacobi identity does not hold (but the antisymmetry is true). By straightforward calculations, we have (ql + q−l)[ [ Lm, Ln](qm−n, qn−m), Ll ](qm+n−l, ql−m−n) + cyclic permutations = 0, (ql + q−l)[ [ Lm, Ln](qm−n, qn−m), Ml ](qm+n−l, ql−m−n) + cyclic permutations = 0. (2.6) (2.7) Define on Wq a linear map α by α(Ln) = (qn + q−n)Ln, α(Mn) = (qn + q−n)Mn. Then, using the q-deformed Jacobi identities (2.6) and (2.7), we obtain the following result. Theorem 2.4 The triple (Wq, [·, ·]q, α) forms a Hom-Lie algebra. In [12] we provided a computation of one-dimensional central extensions of Wq. Hence, according to Corollary 2.3, we can determine the second cohomology group of the q-deformed W (2, 2) algebra Wq as follows: Proposition 2.5 H 2(Wq, C) = Cβ ⊕ Cγ, where β(Lm, Ln) = δm,−n γ(Lm, Mn) = δm,−n [m − 1]q[m]q[m + 1]q [2]q[3]qhmiq [m − 1]q[m]q[m + 1]q [2]q[3]qhmiq , , β(Lm, Mn) = β(Mm, Mn) = 0, γ(Lm, Ln) = γ(Mm, Mn) = 0, and where hmiq = qm + q−m, for all m, n ∈ Z. 3. Derivations of Hom-Lie algebras and q-deformed W (2, 2) Lie algebra This section is devoted to discuss derivations of graded Hom-Lie algebras. We extend to Hom-Lie algebras some concepts and results of derivations of finitely generated Lie algebras with values in graded modules studied in [2]. As application we compute all αk-derivations and particularly the first cohomology group of the q-deformed W (2, 2) algebra. 6 Definition 3.1 Let (L, α) be a Hom-Lie algebra. A representation of L is a triple (V, ρ, β), where V is a C-vector space, β ∈ End(V ) and ρ : L → End(V ) is a C-linear map satisfying ρ([x, y]) ◦ β = ρ(α(x)) ◦ ρ(y) − ρ(α(y)) ◦ ρ(x), for all x, y ∈ L. V is also called a Hom-L-module, denoted by (V, β) for convenience. One recovers the definition of a representation in the case of Lie algebras by setting α = idL and β = idV . For any x ∈ L, define ad : L → End(L) by adx(y) = [x, y] for all y ∈ L. Then (L, ad, α) is a representation of L, which is called the adjoint representation of L. Definition 3.2 Let (V, βV ) and (W, βW ) be two Hom-L-modules. A linear map f : V → W is called a morphism of Hom-L-modules if it satisfies f ◦ βV = βW ◦ f, f (x · v) = x · f (v), for all x ∈ L, v ∈ V. Let G be an abelian group, (L = ⊕g∈GLg, [·, ·], α) be a G-graded Hom-Lie algebra. An Hom-L-module V is said to be G-graded if V = ⊕g∈GVg and LgVh ⊆ Vg+h for all g, h ∈ G. For any nonnegative integer k, denote by αk the k-times composition of α, i.e., αk = α ◦ α ◦ · · · ◦ α. (3.1) In particular, α0 = id and α1 = α. Then we can define αk-derivations of L with values in its Hom-L-modules. Definition 3.3 A linear map D : L → V is called an αk-derivation if it satisfies D ◦ α = α ◦ D, D[x, y] = αk(x) · D(y) − αk(y) · D(x), for all x, y ∈ L We recover the definition of a derivation by setting k = 0 in the definition above. Hence, an α0-derivation is often simply called a derivation in the present paper. We say that an αk-derivation D has degree g (denoted by deg(D) = g) if D 6= 0 and D(Lh) ⊆ Vg+h for any h ∈ G. Let D be an αk-derivation. If there exists v ∈ V such that D(x) = αk(x) · v for all x ∈ L, then D is called an inner αk-derivation. Denote by Derαk(L, V ) and Innαk(L, V ) the 7 space of αk-derivations and the space of inner αk-derivations, respectively. In particular, let Der(L, V ) and Inn(L, V ) denote the space of derivations and the space of inner derivations, (cid:12) deg(D) = g} ∪ {0}. The first respectively, and write Der(L, V )g := {D ∈ Der(L, V ) (cid:12) cohomology group of L with coefficients in V is defined by H 1(L, V ) := Der(L, V )/Inn(L, V ). (3.2) Remark 3.4 The set Derαk (L, V ) (resp. Innαk (L, V )) is not close under map composition or commutator bracket. But the space of all such αk-derivations ⊕k≥0Derαk(L, V ) (resp. ⊕k≥0Innαk (L, V )) form an Lie algebra via commutator bracket. Now let L be a G-graded Hom-Lie algebra which is finitely generated. In the following we present two results, which can be seen as Hom versions of those obtained in [2]. Proposition 3.5 Let V be a G-graded Hom-L-module. For every D ∈ Der(L, V ), we have D = Pg∈G Dg, (3.3) where Dg ∈ Der(L, V )g and where there are only finitely many Dg(u) 6= 0 in the equation D(u) = Pg∈G Dg(u), for any u ∈ L. Proof. For any g ∈ G, define a homogeneous linear map Dg : L → V as follows: for any u ∈ Lh with h ∈ G, write D(u) = Pp∈G up with up ∈ V , then set Dg(u) = ug+h. Clearly, Dg is well defined and Dg ∈ Der(L, V )g. Also, (3.3) is true. (cid:3) Proposition 3.6 Let V be a G-graded Hom-L-module such that (a) H 1(L0, Vg) = 0, for g ∈ G \ {0}. (b) HomL0(Lg, Vh) = 0, for g 6= h. Then Der(L, V ) = Der(L, V )0 + Inn(L, V ). Proof. Let D be a derivation from L into its Hom-L-module V . According to (3.3) we can decompose D into its homogeneous components D = Pg∈G Dg with Dg ∈ Der(L, V )g. Suppose that g 6= 0. Then DgL0 is a derivation from L0 into the Hom-L0-module Vg. By virtue of (a), DgL0 is inner, i.e., there exists vg ∈ Vg such that Dg(u) = u · vg for all u ∈ L0. Consider ψg : L → V defined by ψg(x) := Dg(x)−x·vg for all x ∈ L. Then ψg is a derivation 8 of degree g which vanishes on L0. Hence ψg is a morphism of Hom-L0-modules and condition (b) entails the vanishing of ψg on Lh for every h ∈ G. Consequently, Dg ∈ Inn(L, V ), which completes the proof. (cid:3) In the following we focus on the q-deformed W(2,2) algebra Wq as a Hom-Lie algebra (Wq, [·, ·]q, α) defined in Theorem 2.4. Obviously, Wq is Z-graded by Wq = ⊕n∈ZW n q , where W n q = spanC{Ln, Mn}. Note that M := spanC{Mn} is an ideal of (Wq, α), or in other words, M is an adjoint Hom-Wq-module. In addition, Wq is finitely generated by {L1, L−1, M1}. Let D be an αk-derivation of Wq. For all m, n ∈ Z, we have (qm + q−m)k[D(Ln), Lm]q + (qn + q−n)k[Ln, D(Lm)]q = [m − n]qD(Lm+n), (3.4) (qm + q−m)k[D(Ln), Mm]q + (qn + q−n)k[Ln, D(Mm)]q = [m − n]qD(Mm+n). (3.5) Now we aim to determine all αk-derivation of Wq. First, we compute the (α0-)derivations of Wq. Denote by Der(Wq) and Inn(Wq) the set of all derivations and the set of all inner derivations, respectively. Let Der(Wq)m be the set of derivations of degree m. Lemma 3.7 H 1(W 0 q , W n q ) = 0 for any nonzero integer n. Proof. Note that [L0, X0]q = 0, for any X0 ∈ W 0 q = span{L0, M0}. Let D be any element in Der(W 0 q , W n q ). Then it follows D(L0) ∈ W n q . Applying D to [L0, X0]q = 0, we have [n]qD(X0) = [L0, D(X0)] = [X0, D(L0)]. Consequently, D(X0) = [X0, v] with v = 1 [n]q D(L0) in W n q . In other words, D is an inner derivation from W 0 q into its adjoint module W n q . (cid:3) Lemma 3.8 HomW 0 q (W m q , W n q ) = 0 for m 6= n. Proof. Let f ∈ HomW 0 q (W m q , W n q ) with m 6= n. For any Xm ∈ W m q , we have (qn + q−n)(cid:0)f (Xm)(cid:1) = α(cid:0)f (Xm)(cid:1) = f(cid:0)α(Xm)(cid:1) = (qm + q−m)f (Xm), leading to f (Xm) = 0, since m 6= n. Hence, f = 0. (cid:3) Now according to Proposition 3.6, we have the following result: Proposition 3.9 Der(Wq) = Der(Wq)0 + Inn(Wq). 9 Thanks to Proposition 3.9, the study of Der(Wq) reduces to that of its constitute of degree zero. Let D be an element of Der(Wq)0. For any integer n, assume that D(Ln) = anLn + bnMn, D(Mn) = cnLn + dnMn, (3.6) where the coefficients are complex numbers. Applying D to [L0, Ln]q = [n]qLn, one can obtain D(L0) = 0, i.e., a0 = b0 = 0. Using (3.4), we have am+n = am + an, bm+n = bm + bn, for m 6= n. Let m = −n. Then we have a−m = −am, b−m = −bm, for m 6= 0. Furthermore, we have am = ma1, bm = mb1, for all m ∈ Z. Similarly, using (3.5) we have cm+n = cn, dm+n = am + dn, for m 6= n, from which it follows cm = c0, dm = ma1 + d0, for all m ∈ Z. Applying D to [M1, M0] = 0, we have c0 = 0. It follows that cm = 0 for all m ∈ Z. Hence, there exist a, b, d ∈ C such that D(Ln) = n(aLn + bMn), D(Mn) = (na + d)Mn, for all n ∈ Z. (3.7) From the discussions above we obtain the following result: Proposition 3.10 All the derivations of Wq is Der(Wq) = spanC{D} ⊕ Inn(Wq), where D is defined by (3.7). Corollary 3.11 The first cohomology group of Wq with values in its adjoint module is one-dimensional. 10 Next we compute the α1-derivations of Wq. Let D be an α1-derivation of degree s. Assume that D(Ln) = as,nLn+s + bs,nMn+s, D(Mn) = cs,nLn+s + ds,nMn+s, where the coefficients are complex numbers. Then from equation (3.4) we obtain [m − n]qas,m+n = (qm + q−m)[m − s − n]qas,n + (qn + q−n)[m + s − n]qas,m, (3.8) [m − n]qbs,m+n = (qm + q−m)[m − s − n]qbs,n + (qn + q−n)[m + s − n]qbs,m. (3.9) We first consider the case of s 6= 0. Taking m = 0 in (3.8), we have (2[s + n]q − [n]q)as,n = (qn + q−n)[s − n]qas,0. Furthermore, as,n = (qn + q−n)[s − n]q (2[s + n]q − [n]q) as,0. (3.10) Plugging (3.10) into (3.8), we have (qm+n + q−m−n)[m − n]q[s − m − n]q 2[s + m + n]q − [m + n]q as,0 = + (qm + q−m)(qn + q−n)[m − s − n]q[s − n]q 2[s + n]q − [n]q as,0 (qm + q−m)(qn + q−n)[m + s − n]q[s − m]q 2[s + m]q − [m]q as,0. Let m = s in the equation above, we have (qs+n + q−s−n)[s − n]q[−n]q 2[2s + n]q − [s + n]q as,0 = (qs + q−s)(qn + q−n)[s − n]q[−n]q 2[s + n]q − [n]q as,0. (3.11) Then taking n = −s in (3.11), we get [2s]q[s]q [s]q as,0 = (qs + q−s)2[2s]q[s]q [s]q as,0. It follows as,0 = 0 since s 6= 0. Then we have as,n = 0 for n ∈ Z and s 6= 0 by (3.10). Similarly, from (3.9) we can deduce that bs,n = 0 for s 6= 0 and n ∈ Z. In the case of s = 0, we simply write a0,n as an. Then it can be deduced from (3.8) that am+n = (qm + q−m)an + (qn + q−n)am, for m 6= n. (3.12) 11 Let m = 0 in (3.12), we have an = −(qn + q−n)a0, which implies an = a−n for n > 0. Taking m = −n in (3.12), we have a0 = (qn + q−n)(an + a−n). (3.13) (3.14) Substituting (3.13) into (3.14), we have a0 = 0 and an = 0 for all n ∈ Z. Similarly, we can deduce that b0,m = 0 for all m ∈ Z by using (3.9) where s = 0. Hence, we have proved that as,m = bs,m = 0, for all m, s ∈ Z, or, in other words, we get D(Lm) = 0 for m ∈ Z. It remains to determine D(Mn) for all n ∈ Z. Using D(Ln) = 0, we can deduce from (3.5) that [m − n]qcs,m+n = (qn + q−n)[m + s − n]qcs,m, [m − n]qds,m+n = (qn + q−n)[m + s − n]qds,m. Let m = 0 in (3.15) and (3.16), respectively. We have − [n]qcs,n = (qn + q−n)[s − n]qcs,0, −[n]qds,n = (qn + q−n)[s − n]qds,0. Taking n = 0 in (3.15) and (3.16), respectively, one has [m]qcs,m = 2[m + s]qcs,m. [m]qds,m = 2[m + s]qds,m. (3.15) (3.16) (3.17) (3.18) (3.19) (3.20) Taking m = 0 in (3.19) and (3.20), respectively, we have cs,0 = ds,0 = 0 for s 6= 0. Then it follows from (3.17) (resp. (3.18)) that cs,n = 0 (resp. ds,n = 0) for n ∈ Z and s 6= 0. If s = 0, then it follows from (3.15) that [m − n]qc0,m+n = (qn + q−n)[m − n]qc0,m. (3.21) Let n = 0 in (3.21), we have [m]qc0,m = 2[m]qc0,m. It follows that c0,m = 0 for m 6= 0. Taking n = −m in (3.21), we have [2m]qc0,0 = (qm + q−m)[2m]qc0,m, leading us to c0,0 = 0. Similarly, we can deduce from (3.16) that d0,m = 0 for all m ∈ Z. Thereby, the following proposition is proved. 12 Proposition 3.12 If D is an α1-derivation of Wq, then D = 0. With the similar discussions as above, we can compute αk-derivations of Wq for k > 1 and thus we have all the αk-derivations of Wq for k > 0 determined. Proposition 3.13 For k > 0, all the αk-derivations of Wq are zero. References [1] F. Ammar, A. Makhlouf, Hom-Lie superalgebras and Hom-Lie admissible superalgebras, J. Algebra, 324(7) (2010), 1513 -- 1528. [2] R. Farnsteiner, Derivations ans central extensions of finitely generated graded Lie algebras, J. Algebra, 118 (1988), 33 -- 45. [3] J.T. Hartwig, D. Larsson, S.D. Silvestrov, Deformations of Lie algebras using σ-derivation, J. Algebra, 295(2006), 314 -- 361. [4] D. Larsson, S.D. Silvestrov, Quasi-hom-Lie algebras, central extensions and 2-cocycle-like identities, J. Algebra, 288(2005), 321 -- 344. [5] D. Larsson, S.D. Silvestrov, Graded quasi-Lie agebras, Czechoslovak J. Phys., 55 (2005), 1473 -- 1478. [6] A. Makhlouf, S.D. Silvestrov, Hom-algebra structures, J. Gen. Lie Theory Appl., 2(2) (2008), 51 -- 64. [7] A. Makhlouf, S.D. Silvestrov, Notes on 1-parameter formal deformations of Hom-associative and Hom-Lie algebras, Forum Math., 22 (2010), 715 -- 739. [8] Y. Sheng, Reprensentations of Hom-Lie algebras, Algebras and Representation Theory, (2010), 1 -- 18. [9] D.Yau, Enveloping algebras of Hom-Lie algebras, J. Gen. Lie Theory Appl. 2 (2008), 95 -- 108 [10] D.Yau, Hom-algebras and homology, J. Lie Theory, 19 (2009), 409 -- 421. [11] Lamei Yuan, Hom-Lie color algebra structures,Comm. Algebra, 40(2)(2012), 575 -- 592. [12] Lamei Yuan, q-Deformation of W (2, 2) Lie algebra associated with quantum groups, Acta Mathematica Sinica, English Series, (2012), DOI: 10.1007/s10114-012-0544-y. [13] W. Zhang, C. Dong, W -Algebra W (2, 2) and the Vertex Operator Algebra L( 1 2 , 0)⊗L( 1 2 , 0), Comm. Math. Phys., 285 (2009), 991 -- 1004. 13
1403.5295	5	1403	2016-08-04T21:26:44	Gradings on Lie algebras, systolic growth, and cohopfian properties of nilpotent groups	[ "math.RA", "math.GR" ]	We study the existence of various types of gradings on Lie algebras, such as Carnot gradings or gradings in positive integers, and prove that the existence of such gradings is invariant under extensions of scalars. As an application, we prove that if G is a finitely generated nilpotent group, its systolic growth is asymptotically equivalent to its word growth if and only if the Malcev completion of G is Carnot. We also characterize when G is non-cohopfian, in terms of the existence of a non-trivial grading in non-negative integers, and deduce that this property only depends on its real (or even complex) Malcev completion.	math.RA	math	GRADINGS ON LIE ALGEBRAS, SYSTOLIC GROWTH, AND COHOPFIAN PROPERTIES OF NILPOTENT GROUPS YVES CORNULIER Abstract. We study the existence of various types of gradings on Lie alge- bras, such as Carnot gradings or gradings in positive integers, and prove that the existence of such gradings is invariant under extensions of scalars. As an application, we prove that if Γ is a finitely generated nilpotent group, its systolic growth is asymptotically equivalent to its word growth if and only if the Malcev completion of Γ is Carnot. We also characterize when Γ is non-cohopfian, in terms of the existence of a non-trivial grading in non-negative integers, and deduce that this property only depends on its real (or even complex) Malcev completion. 1. Introduction The purpose of this paper is twofold: to discuss the existence of certain kinds of gradings on Lie algebras (and some more general algebras), and to apply the results to the study of several aspects of finitely generated nilpotent groups. 1.1. Cohopfian properties and systolic growth. This paper will study some properties in the case of finitely generated nilpotent groups. In this subsection, we introduce these properties in general. 1.1.1. Systolic growth. Let Γ be a finitely generated group, and endow it with the word metric with respect to some finite generating subset S. If Λ ⊂ Γ, define its systole sysS(Λ) to be inf{gS : g ∈ Λ r {1}} (which is +∞ in case Λ = {1}). Define, following [Gro] its systolic growth as the function σΓ,S mapping n to the smallest index of a subgroup of systole ≥ n (hence +∞ if there is no such subgroup). Note that Γ is residually finite if and only if σΓ,S(n) < ∞ for all n, and a standard argument shows that the asymptotic behavior (in the usual sense of growth of groups, see §6) of σΓ,S does not depend on the choice of S; hence we call it the systolic growth of Γ. It is obviously asymptotically bounded below by the growth (precisely, σΓ,S(2n + 1) ≥ bΓ,S(n), where bΓ,S(n) is the cardinal of the n-ball). It is easy to see that Γ and its finite index subgroups have asymptotically equivalent systolic growth. Date: August 4, 2016. 2010 Mathematics Subject Classification. Primary 17B30; Secondary 17B70, 20F18, 20F69, 20E07, 22E25 22E40. Supported by ANR 12-BS01-0003-01 G´eom´etrie des sous-groupes. 1 2 YVES CORNULIER It is natural to wonder when the growth and systolic growth are equivalent. Gromov [Gro, p.334] provides a simple argument, based on congruence subgroups, showing that finitely generated subgroups of GLd(Q) have at most exponential systolic growth (although he states a less general fact). Bou-Rabee and the author [BC] actually prove that all finitely generated linear subgroups (i.e., with a faithful finite-dimensional linear representation over some field) have at most exponential systolic growth, and hence exactly exponential systolic growth when the growth is exponential. For finitely generated linear groups, this thus reduces the question of equivalence of growth and systolic growth to virtually nilpotent groups. Remark 1.1. A notion closely related to systolic growth was introduced by Bou- Rabee and McReynolds [BM] (apparently independently of [Gro]), defining the residual girth of a group in the same way as the systolic growth above, but restricting to normal finite index subgroups. If we denote by σ⊳ Γ,S(n) the resulting function, we obviously have σΓ,S ≤ σ⊳ Γ,S ≤ σΓ,S!. The examples in [BS1] show that σΓ,S, for finitely generated residually finite groups, can be arbitrary large. On the other hand, there is an exponential upper bound for the residual girth of finitely generated linear groups [BC]. if Λ ⊂ Γ, define its normal systole Besides, we can define one more notion: sys⊳ S (Λ) as the infimum of gS, when g ranges over Γ-conjugates of elements in Λ r {1}. Note that it has a geometric interpretation: let G(Γ, S) be the Cayley graph of Γ with respect to S. While sysS(Λ) is the length of the smallest non- trivial based combinatorial loop in the quotient Λ\G(Γ, S) (where non-trivial means it does not lift to a loop in G(Γ, S)), the normal systole sys⊳ S (Λ) is the length of the smallest non-trivial combinatorial loop (not necessarily based); of course when Λ is normal, its normal systole equals its systole. We can then define the uniform systolic growth of Γ as the function σu Γ,S mapping n to the smallest index of a subgroup of normal systole ≥ n. Thus clearly we have σΓ,S ≤ σu Γ,S ≤ σ⊳ Γ,S. I do not know examples for which the uniform systolic growth is not equivalent to the systolic growth; on the other hand simple examples show that it can fail to be equivalent to the residual girth, see Remark 1.9. 1.1.2. Cohopfian properties. Recall that a group is non-cohopfian if it admits a non-surjective injective endomorphism, and cohopfian otherwise. Let us say that a group Γ is dis-cohopfian if it admits an injective endomorphism φ such that Tn≥0 φn(Γ) = {1}; such φ is called a dis-cohopf endomorphism. It appears, for a nontrivial group, as a strong negation of being cohopfian. GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 3 Let us consider an intermediate notion: we say that a group Γ is weakly dis- cohopfian if it admits a sequence of subgroups (Γn), all isomorphic to Γ, with Γn+1 contained in Γn for all n, and Tn Γn = {1}. This is implied by dis-cohopfian, and implies, for a nontrivial group, non-cohopfian. 0 a−1 0 0 b−1 for a ∈ A and b 7→ b0a0ba−1 Example 1.2. The group Z is dis-cohopfian. Slightly less trivially, the infinite dihedral group is dis-cohopfian. Actually, every nontrivial free product A ∗ B is dis-cohopfian (that non-trivial free products are non-cohopfian is well-known; whether they are always dis-cohopfian was asked to me by K. Bou-Rabee): indeed, fix nontrivial elements a0 ∈ A, b0 ∈ B and consider the endomorphism φ defined by a 7→ a0b0ab−1 0 . Let S be the generating set A∪ B for A∗ B, and · the corresponding word length. Then this endomorphism formally maps reduced words to reduced words, and hence φ(x) = 5x for all x; thus φn(x) = 5nx for all x and it follows that Tn≥0 Im(φn) = {1}. The group Z× Z/2Z is non-cohopfian but not weakly dis-cohopfian. Examples of groups that are weakly dis-cohopfian but not dis-cohopfian will be provided in Example 5.16. These are the first such examples among finitely generated nilpotent groups. However, it seems from the discussion in [NP] that they were aware of other examples among polycyclic groups of exponential growth. Using the Frobenius endomorphism, it is also possible to find examples of dis- cohopfian groups with exponential growth, such as suitable finite index subgroups of SLd(Fp[t]), or the lamplighter group (Z/pZ) ≀ Z. Remark 1.3. There are natural analogues of these cohopfian-like properties, where injective endomorphisms are required to have an image of finite index. For finitely generated nilpotent groups (and more generally for virtually polycyclic groups, or even finitely generated solvable groups of finite Prufer rank), injective endo- morphisms automatically have an image of finite index, and thus the notions coincide. It is unknown (see [NP]) if there exists a finitely generated group that is not virtually nilpotent, but admits an injective endomorphism φ with image of finite index, such that Tn≥0 Im(φn) = {1}. 1.2. Gradings on Lie algebras. 1.2.1. Main definitions and results. This subsection is independent of the previ- ous one. It introduces some notions of Lie algebras and some results about these notions, which will be used in the study of nilpotent groups in the next subsec- tion, but purport to be of independent interest, notably for readers interested in the classification of finite-dimensional nilpotent Lie algebras. We denote by R a ground commutative ring (always assumed associative with unit); in most of the discussion, R will be a field. We abbreviate "field of char- acteristic zero" into "Q-field". Given an abelian group A, recall that an A-grading of a Lie R-algebra g is a direct sum decomposition g = Lα∈A gα where each gα is an R-submodule, and satisfying [gα, gβ] ⊂ gα+β for all α, β ∈ A. 4 YVES CORNULIER Gradings are a convenient way to encode various actions: for instance, a grad- ing in Z gives rise to an action of the multiplicative group GL1(R) on g, where r ∈ GL1(R) acts by multiplication by rn on gn. Conversely, under suitable as- sumptions (algebraic action, R a field, g of finite dimension), every GL1-action yields a Z-grading of g. Further conditions on the grading, as below, correspond to further conditions on the action. Every Lie algebra has a trivial grading, namely with g = g0. Here are three conditions on a Z-grading of a Lie algebra g: • the grading is non-negative, that is, in the natural numbers N = {0, 1, 2, . . .} (i.e., gi = 0 for all i < 0). If g admits a non-negative non-trivial grading, we call g semi-contractable; admits a positive grading, we call g contractable; • the grading is positive, that is, in the positive natural numbers N+. If g • the Lie algebra g is generated by g1; we then say that the grading is Carnot. If a Lie algebra g admits a Carnot grading, we call g Carnot; if g is endowed with a Carnot grading, it is called a Carnot-graded Lie algebra (see §1.2.2 for counterexamples among nilpotent Lie algebras). Note that each of these conditions on the grading is implied by the next one; thus g Carnot implies g contractable, which implies (for g 6= 0) g semi- contractable. Note that when g is a finitely generated R-module and is con- tractable, then it is nilpotent and actually our emphasis will be on finite-dimensional nilpotent Lie algebras over fields, especially of characteristic zero. Being Carnot can be redefined in the following way: if g is an arbitrary Lie algebra over the commutative ring R, let (g(i))i≥1 be its lower central series (g(1) = g and g(i+1) = [g, g(i)]). The associated Carnot-graded Lie algebra Car(g) is defined as Li≥1 g(i)/g(i+1) with the naturally induced bracket and grading; the Lie algebra g is Carnot if it is isomorphic (as a Lie R-algebra) to its associated Carnot-graded Lie algebra. A special feature of Carnot gradings is that they are all conjugate under Aut(g); in particular two Carnot-graded Lie algebras are isomorphic as graded Lie algebras if and only if they are isomorphic as Lie algebras. Nevertheless, this is only uniqueness up to conjugacy, and Carnot and Carnot-graded should be distinguished; for instance, for a non-abelian Carnot- graded Lie algebra, the automorphism group as a Lie algebra is larger than the graded automorphism group (see Corollary 3.6 for a precise comparison). Carnot Lie algebras are ubiquitous in the study of Lie algebras and the as- sociated Lie and discrete groups. For instance Pansu [Pan2] proved that any two quasi-isometric simply connected real Lie groups have isomorphic associated Carnot-graded real Lie algebras. The classification of various classes of nilpo- tent finite-dimensional Lie algebras also starts with the Carnot case: for instance Vergne [Ver] classified the Carnot-graded d-dimensional Lie algebras of nilpo- tency length exactly d − 1 over a field of characteristic 6= 2: for d ≥ 2 there are 1 or 2 such Lie algebras, and 2 precisely when d ≥ 6 is even; the corresponding GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 5 classification in the non-Carnot case is out of reach in general, and only known in small dimension. Turning back to gradings, there are further natural conditions not related to positivity, such as • The Z-grading is invertible if g0 = {0}. When the ground Q-field is algebraically closed, these various conditions can be easily and conveniently characterized in terms of the maximal grading, see §3.3. If g is a Lie algebra and R′ is a commutative R-algebra, then every grading on g induces a grading on the Lie R′-algebra R′ ⊗R g, which inherits any of the above conditions. It is natural to wonder whether conversely, under reasonable hypotheses, the existence of a grading on R′ ⊗R g with additional properties implies the existence of a similar grading on g. Let us begin with a few simple counterexamples. The 3-dimensional Lie algebra sl2(C) admits a grading in {−1, 0, 1} with each component of dimension 1. On the other hand, the real Lie algebra so3(R) admits no non-trivial Z-grading (because all its self-derivations are inner and have eigenvalues in iR), and C ⊗R so3(R) and sl2(C) are isomorphic as complex Lie algebras. More subtle counterexamples will be provided in the sequel, but let us begin with positive results. Theorem 1.4 (see Th. 3.15 and Th. 3.25). Being Carnot, contractable, semi- contractable are invariant under taking extensions of Q-fields. That is, if R = K is a Q-field, L is an extension of K and g is a finite-dimensional Lie K-algebra, then g is Carnot (resp. contractable, resp. semi-contractable) if and only if L⊗K g satisfies the same property as a Lie algebra over L. There is an analogy between Theorem 1.4 and Sullivan's result [Sul, Theorem 12.1] that the notion of formality for a nilpotent minimal differential algebra is independent of the ground field of characteristic zero; however we are not aware of a link between these two facts. For the Carnot property and the extension Q ⊂ R, the question whether Carnot goes from R down to Q was raised in 1975 by Johnson [Joh]; a positive solution was written by Dekimpe and Lee but their argument is mistaken (see Remark 3.16). Let us introduce a few ideals canonically associated to a finite-dimensional Lie algebra g over a Q-field. Definition 1.5. The CNI-radical, or (relatively) characteristically nilpotent rad- ical cni(g) of g is the intersection of all kernels of all semisimple self-derivations of g. We say that g is essentially flexible if cni(g) = {0}, and flexible if it admits an invertible self-derivation. This seems to be new notions. We say that an ideal is characteristic if it is invariant under all automorphisms (beware that there exist alternative definitions of characteristic ideals, for instance in the case of Lie algebra). The CNI-radical is a nilpotent characteristic ideal. Classically, g is called characteristically nilpotent 6 YVES CORNULIER if cni(g) = g, or equivalently if every self-derivation of g is nilpotent; the smallest nonzero examples are 7-dimensional, see the survey by Ancochea and Campoamor [AC]. By elementary arguments, the CNI-radical is well-behaved with respect to extensions, in the sense that cni(L ⊗K g) = L ⊗K cni(g), and being essentially flexible or flexible is invariant by taking field extensions. Every characteristically nilpotent characteristic ideal is contained in cni(g); however in a nilpotent Lie algebra, cni(g) is not always characteristically nilpotent, see the example in §3.4. The CNI-radical is also the intersection of fixed point sets in g of all connected sub-tori in Aut(g). When the ground Q-field is algebraically closed, being flexible is equivalent to the existence of an invertible Z-grading. However, the latter property is not invariant under taking field extensions, see §4.2, which relies on a construction of Der´e. Another property of Lie algebras over Q that does not behave well with respect to extensions is to be Anosov; see §4.1. Definition 1.6. The uncontractable radical cni+(g) of g is the intersection of all kernels of all self-derivations of g that are diagonalizable with eigenvalues in N = {0, 1, . . .}. We say that g is essentially contractable if cni+(g) = {0}. Clearly cni(g) ⊂ cni+(g); this is not always an equality; for instance it can happen that cni(g) = {0} but cni+(g) = g: this holds when g is semisimple, but also in the example of a nilpotent real Lie algebra described in §4.2. Note that g is semi-contractable if and only if cni+(g) 6= g. We could make a similar definition allowing eigenvalues in Z, but the resulting ideal (trapped between cni(g) and cni+(g)), which can be checked to be the intersection of kernels of K-diagonalizable self-derivations, would not behave well with respect to field extensions (e.g., in the example of §4.2). On the other hand, the positiveness ensures a good behavior: Theorem 1.7 (see Th. 3.25). Let K be a Q-field and L an extension, and g a finite-dimensional Lie K-algebra. Then cni+(L⊗K g) = L⊗K cni+(g). In partic- ular, g is essentially contractable if and only L ⊗K g is essentially contractable. Note that essentially contractable does not imply contractable, since it does not even imply nilpotent: for example the non-nilpotent 2-dimensional Lie algebra, is essentially contractable. The implication does not even hold for nilpotent Lie algebras: the smallest counterexamples are 7-dimensional and not even flexible (see Remark 3.28). 1.2.2. Small dimension. Let us now put these results in light of the classification of small-dimensional nilpotent Lie algebras. In the following table, we write a statement holding in low dimensions, and in the right column we write the largest dimension for which it holds. Let us begin by a few statements about NLAs (Nilpotent Lie Algebras) mainly related to fields of definition GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 7 statement dim. NLAs with isomorphic complexifications are isomorphic ≤ 5 ≤ 6 ≤ 6 there are finitely many isomorphism classes of NLAs every NLA is defined over Q restriction (K ∗2 6= K ∗) (K = R, C) (K = R, C) The same holds if we restrict to Carnot Lie algebras, for instance the g7,3,1(iλ) in Magnin's classification [Mag] form a 1-parameter family of 7-dimensional Carnot Lie algebras. statement every NLA is Carnot every NLA is contractable dim. ≤ 4 ≤ 6 ≤ 6 ≤ 6 every NLA is either char. nilpotent or essentially contractable ≤ 7 The largest dimension for which a classification (over an algebraically closed It includes 5 1-parameter families and is generally every nonzero NLA is semicontractable every NLA is flexible Q-field) is available is 7. described as a list of about 154 types: • 31 being decomposable as nontrivial direct product, thus contractable; • among the 123 indecomposable types -- 8 types, denoted g7,0,∗ in Magnin's classification [Mag], including one 1-parameter family, consist of characteristically nilpotent NLAs; -- 4 types, denoted g7,1,0∗ in Magnin's classification [Mag], consist of NLAs that are not flexible, but are semicontractable and essentially flexible; -- the other 111 types, including four 1-parameter families, are con- tractable; among them, 36 (including one 1-parameter family) are Carnot. • In dimension ≤ 6, the classification yields, one indecomposable NLA in dimension 3, 1 in dimension 4, 6 in dimension 5 (including 2 non-Carnot), and 20 in dimension 6 (including 10 non-Carnot). The two non-Carnot 5-dimensional NLAs can be described as follows: -- The Lie algebra denoted g5,3 in Magnin's classification and l5,5 in de Graaf's classification, with basis (Xi)1≤i≤5 with nonzero brackets [X1, X3] = X4, [X1, X4] = X5 and [X2, X3] = X5. Its nilpotency It is not Carnot, for instance because its center is 1- length is 3. dimensional but the center of the associated Carnot-graded Lie alge- bra is 2-dimensional. (Note that although l5,5 is indecomposable as a direct product, Car(l5,5) ≃ l5,3 splits as a direct product with the abelian factor generated by X2.) -- The Lie algebra g5,6 (or l5,6) with basis (Xi)1≤i≤5 with nonzero brack- ets [X1, Xi] = Xi+1 (i = 2, 3, 4) and [X2, X3] = X5. Its nilpotency length is 4. Its associated Carnot Lie algebra g5,5 (or l5,7) is defined in the same way except [X2, X3] = 0. 8 YVES CORNULIER 1.3. Systolic growth of nilpotent groups. We now turn to the study of the systolic growth for finitely generated virtually nilpotent groups; as the systolic growth is invariant under passing to finite index subgroups, it is no restriction to focus to torsion-free finitely generated nilpotent groups. The main question raised in §1.1.1 was to understand when the systolic growth and growth are equivalent. This question is solved for nilpotent groups in the following theorem, which pro- vides a geometric characterization of Carnot simply connected Lie groups (among those admitting lattices). Theorem 1.8 (Theorem 6.5). Let G be a simply connected nilpotent real Lie group whose growth rate is polynomial of degree δ, and whose Lie algebra g is definable over Q. Equivalences: (i) the Lie algebra g is Carnot (over R) (ii) every lattice in G has systolic growth ≃ nδ (iii) some lattice in G has systolic growth ≃ nδ (iv) G admits a sequence (Γn) of lattices with systole un → ∞ and covolume (cid:22) uδ n. Note that (iii)⇒(iv) is clear. The arithmeticity of lattices (see [Rag]) shows that definability over Q is equivalent to the existence of a lattice, whence (ii)⇒(iii); more precisely, any lattice yields a Q-structure on g. Assuming (i), we use The- orem 3.15 (i.e., the Carnot case of Theorem 1.4) in order to show that some Carnot grading is defined over Q, which allows to prove (ii). Finally the implica- tion (iv)⇒(i) consists, roughly speaking, in rescaling G, and view some Gromov- Hausdorff limit Ξ of the Γn as a lattice in the asymptotic cone of G and then observe that Γn is isomorphic to Ξ for n large enough. This requires some pre- liminaries to ensure that Ξ is indeed a lattice, and that Γn converges to Ξ in the space of marked groups. The equivalence between (i) and (ii) was suggested by Gromov [Gro, p.333], with, as only comment, the easy checking of (ii) in the case of the Heisenberg group. The proof of (i)⇒(ii) is based on the same construction in general, but as we already mentioned, it makes, beforehand, a crucial use of Theorem 3.15 in its full generality (when the field extension is Q ⊂ R), and Gromov made no hint towards proving that any of the other properties implies (i). Using (iv), any lattice in a non-Carnot simply connected nilpotent Lie group of polynomial growth of degree δ has systolic growth ≫ nδ; it would be interesting to improve this estimate. For instance, for both non-Carnot 5-dimensional Lie algebras l5,5, l5,6 mentioned earlier (before §1.3), we can check that the systolic growth is (cid:22) nδ+1 (with δ the degree of growth, 8 and 11 respectively) and I do In general, the obvious upper bound not know if it is optimal in these cases. σ(n) (cid:22) nc dim(G), where c is the nilpotency length, given by congruence subgroups is easy to improve (see Proposition 6.7), but the precise behavior remains unclear and its study could shed light on how to quantify the lack of being Carnot. GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 9 Remark 1.9. For finitely generated nilpotent groups, while obviously the residual girth σ⊳ is polynomially bounded (as we see using an embedding into upper unipotent integral matrices), it is in general asymptotically much larger that the systolic growth σ: for instance for the Heisenberg group we easily obtain that σ⊳(n) ≃ n6 (while σ(n) ≃ n4). The proof of Theorem 1.8 actually shows that in the Carnot case, the uniform systolic growth (see Remarks 1.1 and 1.9) is asymptotically equivalent to the growth; I do not know whether it is asymptotically equivalent to the systolic growth for all finitely generated nilpotent groups. 1.4. Cohopfian properties for nilpotent groups. This part has a strong similarity with the previous one, since we characterize one property of finitely generated torsion-free nilpotent groups in terms of a property of Lie algebras that we have shown to be invariant under extensions of scalars. However, unlike in the case of systolic growth, we directly obtain the characterization of the cohopfian properties in terms of the rational Lie algebra. The invariance of the Lie algebra properties under extensions of scalars, nevertheless, appears as a way to recognize easily these properties, especially when we only have access to the complexification of the Lie algebra, as in most of the available classifications; they also show that the property does not differ when we consider two lattices in the same simply connected nilpotent Lie group. The most familiar examples of infinite finitely generated torsion-free nilpotent groups fail to be cohopfian; however, Belegradek [Bel] observed that there exists cohopfian examples: for instance, those for which the Malcev Lie algebra is char- acteristically nilpotent (i.e. has a virtually unipotent automorphism group). He gave a criterion for such a group Γ to be non-cohopfian; his criterion [Bel, The- orem 1] is that the real Malcev completion g admits an automorphism mapping log(Γ) into itself, and of determinant of norm greater than 1. However this crite- rion depends on Γ and from this characterization it is by no ways clear whether it changes when Γ is replaced by a finite index subgroup. Actually, being co- hopfian is not inherited by subgroups of finite index (see Appendix A); however the following simple characterization shows that for finitely generated nilpotent groups, it is a commensurability invariant, and even only depends on the real Malcev completion. Theorem 1.10 (Cor. 5.10). Let G be a simply connected nilpotent real Lie group whose Lie algebra g is definable over Q. Equivalent statements: (i) g is semi-contractable; (ii) every lattice of G is non-cohopfian; (iii) some lattice of G is non-cohopfian. Part of this theorem has independently been obtained by Dekimpe and Der´e, namely, the statement that a finitely generated torsion-free nilpotent nilpotent Lie group is co-hopfian if and only if its rational Lie algebra is semi-contractable. 10 YVES CORNULIER This is one half of the the proof of Theorem 5.10; the other (independent) part being that the rational Lie algebra is semi-contractable if and only if the real Lie algebra is semi-contractable, which is part of Theorem 1.4. A naive expectation for proving (iii)⇒(i) would be to consider a non-injective endomorphism and extend it to the Malcev completion, hoping that the resulting endomorphism necessarily has all its eigenvalues of modulus ≥ 1. This is not always the case, as the following simple example shows: G = R2, Γ = Z2 and 2 2(cid:19), whose eigenvalues are 3 ± √5 the endomorphism given by the matrix (cid:18)4 2 (=0.76. . . and 5.23. . . ). A similar statement, with a similar proof, is the following: Theorem 1.11 (Cor. 5.10). Let G be a simply connected nilpotent real Lie group whose Lie algebra g is definable over Q. Equivalent statements: (i) g is contractable; (ii) every lattice of G is dis-cohopfian; (iii) some lattice of G is dis-cohopfian. Remark 1.12. Fix a rational structure on G. When g is known to be contractable (resp. semi-contractable) over Q, it is easy to check that some lattice in G con- tained in GQ is dis-cohopfian (resp. non-cohopfian). However to obtain the con- clusion for every such lattice requires more work, mainly encapsulated in the technical Lemma 5.8. Theorem 1.13 (Cor. 5.15). Let G be a simply connected nilpotent real Lie group whose Lie algebra g is definable over Q. Equivalent statements: (i) g is essentially contractable; (ii) every lattice of G is weakly dis-cohopfian; (iii) some lattice of G is weakly dis-cohopfian. As an example of a corollary of these results, we have the following corollary. Recall that the Hirsch length of a finitely generated nilpotent group Γ is the number of infinite subfactors in any composition series of Γ with cyclic subfactors; it is also the dimension of any simply connected nilpotent Lie group admitting Γ as a lattice. Corollary 1.14. is dis-cohopfian. • Every finitely generated torsion-free nilpotent group of Hirsch length ≤ 6 • If Γ is a finitely generated torsion-free nilpotent group of Hirsch length 7, then either it is cohopfian or weakly dis-cohopfian (dis-cohopfian or not); this does not hold for Hirsch length 8. • Every finitely generated torsion-free 2-step nilpotent group is dis-cohopfian. Questions left open. GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 11 This study of cohopfian properties was limited to torsion-free nilpotent groups It would be interesting to in order to keep the size of the paper reasonable. investigate finitely generated virtually nilpotent groups with no restriction. Question 1.15. Let Γ be a finitely generated virtually nilpotent group, and Γ′ a torsion-free nilpotent subgroup of finite index. Is it true that Γ is cohopfian if and only if Γ′ is cohopfian? The answer is yes for virtually abelian groups, by a non-trivial result of Delzant and Potyagailo [DP, Proof of Theorem D], namely every infinite finitely generated virtually abelian group is non-cohopfian. If we turn to dis-cohopfian and weak dis-cohopfian properties, we need to intro- duce the polyfinite radical W (Γ), which for an arbitrary group Γ is the subgroup generated by all finite normal subgroups of Λ. For a virtually polycyclic group Γ, the subgroup W (Γ) is finite; moreover it follows using [Cor2, Proposition 2.7] that every injective endomorphism of Γ maps W (Γ) onto itself. It particular, non-triviality of W (Γ) is an obstruction for Γ to be weakly dis-cohopfian (and dis-cohopfian). Question 1.16. Let Γ be a finitely generated virtually nilpotent group with W (Γ) = {1}, and Γ′ a torsion-free nilpotent subgroup of finite index. Is it true that Γ is dis-cohopfian (resp. weakly dis-cohopfian) if and only if Γ′ is dis-cohopfian (resp. weakly cohopfian)? In the virtually polycyclic case, finding a general characterization remains widely open. Many semidirect products of the form Zd ⋊ Z were shown to be weakly dis-cohopfian in [NP]. Question 1.17. Can the cohopfian property be characterized, for a virtually poly- cyclic group, in terms of Lie algebras? is it sensitive to passing to finite index subgroups? What if we restrict to those polycyclic groups that are Zariski-dense in a connected algebraic group? What about the weak dis-cohopfian property, if we assume in addition that the polyfinite radical is trivial? For the dis-cohopfian property, we have in mind that a contractable Lie algebra is always nilpotent. This suggests the following: Question 1.18. Is it true that no virtually polycyclic group of exponential growth is dis-cohopfian? Organization of the paper. Many properties of the Lie algebras we are interested in, such as the existence of gradings with given properties, only depend on their group of automorphisms. Therefore it is convenient to study these properties by forgetting the Lie algebra structure and retaining the group of automorphisms, and more generally consid- ering a Zariski-closed subgroup G in GL(V ) and study gradings on V defined by 12 YVES CORNULIER sub-tori in G. This is done in the preliminary Section 2. Next, in Section 3, we specify to group automorphisms of algebras; actually all results in this section were initially written for Lie algebras, but none of the Lie algebras axioms are used, except the existence of a bilinear law; therefore this section is written for arbitrary algebras. This is not the most general context but all applications we have in mind concern Lie algebras and further generalizations (e.g., considering several laws, or ternary laws, etc.) would make the text harder to read. In addi- tion, for the (counter)examples given throughout the text, and in Section 4 for some more consistent ones, we especially focus on Lie algebras. In Section 5, we prove the theorems on cohopfian properties; this makes use of the contractive decomposition which is studied in Sections 2 and 3. Section 6 is mostly independent of the others; it starts with a notion of systolic growth for locally compact groups; it includes the proof of Theorem 1.8 on the systolic growth, or rather the more general Theorem 6.5. Its main bulk is the geometric part of the proof, namely (iv)⇒(i). Acknowledgements. I thank Goulnara Arzhantseva and Yves Benoist for useful conversations, and Pierre de la Harpe and Khalid Bou-Rabee for a number of corrections and suggestions. I thank Igor Belegradek for letting me know about the work of Karel Dekimpe and Jonas Der´e. I thank Jonas Der´e for pointing a mistake in a previous version of this work. Contents Introduction 1. 2. Gradings associated to a Zariski closed subgroup of GLd 3. Generalities, grading and extensions of scalars 4. Counterexamples 5. Cohopfian nilpotent groups 6. Systolic growth and geometry of lattices in nilpotent groups Appendix A. Cohopfian does not pass to finite index References 1 12 16 28 31 38 45 46 2. Gradings associated to a Zariski closed subgroup of GLd 2.1. Linear algebraic K-groups. In this section, we let K be a Q-field, i.e., a field of characteristic zero. The facts we will use about linear algebraic K-groups G (whose unit component is denoted by G◦) are the following. • G admits a maximal torus that is defined over K (such tori are not nec- essarily conjugate over K); more generally, every K-defined torus in G is contained in a maximal torus that is defined over K (see [Con, A.1.2]). GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 13 • [BT, §8.2] G admits maximal K-split tori, which are all conjugate un- der G◦(K) (but these are not necessarily maximal tori). Their common dimension r is called the K-rank of G. • [BS2, §5.1] G admits a Levi factor R defined over K, so that G = U ⋊ R with U the unipotent radical; all such R are conjugate under U(K); every reductive K-subgroup of G (not necessarily connected) is contained in a K-defined Levi factor. Lemma 2.1. Let G′ ⊂ G ⊂ GL(V ) be closed subgroups. Then all pairs (T ′, T ), where T ′ is a maximal torus in G′ and T a maximal torus in G containing T ′, are conjugate under G◦. 1, T1) and (T ′ 1g−1 = T ′ Proof. Let (T ′ such that gT ′ Then T2 and T3 are maximal tori of N containing T ′ such that hT3h−1 = T2. Since hT ′ (T ′ 2, T2) be two such pairs. Then there exists g ∈ (G′)◦ 2 in G. 2. Hence there exists h ∈ N ◦ 1, T1)(hg)−1 = (cid:3) 2. Define T3 = gT1g−1. Let N be the normalizer of T ′ 2, T2). 2h−1 = T ′ 2, it follows that hg(T ′ 2.2. Maximal split tori and gradings. Let V be the affine n-space, i.e., V (K) = K n for every field K. Let T ⊂ GLd = GL(V ) be a K-split torus, say r-dimensional. Then T defines a grading of V in the group X(T ) ≃ Zr of multiplicative characters of T . Here Vχ = {v ∈ V ∀t ∈ T : tv = χ(t)v}. Note that T being K-split means that all χ ∈ X(T ) are K-defined. Let G ⊂ GL(V ) be a K-closed subgroup, and r its K-rank. Every maximal K-split torus in G thus yields a grading of V in Zr. Moreover, any two such gradings (for two choices of maximal split tori and identification of their group of characters with Zr) are conjugate under G◦(K) and GLd(Z), in the sense that n)n∈Zr are two such gradings, then there exists s ∈ G◦(K) and if (gn)n∈Zr and (g′ f ∈ GLr(Z) such that g′ n = s(gf (n)) for all n ∈ Zr. 2.3. Positive weights, contractive decomposition and fine tori. (The forth- coming notions do not depend on a field of definition.) If T ⊂ GL(V ) is a torus, it defines a grading V = Pα∈X(T ) Vα. We say that α ∈ X(T ) is a weight if Vα 6= 0. We say that a homomorphism X(T ) → R is non-negative if it maps weights to non-negative numbers. We say that α ∈ X(T ) is positive (or T -positive if we want to emphasize T ) if there exists a non-negative homomorphism f such that f (α) > 0 (such a homomorphism can then be chosen to be valued in Z). Definition 2.2. The contractive decomposition associated to T is the decompo- sition V = V[0] ⊕ V[+], where V[+] is the sum of all Vα when α ranges over positive weights, and V[0] is the sum of Vα when α ranges over non-positive weights. We write V T [+] if we want to emphasize T . [0] and V T 14 YVES CORNULIER The dimensions of V[+] and V[0] are called the contracted dimension and un- contracted dimension of (V, T ), or of (V, G) whenever G admits T as a maximal torus. If T ′ ⊂ T is a subtorus, we have V T ′ [0] ⊃ V T [0] and V T ′ [+] ⊂ V T [+]. Definition 2.3. Given T ′ ⊂ T ⊂ GL(V ), the torus T ′ is fine in T if the re- striction map X(T ′) → X(T ) maps positive T -weights to positive T ′-weights, or equivalently if V T ′ [+], or equivalently if V T ′ [0] = V T [0]. [+] = V T If G, G′ ⊂ GL(V ) are closed subgroups, we say that G′ ⊂ G is fine if, denoting by T ′ some maximal torus in G′ and T some maximal torus of G containing T ′, we have T ′ fine in T (this does not depend on T, T ′). The following lemma is immediate: Lemma 2.4. Given G′′ ⊂ G′ ⊂ G, the subgroup G′′ is fine in G if and only if G′′ is fine in G′ and G′ is fine in G. Definition 2.5. Let T ⊂ GL(V ) be a torus. A T -fine cocharacter is a cocharacter GL1 → T whose associated grading is an N-grading and satisfies V0 = V T [0]. Every grading obtained this way is called a fine N-grading for T . If G ⊂ GL(V ) is an arbitrary closed subgroup, a G-fine cocharacter is a fine cocharacter of some maximal torus in G. (cid:3) It follows from the definition that the image of every T -fine cocharacter is fine in T . There always exist T -fine cocharacters; more precisely: Lemma 2.6. Every torus T admits a T -fine cocharacter; Proof. For each positive weight α, choose a non-negative homomorphism fα : X(T ) → Z such that fα(α) > 0. Then f = Pα fα is a non-negative homomor- phism that is positive on all positive weights. This homomorphism is induced by some cocharacter, and it satisfies the required properties. (cid:3) Theorem 2.7. Let G ⊂ GL(V ) be a K-closed subgroup. Let R be a K-defined reductive Levi factor; let D be the maximal K-split torus in the center Z(R) of R. Then D ⊂ G is fine. Proof. First assume that G is reductive and connected. Write G = ST , where S is the semisimple part and T = Z(G)◦ the maximal normal torus in G. Write T = AD, where A is the maximal K-anisotropic torus and D the maximal K- split torus. Write M = SA, so that G = MD. Decompose V as a direct sum Li Vi of K-irreducible components with respect to the action of G. On each Vi, the action of M has determinant 1, because M admits no nontrivial K- defined multiplicative character, and the action of D is scalar, given by a weight αi ∈ X(D). Fix a maximal torus L in M (possibly not K-defined). Write the weights (counted with multiplicity) of L×D in Vi as (β, αi) ∈ X(L)×X(D), where β ranges over a set Pi of weights of L. Since the action of L has determinant 1, we GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 15 have Pβ∈Pi β = 0. Let (β0, αj) be a positive weight. This means that there exists a homomorphism (f, g) on X(L × D), valued in Z, mapping (β0, αj) to a positive number and all weights (β, αi) to non-negative numbers. The latter reads: for all i and every β ∈ Pi, we have f (β) + g(αi) ≥ 0, and f (β0) + g(αj) > 0. Summing over β, we obtain #(Pi)g(αi) ≥ 0, and hence g(αi) ≥ 0. Also for i = j this provides g(αj) > 0. This shows that g is non-negative on weights and positive on αj. This proves that whenever (β0, αj) is a positive weight of LD, then αj is a positive weight of D. This proves that D ⊂ LD is fine; since LD is a maximal torus in G, it is fine in G, and hence it follows that D is fine in G. Now assume that G is reductive, but not necessarily connected. Define S, T, A, M from G◦ as above, and let D′ be the maximal K-split torus in Z(G◦). Then G◦ = MD′, and D ⊂ D′. By the connected case, D′ is fine in G◦, and hence in G. So we have to show that D is fine in D′. By Lemma 2.6 (or its proof) there exists a homomorphism f : X(D′) → Z mapping weights to non-negative integers and mapping positive weights to positive integers. The finite group G/G◦ acts on D′ by conjugation, inducing an action on X(D′). This actions permutes the weights and permutes the positive weights. Therefore if we average f by defining ¯f = Ps∈G/G◦ s· f , then ¯f is a homomorphism X(D) → Z that is non-negative on weights, positive on positive weights, and G-invariant. Therefore the correspond- ing cocharacter is fine in D′ and has its image contained in the center of G, and hence in D. This proves that D is fine in D′, and hence in G. Finally, if G is arbitrary, then by the reductive case D is fine in R, and it is (cid:3) always true (and obvious) that R is fine in G. Hence D is fine in G. Corollary 2.8. If G is K-defined, every maximal K-split torus is fine in G. Proof. Let R and D be given as in Theorem 2.7. Then D is fine in G by the theorem. Let D′ be a maximal K-split torus of G containing D. Then D′ ⊂ G is fine as well. Corollary 2.9. If G ⊂ GL(V ) is K-defined and T is a maximal K-split torus, then the (un)contractive dimension of (V, G) equals the (un)contractive dimension of (V, T ). (cid:3) Note that the latter corollary is trivial when T is a maximal torus. A restatement of Corollary 2.8 is the following: Corollary 2.10. If T ⊂ GL(V ) is a K-defined torus, then there is a K-defined T -fine cocharacter. Proof. Let T ′ be the maximal K-split torus in T . By Lemma 2.6, T ′ admits a K-defined fine cocharacter σ. By Corollary 2.8, T ′ is fine in T . Hence σ is also fine as a cocharacter of T . (cid:3) Another application of Theorem 2.7 concerns finite subgroups. 16 YVES CORNULIER Corollary 2.11. Let S be a reductive K-subgroup of G (possibly not connected, e.g., finite). Then every maximal K-split torus in the centralizer of S is fine in G. Equivalently, there exists a K-defined fine cocharacter whose image centralizes S. Proof. Let R be a K-defined Levi factor containing S. By Theorem 2.7, the maximal K-split torus D in Z(R) is fine in G. Let C be the centralizer of S in G. Then D ⊂ C. Let D′ be a maximal K-split torus in C containing D. Since D is fine in G, so is D′. (cid:3) Let us also provide an explicit corollary in extremes cases of the contractive decomposition. Define a cocharacter GL1 → GL(V ) to be non-negative if it defines a grading in N, and positive if moreover 0 is not a weight. Corollary 2.12. Let G ⊂ GL(V ) be a K-closed subgroup, and S a K-closed reductive subgroup. • If G has a positive cocharacter, then it admits a positive K-defined cochar- • if G has a non-trivial non-negative cocharacter, then it admits a K-defined acter valued in the centralizer of S; non-trivial non-negative cocharacter valued in the centralizer of S. Note that for a cocharacter, the condition that it is valued in the centralizer of S implies that the corresponding grading is preserved by S. 3. Generalities, grading and extensions of scalars 3.1. Arbitrary algebras. Let R be a scalar ring, that is, an associative, unital, commutative ring. In this section, we call R-algebra an R-module endowed with an arbitrary R-bilinear law, with no further assumption; it in particular includes Lie algebras, associative non-unital algebras and various generalizations. If g is an R-algebra, we define its lower series as follows: g(1) = g, and, for k ≥ 2, g(k) = Xi,j≥1,i+j=k g(i)g(j). This is a bilateral ideal. For k ≥ 0, if g(k+1) = 0, then g is called k-step nilpotent; if this holds for some k, then g is called nilpotent; its nilpotency length is the smallest k for which this holds. When g is a Lie algebra, it is the usual lower central series. If (A, +) is a magma (a set endowed with an arbitrary binary law), a grading of an algebra g in A (or A-grading) is a direct sum decomposition g = Lα∈A gα of g as an R-module, such that gαgβ ⊂ gα+β. The weights of g are those α ∈ A such that gα 6= {0}. If V is an arbitrary A-graded R-module, then any graded subquotient of V inherits an A-grading. In particular, if g is graded in A, then g/g(2) inherits a grading in A; its weights are called principal weights of g. If g is nilpotent, then GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 17 it can be checked that the set of weights is contained in the submagma generated by principal weights. We can extend to arbitrary algebras the definitions of Lie algebras. Definition 3.1. Let K be a Q-field and let g be a K-algebra that is finite- dimensional as vector space. • cni+(g) is the intersection of kernels of K-diagonalizable self-derivations of g with all eigenvalues in the non-negative integers N; • cni(g) is the intersection of kernels of semisimple self-derivations of g; • g is contractable if it admits an algebra grading in positive integers, or equivalently if it admits a self-derivation with only positive integers as eigenvalues; • g is semi-contractable if cni+(g) 6= g, or equivalently if g admits a non- trivial algebra grading in non-negative integers, or still equivalently if it admits a self-derivation with only non-negative integers as eigenvalues, and at least one positive eigenvalue; • g is characteristically nilpotent if cni(g) = g; • g is flexible if it admits an invertible self-derivation; • g is essentially flexible if cni(g) = {0}; • g is essentially contractable if cni+(g) = {0}. Remark 3.2. In the realm of Lie algebras, characteristically nilpotent is a classical notion and terminology. Contractable is a classical notion but has no common ter- minology (it is sometimes called "graded", but this is also used to mean "Carnot" which is a strictly stronger notion). Even for Lie algebras, essentially flexible and essentially contractable are new notions; the latter is motivated by Theorem 1.13. 3.2. Carnot algebras. Here the algebras are over an arbitrary commutative associative unital ring. Carnot Lie algebras and associated Carnot-graded Lie algebras are important objects, appearing in different places (possibly first in Leger's paper [Leg]) un- der many more names ("graded", "naturally graded", "homogeneous", "quasi- cyclic"), which are often inconvenient and ambiguous; the non-ambiguous "fun- damental graded" is also used by some authors (with negative gradings). The use of the word "Carnot" in this context is common in sub-Riemannian and conformal geometry. Here we introduce them in the context of arbitrary algebras. Definition 3.3. A Carnot grading on an algebra g is a algebra grading of g in N such that g is generated by g1 as an algebra. An algebra is Carnot-graded if it endowed with a Carnot grading, and Carnot if it admits one Carnot grading. Let g be an algebra, and let (g(i))i≥1 be its lower series. The product maps g(i) × g(j) into g(i+j) for all i, j; in particular, denoting vi = g(i)/g(i+1), it induces a bilinear map vi × vj → vi+j for all i, j. 18 YVES CORNULIER Definition 3.4. The direct sum Car(g) = Li≥1 vi, endowed with the bilinear law, is a graded algebra, called the associated Carnot-graded, or associated graded algebra to g. The graded algebra Car(g) is indeed Carnot-graded: this follows from the surjectivity of Li+j=n vi ⊗ vj → vn for all n ≥ 2, which implies by induction that gn is contained in the subalgebra generated by g1 for all n ≥ 1. Proposition 3.5. An algebra g is Carnot if and only if it is isomorphic (as an algebra) to its associated Carnot-graded algebra Car(g). Moreover if these conditions hold, then morphic; • for any Carnot grading on g, the graded algebras g and Car(g) are iso- • for any two Carnot gradings on g, there is a unique automorphism map- ping the first to the second, and inducing the identity modulo g(2). Proof. If g is isomorphic to Car(g), then it admits a Carnot grading (inherited from Car(g)). For the second sentence, we observe that any Carnot-grading on g defines, in a canonical way, an isomorphism from g to Car(g), which in restriction to gi is the restriction of the projection pi : g(i) → g(i)/g(i+1). This proves the first item. For the second item, suppose that we have two Carnot gradings (gi) and (g′ i), defining as above isomorphisms f, g : g → Car(g). Then for every x ∈ g1, x′ ∈ g′ 1 we have f (x) = p1(x) and g(x′) = p1(x′). Hence, if we define h = g−1f , then h ∈ Aut(g) and p1(h(x)) = g(h(x)) = f (x) = p1(x). This shows that h equals the identity modulo [g, g]. Then h maps the first grading to the second grading. The uniqueness is clear. (cid:3) Corollary 3.6. Let g be a Carnot-graded algebra over R. Denote by Aut(g) its automorphism group as an algebra and Aut(g)0 its automorphism group as graded algebra. Let Aut(g)≥1 be the group of automorphisms of the Lie algebra g inducing the identity on g/g(2). Then Aut(g)≥1 is a normal subgroup and Aut(g) = Aut(g)0 ⋉ Aut(g)≥1. Proof. The group Aut(g)≥1 is clearly normal. If we have an automorphism φ of g, then it maps the given Carnot grading to another one. By the last assertion of Proposition 3.5, there exists u ∈ Aut(g)≥1 mapping the first Carnot grading to the second one. Hence v = u−1φ ∈ Aut(g)0, and φ = uv. (cid:3) If we work in a certain category of algebras g (e.g., associative, Lie,. . . ) it is useful to know that Car(g) is also in the same category. The following proposition proves it in many cases. If P = P (X1, . . . , Xk) is any non-associative polynomial (with scalars in R) in the formal variables X1, . . . , Xk, we say that P is an identity for an algebra g if P (x1, . . . , xk) = 0 for all xi ∈ g. GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 19 Proposition 3.7. Let P be an identity for g. If P is either multilinear or P (X) = X 2, then it is an identity for Car(g). Proof. We need to show that P (x1, . . . , xk) = 0 for all x1, . . . , xk ∈ Car(g). By multilinearity, we can assume that each xi is homogeneous, say of degree ni. Let yi be a lift of xi in g(ni). For πj be the projection from g(j) to g(j)/g(j+1) and have πn(Q(z1, . . . , zk)) = Q(πn1(z1), . . . , πnk (zk)). In particular, we deduce that P (x1, . . . , xk) = 0. let n = P ni. Then for every k-multilinear polynomial Q and all zi ∈ g(ni), we For the case of P (X) = X 2, consider an element Pj≥1 xj with xj ∈ g(j)/g(j+1) (almost all zero) and let yj be a lift of xj in g(j) (chosen to be 0 if xj = 0). Then (xkxℓ + xℓxk) πk+ℓ(ykyℓ + yℓyk) (cid:16)X xj(cid:17)2 j +Xk<ℓ π2j(y2 =X x2 =Xj =Xj j ) +Xk<ℓ j ) +Xk<ℓ j = (yk + yℓ)2 = 0 for all j, k, ℓ. π2j(y2 πk+ℓ((yk + yℓ)2 − y2 k − y2 ℓ ) = 0, (cid:3) because y2 Remark 3.8. To be multilinear is a useful property for an identity: implies that it passes to extensions of scalars. See the discussion in [Sch]. indeed, it Example 3.9. Here are some further types of algebras defined by a multilinear identity: ciator); • Associative algebras: A(x, y, z) = (xy)z − x(yz) (A is known as the asso- • Pre-Lie algebras: P L(x, y, z) = (xy)z−x(yz)−(xz)y+x(zy) = A(x, y, z)− • Novikov algebras: P L and N(x, y, z) = (xy)z − (xz)y; • Leibniz algebras: L(x, y, z) = (xy)z − x(yz) − (xz)y. A(x, z, y); Some more are defined by an identity with a quadratic variable, which can, by the obvious polarization formulas, be converted to a multilinear identity when 2 is invertible: A is the associator; ((yz)x)x − ((zx)x)y. • alternative algebras: AL1(x, y) = A(x, x, y), AL2(x, y) = A(y, x, x), where • Malcev algebras: C ′(x, y) = xy+yx and M(x, y, z) = (xy)(xz)−((xy)z)x− Another classical notion is that of Jordan algebra; classically it is defined as an algebra satisfying the identity C(x, y) = xy − yx (i.e., commutative) and the identity Jo(x, y) = (xy)(xx) − x(y(xx)) = A(x, y, xx); if 6 is invertible, a simple verification is that this is equivalent to satisfy the multilinear identities C and Jo′(x, y, z, w) = A(x, y, zw) + A(z, y, wx) + A(w, y, xz). 20 YVES CORNULIER Lemma 3.10. Let K be a field. Let g be a nilpotent algebra over K; if K has positive characteristic p, assume that its nilpotency length c is at most p+1. Then g is Carnot over K if and only if g has a self-derivation inducing the identity on g/g(2). Proof. The "only if" part is clear, since any Carnot grading defines such a deriva- tion, defined to be the multiplication by i on gi. Conversely, assume that there exists such a self-derivation δ (and keep in mind that we do not assume g to be finite-dimensional). For any integer i ≥ 1, define g[i] as the eigenspace of δ for the eigenvalue i. By assumption, we have δ(x) − x ∈ g(2) for all x ∈ g. By induction, we deduce that δ(x) − ix ∈ g(i+1) for all i ≥ 1 and x ∈ g(i). Then by a descending induction on 1 ≤ i ≤ c, we see that (δ − i) . . . (δ − c) vanishes on g(i). Eventually for i = 1, this means that ∆ = (δ − 1) . . . (δ − c) vanishes on g. First assume that K has characteristic zero, or characteristic p ≥ c. Then 1, . . . , c are distinct in K. This implies that the eigenspaces g[i] for 1 ≤ i ≤ c span their direct sum, and the vanishing of ∆ implies that they actually span g. Thus defining gi = g[i] yields a Carnot grading. Now assume the remaining case, namely p = c− 1. Observe that 0 = ∆ can be rewritten as (δ − 1)2(δ − 2) . . . (δ − p − 1). If we define gh1i = ghpi as the kernel of (δ − 1)2 and ghii = g[i] for 2 ≤ i ≤ c − 1, then the ghii for 1 ≤ i ≤ p − 1 generate their direct sum, which by the vanishing of ∆ equals g, and ghiighji ⊂ ghi+ji for all integers i, j ≥ 1. Then define g1 as a supplement subspace of g(p) in gh1i, define gp = g(p), and gi = ghii for 2 ≤ i ≤ p − 1. Then g = Lp i=2 gi, and, noting that gp = ghpi ∩ g(2), we see that (gi) is an algebra grading; thus it is a Carnot grading. i=1 gi and g(2) = Lp (cid:3) Remark 3.11. In characteristic p, the nilpotency length condition in Lemma 3.10 is optimal: consider the Lie algebra g with basis (U1, X1, . . . , Xp+2, Y1, Z1), with nonzero brackets [U1, Xi] = Xi+1 for 1 ≤ i ≤ p+1 and [Y1, Z1] = Xp+2. It is easily checked not to be Carnot. On the other hand, it admits a grading in Z/pZ where g1 has basis (U1, X1, Xp+1, Y1, Z1), where g2 has basis (X2, Xp+2) and gi has basis (Xi) for 3 ≤ i ≤ p. This grading induces a derivation which is the identity on the abelianization (which has the basis (U1, X1, Y1, Z1)). On the other hand, we do not know whether the assumptions in positive characteristic can be relaxed for Theorem 3.15. Example 3.12. Every 2-step nilpotent algebra g is Carnot, as any supplement sub- space of g(2) yields a Carnot grading. However, the 3-step nilpotent 5-dimensional Lie algebra g5,3 from §1.2.2 is not Carnot. Example 3.13. Every nilpotent algebra of dimension at most 3 over a field is Carnot, by an easy verification. While all 4-dimensional nilpotent Lie algebras, over an arbitrary field, are Carnot, the following 4-dimensional commutative associative nilpotent algebra, defined over any ground field, is not Carnot: The GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 21 algebra g defined with basis (x, y, z, w) and nonzero products x2 = z, xz = zx = w, y2 = w (alternatively described K[x, y]+/(xy, x3− y2), where K[x, y]+ = (x, y)K[x, y] is the free associative (non-unital) commutative ring on 2 generators, or also described as the unique maximal ideal in K[x, y]/(xy, x3 − y2)). The next example also provides a non-Carnot 4-dimensional Leibniz algebra. Example 3.14. Here is a 4-dimensional nilpotent Leibniz algebra in which every self-derivation is nilpotent: it has the basis (xi)1≤i≤4 with nonzero products x2 1 = x3, x2x1 = x3, x2 2 = x4, x3x1 = x4. (It is denoted R4(0) in [AOR, Theorem 3.2].) Indeed, calling it g, a computation shows that every derivation maps g into g(2). Theorem 3.15. Let K ⊂ K ′ be fields of characteristic zero. Let g be a finite- dimensional algebra over K. Then g is Carnot over K if (and only if ) K ′ ⊗K g is Carnot over K ′. More generally, this holds in positive characteristic p when either • the nilpotency length of g is at most p + 1, or • K is an infinite and perfect field. Proof. We use the criterion of Lemma 3.10. It is convenient to take a slightly = L ⊗K g for every schematic point of view and write g for the functor L 7→ g field extension K ⊂ L. Let D be the affine space of self-derivations of g that induce the identity on g/g(2): that is, DL is the (possibly empty) affine space of self-derivations δ of g L such that δ(x) − x belongs to g(2) . Then D is a K-defined affine subspace of the space of linear endomorphisms of g. Then by assumption DK ′ is non-empty. It follows that DK is non-empty as well. In other words, g admits a self-derivation inducing the identity on g/g(2). for all x ∈ g L L L Now assume that K is infinite perfect with no further restriction, and that g has nilpotency length ℓ ≥ 1 and g⊗K K ′ is Carnot over K ′. We can assume g 6= {0}. Let H be the algebraic subgroup of GL(g) consisting of those automorphisms h inducing a scalar multiplication modulo g(2), say by the scalar χ(h). This is a K-defined subgroup, and χ is a K-defined multiplicative character on H. By a straightforward induction, the action of h ∈ H on g(i)/g(i+1) is given by multiplication by χ(h)i, for all i ≥ 1. In particular, we have Ph(h) = 0, where Ph is the polynomial Qℓ i=1(X − χ(h)i). Since by Rosenlicht's theorem (which applies as K is perfect and infinite), H ◦(K) is Zariski-dense in H ◦(K ′), and since the Zariski-open subset Ω = {h ∈ H ◦(K ′), χ(h)(2ℓ)! 6= 1} is nonempty (because H is Carnot over K ′), we can find in H ◦(K)∩ Ω some element g. Then χ(g), . . . , χ(g)k are pairwise distinct, so Pg has no multiple root and thus g is diagonalizable with eigenvalues χ(g), . . . , χ(g)ℓ. The corresponding eigenspace decomposition is defined over K and defines a Carnot grading. (cid:3) Remark 3.16. The assertion that if g is a finite-dimensional Lie algebra over Q and R⊗Qg is Carnot over R, then g is Carnot, was done by Dekimpe and Lee [DL, 22 YVES CORNULIER Corollary 4.2]. However, their proof is mistaken. It is based on the assertion [DL, Lemma 4.1] that in a Carnot Lie algebra, every supplement subspace of [g, g] can be chosen to be g1 for some Carnot grading. However, this assertion is false. For instance, it fails when g is the 5-dimensional Lie algebra with basis (X1, . . . , X5) and nonzero brackets [X1, X2] = X3 and [X1, X3] = X4 (note that X5 generates a direct factor). Then there is no Carnot grading for which g1 is the subspace v with basis (X1, X2, X5 + X3). The error in [DL, Lemma 4.1] is the claim, not verified, that [v, v] and [v, [v, v]] have an intersection reduced to zero. Let us also mention the following fact, which was asserted for Lie algebras when S is finite by Dekimpe and Lee [DL, Proposition 4.3], but with a mistaken proof: indeed their proof consists in proving that [g, g] admits an S-invariant supplement, but this is not enough (for the same reason as in Remark 3.16). Proposition 3.17. Let g be a finite-dimensional Carnot algebra over the Q-field K. Let S be a subgroup of automorphisms of g, with a reductive Zariski closure. Then there exists an S-invariant Carnot grading. Proof. We use the conventions of the proof of Theorem 3.15. The group Aut(g) naturally acts on the space of self-derivations of g, by (g · δ)(x) = gδ(g−1x). This action preserves the affine subspace D of self-derivations that induce the identity on g/g(2), which is by assumption non-empty since g is Carnot. Since the Zariski closure of S, which we denote by S, is reductive and K has characteristic zero, the affine subspace DS of points in D fixed by S is non-empty; since it is defined over K, it admits a K-point, and thus DS is non-empty as well. This defines (by Lemma 3.10) a Carnot-grading on g that is S-invariant. (cid:3) 3.3. Maximal grading. Let K be a Q-field (i.e., a field of characteristic zero). Let g be a K-algebra that is finite-dimensional as vector space (with no assump- tion such as associativity). Definition 3.18. The K-rank of g, denoted r, is the the K-rank of Aut(g). The gradings on g in Zr induced by maximal K-split tori of Aut(g) (as in §2.2) are called K-maximal gradings of g. Remark 3.19. The systematic use of this notion of maximal grading in the case of Lie algebras can be found, for instance, in [Fav], where it is called "system of weights". When there is a definition of inner derivation (e.g., in Lie algebras), or inner automorphism (e.g., in associative algebras), a notion of (inner) maximal grading can be defined in terms of inner derivations, or inner automorphisms. In the case of Lie algebras, it is known as the Cartan grading. We will not use this inner notion here. Proposition 3.20. Every K-maximal grading on g is an algebra grading in Zr, and for each such grading, Zr is additively generated by the weights. GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 23 They are conjugate under Aut(g)◦(K) and GLd(Z): if (gn)n∈Zr and (g′ n = u(gf (n)) for all n ∈ Zd. n)n∈Zr are maximal gradings, then there exists u ∈ Aut(g)◦(K) and f ∈ GLd(Z) such that g′ The K-maximal gradings are the finest gradings of g in torsion-free abelian groups, in the sense that for every algebra grading (g′ α)α∈A of g in a torsion- free abelian group A, there exists a K-maximal grading (gn)n∈Zr on g and a homomorphism f : Zr → A such that g′ Proof. The first statement is straightforward. The second statement (uniqueness up to conjugacy) is an immediate consequence of the fact that all maximal K-split tori in G = Aut(g) are conjugate under Aut(g)0(K). α = Ln∈f −1({α}) gn for all α ∈ A. For the last statement, it is no restriction to assume that A is generated by the weights; in particular A is finitely generated and we can suppose that A = Zs. Consider the action of GLs 1 on g, such that (λ1, . . . , λs) acts on gn by multipli- cation by Q λni i , for n = (n1, . . . , ns) ∈ Zs. Since A is generated by weights, this action is faithful. The image of GLs 1 in Aut(g)◦ is a K-split torus T ′, hence is contained in a maximal K-split torus T ; consider the corresponding maxi- mal grading. Then the embedding T ′ ⊂ T induces a surjective homomorphism f : Zr = X(T ) → X(T ′) with the required property. (cid:3) Some properties of the maximal grading are more sensitive to the context: for instance, if K is an algebraically closed Q-field and g is a Lie algebra, then g0 is always nilpotent. Various properties such as those stated in the introduction can be restated in terms of maximal gradings. Proposition 3.21. Let g be a K-algebra that is finite-dimensional as vector space, and endow it with a K-maximal grading in Zr where r is the K-rank of g. First assume that K is algebraically closed. Then (1) g is characteristically nilpotent ⇔ r = 0 ⇔ 0 is the only weight of g for the (2) g is flexible if and only if g0 = 0; (3) g is essentially flexible if and only if the only characteristic ideal of g contained maximal grading; in g0 is zero; (4) the CNI-radical of g is the largest characteristic ideal of g contained in g0. When K is an arbitrary Q-field, the first two assertions have the following more general form: (1') g has no nonzero K-diagonalizable derivation ⇔ r = 0 ⇔ 0 is the only (2') g has an invertible K-diagonalizable derivation if and only if g0 = 0. weight of g for the K-maximal grading; Proof. (1') is immediate. For (2'), if g admits an invertible K-diagonalizable derivation, then it induces a grading of g in the additive group of K, with g0 = {0}. Since the K-maximal grading is a refinement of the latter grading, it also 24 YVES CORNULIER satisfies g0 = {0}. Conversely if 0 is not a weight for the K-maximal grading, pick a homomorphism Zr → Z mapping no weight to 0; then 0 is not a weight of the resulting grading in Z, which therefore induces an invertible derivation by multiplication by k on gk. (1) and (2) follow as particular cases. Let us prove (4), which admits (3) as a particular case. There exists a ho- momorphism f : Zr → Z not vanishing on any nonzero weight. This defines a self-derivation of g, defined to be multiplication by f (α) on gα. This derivation is diagonalizable with integral eigenvalues and the kernel of this derivation is g0 (because the ground field is a Q-field). Hence cni(g) ⊂ g0. Let h be the largest characteristic ideal contained in g0. It follows that cni(g) ⊂ h. To show the re- verse inclusion, suppose that x ∈ h and assume by contradiction that x /∈ cni(g). Then there exists a semisimple derivation D such that x /∈ Ker(D). This deriva- tion induces a grading of g in the additive group of the algebraically closed field K, for which the set of weights is the spectrum Spec(D) of D. We can find a homomorphism from the subgroup generated by Spec(D) to Z that is injective on Spec(D) ∪ {0}. This defines a grading (g′ n)n∈Z of g in Z for which x /∈ g′ 0. Then there exists a maximal grading (g′′ 0 ⊂ g′ 0 and hence x /∈ g′′ 0. Since all maximal gradings are conjugate by automorphisms, there exists η ∈ Aut(g) such that η(g′′ 0) = g0. Hence η(x) /∈ g0. This contradicts the fact that h is a characteristic ideal contained in g0. α) refining this grading, and thus g′′ (cid:3) Note that these properties are very sensitive on the ground field. On the other hand, we are going to check, using Corollary 2.8, that several properties related to positivity are invariant under extension of scalars. It is convenient to use the following language: when a fixed algebra g is graded in Rd, let us call a linear form on Rr positive if it maps all weights of g to non-negative numbers; let us call an element of Rr positive if it is mapped to a positive number by some positive linear form, and non-positive otherwise. Proposition 3.22. Let g be a finite-dimensional K-algebra, endowed with a K- maximal grading. • g is contractable ⇔ the closed convex cone generated by weights is salient ⇔ there exists a linear form on Rr sending all weights to positive numbers ⇔ all weights of g are positive. • g is semicontractable ⇔ the interior of the convex hull of weights does not contain 0 ⇔ all weights lie in a closed linear half-space ⇔ there exists a nonzero positive linear form on Rr ⇔ g admits at least one positive weight • g is Carnot ⇔ g is nilpotent and there exists a linear form on Rr map- ping all principal weights to 1 ⇔ the affine subspace of Rr generated by principal weights does not contain 0. Proof. If g is contractable, then it admits a grading in Z for which all weights are positive. This grading has to be obtained by projection from the maximal GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 25 grading; hence there exists a homomorphism Zr → Z mapping all weights to positive numbers. This holds if and only if the convex cone generated by weights is salient. Conversely if there exists such a homomorphism, the resulting grading in Z has only positive weights and hence g is contractable. The characterization of semicontractable goes along the same lines (keeping in mind that Rr is spanned by weights). If g is Carnot, then it is nilpotent and fixing a Carnot grading, the resulting homomorphism Zr maps all principal weights to 1. Conversely, suppose g is nilpotent and that some linear map Rr → R maps all principal weights to 1, consider the resulting grading of g in R; for this grading, 1 is the only principal weight. Since g is nilpotent, it is generated by the gα where α ranges over principal weights, and hence is generated by g1. Thus g is Carnot. (cid:3) Definition 3.23. Let g be a finite-dimensional K-algebra, and fix a maximal K-split torus in Aut(g). The resulting contractive decomposition g = g[0] ⋉ g[+] (Definition 2.2) is called K-contractive decomposition of g. The dimensions of g[0] and g[+] are called the uncontracted and contracted dimensions of g. We say that an algebra grading of g in N is fine if g = g0 ⋉(cid:0)Ln>0 gn(cid:1) is a K- contractive decomposition, or equivalently if dim(g0) is equal to the uncontracted dimension of g. Here we use the semidirect product notation since g[0] is a subalgebra and g[+] is a bilateral ideal. Proposition 3.24. Let g be a finite-dimensional K-algebra and g = g[0] ⋉ g[+] a K-contractive decomposition. Then • g is contractable if and only if g = g[+] (i.e., g[0] = {0}) • g is semicontractable if and only if g[+] is nonzero • cni+(g) is the largest characteristic bilateral ideal of g contained in g[0], and is also the largest bilateral ideal of g invariant under Aut◦(g). • g is essentially contractable if and only if g[0] does not contain any nonzero characteristic bilateral ideal, if and only if it does not contain any bilateral ideal invariant under Aut(g)◦(K). Proof. Everything follows from the definitions, except the facts referring to Aut(g)◦ = Aut(g)◦(K). Let us check the last characterization of cni+(g) (the last charac- terization of being essentially contractable following immediately). Let n be the maximal Aut(g)◦-invariant bilateral ideal contained in g[0]; we have to show that n is a characteristic ideal. Let α be an automorphism of g. Since the contractive decomposition is unique modulo Aut(g)◦, there exists β ∈ Aut(g)◦ such that βα(g[0]) = g[0]. Thus βα(n) is an Aut(g)◦-invariant bilateral ideal contained in g0; hence we have βα(n) = n; thus composing by β−1 we get α(n) = n. (cid:3) 26 YVES CORNULIER Theorem 3.25. Let K ⊂ L be an extension of Q-fields. Let g be a finite- dimensional K-algebra; write gK = g and gL = L ⊗K g (viewed as L-algebra). Then (1) gK is contractable if and only if gL is contractable (2) gK is semicontractable if and only if gL is semicontractable (3) gK is essentially contractable if and only if gL is essentially contractable (4) if gK = g[0] ⋉ g[+] is a K-contractive decomposition of gK, then, denoting [∗] = L ⊗K g[∗] for ∗ ∈ {0, +}, we have that gL gL [+] is a contractive decomposition of the L-algebra gL = L⊗K g; in particular the uncontracted and contracted dimension of g are invariant by field extension of scalars. (5) the uncontractable radical cni+(gL) is equal to cni+(g)L = L ⊗K cni+(g) Proof. Let g have dimension d, and let G be its automorphism group, viewed as an algebraic K-subgroup of GLd. ⋉ gL [0] Then (4) immediately follows from Corollary 2.10. In view of Proposition 3.24, this immediately yields (1) and (2). By Proposition 3.24, cni+(g) is the largest Aut(g)◦(K)-invariant bilateral ideal of g contained in g[0]. Thus cni+(g) = Tα∈Aut(g)◦(K) α(g[0]). In particular, for every [0]). By Zariski density of Aut◦(g)(K), α ∈ Aut(g)◦(K), we have cni+(g)L ⊂ α(gL [0]). Thus cni+(g)L ⊂ Tα∈Aut(g)◦(L) α(gL we deduce that we have cni+(g)L ⊂ α(gL [0]). Let us show the reverse inclusion. We can write cni+(g) = Tn i=1 αi(g[0]) with αi ∈ Aut(g)◦(K). Hence we obtain cni+(g)L = Tn [0]) ⊃ \α∈Aut(g)◦(L) [0]) ⊃ cni+(g)L, cni+(g)L = αi(gL i=1 αi(gL [0]). Thus n \i=1 α(gL so that all inclusions are equalities. Therefore, again using the characterization of Proposition 3.24, we obtain cni+(gL) = cni+(g)L, proving (5), and (3) follows. (cid:3) Theorem 3.26. Let g be a finite-dimensional K-algebra and S ⊂ Aut(g) a subgroup of automorphisms with a reductive Zariski closure. Then g admits an S-invariant fine grading in N. Proof. This follows from Corollary 2.11. (cid:3) It yields the following corollary as particular cases: Corollary 3.27. Under the assumptions of Theorem 3.26, • if g is contractable, then it admits an S-invariant positive grading; • if g is semicontractable, then it admits an S-invariant non-trivial non- negative grading. This proves conjectures of Dekimpe and Der´e [DD, Sec. 4 and 5]. Remark 3.28 (On the contractive decomposition). If g is the 2-dimensional non- abelian Lie algebra, then in the contractive decomposition, both g[+] and g[0] are GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 27 1-dimensional; for all Lie algebras up to dimension 3, g[+] is actually equal to the nilpotent radical (for any contractive decomposition). For nilpotent Lie algebras, the smallest examples for which g[+] is not either {0} or g are the Lie algebras denoted g7,1,∗ in Magnin's classification [Mag], where ∗ ∈ {01(i), 01(ii), 02, 03} (see also Example 5.16). For the first two, g[0] has dimension 2; for the last two, g[0] has dimension 1; in all these four cases, cni+(g) is zero, and g[+] is a characteristic ideal, i.e., does not depend on the choice of maximal grading. Nevertheless, it is not true in general that g[+] is a characteristic ideal, see §3.4. 3.4. An example. Here we give a simple example showing that • the CNI-radical of a nilpotent finite-dimensional Lie algebra is not always • the CNI-radical of a direct product of two Lie algebras can be strictly • if g = g[0] ⋉ g[+] is a contractive decomposition, then g[+] is not necessarily contained in the product of the CNI-radicals of the given factors. characteristically nilpotent. a characteristic ideal. Proposition 3.29. Let a be a nonzero abelian Lie algebra and g a characteristi- cally nilpotent Lie algebra. Then cni(g × a) = cni+(g × a) = [g, g] × {0}. Proof. Let d be a semisimple derivation. Then d maps the derived subalgebra [g, g] × {0} into itself, and maps the center z(g) × a into itself. Since g is charac- teristically nilpotent, we have z(g) ⊂ [g, g]. Thus d maps [g, g]× a into itself. Let b be a d-stable supplement subspace of [g, g]× a in g× a. Thus h = [g, g]⊕ b is a d-stable supplement subspace of {0} × a. Then h is isomorphic to g (since both are isomorphic to (g × a)/a). Thus h is characteristically nilpotent. Hence d is zero on h. In particular, d is zero on [g, g] × {0}. Thus [g, g] × {0} ⊂ cni(g). If x0 belongs to g r [g, g], then there exists a homomorphism f : g → a such that f (x0) 6= 0, and hence (x, y) 7→ (x, y + f (x)) maps (x0, 0) outside g × {0}, so that (x0, 0) /∈ cni+(g) since the latter is a characteristic ideal. Therefore cni+(g) ⊂ [g, g] × {0}. Since cni(g) ⊂ cni+(g), this completes the proof. Corollary 3.30. There exist nilpotent Lie algebras over any Q-field whose CNI- radical is not characteristically nilpotent. The minimal dimension in which such Lie algebras exist is 8. Conversely, it is clear that cni+(g) ⊂ g × {0}. (cid:3) Proof. If g is a nonzero characteristically nilpotent Lie algebra of minimal dimen- sion (namely 7), then [g, g] is not characteristically nilpotent, and hence if a is abelian and nonzero then cni(g × a) ≃ [g, g] is not characteristically nilpotent. (According to the classification in dimension 7, [g, g] is then of dimension 4 or 5: it is 4 for only one example, namely g7,0,8 in Magnin's classification [Mag], and 5 for all others including the infinite family.) Thus picking g of dimension 7 and a of dimension 1, we get 8-dimensional examples. 28 YVES CORNULIER Conversely, since according to the classification, in dimension ≤ 6, every Lie algebra g has a zero CNI-radical, and in dimension 7 the CNI-radical is always {0} or g, the smallest possible dimension of a nilpotent algebra whose CNI-radical is not characteristically nilpotent is ≥ 8. (cid:3) 4. Counterexamples Here we indicate a few properties of Lie algebras, also related to tori of auto- morphisms, but which are not well-behaved with respect to extensions. 4.1. Anosov nilmanifolds. Let Γ be a finitely generated torsion-free nilpotent group. Let GQ and GR be its rational and real Malcev completions, and gQ and gR the corresponding Lie algebras. An Anosov automorphism of (GR, Γ) is an automorphism of G preserving Γ, such that the corresponding automorphism of gR has no complex eigenvalue of modulus 1. The group Γ is called Anosov if there exists an Anosov automorphism of (GR, Γ). It has long been observed that this only depends on GQ. More precisely, we have: Proposition 4.1. The group Γ is Anosov if and only if there exists a Q-defined torus in Aut(GQ) that is R-split, Q-anisotropic, and which has no nonzero in- variant vector in gQ. Proof. Let φ be an Anosov automorphism of (GR, Γ), and let ξ be the corre- sponding automorphism of gR. Let A be the Zariski closure of hξi. Let T be the indeed if V ⊂ gQ is a maximal torus in A. We claim that T is Q-anisotropic: nonzero eigenspace of the maximal Q-split torus in A, then ξ preserves a lattice in V , and hence acts with determinant one, which implies that the eigenspace is for the trivial character. Let D be the maximal R-split torus in T , and let W be the set of vectors in gR fixed by D; since A is abelian, it preserves W . Then the action of TR on W being trivial on D, it factors through (A/D)R, and A/D is unipotent-by-(R-anisotropic), and hence is by distal matrices (i.e., with all eigenvalues of modulus 1). Since ξ has no such eigenvalues, this forces W = {0}. Hence D is as required. Conversely, let T be an R-isotropic, Q-anisotropic torus with no nonzero invari- ant vector in gQ. Fix an identification of gQ with Qd, so that T (Z) makes sense. Then T (Z) is a lattice in T (R), and is Zariski-dense in T . Then g 7→ det(g2 − 1) is a nonzero regular map on T and hence does not vanish on T (Z). Let ξ be any element in T (Z) with det(ξ2 − 1) 6= 0. Since ξ is diagonalizable over R, this im- plies that ξ has no complex eigenvalue of modulus 1. Let φ be the automorphism of GQ induced by ξ. To finish the proof, let us show that some power of φ preserves Γ. Let Γ′ ⊂ GQ be a full lattice containing Γ as a subgroup (of finite index), where full lattice means that log(Γ′) is a Lie subring of gQ. In gQ, there are finitely many lattices M such that [M ∩ Zd : Zd][M ∩ Zd : M] is given. In particular, since Zd is a fixed point of hξi for its action on the set of lattices in Qd, the orbits of the action of GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 29 hξi on the set of lattices in Qd are finite. Thus some positive power ξm stabilizes log(Γ′). Thus φm preserves log(Γ′) Now we can perform the same argument in the set of lattices of GQ, so that some positive power φmm′ preserves Γ. (The latter argument is a variation of the proof of Lemma 1.1 and Corollary 1.2 in [Dan].) (cid:3) This characterization shows in particular that being Anosov, for finitely gen- erated torsion-free nilpotent groups, is invariant under taking subgroups and overgroups of finite index; this was proved in [Dan], but the above convenient criterion was not explicitly written there. Thus we can define Anosov for an arbi- trary Q-subgroup H of GLd by the existence of a subtorus in H as in Proposition 4.1. Example 4.2. There exist two lattices in the same simply connected nilpotent Lie group, one being Anosov and the other not. Namely, consider the 3-dimensional Heisenberg group H3(R) and G(R) = H3(R) × H3(R). Then G admits several Q-forms, one of which is the obvious one, whose Q-points form H3(Q) × H3(Q); this one is not Anosov, because it can be checked that the restriction of its automorphism group to the 2-dimensional center is a Q-split torus, and thus any Q-anisotropic torus acts as the identity on the center. On the other hand, all other Q-forms are Anosov: each such form admits H3(Q[√n]) as group of Q-points, for some square-free integer n ≥ 2, and a standard argument [Lau] provides the desired Anosov automorphism: its Lie algebra can be written as q = h3(Q[√n]), and can be endowed with a Carnot grading q = q1 ⊕ q2 (as a 3-dimensional Lie algebra over Q[√n]). Then we have a torus of Q[√n]-linear automorphisms, acting by multiplication by ti on qi, for t ranging over the group of elements of norm 1 in Q[√n]∗. Now view this as a torus of Q-linear automorphisms of the 6-dimensional Lie Q-algebra q. This is a Q-anisotropic torus satisfying the conditions of Proposition 4.1. Also, this example shows that the property that G admits at least one Anosov lattice cannot be read on the complexification of G: indeed as just observed, H3(R)×H3(R) admits an Anosov lattice, but it has the same complexification as the real group H3(C), while the latter admits no Anosov lattice, because its group of automorphisms of determinant 1 acts on the center through an R-anisotropic torus. 4.2. Existence of an invertible grading is not invariant under taking extensions. Proposition 4.3. There exists a finite-dimensional real nilpotent Lie algebra h thatv admits an invertible Z-grading, and admitting a rational form with no such grading. Proof. In [Der, §4], J. Der´e constructs a finite-dimensional real nilpotent Lie alge- bra as a suitable quotient of the free 6-nilpotent real Lie algebra on 4 generators. 30 YVES CORNULIER Simple computations (see Lemma 4.4) show that the derivation Lie algebra of this Lie algebra is solvable and acts on the 4-dimensional abelianization as those diagonal matrices of trace 0. In particular, any maximal torus of the automor- phism group is 3-dimensional, R-split, and induces a grading in Z3 of the Lie algebra. Der´e checks [Der, Proposition 4.5] that this Lie algebra admits an Anosov rational form. In particular, it admits an invertible Z-grading over C, and hence over R since the maximal tori of automorphisms are R-split. A more careful look at his proof shows that the integral points of the torus of automorphisms defining the rational structure contains a discrete abelian group of rank 3. In particular, this torus (for this given Q-form) has to be Q-anisotropic, and hence so are all Q-defined tori of the automorphism group of this Q-form; in particular, this Q-form has no nontrivial grading in Z. (cid:3) In the above proof, the (real, but the following works over any field K) Lie al- gebra g is defined with 4 generators X1, . . . , X4, killing all commutators of length ≥ 7, all [Xi, [Xj, [Xk, Xℓ]]] whenever {i, j, k, ℓ} = {1, 2, 3, 4}, as well as the four elements obtained as cyclic conjugates of [X2, X4]−[[X4, [X3, X4]], [[X2, [X1, X2]]] (note that there are only two such conjugates up to sign). Lemma 4.4. The derivations of g act on g/[g, g] as diagonal matrices in the given basis. Proof. It is enough to prove this for g/g(5), which is defined with the generators X1, . . . , X4, by killing the elements all commutators of length ≥ 5, killing [X1, X3] and [X2, X4] and all [Xi, [Xj, [Xk, Xℓ]]] whenever {i, j, k, ℓ} = {1, 2, 3, 4}. Since we only kill monomials, all diagonal matrices on (X1, X2, X3, X4) extend to derivations of g/g(5). The corresponding 4-dimensional diagonal torus defines a grading of g/g(5) in Z4 for which, denoting by (ei) the canonical basis of Z4, Xi has degree ei. In turn, this defines a grading on the Lie algebra d of derivations of g. Let us check that dej−ei = 0 for all i 6= j (this immediately entails the result). If f ∈ dej−ei, then f (Xi) ∈ Kej, and f (Xk) = 0 for k 6= i. If by contradiction f 6= 0, we can assume up to multiplication that f (Xi) = Xj. Let k, ℓ be the two other elements, and assume that ℓ 6= j + 2 mod 4. Then 0 = f ([Xk, [Xj, [Xℓ, Xi]]]) = [Xk, [Xj, [Xℓ, Xj]]] We get a contradiction by observing that [Xk, [Xj, [Xℓ, Xj]]] is a nonzero el- ement of g/g(5). The latter fact holds because the only elements of degree 2ej + ek + eℓ in the ideal generated by the given relators are the scalar mul- tiples of [Xj, [Xj, [Xk, Xℓ]]] if k = ℓ + 2 mod 4, and the plane generated by [Xj, Xℓ, [Xj, Xk]]] and [Xℓ, Xj, [Xj, Xk]]] if j = k + 2 mod 4). (cid:3) Der´e directly checks that diagonal derivations of g have trace 0 in the abelian- ization, which, unlike the lemma, does not hold for the Carnot Lie algebra g/g(5). GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 31 5. Cohopfian nilpotent groups 5.1. Grading associated to a rational torus. Let A ⊂ GLd = GL(V ) a Q-defined abelian subgroup. Hence its identity component A◦ decomposes as AsAaU, the almost product of its maximal Q-split and Q-anisotropic subtori, and of its unipotent radical U. Then A canonically defines a grading of V in the group of Q-defined multiplicative characters X(As), where Vχ = {v ∈ V : ∀g ∈ As, g(v) = χ(g)v}. The weights of V are by definition those χ such that Vχ 6= {0}. Beware that the projection A◦ → As is not well-defined (because of the possi- ble nontrivial finite intersection As ∩ Aa), and hence that characters of As do not all define characters of A. Still, we consider a projection map A◦(C) → As(C), denoted g 7→ g, such that gg−1 ∈ (UAa)(C). Then g is only determined up to multiplication by an element of the finite group (As ∩ Aa)(C). As a consequence, if g ∈ A(C), then χ(g) ∈ C∗ is only defined up to multiplication by some root of unity. In particular, its modulus χ(g) is a well-defined continuous homomor- phism, denoted χ, from A◦(C) to the group R+ of positive real numbers. The latter extends to a unique continuous homomorphism from A(C) to the group of positive real numbers, again denoted χ (the uniqueness is clear and the existence follows from the injectivity of R as an abstract Z-module). Now let ξ be a fixed element of A(Q). Then we define χ ∈ X(As) to be non-negative (resp. positive, resp. distal) if χ(ξ) ≥ 1 (resp. χ(ξ) > 1, resp. χ(ξ) = 1). We define V≥0, V>0, V≈0 as the sum of Vχ where χ ranges over non-negative, resp. positive, resp. distal weights. (All these notions are relative to the choice of ξ, and do not change if ξ is replaced by a positive power of itself.) In Qd, by lattice we mean any subgroup isomorphic to Zd. Given a group G acting on a set X we say that g ∈ G stabilizes Y ⊂ X if gY ⊂ Y and preserves Y if gY = Y . Lemma 5.1. Let ξ be an element in GLd(Q). • ξ stabilizes some lattice if and only if there is a bound on denominators • ξ preserves some lattice if and only if there is a bound on denominators in the matrices ξn for n ≥ 0 in the matrices ξn for n ∈ Z. Proof. If ξ stabilizes (resp. preserves) a lattice, then up to conjugation we can suppose that this matrix is integral (resp. in GLd(Z)) and hence there is a bound on the denominators as stated. Conversely, if there is a bound on denominators of ξn for n ≥ 0 (resp. for n ∈ Z), the subgroup generated by the ξnZd for n ≥ 0 (resp. for n ∈ Z) is a lattice, and is stabilized (resp. preserved) by ξ. (cid:3) Now consider the above grading, written additively, so that V0 is the set of vectors fixed by As. Lemma 5.2. If the Zariski closure of hξi contains A◦, then V≈0 = V0. 32 YVES CORNULIER Proof. It is no restriction to assume that ξ ∈ A◦. Let χ be a distal weight and let us show that χ = 0. Then AaU acts on Vχ with determinant 1. For g ∈ A◦, write g = gg−1. Then g acts on Vχ by multiplication by χ(g), while g acts on Vχ with determinant 1. In particular, if δ = dim(Vχ), the determinant of the action of A◦ on Vχ is given by g 7→ χ(g)δ, which has to be a rational number whenever g ∈ A◦(Q). Since χ is distal, we have 1 = χ(ξ) = χ( ξ). Thus χ( ξ)δ is a rational number of modulus 1 and hence is equal to ±1, and hence χ( ξ)2δ = 1. Since χ is also distal with respect to ξn for all n ∈ Z, this shows that χ( ξn)2δ = 1 for all n ∈ Z. In other words, for all n ∈ Z, we have det(ξnVχ)2 = 1. By Zariski density, we deduce that for all g ∈ A◦, we have det(gVχ)2 = 1. On the other hand, if g ∈ As, we have det(gVχ)2 = χ(g)2δ. Hence χ(g)2δ = 1 for all g ∈ As. By connectedness of As, we deduce χ(g) = 1 for all g ∈ As, which in additive notation means that χ = 0. (cid:3) Lemma 5.3. • if ξ stabilizes a lattice then V = V≥0 • if ξ preserves a lattice then V = V≈0 If ξ stabilizes a lattice Λ, the converse of the second implication holds; more precisely if V = V≈0 then ξ preserves Λ. Proof. Let us show the implications. It is no restriction to assume that V = Vχ for some χ ∈ X(As). If ξ preserves a lattice then det(ξ) = ±1. Thus det( ξ) = ±1. Since ξ acts by scalar multiplication by χ( ξ), the latter is a root of unity and hence χ(ξ) = 1, which means by definition that χ is a distal weight. Similarly, if ξ stabilizes a lattice, then det(ξ) is a nonzero integer, and hence det( ξ) = det(ξ) is a positive integer and hence is ≥ 1, which implies that χ(ξ) ≥ 1, which means by definition that χ is a non-negative weight. Now let us show the partial converse. Assume that V = V≈0 and that ξ stabilizes the lattice Λ. Again, we can suppose that V = Vχ; then χ is distal. The assumption implies that det(ξ) is a nonzero integer m. On the other hand, det(ξ) = det( ξ), and ξ is the scalar multiplication by χ( ξ). Hence χ( ξ)δ = m, where δ = dim(Vχ). Since χ is distal, χ( ξ) = 1. It follows that m = ±1. Since ξ stabilizes the lattice Λ and det ξ = ±1, we deduce that ξ preserves Λ. Remark 5.4. The converse of both implications of Lemma 5.3 are false. For instance, let ξ be the companion matrix of X 2 + (1/2)X + 1. Then ξ is irreducible on Q2, and det(ξ) = 1; hence the Zariski closure A of hξi is a Q-anisotropic torus; thus V = V0 for this choice of A. But the spectral radius of ξ and ξ−1 in the 2-adic field Q2 is greater than 1. In particular, (ξn)n≥0 has unbounded denominators (actually, large powers of 2) and thus ξ stabilizes no lattice. Proposition 5.5. Assume that the Zariski closure of hξi contains A◦, and that ξ stabilizes a lattice Λ in Qd. Let A be the Zariski closure of hξi. Then Tn≥0 ξn(Λ) is equal to Λ ∩ V0 and is preserved by ξ. Moreover, if Λ′ is another lattice in Qd, then Tn≥0 ξn(Λ′) is a lattice in V0. (cid:3) GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 33 Proof. Let ξ′ be a positive power of ξ contained in A◦. If the result is proved for ξ′ then it immediately follows for ξ. Hence we can suppose that ξ ∈ A. That Λ ∩ V0 is preserved by ξ follows from the partial converse statement in Lemma 5.3. Hence it remains to check that T ξm(Λ) ⊂ V0. It is enough to assume that V = Vχ for χ not distal and show that Λ′ = Tn≥0 ξn(Λ) = {0}. Otherwise, let W be the subspace spanned by Λ′; then W = Wχ and since χ is not distal, it is positive, i.e., χ( ξ) > 1. Thus det(ξ) = det( ξ) = χ( ξ)dim(W ) > 1 since dim(W ) > 0; this is a contradiction since ξ preserves the lattice Λ′ of W ′. For the statement about another lattice, it is clear if Λ′ = rΛ for some nonzero rational number r, and the general case follow since any lattice Λ is contained in (1/m)Λ and contains mΛ for some positive integer m. (cid:3) Now let again A and ξ be as above. We have a grading of V in X(As); through the homomorphism χ 7→ log(χ(ξ)) from X(As) to R, we obtain a new grading of V in R, which we call the absolute grading of V associated to (A, ξ). Actually, since given χ, the value of χ(ξ) only depends on the restriction of χ to the Zariski closure of hξi, we see that this absolute grading only depends on ξ (and not on the subgroup A containing ξ), so we call it the absolute grading of V associated to ξ. To avoid confusion here, we denote it by (V ♯ 0 = V≈0; in particular if ξ generates a Zariski-dense subgroup of A then V ♯ 0 = V0 (Lemma 5.2). It follows from the definition that all weights in the absolute grading are non-negative real numbers if and only if V = V≥0, which holds if ξ stabilizes a lattice, by Lemma 5.3. r = V ♯ r (ξ) to emphasize ξ, we can note that V ♯ r )r∈R. Then V ♯ If we write V ♯ r (ξm) = V ♯ mr(ξ). In particular, V ♯ 0 does not change if ξ is replaced by a nontrivial power. Lemma 5.6. If ξ is an automorphism of a finite dimensional algebra g over Q, then the absolute grading defined by ξ is an algebra grading. Proof. If A is the Zariski closure of hξi, then A ⊂ Aut(g), and then the grading defined by As is an algebra grading. Since the absolute grading of ξ is a projection of this grading by a group homomorphism, it is also an algebra grading. (cid:3) Proposition 5.7. Let Γ be a torsion-free finitely generated nilpotent group, G its rational Malcev completion, and g the Lie algebra of G. Let φ be an injective endomorphism of Γ, and Φ the automorphism of G extending φ. Let ξ be the r with the corresponding absolute grading. Let G{1} = G{1}(φ) the rational subgroup of G corresponding to g♯ corresponding automorphism of the Lie algebra g, and endow g = Lr∈R g♯ 0. Then Tn≥0 φn(Γ) is equal to Γ ∩ G{1}. Proof. We first check that Tn≥0 φn(Γ) ⊂ G{1}. The subset log(Γ) is contained in some lattice L of g, and by Proposition 5.5, we have Tn≥0 ξn(L) ⊂ g0 = g♯ Hence Tn≥0 φn(Γ) ⊂ G{1}. {0}. 34 YVES CORNULIER On the other hand, since φ(Γ ∩ G{1}) ⊂ Γ ∩ G{1} and the restriction of ξ to g0 has determinant ±1, this inclusion is an equality. Accordingly Γ ∩ G{1} is contained in Tn≥0 φn(Γ). 5.2. A lemma for constructing endomorphisms of lattices. By unipotent Q-group, we mean the group of Q-points of some unipotent linear algebraic Q- group, or equivalently the group obtained from a finite-dimensional nilpotent Lie algebra over Q when it is endowed with the group law defined by Baker-Campbell- Hausdorff formula. The dimension of a Q-group is defined in the obvious way. By lattice in a Q-group G, we mean any finitely generated subgroup of Hirsch length equal to dim(G); every unipotent Q-group admits lattices, and they are all commensurate to each other. Also note that if GR is the real completion of G, then lattices of G are the lattices of GR (in the usual sense) that are contained in G. (cid:3) Lemma 5.8. Let g be a finite-dimensional nilpotent Lie algebra over Q, graded in the non-negative integers N. For t ∈ Q, let δ(t) be the automorphism of g defined as the multiplication by tn on gn. Let G be the corresponding unipotent Q-group and also denote by δ(t) the corresponding automorphism of G. Let Γ be a lattice in G. Then there exists an integer m ≥ 2 such that δ(m) stabilizes Γ, in the sense that δ(m)(Γ) ⊂ Γ. More precisely, there exists k0 such that every integer m ∈ k0N + 1 satisfies this condition. Proof. It is convenient to identify the Lie group to the Lie algebra through the exponential map; thus the group law ∗ is given by the Baker-Campbell-Hausdorff (BCH) formula, and the group powers are given by x∗m = mx for all m ∈ Z. Moreover the identification of δ(t) : g → g and δ(t) : G → G is coherent with this identification. Fix a basis (ei) of g, with ei homogeneous of degree ni; we can multiply these basis elements by large enough integers so as to ensure that all structural con- stants of the Lie algebra are integers. Using this basis, identify g with Qd. There exist integers k, k′ ≥ 1 such that k′Zd ⊂ Γ ⊂ k−1Zd. Fix g ∈ g and an integer m ≥ 0. We need to describe g∗ (δ(m + 1)(g))∗−1 = g∗ (−δ(m+1)(g)); it is given by the BCH-formula as a certain sum g−δ(m+1)(g)+h, where h is a Z[1/s]-linear combination of iterated brackets [c1, . . . , cℓ, [g, δ(m + 1)(g))] . . . ], where 0 ≤ ℓ ≤ d−2 (actually ℓ is at most the nilpotency length minus 2), and s is a common denominator for terms in the BCH-formula of degree at most the nilpotency length of g (we can choose s = d! although it is far from optimal), and the ci are in {g, δ(m + 1)(g)}. We write g = Pn≥0 gn, according to the decomposition g = g0 ⊕ (cid:0)Ln≥1 gn(cid:1). Then GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 35 δ(m + 1)(g) = Mn≥0 (1 + m)ngn = g +Mn≥0 ((1 + m)n − 1)gn. m (1+m)n−1 Since m divides ((1 + m)n − 1) for all n ≥ 0, we can write δ(m + 1)(g) = g + mg′. gn. Hence g−δ(m+1)(g) = −mg′, and [g, δ(m+1)(g)] = Here g′ = Ln≥1 m[g, g′]. Now assume that g ∈ Γ, so that g belongs to k−1Zd. It also follows that all gn belong to k−1Zd, and hence δ(m + 1)(g) and g′ belong to k−1Zd as well. Hence [c1, . . . , cℓ, [g, δ(m + 1)(g))] . . . ] belongs to mk−dZd. Therefore, if we assume that skdk′ divides m, we deduce that g ∗ (δ(m + 1)(g))∗−1 belongs to k′Zd and hence belongs to Γ. Since g ∈ Γ as well, it follows that δ(m + 1)(g) ∈ Γ. Hence we get the conclusion with k0 = skdk′. (cid:3) 5.3. The main theorem about cohopfian properties. Theorem 5.9. Let Γ be a torsion-free finitely generated nilpotent group, G its rational Malcev completion, and g the Lie algebra of G. Let k be the uncontracted dimension of g (Definition 3.23). Then the smallest possible Hirsch length for if g = g[0] ⋉ g[+] is a contractive decomposition of g and G[0] is the subgroup of G corresponding to g[0], then there exists an injective endomorphism φ of G such Tn≥0 φn(Γ), when φ ranges over injective endomorphisms of Γ, is k. Actually, that T φn(Γ) = G[0] ∩ Γ. of Tn≥0 φn(Γ) is, in the notation of Proposition 5.7 and using this proposition, Proof. Let φ be an injective endomorphism of Γ. Then the Hirsch length h(Γ, φ) the dimension of g♯ 0. Let M be the subgroup of R generated by weights of g in the absolute grading defined by ξ. This is a finitely generated torsion-free abelian group; hence it has a basis. Then by a small perturbation of those basis elements, we find a homomorphism M → R mapping all positive weights of g to positive rational numbers, and then after multiplication we find a homomorphism M → R mapping all weights to positive integers. Thus we have a Lie algebra grading of g in N, denoted (gn)n∈Z, for which g0 = g♯ 0. This implies that there exists a contractive decomposition g = g[0] ⋉ g[+] of g such that g0 ⊃ g[0]. Hence h(Γ, φ) = dim(g♯ To show that k is achieved, let g = g[0] ⋉ g[+] be a contractive decomposition of g: then there exists a Lie algebra grading of g in N such that g0 = g[0]. Define δ(t) as in Lemma 5.8, namely to be the multiplication by ti on gi. Then this lemma asserts that there exists an integer m ≥ 2 such that δ(m) maps log(Γ) into itself. Thus the corresponding automorphism ∆m = exp ◦δ(m) ◦ log of G maps Γ into itself. Then Tn≥0 ∆n m(Γ) = Γ ∩ G[0]: although this is a particular case of Proposition 5.7, this can be seen directly since δ(m) is diagonalizable over Q. Hence h(Γ, ∆m) = k. (cid:3) 0) ≥ k. 36 YVES CORNULIER Corollary 5.10. Let Γ be a finitely generated torsion-free nilpotent group and g its rational Lie algebra. Then • Γ is non-cohopfian if and only if g is semi-contractable; • Γ is dis-cohopfian if and only if g is contractable. Proof. If φ is an injective endomorphism of Γ, define h(Γ, φ) as the Hirsch length of T φn(Γ), and k as the uncontracted dimension of g, so the theorem asserts that k = minφ(Γ, φ). Then g is semi-contractable if and only if k < dim(g), and is contractable if and only if k = 0. On the other hand, we have h(Γ, φ) < dim(g) if and only if φ is non-surjective (observing that T ξn(Γ) cannot be a proper subgroup of finite index), and h(Γ, φ) = 0 if and only if T φn(Γ) = {1}. Thus the theorem implies the corollary. Proof of Theorem 1.10. (ii)⇒(iii) is clear, since G admits a lattice. (iii)⇒(i). Let Γ be a non-cohopfian lattice. Let GQ be the group of Q-points for the Q-form defined by Γ, and gQ its Lie algebra. Then by Corollary 5.10, gQ is semi-contractable. Thus g ≃ R ⊗Q gQ is semi-contractable. (i)⇒(ii) Let Γ be a lattice. Let GQ be the group of Q-points for the Q-form defined by Γ, and gQ its Lie algebra. Then since g ≃ R ⊗Q gQ and g is semi- contractable, by Theorem 3.25 it follows that gQ is semi-contractable. Thus by Corollary 5.10, Γ is not cohopfian. (cid:3) (cid:3) Proof of Theorem 1.11. The proof follows mutatis mutandis the same steps as the previous ones, replacing "non-cohopfian" with "dis-cohopfian" and "semi- contractable" with "contractable". (cid:3) 5.4. Generalities about being weakly dis-cohopfian. For short, call Γ-chain any descending sequence (not necessarily with strict inclusions) of subgroups (Γn) of Γ, all isomorphic to Γ, with Γ0 = Γ. Note that for each such chain, there exists a sequence (φn)n≥1 of injective endomorphisms of Γ such that Γn = φ1 . . . φn(Γ) for all n. Lemma 5.11. Let Γ be a countable group and W a subgroup. Assume that (1) for every injective endomorphism φ of Γ, we have φ(W ) = W , and (2) for every γ /∈ W , there exists an injective endomorphism φ of Γ such that γ /∈ φ(Γ). Then the following two assertions hold (a) for every Γ-chain (Γn) we have W ⊂ T Γn; (b) there exists a Γ-chain (Γn) such that T Γn = W . Proof. (a) is an immediate consequence of (1). Let us prove (b). If W = Γ, there is nothing to do; otherwise, let (gn)n≥1 be a (possibly non-injective) enumeration of Γ r W , and define by induction a Γ-chain (Γn) such that gn /∈ Γi for all n ≥ 1. By definition Γ0 = Γ. Assume n ≥ 1 and Γi is defined for i < n. If gn /∈ Γn−1, we GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 37 choose Γn = Γn−1. Otherwise we apply (2) to Γn−1 and γ = gn to define Γn as the image of an injective endomorphism of Γn−1 whose image does not contain gn. Since all Γn contain W by (1) and no gn belongs to Ti Γi, we obtain Ti Γi = W . Accordingly, (Γn) is a Γ-chain with the required property. (cid:3) Corollary 5.12. A countable group Γ is weakly dis-cohopfian if and only if for every g ∈ Γ r{1} there exists a subgroup of Γ isomorphic to Γ and not containing g. Proof. This is the particular case of Lemma 5.11, when W = {1}, noting that (1) then holds automatically. (cid:3) Remark 5.13. A variant of Lemma 5.11 holds, namely when in both the assump- tions and conclusions, "injective endomorphism" is replaced with "injective en- domorphism with image of finite index" and "Γ-chain" is replaced with "Γ-chain (Γn) such that all Γn have finite index, the proof being the same. 5.5. Weakly dis-cohopfian nilpotent groups. (Refer to §5.4 for the meaning of Γ-chain.) Theorem 5.14. Let Γ be a finitely generated torsion-free nilpotent group, G its rational Malcev completion, and g the Lie algebra of G. Let cni+(G) be the subgroup corresponding to cni+(g). Then for every Γ-chain (Γn), we have Tn Γn ⊃ cni+(G) ∩ Γ, with equality for some choice of (Γn). Proof. Let φ be an injective endomorphism of Γ, φ its unique extension to an automorphism of G, and ξ = log ◦ φ ◦ exp the corresponding automorphism of g. Since ξ preserves cni+(g), φ preserves cni+(G). Hence φ stabilizes cni+(G) ∩ Γ. Since the determinant of the restriction of φ to cni+(g) is ±1, it follows that φ preserves cni+(G) ∩ Γ. Accordingly, cni+(G) ∩ Γ ⊂ Tn Γn. Let us now show the existence statement. Let g = g[0] ⋉ g[+] be a contractive decomposition of g, and G[0] the subgroup corresponding to g[0]. Then by Theo- rem 5.9, there exists an injective endomorphism φ of Γ such that Tn≥0 φn(Γ) is equal to G[0] ∩ Γ. Define W = cni+(G) ∩ Γ. By the previous verification, φ′(W ) = W for every injective endomorphism φ′ of Γ, which is the assumption (1) of Lemma 5.11. Let us check (2) of the same lemma. Fix x ∈ Γ r W . Thus x /∈ cni+(G), that is, log(x) /∈ cni+(g). Thus there exists a contractive decomposition of g such that log(x) /∈ g[0]. Given a non-negative Lie algebra grading of g such that g0 = g[0], Lemma 5.8 provides an injective endomorphism ξ of Γ such that T ξn(Γ) ⊂ exp(g0). In particular x does not belong to the image of ξn for some n. (Γn) we have T Γn = W . This means that (2) holds, and hence Lemma 5.11 implies that for some Γ-chain (cid:3) Corollary 5.15. Let Γ be a finitely generated torsion-free nilpotent group, and g its rational Lie algebra. Then Γ is weakly dis-cohopfian if and only if cni+(g) = {0}. 38 YVES CORNULIER Proof of Theorem 1.13. Again, the proof follows mutatis mutandis from that of Theorem 1.10, replacing the use of Corollary 5.10 by Corollary 5.15. (cid:3) Example 5.16. Let g be one of the four complex 7-dimensional Lie algebras re- ferred to in Remark 3.28. For instance, the Lie algebra g7,1.02 can be defined from a basis (Z1, A2, A3, A4, B5, B6, C7) by the nonzero brackets [Z1, A2] = A3, [Z1, A3] = A4, [A2, A3] = B5, [A2, A4] = B6, [A2, B5] = C7, [A2, B6] = C7, [Z1, B5] = B6, [A3, B5] = −C7. (The grading on N for which Z1 has degree 0, Ai has degree 1, Bi has degree 2, and C7 has degree 3, is a maximal grading.) These Lie algebras being defined using rational coefficients, we can find a lattice Γ in the real group. Then cni+(g) = {0} but g is not contractable. Thus Γ is weakly dis-cohopfian but not dis-cohopfian. 6. Systolic growth and geometry of lattices in nilpotent groups 6.1. Generalities on the systolic growth. If f, g are non-negative functions defined on the integers or reals, we say that f (cid:22) g, or that f is asymptotically bounded above by g, if there exists a positive constant C such that f (x) ≤ Cg(Cx) + C for all x large enough. If f (cid:22) g (cid:22) f , we say that f, g are asymptot- ically equivalent and write f ≃ g. Also, we write f ∼ g if f = O(g) and g = O(f ) (for instance, 2n ≃ 3n but 2n ≁ 3n). We first generalize the notion of systolic growth from the introduction to com- pactly generated locally compact groups. Definition 6.1. Let G be a locally compact group. We say that G is residually systolic if for every compact neighborhood S of 1, there exists a lattice Γ in G such that Γ ∩ S = {1}. When G is discrete, residually systolic just means residually finite. In gen- eral, residually systolic implies the existence of lattices, and in particular implies unimodular. If G admits a residually finite lattice, then it is residually systolic. Let now G be a locally compact, compactly generated group, endowed with the word length · with respect to some compact generating subset, or any equivalent length. We fix a left Haar measure on G, so that every discrete subgroup has a well-defined covolume (possibly infinite). Definition 6.2. If Γ ⊂ G, we define its systole as sys(Γ) = inf{γ : γ ∈ Γ r {1}}, with inf ∅ = +∞. The systolic growth of G is the function σ = σG,· mapping r to the infimum of covolumes of discrete subgroups of G of systole ≥ r (+∞ if there is no such subgroup). GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 39 Note that σ(r) < ∞ for all r if and only if G is residually systolic, i.e., G admits lattices of arbitrary large systole. The growth is a lower bound for the systolic growth: if v(r) is the volume of the open r-ball, then σ(2r) ≥ v(r). Note that an alternative definition would restrict to cocompact lattices; if we focus on nilpotent groups or discrete groups, of course this makes no difference. Definition 6.3. If Γ ⊂ G, define its normal systole as sys⊳(Γ) = inf{gγg−1 : γ ∈ Γ r {1}, g ∈ G}. The normal systolic growth of G is the function σ⊳ = σ⊳ G,· mapping r to the infimum of covolumes of discrete subgroups of G of systole ≥ r (+∞ if there is no such subgroup). Obviously, we have sys⊳ ≤ sys, and σ⊳ ≥ σ. For standard reasons, the asymp- totic behaviors of σ and σ⊳ do not depend on the choice of word length and Haar measure. Lemma 6.4. The ≃-asymptotic growth of the normal systolic growth is a com- mensurability invariant for finitely generated groups. Proof. Suppose Γ′ ⊂ Γ has finite index. We endow Γ with a word length · , and endow Γ′ with the restriction of this word length, which is equivalent to the word length on Γ′, and define the systoles with these lengths. sys⊳ If Λ is a finite index subgroup in Γ, then [Γ′ : Λ∩Γ′] ≤ [Γ : Λ] and sys⊳ Γ′(Λ∩Γ′) ≥ Γ (Λ). This implies that σ⊳ Let us now prove an inequality in the other direction. Write Γ = F Γ′ with F Γ′ ≤ σ⊳ Γ . finite. If g ∈ Γ, define [g]Γ = inf h∈Γ hgh−1, and [g]Γ′ = inf h∈Γ′ hgh−1. Then [g]Γ′ − k ≤ [g]Γ ≤ [g]Γ′, k = 2 sup h∈F h It follows that for every finite index subgroup Λ of Γ′, we have sys⊳ Γ′(Λ) − k ≤ sys⊳ Γ (Λ) ≤ sys⊳ Γ′(Λ), Γ (n − k) ≤ [Γ : Γ′]σ⊳ and this implies σ⊳ we deduce σ⊳ Γ ≃ σ⊳ Γ′. Γ′(n) for all n ≥ k, thus σ⊳ Γ (cid:22) σ⊳ Γ′. Finally (cid:3) 6.2. The main theorem on systolic growth and first part of the proof. Theorem 6.5. Let G be a connected unipotent Q-group. Let δ = Xi≥1 i dim(G(i)/G(i+1)) be the growth rate of the simply connected nilpotent Lie group GR. Let gQ and gR = R ⊗Q gQ be the corresponding Lie algebras. Let σ and σ⊳ be the systolic growth and normal systolic growth of GR. Equivalences: (i) the Lie algebra gR is Carnot over R; (i+) the Lie algebra gQ is Carnot over Q 40 YVES CORNULIER (ii−) either G = {1}, or every lattice in GQ (i.e., lattice of GR contained in GQ) admits an injective endomorphism φ such that, for some integer m ≥ 2, we have sys(Im(φn)) ∼ sys⊳(Im(φn)) ∼ mn and covol(Im(φn)) = λmδn for some λ > 0 and all n ∈ N; (ii) every lattice in GQ has systolic growth ≃ nδ; (iii) some lattice in GQ has systolic growth ≃ nδ; (iv−) σ(r) ≃ rδ; (iv) lim inf σ(r)/rδ < ∞; (v) every lattice in GQ has normal systolic growth ≃ nδ; (vi) GR has normal systolic growth ≃ rδ. Note that Theorem 1.8 holds as a consequence. We are going to prove the equivalences of all properties except (vi) by proving the cycle of implications comments, which follow here: (i) •⇒(i+) •⇒(ii−) ◦⇒(v) ◦⇒(ii) ◦⇒(iii) ◦⇒(iv−) ◦⇒(iv) •⇒(i), which along with the implications (ii) ◦⇒(vi) ◦⇒(iv−) proves all the equivalences. Here •⇒ means an implication requiring a proof, that is part of the work of the paper, while ◦⇒ means an implication which are trivial or only require a few • (ii−)⇒(v) and (ii−)⇒(vi): formally speaking, (ii−) implies that nδ is an asymptotic lower bound for the normal systolic growth of both G and its lattices. Since conversely the growth of both G and its lattices is ≃ nδ and is an asymptotic lower bound for the normal systolic growth, this yields the conclusion. • (v)⇒(ii) and (vi)⇒(iv−): just use that the systolic growth is asymptoti- cally trapped between the growth and the systolic growth; similarly, for (iii)⇒(iv−), just use that the systolic growth of GR is asymptotically trapped between the growth of GR and the systolic growth of any of its lattices. be equivalent to the assumption that g is definable over Q. • (ii)⇒(iii): this follows from the existence of a lattice, which is known to It remains to consider the implications •⇒. Since (i)⇒(i+) is part of Theorem 3.15, the only two remaining are (i+)⇒(ii−) and (iv)⇒(i). Proof of Theorem 6.5, (i+)⇒(ii−). Write the Carnot grading as gQ = L(gi)Q. Write g = gR and gi = (gi)R. Fix norms on the gi, and define, for x = P xi ∈ g = L gi, the "length" ℓ(x) = supkxik1/i. By Guivarch's estimates [Gui], the word length on G and on its lattices are equivalent to ℓ ◦ log. Therefore in the sequel of the proof, we use · to denote ℓ ◦ log, although it is not necessarily a length (it may fail to be sub-additive, but this does not matter), and use it in the definition of systole. If x ∈ g r {0}, write i(x) = min{i : xi 6= 0}. We can assume that g 6= {0}. Let Γ be a lattice in GQ, and define Γ′ = log(Γ). Define δ(m) as the automorphism of g (and gQ) defined to be the multiplication GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 41 by mi on gi. By Lemma 5.8, there exists an integer m ≥ 2 such that δ(m) maps Γ′ into itself. Let φ = exp ◦δ(m) ◦ log be the corresponding automorphism of G. Then φ(Γ) has index mδ in Γ, and it follows that covol(φn(Γ)) = mnδcovol(Γ) for all n ∈ Z. It remains to estimate the systole and normal systole of φn(Γ). We have ℓ(δ(m)h) = mℓ(h) for all h ∈ g, and thus sys(φn(Γ)) = mnsys(Γ). This also yields sys⊳(φn(Γ)) ≤ mnsys(Γ). To obtain a lower bound for sys⊳(φn(Γ)), con- sider any g ∈ Γ′ r {0} and write i = i(g), and define πi as the projection on gi. Then for every h ∈ GR, we have πi(hgh−1) = πi(g) (we identify the group and the Lie algebra in order to avoid cumbersome notation), and thus ℓ(hgh−1) ≥ kgik1/i. Thus for any n ≥ 0, any g ∈ Γ′ r {0}, we have ℓ(hφn(g)h−1) ≥ mnkgik1/i. If r{0} kgi(g)k1/i(g), then k > 0 (the infimum is attained, by an we define k = inf g∈Γ′ easy argument using that the projection of Γ on G(i)/G(i+1) is discrete for all i), and we thus have ℓ(hφn(g)h−1) ≥ kmn for all h ∈ GR and g ∈ Γ′ r {0}. Thus sys⊳(φn(Γ)) ≥ kmn for all n ∈ N. (cid:3) Before proceeding to the last and most difficult implication (iv)⇒(i) in §6.3, we provide a few auxiliary results. Corollary 6.6. Let Γ be a finitely generated, virtually nilpotent group, of growth ≃ nδ. Suppose that the rational Lie algebra of some/any torsion-free nilpotent finite index subgroup is Carnot. Then the systolic growth and normal systolic growth of Γ are ≃ nδ. This follows from the theorem (namely the already-proved implication (i+)⇒(ii)), along with the commensurability invariance of the systolic and normal systolic growth, the latter being checked in Lemma 6.4. In general, let us provide an upper bound on the systolic (and normal systolic growth) of an arbitrary finitely generated torsion-free nilpotent group. Proposition 6.7. Let Γ be a finitely generated torsion-free nilpotent group, of growth ≃ nδ and nilpotency length c. Let g be its rational Lie algebra. Then the systolic and normal systolic growth of Γ are (cid:22) nD, where D = c dim([g, g]) + dim(g/[g, g]) ≤ c dim(g). Proof (sketched). Let g1 be a supplement subspace of [g, g] in g, and fix a basis of g adapted to the decomposition g = g1 ⊕ [g, g], such that the Lie algebra constants n as g1(nZ) ⊕ [g, g](ncZ), are divisible by c!. This implies that if we define Λ′ then by the Baker-Campbell-Hausdorff formula (whose denominators divide c!), Λn = exp(Λ′ n) is a subgroup (hence a lattice) in G. The index of Γn is nD and its systole and normal systole (in G) are ∼ n. Hence Γn ∩ Γ has index ≃ nD and normal systole ∼ n. This proves that the systole and normal systole of Γ are (cid:22) nD. Note that this upper bound is optimal when c ≤ 2, but far from optimal when c > 2 and g is Carnot. I do not know if it is close to optimal in the "worst" cases. (cid:3) 42 YVES CORNULIER 6.3. The geometric part of the proof. We now complete the proof of Theorem 6.5, by proving (iv)⇒(i). We note that here no particular choice of Q-structure on g is relevant. Hence we use the notation as in Theorem 1.8, and will assume (iv) (which is equivalent to (iv of Theorem 6.5)), namely that G admits a sequence (Γn) of lattices with systole un → ∞ and covolume (cid:22) uδ n, and aim at proving that the real Lie algebra g is Carnot. We need some further definitions pertaining to the geometry of lattices. For the moment, let G be a group endowed with a left-invariant distance. Let Br be the closed r-ball in G. Definition 6.8. Given a subgroup Γ of G, define its packing packG(Γ) to be supg∈G d(g, Γ) ∈ [0,∞]. Also define its generating radius gerG(Λ) as the infimum of the r such that Λ is generated by Λ ∩ Br. All these definitions are understood with the usual conventions: the infimum of the empty set and the supremum of an unbounded subset of positive reals are +∞. Lemma 6.9. Let G be a locally compact group with a continuous left-invariant geodesic distance d and let Γ be a cocompact lattice. Then gerG(Γ) ≤ 2packG(Γ); moreover any element γ ∈ Γ with d(γ, 1) ≤ n is a product of n elements of B2packG(Γ)+1 ∩ Γ. Proof. Fix an integer m ≥ 1. Given γ ∈ Γ with d(γ, 1) ≤ n, consider a geo- desic joining γ to 1. On this geodesic choose points 1 = x0, . . . , xmn = γ with d(xi−1, xi) ≤ 1/m for all i = 1 . . . k. There is γi in Γ with d(xi, γi) ≤ packG(Γ) for all i, where we choose γ0 = 1 and γk = γ. Hence γ = Qk i−1γi, and d(1, γi−1γi) ≤ 2packG(Γ) + 1/m for all i. Hence gerG(Γ) ≤ 2packG(Γ) + 1/m; since this holds for all m we deduce gerG(Γ) ≤ 2packG(Γ); on the other hand taking m = 1 in the above argument shows that γ is a product of n elements from B2packG(Γ)+1 ∩ Γ. (cid:3) Lemma 6.10. Let V be a Euclidean space and Λ a lattice (in this case, gerV (Λ) is often denoted λdim(V )(Γ) in the literature). Then i=1 γ−1 2 dim(V ) packV (Λ) ≤ gerV (Λ) ≤ 2packV (Λ). Proof. The right-hand inequality is borrowed from Lemma 6.9. For the left-hand inequality, V has a basis (ei) with ei ∈ Λ and keik ≤ gerV (Λ). If x ∈ V , we write x = P αiei; hence we can decompose x = w + y with w ∈ Λ and y = P βiei with βi ≤ 1/2 for all i. Hence kyk ≤ dim(V )gerV (Λ)/2, whence packV (Γ) ≤ dim(V )gerV (Λ)/2. (cid:3) Assume now that G is a simply connected nilpotent Lie group, endowed with a left-invariant Riemannian metric. Guivarch [Gui] established that the growth GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 43 rate of G and of its lattices is ≃ nδ, where δ is characterized in terms of the lower central series by δ = Pi≥1 i dim(g(i)). The arithmeticity of lattices (see [Rag]) implies in particular that for every lattice in G, its projection on G/[G, G] is also a lattice. We endow V = G/[G, G] with the Euclidean metric defined by identifying g/[g, g] with the orthogonal of [g, g]. Let p : G → G/[G, G] be the projection, which is 1-Lipschitz. Let λ be a left Haar measure on G. Lemma 6.11. There exists a constant C (depending only on G and its Rie- mannian metric) such that for every lattice Γ in G and Λ = p(Γ), we have packG(Γ) ≤ CpackV (Λ). Proof. We argue by induction on the nilpotency length c of G. If c = 1 the result is trivial (with C = 1). Otherwise, the c-iterated commutator induces an alternating multilinear form from V c to G(c), and more precisely a surjective linear map F from ΛcV onto G(c). If we endow both V and G(c) with their intrinsic Riemannian (Euclidean) metric, There exists a constant C0 such that i=1 kvik for all v1, . . . , vc ∈ V . F (v1, . . . , vc) ≤ C0Qc Note that F (Λ⊗c) is a lattice in G(c), of finite index in Γ ∩ G(c). Moreover, it is generated by the image of the generators of Λ⊗c, and therefore is generated by elements of norm ≤ C0gerV (Λ)c. Thus, using twice Lemma 6.10, we successively obtain packG(c)(F (Λ⊗c)) ≤ C1gerV (Λ)c for some constant C1 = (dim G(c))C0/2 and then, packG(c)(F (Λ⊗c)) ≤ C2packV (Λ)c with C2 = 2cC1. On the other hand, denote by p′ the projection G → G′ = G/G(c). If Γ′ = p(Γ), we have, by induction, packG′(Γ′) ≤ C3packV (Λ) for some constant C3 depending only on G and its fixed Riemannian metric. Thus if x ∈ G, there exists γ ∈ Γ such that d(p′(x), p′(γ)) ≤ C3packV (Λ). If we lift a minimal geodesic joining 1 to p(x−1γ), we obtain y ∈ G such that p(y) = p(γ−1x) and d(1, y) ≤ C3packV (Λ). Since y−1γ−1x ∈ G(c), there exists γ′ ∈ F (Λ⊗c) ⊂ Γ with dG(c)(γ′−1y−1γ−1x, 1) ≤ C2packV (Λ)c. Here dG(c) is the intrinsic distance of G(c), which by Guivarch's estimates is distorted in such a way that d(w, 1) ≤ C4dG(c)(w, 1)1/c for all w ∈ G(c). Hence, writing s = γ′−1y−1γ−1x, we have d(1, s) ≤ C4C 1/c We have x = γyγ′s = γγ′ys, because γ′ is central. Hence we have 2 packV (Λ). d(ys, 1) ≤ CpackV (Λ); C = C3 + C4C 1/c 2 thus d(x, Γ) ≤ CpackV (Λ) and accordingly packG(Γ) ≤ CpackV (Λ). Lemma 6.12. For every lattice Γ in G with systole ≥ 2r + 1, we have, denoting again Λ = p(Γ) ⊂ V = G/[G, G], the following lower bound on its covolume (cid:3) covolG(Γ) ≥ 2r + 1 packV (Λ)λ(Br) . Proof. Define a possibly finite sequence of cosets Wi of Λ by W1 = Λ, and, assuming W1, . . . , Wi are defined, if d(x,S1≤j≤i Wj) < 2r + 1 for all x ∈ V , then 44 YVES CORNULIER 2r + 1 and we define Wi+1 = x + Λ. stop; otherwise there exists, by connectedness, x ∈ V such that d(x,S1≤j≤i Wj) = Since the Wi are at pairwise distance ≥ 2r + 1, the process stops, say at i = kr. Since for 2 ≤ i ≤ kr every point in Wi is at distance 2r + 1 to a point in Sj<i Wj and every point in V is at distance ≤ 2r + 1 to a point in Si Wi, it follows that every point in V is at distance ≤ kr(2r + 1) of some point in Λ. In other words, packV (Λ) ≤ kr(2r + 1). Fix xi ∈ Wi and lift it to some element gi ∈ G. Define X = S1≤i≤kr xiBr. This is a disjoint union, since the xi are at pairwise distance ≥ 2r + 1. Moreover, the indeed if gibγ = gjb′γ′ with b, b′ ∈ Br and Xγ for γ ∈ Γ are pairwise disjoint: γ 6= γ′ ∈ Γ, then, projecting, we obtain xi − xj + p(b) − p(b′) = p(γ−1γ′) ∈ Λ. Since kp(b) − p(b′)k ≤ 2r, this forces i = j. Thus gi = gj, hence bγ = b′γ′. Hence b−1b′ = γγ′−1 ∈ Γ; since the systole of Γ is ≥ 2r + 1, this implies γ = γ′, contradiction. This proves that the covolume of Γ is at least equal to the volume of X, and hence is ≥ krλ(Br). Combining both inequalities yields the lemma. (cid:3) n, we deduce that packV (Λn) (cid:22) un, proving the claim. Conclusion of the proof of (iv)⇒(i). Let now (Γn) be a sequence of lattices in G, satisfying sys(Γn) ≥ 2un + 1 and covol(Γn) (cid:22) uδ n. Define Λn = p(Γn) as the projection of Γn on V = G/[G, G]. We first claim that we have packV (Λn) (cid:22) un. Indeed, we have, by Lemma 6.12, covol(Γn) ≥ packV (Λn)λ(Bun)/(2un + 1). Since by assumption covol(Γn) ≃ λ(Bun) ≃ uδ Lemma 6.11 combined with the above claim implies that packG(Γn) (cid:22) un, say packG(Γn) ≤ (Cun − 1)/2 (for n large enough). It follows from Lemma 6.9 that Γn is generated by the elements in BCun ∩ Γn, in such a way that for any integer R ≥ 1, any element in the BRun ∩ Γn is a product of at most R elements in BCun ∩ Γn. If we divide the distance in G by un, the lattice Γn endowed with the resulting distance has the property that its systole is ≥ 2 + 1/un and that every element in the R-ball is product of at most R elements in the C-ball, and the packing of Γn in (G, (1/un)d) is bounded independently of n. By Pansu's thesis [Pan1], the (G, (1/n)d) converge in the sense of Gromov- Hausdorff to a (real) Carnot simply connected nilpotent Lie group endowed with a Carnot-Carath´eodory metric, with the same dimension as G (that H is isometric to a Carnot group is due to Pansu; that H inherits the group law as limit of the laws from G is proved in [Cor1]). Denote by BH(r) the closed r-ball in H. Fix any non-principal ultrafilter ω on the positive integers. The metric ul- tralimit Ξ of the sequence (Γn, (1/un)d) is a discrete subset of H, with systole ≥ 2, with the property that any element of Ξ ∩ BH(R) is a product of at most R elements of Ξ ∩ BH(C), for all R ≥ 1, and any element of H is at bounded distance to some element of Ξ. The fact that Ξ is a subgroup follows from the refinement in [Cor1] of Pansu's result mentioned above. Thus Ξ is a lattice in H. GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 45 Recall that a marked group on k generators is a group endowed with a map (called marking) s from {1, . . . , k}, whose image generates the group, and a net (Mi, si) converges to (M, s) if, denoting by Ni (resp. N) the kernel of the unique homomorphism Fk → Mi extending si (resp. Fk → M extending s), we have the convergence 1Ni → 1N pointwise on the set of functions Fk → {0, 1}; see [CG] for more details. The sequence (#(B(Cun) ∩ Γn)) being bounded, let k be an upper bound and for each n, choose a surjective map sn from {1, . . . k} onto B(Cun) ∩ Γn. Since B(Cun) ∩ Γn generates Γn, this provides a marking of Γn. Define s(i) = limn→ω sn(i) ∈ Ξ for 1 ≤ i ≤ k. Then s defines a marking of Ξ: indeed every element of the R-ball in (Γn, (1/un)d) is a product of ≤ R elements in the C-ball; this fact passes to the ultralimit to show that every element in the R-ball of Ξ is a product of ≤ R elements of S, and thus the image of s generates Ξ. then shows that (Γn, sn) tends to (Ξ, s) for the topology of marked groups. A straightforward argument (using that these groups are uniformly discrete) Since Ξ is finitely presented, eventually Γn lies as a quotient of Ξ (by [CG, Lemma 2.2]), in the sense that there exists I ∈ ω such that for all n ∈ I, the kernel of Fk → Γn contains the kernel of Fk → Ξ. Since both Ξ and Γn are torsion-free of the same Hirsch length, we deduce that Γn is isomorphic to Ξ for every n ∈ I. The rigidity of nilpotent lattices [Rag] asserts that if two simply connected nilpotent Lie groups admit isomorphic lattices, then they are isomorphic. It follows that G is isomorphic to H, and hence that G is Carnot (i.e., g is Carnot over R). (Note that the fact that Γn is generated by elements of length (cid:22) un -- and hence bounded length after rescaling -- played a crucial role: otherwise the ultralimit of the (Γn, (1/un)d) could have been of Hirsch length less that that of G, yielding no conclusion.) (cid:3) Appendix A. Cohopfian does not pass to finite index It was asserted in [CM] that it is an open question whether cohopfian is inher- ited by subgroups of finite index; however, it seems that this question is already settled in the negative: a general criterion for being cohopfian, for finitely pre- sented groups with infinitely many ends, with cohopfian vertex groups, was pro- vided by Delzant and Potyagailo. It turns out that this criterion can be applied to virtually free groups. The general theorem being a bit technical, let us provide an easier instance, with a self-contained proof. Proposition A.1. Let A, B be cohopfian groups with Property FA, with a com- mon cohopfian subgroup C. Suppose that C is equal to its normalizer in both A and B. Then the amalgam G = A ∗C B is cohopfian. In particular, there ex- ists some cohopfian group among non-elementary virtually free finitely generated groups, and being cohopfian does not pass to finite index subgroups. 46 YVES CORNULIER Recall that a group has Property FA if each of its actions on a tree fixes a vertex or an edge. For instance, finite groups have Property FA, and this is enough for our purposes. Proof. We can suppose that A 6= C 6= B since otherwise the result is trivial. Let G act on its Bass-Serre tree T , with a fundamental domain consisting of an edge e with stabilizer C and vertices a and b with stabilizers A and B respectively. We observe that the only vertices fixed by C are a and b. Indeed, otherwise it would fix another neighbor of a or b; but the self-normalization assumption rules this out. Let φ be an injective endomorphism of G. Then φ(A) and φ(B) fix vertices a′ and b′ respectively, and in particular φ(C) fixes the segment [a′, b′]. If a′ = b′, we obtain an embedding of G into either A or B, which contradicts that A and B are cohopfian. Otherwise, after possibly composing φ with an inner automorphism, we can suppose that e is contained in [a′, b′]. Then φ(C) fixes e, and hence φ(C) ⊂ C. Since C is cohopfian, we deduce from the previous observation that φ(C) = C. In particular, since φ(C) fixes [a′, b′], we deduce that {a, b} = {a′, b′}. Replacing φ by φ2 if necessary, we have (a, b) = (a′, b′). Thus φ(A) ⊂ A, and φ(B) ⊂ B; by the cohopfian assumption, we have φ(A) = A and φ(B) = B, and the surjectivity of φ follows. The second statement follows by picking, for instance, the double S3 ∗S2 S3, where Sn is the symmetric group on n elements (as a finite group, S3 has Property FA). This group is cohopfian by the first statement, but admits a non-abelian free subgroup of finite index (of index 6), which is not cohopfian. (cid:3) Remark A.2. In contrast, being Hopfian, for finitely generated groups (or more generally groups with finitely many subgroups of each given index) is inherited by finite index subgroups [Hir, Co. 2]. On the other hand, I do not know if the cohopfian property passes to overgroups of finite index, including in the case of finitely generated groups. References [AC] [AOR] [BC] [Bel] [BM] [BS1] [BS2] J. Ancochea, R. Campoamor. Characteristically nilpotent Lie algebras: a survey. Extracta Math. 16 (2001), no. 2, 153 -- 210. S. Albeverio, B. Omirov, I. Rakhimov. Classification of 4-dimensional nilpotent com- plex Leibniz algebras. Extracta Math. 21(3) 197 -- 210 (2006). K. Bou-Rabee and Y. Cornulier. Systolic growth of linear groups. Proc. Amer. Math. Soc. 144 (2016), no. 2, 529 -- 533. I. Belegradek. On co-Hopfian nilpotent groups. Bull. London Math. Soc. 35(6) (2003) 805 -- 811. K. Bou-Rabee and D. McReynolds. Asymptotic growth and least common multiples in groups. Bull. London Math. Soc. 43(6) (2011) 1059 -- 1068. K. Bou-Rabee and B. Seward. Arbitrarily large residual finiteness growth. J. Reine Angew. Math. 710 (2016), 199 -- 204. A. Borel and J-P. Serre. Th´eor`emes de finitude en cohomologie galoisienne. Comment. Math. Helv. 39 (1964), 111 -- 169. [BT] [CG] [CM] [Con] [Cor1] [Cor2] [Dan] [DD] [Der] [DL] [DP] [Fav] [Gra] [Gro] [Gui] [Hir] [Joh] [Lau] [Leg] [Mag] [NP] [Pan1] [Pan2] [Rag] GRADINGS, SYSTOLIC GROWTH AND COHOPFIAN GROUPS 47 A. Borel and J. Tits. Groupes r´eductifs. Publ. Math. IHES 27, 55-150, 1965. C. Champetier and V. Guirardel. Limit groups as limits of free groups. Israel J. Math., 146 (2005) 1 -- 75. A. Cain, V. Maltcev. Hopfian and co-hopfian subsemigroups and extensions. Demon- str. Math. 47 (2014), no. 4, 791 -- 804. B. Conrad. Reductive group schemes. Pages 93-444 in: Autour des sch´emas en groupes. Vol. I. A celebration of SGA3. Lecture notes from the Summer School held at the CIRM, Luminy, August 29 -- September 9, 2011. Panoramas et Synth`eses, 42/43. SMF, Paris, 2014. Y. Cornulier. Asymptotic cones of Lie groups and cone equivalences. Illinois J. Math. 55(1) (2011), 237 -- 259. Y. Cornulier. Commability and focal locally compact groups. (2013) To appear in Indiana Univ. Math. J. arXiv:1306.4194. S.D. Dani. Nilmanifolds with Anosov automorphism. J. London Math. Soc. (2) 18 (1978) 553 -- 559. K. Dekimpe and J. Der´e. Expanding maps and non-trivial self-covers on infra- nilmanifolds. Topol. Methods Nonlinear Anal. 47, Number 1 (2016), 347 -- 368. J. Der´e. Gradings on Lie algebras with applications to infra-nilmanifolds. ArXiv 1410.3713v3 (2016) K. Dekimpe and K.B. Lee. Expanding maps on infra-nilmanifolds of homogeneous type. Trans. Amer. Math. Soc., 2003, 355 (3), pp. 1067 -- 1077. T. Delzant and L. Potyagailo. Endomorphisms of Kleinian groups. Geom. Funct. Anal. 13 (2003) 396 -- 436. G. Favre. Syst`eme de poids sur une alg`ebre de Lie nilpotente. Manuscripta Math. 9 (1973), 53 -- 90. W. de Graaf. Classification of 6-dimensional nilpotent Lie algebras over fields of char- acteristic not 2. Journal of Algebra 309 (2007) 640 -- 653. M. Gromov. Systoles and intersystolic inequalities. Actes de la table ronde de g´eom´etrie diff´erentielle (Luminy, 1992), 291 -- 362, S´emin. Congr., 1, Soc. Math. France, Paris, 1996. Y. Guivarc'h. Croissance polynomiale et p´eriodes des fonctions harmoniques. Bull. Soc. Math. France 101 (1973) 333 -- 379. R. Hirshon. Some theorems on hopficity. Trans. Amer. Math. Soc. 141 (1969) 229 -- 244. R.W. Johnson. Homogeneous Lie Algebras and Expanding Automorphisms. Proc. Amer. Math. Soc. 48(2) (1975), 292 -- 296. J. Lauret. Examples of Anosov diffeomorphisms. J. Algebra 262 (2003), no. 1, 201 -- 209; corrigendum J. Algebra 268 (2003), no. 1, 371 -- 372. G. Leger. Derivations of Lie algebras III. Duke Math. J. Volume 30, Number 4 (1963), 637 -- 645. L. Magnin. Determination of 7-dimensional indecomposable nilpotent complex Lie algebras by adjoining a derivation to 6-dimensional Lie algebras Algebras and Rep- resentation Theory, vol.13, Number 6, p. 723 -- 753, 2010. V. Nekrashevych and G. Pete. Scale-invariant groups. Groups Geom. Dyn. 5 (2011), no. 1, 139 -- 167. P. Pansu. Croissance des boules et des g´eod´esiques ferm´ees dans les nilvari´et´es. Er- godic Theory Dyn. Syst. 3 (1983) 415 -- 445. P. Pansu. M´etriques de Carnot-Carath´eodory et quasiisom´etries des espaces sym´etriques de rang un, Ann. of Math. 129(1) (1989), 1 -- 60. M.S. Raghunathan. Discrete subgroups of Lie groups. Ergebnisse der Mathematik und ihrer Grenzgebiete 68. Springer-Verlag, 1972. 48 [Sch] [Sul] [Ver] YVES CORNULIER R. Schafer. An introduction to nonassociative Algebras. Dover, 1966. D. Sullivan. Infinitesimal computation in topology. Publ. IHES 47 (1977) 269 -- 331. M. Vergne. Cohomologie des alg`ebres de Lie nilpotentes. Applications a l'´etude de la vari´et´e des alg`ebres de Lie nilpotentes, Bull. Soc. Math. France 98 (1970), 81 -- 116. CNRS -- D´epartement de Math´ematiques, Universit´e Paris-Sud, 91405 Orsay, France E-mail address: [email protected]
1308.4607	2	1308	2013-08-26T15:56:34	A technical lemma for congruences of finite lattices	[ "math.RA" ]	The classical Technical Lemma for congruences is not difficult to prove but it is very efficient in its applications. We present here a Technical Lemma for congruences on \emph{finite lattices}. This is not difficult to prove either but it has already has proved its usefulness in some applications.	math.RA	math	A technical lemma for congruences of finite lattices G. Gratzer Abstract. The classical Technical Lemma for congruences is not difficult to prove but it is very efficient in its applications. We present here a Technical Lemma for congruences on finite lattices. This is not difficult to prove either but it has already has proved its usefulness in some applications. Here is the classical Technical Lemma for congruences, see G. Gratzer and E. T. Schmidt [8] and F. Maeda [10]. Lemma 1. A reflexive binary relation α on a lattice L is a congruence relation iff the following three properties are satisfied for any x, y, z, t ∈ L: (i) x ≡ y (mod α) (ii) Let x ≤ y ≤ z; then x ≡ y (mod α) and y ≡ z (mod α) imply that x ≡ z iff x ∧ y ≡ x ∨ y (mod α). (mod α). (iii) x ≤ y and x ≡ y (mod α) imply that x∨t ≡ y∨t (mod α) and x∧t ≡ y ∧ t (mod α). I stated and proved this Technical Lemma in all my lattice theory books, see for instance, [3, Lemma I.3.8], [4, Lemma 1.1], and [5, Lemma 11]. Witness the numerous references to this lemma, for instance, in [5]. The lemma is not difficult to prove, but it surely saves a lot of computation wherever we need to prove that a reflexive binary relation is a congruence relation. In some recent research, G. Cz´edli and I, see [1], [2] and [6], [7] spent quite an effort in proving that some equivalence relations on a planar semimodular lattices with intervals as equivalence classes are congruences. The number of cases we had to consider was dramatically cut by the following result. Lemma 2. Let L be a finite lattice. Let δ be an equivalence relation on L with intervals as equivalence classes. Then δ is a congruence relation iff the following condition and its dual hold: If x is covered by y, z ∈ L and x ≡ y (mod δ), then z ≡ y ∨ z (mod δ). (C∨) Proof. First, we prove the join-substitution property: (mod δ), then if x ≤ y and x ≡ y x ∨ z ≡ y ∨ z (mod δ). (1) This is trivial if y = z, so we assume that y 6= z. Clearly, we can also assume that x < y and x < z. Let U = [x, y ∨ z]. We induct on len U , the length of U . 2010 Mathematics Subject Classification: Primary: 06B10. Key words and phrases: finite lattice, congruence. 2 G. Gratzer Algebra univers. Using the fact that the intersection of two convex sublattices is either ∅ or a convex sublattice, it follows that, for every interval V of L, the classes of δ⌉V are intervals. Hence, we can assume that x = y ∧ z; indeed, otherwise the induction hypothesis applies to V = [y ∧ z, y ∨ z] and δ⌉V , since len V < len U , yielding (1). Note that len U ≥ 2. If len U = 2, then (1) is stated in (C∨). So we can also assume that len U > 2. Pick the elements y1, z1 ∈ L so that x ≺ y1 ≤ y and x ≺ z1 ≤ z. The elements y1 and z1 are distinct, since y1 = z1 would contradict that x = y ∧ z = y1 ∧ z1. Let w = y1 ∨ z1. Since the δ-classes are intervals, x ≡ y1 (mod δ), therefore, (C∨) yields that z1 ≡ w (mod δ). (2) Let I = [y1, y ∨ z] and J = [z1, y ∨ z]. Then len I, len J < len U . Hence, the induction hypothesis applies to I and δ⌉I, and we obtain that w ≡ y ∨ w (mod δ). Combining this with (2), by the transitivity of δ, we conclude that z1 ≡ y ∨ w (mod δ). (3) Therefore, applying the induction hypothesis to J and δ⌉J, we conclude from (3) that x ∨ z = z ∨ z1 ≡ z ∨ (y ∨ w) = y ∨ z (mod δ), proving (1). Second, we get the meet-substitution property by duality. (cid:3) Observe that for a finite semimodular lattice (C∨) states that we have to check the join-substitution property only in covering-square sublattices. Note that Lemma 2 holds in any lattice L in which every interval has a finite length. I hope that others, working with congruences of finite lattices, will also find this new Technical Lemma useful. References [1] Cz´edli, G.: Patch extensions and trajectory colorings of slim rectangular lattices. Algebra Universalis (in press) [2] Cz´edli, G.: A note on congruence lattices of slim semimodular lattices. Algebra Universalis (in press) [3] Gratzer, G.: General Lattice Theory, second edition. New appendices by the author with B. A. Davey, R. Freese, B. Ganter, M. Greferath, P. Jipsen, H. A. Priestley, H. Rose, E. T. Schmidt, S. E. Schmidt, F. Wehrung, and R. Wille. Birkhauser, Basel (1998) [4] Gratzer, G.: The Congruences of a Finite Lattice, A Proof-by-Picture Approach. Birkhauser, Boston (2006) [5] Gratzer, G.: Lattice Theory: Foundation. Birkhauser, Basel (2011) [6] G. Gratzer, The order of principal congruences of a lattice. Algebra Universalis 70, 95 -- 105 (2013) [7] Gratzer, G.: Congruences of fork extensions of lattices. Acta Sci. Math. (Szeged) (submitted) [8] Gratzer, G. and Schmidt, E. T.: Ideals and congruence relations in lattices. Acta Math. Acad. Sci. Hungar. 9, 137 -- 175 (1958) Vol. 00, XX A technical lemma for congruences of finite lattices 3 [9] Gratzer, G. and Wehrung, F. eds.: Lattice Theory: Empire. Special Topics and Applications. Birkhauser, Basel [10] Maeda, F.: Kontinuierliche Geometrien. Springer-Verlag,Heidelberg (1958) G. Gratzer Department of Mathematics, University of Manitoba, Winnipeg, MB R3T 2N2, Canada e-mail, G. Gratzer: [email protected]
1011.4133	1	1011	2010-11-18T04:42:34	Primitive algebraic algebras of polynomially bounded growth	[ "math.RA" ]	We show that if $k$ is a countable field, then there exists a finitely generated, infinite-dimensional, primitive algebraic $k$-algebra $A$ whose Gelfand-Kirillov dimension is at most six. In addition to this we construct a two-generated primitive algebraic $k$-algebra. We also pose many open problems.	math.RA	math	PRIMITIVE ALGEBRAIC ALGEBRAS OF POLYNOMIALLY BOUNDED GROWTH JASON P. BELL, LANCE W. SMALL, AND AGATA SMOKTUNOWICZ Abstract. We show that if k is a countable field, then there exists a finitely generated, infinite-dimensional, primitive algebraic k-algebra A whose Gelfand-Kirillov dimension is at most six. In addition to this we construct a two-generated primitive algebraic k-algebra. We also pose many open problems. 0 1 0 2 v o N 8 1 ] . A R h t a m [ 1 v 3 3 1 4 . 1 1 0 1 : v i X r a 1. Introduction In recent years there has been renewed interest in the construction of algebraic algebras that are not locally finite-dimensional; that is, algebras that are algebraic over their base fields, but which have the property that some finitely generated subalgebra is infinite- dimensional. The first construction of such an algebra was made by Golod and Shafare- vich [6], which provided a counterexample to a famous conjecture of Kurosh [11]. The third-named author [17, 18, 19] has produced a variety of important counterexamples to conjectures about algebraic algebras, which go beyond the original construction of Golod and Shafarevich. There is an intimate connection between Kurosh's conjecture in ring theory and its group- theoretic counterpart, Burnside's conjecture. Indeed, Golod and Shafarevich were able to use their construction to produce a finitely generated infinite torsion group, providing the first counterexample to the Burnside problem. Gromov [7] showed that finitely generated groups of polynomially bounded growth are nilpotent-by-finite. In particular, there cannot exist a finitely generated infinite torsion group of polynomially bounded growth. In light of this theorem, it is natural to ask whether an analogous result holds for rings. Surprisingly, Lenagan and Smoktunowicz [13] showed that over a countable field there exists a finitely generated infinite-dimensional algebraic algebra whose Gelfand-Kirillov dimension (defined in §2) is at most 20. By refining estimates used in this paper, Lenagan, Smoktunowicz, and Young [14] showed that the bound of 20 on the Gelfand-Kirillov dimension could be lowered to three. Despite all of the recent progress on Kurosh-type problems, there has been little progress -- either with or without restrictions on Gelfand-Kirillov dimension -- on the related problem Date: Oct. 1, 2010. 2000 Mathematics Subject Classification. 16P90. Key words and phrases. primitive algebras, Kurosh problem, algebraic algebras, Gelfand-Kirillov dimension. Research supported by NSERC grant 31-611456. The research of the third author was supported by Grant No. EPSRC EP/D071674/1. 1 2 JASON P. BELL, LANCE W. SMALL, AND AGATA SMOKTUNOWICZ of whether or not there exists an algebraic division ring that is finitely generated and infinite-dimensional as an algebra over its center. In fact, most of the constructions have had a large, nil Jacobson radical and are very far from being division rings. A resolu- tion of this problem appears to be far away, but one can nevertheless hope to make some progress by producing counter-examples in small increments that are in some sense in- creasingly closer to being division rings. As a partial step towards producing a finitely generated algebraic infinite-dimensional division ring, one can ask whether or not there is a finitely generated infinite-dimensional primitive algebraic algebra. This question was initially posed by Kaplansky [9, Problem 15] and answered by the first- and second-named authors [2]. In this paper, we show that one can in fact construct such an example in which the Gelfand-Kirillov dimension is at most six. Theorem 1.1. Let k be a countable field. Then there exists a finitely generated infinite- dimensional algebraic primitive k-algebra A whose Gelfand-Kirillov dimension is at most six. In order to do this construction, we use an (unpublished) affinization construction due to the second-named author, along with growth estimates of the first-named author [1], and the construction of Lenagan, et al. [14]. In addition to this, we also consider the question of whether or not there are primitive algebra algebras that are two-generated. We note that the algebra constructed in Theorem 1.1, while finitely generated, needs several generators. It is natural to consider whether one can construct such algebras with fewer generators. We are able to prove the following result. Theorem 1.2. Let k be a countable field. Then there is an infinite-dimensional primitive algebraic k-algebra that is generated by two elements. The outline of the paper is as follows. In §2, we define Gelfand-Kirillov dimension and describe some basic facts about it. In §3, we give a very general affinization construction, which allows one to construct finitely generated algebras with many different properties. In §4, we apply our construction and use the construction of Lenagan et al. to prove Theorem 1.1. In §5 we describe another affinization construction and prove Theorem 1.2. Finally, in §6 we pose some open problems which are related to our investigations. 2. Gelfand-Kirillov dimension In this section, we define Gelfand-Kirillov dimension and give some basic facts about it. Given a field k and a finitely generated k-algebra A, the Gelfand-Kirillov dimension of A (GK dimension, for short) is defined to be GKdim(A) := lim sup n→∞ log(cid:0)dim V n(cid:1)(cid:14) log n, where V is a finite dimensional k-subspace of A which contains the identity of A and which generates A as an k-algebra. As it turns out, Gelfand-Kirillov dimension is independent of choice of V [10, pp. 6]. In the case that A is not finitely generated, the GK dimension of PRIMITIVE ALGEBRAIC ALGEBRAS OF POLYNOMIALLY BOUNDED GROWTH 3 A is defined to be the supremum of the GK dimensions of finitely generated subalgebras of A. We give some of the basic properties about Gelfand-Kirillov dimension. Proposition 2.1. Gelfand-Kirillov dimension has the following properties: (1) the Gelfand-Kirillov dimension of a finitely generated commutative algebra is the same as its Krull dimension [10, pp. 39]; (2) there are no algebras whose GK dimension is strictly between 0 and 1; (3) there are no algebras whose GK dimension is strictly between 1 and 2 [10, pp. 18]; (4) for every α ≥ 2 there exists a finitely generated algebra of GK dimension α [10, pp. 162]; (5) if A and B are algebras such that B is either a finite left or right A-module, then GKdim(A) = GKdim(B) [3]; (6) if A and B are two algebras then GKdim(A ⊗ B) ≤ GKdim(A) + GKdim(B) [10, pp. 28]. We refer the reader to the book of Krause and Lenagan [10] for additional facts about Gelfand-Kirillov dimension. 3. Affinization In this section, we describe a general affinization construction. This construction takes a countably generated algebra T over a field k and builds a finitely generated k-algebra A with the property that eAe ∼= T for some idempotent e of A. Our construction has the additional property that we can bound the Gelfand-Kirillov dimension of the algebra A in terms of the Gelfand-Kirillov dimension of T . Notation 3.1. Throughout this section, we fix the following notation: (1) we let k be a field; (2) we let T be a prime, countably generated k-algebra; (3) we let B be a prime infinite-dimensional finitely generated k-algebra; (4) we let R = B ⋆ k[y], the free product of B and k[y] in the category of finitely generated k-algebras; (5) we let S = (cid:18) k + Ry R R (cid:19) . Ry The ring S is generated as a k-algebra by (cid:18) 0 0 0 c (cid:19) ,(cid:18) 1 0 0 0 (cid:19) ,(cid:18) 0 1 0 0 (cid:19) , and (cid:18) 0 0 y 0 (cid:19) , where c ranges over elements in a finite generating set for the k-algebra B. Hence S is a finitely generated k-algebra. Given these data, we show that we can construct a finitely generated prime k-algebra A with the following properties: 4 JASON P. BELL, LANCE W. SMALL, AND AGATA SMOKTUNOWICZ (1) A has an idempotent e such that eAe ∼= T ; (2) A/AeA is a homomorphic image of B; (3) GKdim(A) ≤ 2GKdim(B) + GKdim(T ). We call such a ring an affinization of T with respect to B. To describe this construction, note that k + Ry is a free k-algebra on the infinitely many generators {cy c ∈ C}, where C is a basis for B as a k-vector space. It follows that we have a surjective ring homomorphism (1) Let (2) Φ : k + Ry → T. P = ker(Φ) and let ei,j denote the matrix with a 1 in the (i, j)-entry and zeros everywhere else. Notice P is a prime ideal. Observe that Q′ := S(P e1,1)S satisfies e1,1Q′e1,1 = P . Using Zorn's lemma we can choose an ideal Q in S maximal with respect to the property that (3) e1,1Qe1,1 = (cid:18) P 0 0 0 (cid:19) . By maximality, we have that Q is prime. Throughout this paper, we let ei,j denote the image of ei,j is S/Q. Observe that Q is in fact uniquely determined. To see this, suppose that Q′ is another such ideal. Then By maximality of Q and Q′ we have that Q = Q + Q′ = Q′ and so Q = Q′. We note that e1,1(Q + Q′)e1,1 = e1,1Qe1,1 + e1,1Q′e1,1 = P (4) Similarly, (5) and (6) Q ⊇ (cid:18) 0 Ry 0 (cid:19)(cid:18) P 0 0 0 0 (cid:19)(cid:18) 0 0 0 R (cid:19) = (cid:18) 0 0 RyP R (cid:19) . 0 Q ⊇ P Re1,2, Q ⊇ RyP e2,1. The algebra S/Q has the property that (7) e1,1(S/Q)e1,1 ∼= T. We call S/Q the affinization of T with respect to Φ and B, and we denote it by A(T, B, Φ). Since the prime ideal Q is uniquely determined by Φ, the algebra A(T, B, Φ) is uniquely determined by T , B, and Φ. Observe that if e denotes the image of e1,1 in S/Q = A(T, B, Φ), then by construction we have: (1) A(T, B, Φ) is prime; PRIMITIVE ALGEBRAIC ALGEBRAS OF POLYNOMIALLY BOUNDED GROWTH 5 (2) eA(T, B, Φ)e ∼= T ; (3) A(T, B, Φ)/AeA is a homomorphic image of B. We now prove our main result of this section. Proposition 3.2. Assume the notation given in Notation 3.1. There exists a homomor- phism Φ : k + Ry → T such that A(T, B, Φ) has Gelfand-Kirillov dimension at most 2GKdim(B) + GKdim(T ). Proof. We let α and β denote respectively the Gelfand-Kirillov dimension of T and B. If α or β is infinite, there is nothing to prove, thus it is no loss of generality to assume that α, β < ∞. Let P and Q be as in equations (2) and (3). Let W0 be a finite-dimensional subspace of B that contains 1B and generates B as a k-algebra, let W = W0 + ky ⊆ R, (8) and let V ⊆ S be the generating subspace of S given by We have V ⊆ (cid:18) k + W y W W y W (cid:19) . V n ⊆ (cid:18) k + W ny W n W ny W n (cid:19) . We shall construct a homomorphism that will give an affinization with the desired upper bound on the Gelfand-Kirillov dimension. Let B = {1, u1, u2 . . .} ⊆ T be a basis for T as a k-vector space. For each j ≥ 1, define Uj to be the vector space spanned by the first j + 1 elements of B; that is, (9) Uj = k + ku1 + · · · + kuj. Since T has GK dimension α, we have Hence there exists a positive integer mj such that lim sup n→∞ log(cid:0)dim (Uj)n(cid:1). log n ≤ α. dim (Uj)n < nα+1/j for all n ≥ mj. By increasing mj if necessary, we may assume that mj ≥ mj−1 for all j ≥ 2. Pick vj ∈ W mj for each j ≥ 1. By construction, v1, v2, . . . are linearly independent and thus can be extended to a basis C for B. We define \ W mj −1 0 0 by Φ : k + Ry → T Φ(cid:0)cy(cid:1) = (cid:26) uj 0 and we extend by linearity. if c = vj for some j ≥ 1, if c ∈ C \ {v1, v2, . . .}, 6 JASON P. BELL, LANCE W. SMALL, AND AGATA SMOKTUNOWICZ Consider dim(cid:18) k + W ny 0 0 (cid:19) . 0 Let ε > 0. Then there exists some number j0 such that dim(W n) < nβ+ε for all n ≥ mj0. Suppose mj ≤ n < mj+1 for some j ≥ j0. Then since P e1,1 = e1,1Qe1,1, an element of W nye1,1 is determined by its behavior modulo P e1,1. Since n < mj+1, we have Hence Φ(cid:0)W ny(cid:1) ⊆ (Uj)n. dim W nye1,1 ≤ dim (Uj)n ≤ nα+1/j. We now compute the dimension of W ne2,2. Notice that any element of R can be expressed as a linear combination of elements of C, elements of the form c1yc2, with c1, c2 ∈ C, and elements of the form c1ywyc2, where w is a word over the alphabet C ∪ {y}, and c1, c2 ∈ C. Hence anything in W n is contained in 0 + W n W n 0 yW n 0 + W n 0 yW nyW n 0 . Thus dim(W n) ≤ dim(W n 0 ) + (dim(W n 0 ))2 + (dim(W n 0 ))2 · dim(W ny). Observe that RyP Re2,2 ⊆ Q and hence the image in e2,2A(T, B, Φ)e2,2 of an element of the form c1ywyc2e2,2, with c1, c2 ∈ C and w a word over the alphabet C ∪ {y}, is completely determined by the behavior of wy mod P . As Φ(cid:0)W ny(cid:1) ⊆ (Uj)n, we have for n ≥ mj, 0 yW nyW n 0 e2,2 ≤ dim(Uj)ndim(W n 0 )2 ≤ nα+2β+2ε+1/j. dim W n Since Thus dim W ne2,2 ≤ nβ+ε + n2β+2ε + nα+2β+ε+1/j = O(n2β+α+2ε+1/j). As j → ∞ as n → ∞ and ε > 0 is arbitrary, we see that dim W ne2,2 = O(n2β+α+ε) for every ε > 0 Let D denote the "diagonal" of A(T, B, Φ) and let C denote the "upper-triangular part" of A(T, B, Φ). We have just shown that has GK dimension at most α + 2β + ε. D ∼= (cid:0)(k + Ry)/P(cid:1) ⊕(cid:0)R/RyP R(cid:1) Observe that C = D + e1,2D and hence B has GK dimension at most α + 2β + ε [10, Lemma 4.3]. Finally, note that and thus A(T, B, Φ) = B + B(ye2,1) A(T, B, Φ) PRIMITIVE ALGEBRAIC ALGEBRAS OF POLYNOMIALLY BOUNDED GROWTH 7 has GK dimension at most α + 2β + ε, [10, Lemma 4.3]. Since ε > 0 is arbitrary, we conclude that A(T, B, Φ) has GK dimension at most α + 2β. (cid:3) In this section we prove Theorem 1.1 4. Algebraic algebras Proof of Theorem 1.1. Note that Lenagan, Smoktunowicz, and Young [14] have shown that one can construct a finitely generated infinite-dimensional algebraic algebra A of Gelfand- Kirillov dimension at most three over any countable field k. We note that this algebra has a prime infinite-dimensional homomorphic image B. Indeed, the prime radical (the intersection of all prime ideals) of an arbitrary algebra is always locally nilpotent and hence has Gelfand-Kirillov dimension zero. Consequently the prime radical doesn't equal the whole algebra A. Note that finitely dimensional nil algebras are nilpotent, therefore A/I is infinite dimensional for any prime ideal in A. The algebra B is necessarily algebraic, finitely generated, and has Gelfand-Kirillov dimension at most three. We let T be a countably generated infinite-dimensional primitive k-algebra of Gelfand- Kirillov dimension zero. We note that an example of such an algebra is given by the first- and second-named authors [2]. By Proposition 3.2, there exists a prime finitely generated k-algebra A with the following properties: (1) GKdim(A) ≤ 2GKdim(B) + GKdim(T ) ≤ 6; (2) there is an idempotent e ∈ A such that eAe ∼= T ; (3) A/AeA is a homomorphic image of B. The second property gives that A as primitive [12, Theorem 1], as A is a prime ring with a primitive corner. We next claim that AeA is a locally finite two-sided ideal of A. To see this, note that any finite-dimensional subspace of AeA is contained in a subspace of the form W eW for some finite-dimensional subspace of A. Then As eAe ∼= T , we see that eW 2e ∼= W ′ for some finite-dimensional subspace W ′ of T . As T is locally finite, we have that (W ′)p = (W ′)p+1 for some natural number m and hence (W eW )m ⊆ W (eW 2e)m−2eW. (W neW n)p+2 = (W neW n)p+3 , giving that AeA is locally finite. Since A/AeA is a homomorphic image of B, it is algebraic and AeA is a locally finite (cid:3) two-sided ideal, we see that A is algebraic. The result follows. 5. Affinization with two generators In this section, we briefly describe another affinization construction. This construction is a generalization of a construction of Markov [15]. We rely heavily on both the ideas and 8 JASON P. BELL, LANCE W. SMALL, AND AGATA SMOKTUNOWICZ notation from the recent construction of Lenagan, Smoktunowicz, and Young [14]. Using these ideas we are able to construct an infinite-dimensional primitive algebraic algebra generated by just two elements. We point out that our construction only works over countable fields. Notation 5.1. Throughout this section, we fix the following notation: (1) we let k be a countable field; (2) we let T be a prime, countably generated k-algebra with unity; (3) let k{x, y} denote the free k-algebra on two generators; (4) we let B denote an infinite-dimensional k-algebra of the form B = k{x, y}/(y2, I) where I ⊆ (x, xyx)k{x, xyx} ⊆ k{x, y}. Theorem 5.2. Assume the notation from Notation 5.1. Then there exists k-algebra A generated by two elements x and y such that that y3 = y2, y2Ay2 ∼= T and A/(y2) is a homomorphic image of B. Proof. Let R = k{x, y}/(y3 − y2, I). Note that (y3 − y2, I) denotes the ideal of k{x, y} generated by I and by the element y3 − y2. Let r1, r2, . . . be a basis for P0≤i,j≤n k{x, xyx}/I ∩ k{x, xyx} as a k-vector space. Observe that the elements y2r1y2, y2r2y2 . . . are generators of a countably-generated uni- tal free noncommutative k-algebra, which we denote by C. It follows that we have a surjective ring homomorphism (10) Let Φ : C → T. P = ker(Φ) (11) Notice P is a prime ideal in C. Let Q′ = RP R + RP + P R + P . Then Q′ is an ideal in R and y2Q′y2 = P . Using Zorn's lemma we can choose an ideal Q in R maximal with the property that y2Qy2 = P . Then Q is a prime ideal, because if Q ⊆ Q1 and Q ⊆ Q2 for some ideals Q1, Q2 in R then Q1Q2 ⊆ Q gives y2Q1y2Cy2Q2y2 ⊆ y2Qy2 = P , and so either y2Q1y2 = P or y2Q2y2 = P , as P is a prime ideal of C. Observe also that C = y2Ry2, because y2 −1 annihilates the homogeneous maximal ideal of C when we regard C as a subalgebra of R. Let A = R/Q. Observe that by construction we have A = R/Q is prime; furthermore, A is generated by elements x, y and A/(y2) is a homomorphic image of B. Finally, we clearly have y2Ay2 ∼= T . The result follows. (cid:3) We next prove a technical lemma. All of the groundwork needed for this result was done by Lenagan, Smoktunowicz, and Young [14]. In order to avoid unnecessary repetition, we will make use of the notation and proofs from their paper and a careful reading of this paper is essential for a full understanding of this lemma. Lemma 5.3. Let k be a countable field. Then there exists an infinite-dimensional k-algebra R generated by two elements x, y such that y2 = 0 and with the property that the ideal (x, y) in R is nil. Furthermore, one can choose R to have the property that its GK dimension is at most 3. PRIMITIVE ALGEBRAIC ALGEBRAS OF POLYNOMIALLY BOUNDED GROWTH 9 Proof. We use the construction and notation from the paper of Lenagan, Smoktunowicz, and Young [14]. Let Vi, Ui, H(n) be as in Theorem 3 from this paper. Observe that if we take (12) and (13) and (14) V1 = {x, y}, U1 = {∅} V2 = {xx, xy}, U2 = {yy, yx} V4 = {xxxx, xxxy}, (15) U4 = {y4, yyxy, yyxx, yyyx, xyyy, xxyy, yxyy, yxxx, yxxy, yxyx, xyyx, xxyx, xyxy, xyxx} and apply Theorem 3 of [14], then H(i)y2H(8 − i − 2) ⊆ U(4)H(4) + H(4)U(4) and thus y2 ∈ E where E is the ideal defined in [14] with the property that the image of the ideal (x, y) in k{x, y}/E is nil and k{x, y}/E is an algebra of GK dimension not exceeding three. Thus the image of y2 in k{x, y}/E is zero. (cid:3) Proposition 5.4. Let k be a countable field. Then there is a two-sided ideal I of k{x, y} satisfying the following properties: (1) I is generated by elements from (x, xyx)k{x, xyx}; (2) the k-algebra B = k{x, y}/(y2, I) is infinite-dimensional as a k-vector space; (3) the image of the ideal (x, y) in B is nil. Proof. Let R be the algebra defined in Lemma 5.3. Then R = k{x, y}/E for some ideal E of k{x, y}. Because the image of the homogeneous maximal ideal (x, y) of k{x, y} is nil in R, there is p such that xp ∈ E. Since N is infinite-dimensional as a k-vector space and y2 ∈ E we see that p > 2. Let B = k{x, y}/({xEx, y2, xp, xyxiExjyx : 0 ≥ i, j < p}). Then B satisfies the conclusion of the statement of the proposition. (cid:3) Proof of Theorem 1.2. Let T be infinitely generated locally finite primitive algebra with unity over k. Let B be an algebra satisfying the conclusion of Proposition 5.4. By Theorem 5.2, there exists prime algebra A generated by two elements x, y such that y4 = y2 and y2Ay2 is isomorphic to T . Since y2 is idempotent and T is primitive, we see that A is primitive by a result of Lanski, Resco, Small [12]. It only remains to show that A is algebraic. Note that A/(y2) is a homomorphic image of B and hence is algebraic. Furthermore y2 generates a locally finite ideal of A, and hence A is algebraic over k. The result follows. (cid:3) 10 JASON P. BELL, LANCE W. SMALL, AND AGATA SMOKTUNOWICZ 6. Questions In this section, we pose some questions related to algebraic algebras and our main result. Question 1. Does there exist some real number α > 2 such that for every β ≥ α there exists a finitely generated nil algebra whose GK dimension is exactly β? It is the opinion of the authors that by suitably modifying the homomorphism φ which is used in our construction, one should in fact be able to construct finitely generated nil primitive algebras of every GK dimension larger than or equal to six. Question 2. Does there exist a finitely generated, infinite-dimensional k-algebra that is a division ring of finite GK dimension? Even answering this question for algebras of quadratic growth would be an impressive result. In fact, there are no known examples of division rings that are algebraic over their centers that do not have the property that each finitely generated subalgebra is finite- dimensional over its center. Question 3. Does there exist a finitely generated infinite-dimensional unital simple alge- braic algebra of finite GK dimension? We have constructed a finitely generated infinite-dimensional unital primitive algebra of finite GK dimension. The next logical step is to attempt to modify this construction somehow to create a simple algebra with the aforementioned properties. Unfortunately, there is an obstacle that one immediately encounters; namely, algebras constructed via the affinization method always have a nonzero proper two-sided ideal generated by the image of ye2,2; moreover, any homomorphic image under which this ideal becomes zero is a homomorphic image of B. The third-named author [19], on the other hand, has constructed a simple nil algebra (which clearly cannot be unital). Observe that by Nakyama's lemma a finitely generated Jacobson radical algebra cannot be simple. Question 4. Does there exist a finitely presented infinite-dimensional algebraic algebra of finite GK dimension? We note that all constructions of infinite-dimensional algebra algebras so far have re- quired infinite sets of relations in order to obtain algebraic algebras. It is conjectured that the corresponding question for groups (with the polynomial growth restriction removed), namely, whether or not there exists a finitely presented infinite torsion group, has a neg- ative answer. A negative answer to the above question would both lend verisimilitude to the group theoretic conjecture and may also provide techniques which could eventually be used to prove this conjecture. Question 5. Does there exist a finitely generated infinite-dimensional algebraic algebra of finite GK dimension over an uncountable base field? One of the main problems with trying to do the Lenagan and Smoktunowicz construction over an uncountable field k is that a countable enumeration of the elements of the free k- algebra on two generators is required. This can be relaxed somewhat, but it does not seem PRIMITIVE ALGEBRAIC ALGEBRAS OF POLYNOMIALLY BOUNDED GROWTH 11 possible to modify their construction to obtain an infinite-dimensional algebraic algebra of finite GK dimension over an uncountable base field. Question 6. Does there exist a finitely generated algebraic algebra of GK dimension two? Lenagan, Smoktunowicz, and Young [14] produced a finitely generated infinite-dimensional algebraic algebra (over a countable base field) whose GK dimension is at most three. It should be noted that finitely generated algebras of GK dimension strictly less than 2 satisfy a polynomial identity by Bergman's gap theorem [10, Theorem 2.5] along with a theorem of Small, Stafford, and Warfield [16]. Consequently, the question of what is the infimum over all Gelfand-Kirillov dimensions of finitely generated infinite-dimensional algebraic algebras is still unresolved. Question 7. Can one give a construction as done in Theorem 1.2 when k is uncountable? Can one modify the construction in Theorem 1.2 to give an algebra of finite GK dimension? In the authors' opinion, the first part should not be too difficult, but it will probably require a deep understanding of the paper [14]. The second part will require delicate estimates, but is probably quite doable, however the calculations involved appear to be more complicated and less elegant than those used in the construction of Theorem 3.2. Question 8. Are there other interesting algebras of low GK dimension that can be con- structed using Proposition 3.2? As an example, let k be a field and let T = k[x1, x2, . . .]/I, where I is the ideal generated by {xi+1 i ≥ 1}. Then T has prime radical J generated by the image of (x1, x2, . . .), which is nil and not nilpotent. Observe that J is the sum of all nilpotent ideals of T and hence T has no maximal nilpotent ideal. : i Since T has GK dimension 0, we can apply Proposition 3.2, taking B to be a nil ring of GK dimension at most 3, as we did in the proof of Theorem 1.1, to obtain an algebra A whose GK dimension is at most six. Moreover, by using basic facts about corners, one can see that A is a finitely generated k-algebra with a non-nilpotent prime radical and without a maximal nilpotent ideal. This example answers a question of Lvov [5, Question 2.69]. In a similar manner, one should be able to construct strange examples of 2-generated Jacobson radical algebras that are not nil. References [1] Jason P. Bell, Examples in finite Gelfand-Kirillov dimension, J. Algebra 263 (2003), no. 1, 159 -- 175. [2] Jason P. Bell, Lance W. Small, A question of Kaplansky, Special issue in celebration of Claudio Procesi's 60th birthday, J. Algebra 258 (2002), no. 1, 386 -- 388. [3] W.Bohro, H. Kraft, Uber die Gelfand-Kirillov Dimension, Math.Ann. 220 (1) (1976), 1 -- 24. [4] Daniel R. Farkas, Lance W. Small, Algebras which are nearly finite dimensional and their identities, Israel J. Math 127 (2002), 245-251. [5] V. T. Filipov, V.K. Kharchenko, I.P.Shestakov (Editors), Dniester Notebook, Unsolved Problems in the Theory of Rings and Modules, Mathematics Institute, Russian Academy of Sciences, Siberian Branch, Novosibirsk, Fourth Edition, 1993. 12 JASON P. BELL, LANCE W. SMALL, AND AGATA SMOKTUNOWICZ [6] E.S. Golod and I.R. Shafarevich, On the class field tower, Izv. Akad. Nauk. SSSR Mat. Ser. 28 (1964), 261 -- 272. (in Russian) [7] M. Gromov, Groups of polynomial growth and expanding maps, Publ. Math. IHES 53 (1981), 53 -- 73. [8] Nathan Jacobson, Structure of Rings, Amer. Math. Soc. Coll., vol. 37, rev. ed., 1964. [9] Irving Kaplansky, "Problems in the theory of rings" revisited, Amer. Math. Monthly 77 (1970), no. 5, 445 -- 454. [10] Gunter R. Krause and Thomas H. Lenagan, Growth of algebras and Gelfand-Kirillov dimension. Revised edition. Graduate Studies in Mathematics, 22. American Mathematical Society, Providence, RI, 2000. [11] A. Kurosh, Ringtheoretische Probleme die mit dem Burnsideschen Problem uber periodische Gruppen in Zussammenhang stehen, Bull. Acad. Sci. URSS. S´er. Math. [Izvestia Akad. Nauk SSSR] 5 (1941), 223 -- 240. [12] Charles Lanski, Richard Resco, and Lance Small, On the primitivity of prime rings. J. Algebra 59 (1979), no. 2, 395 -- 398. [13] T. H. Lenagan, A. Smoktunowicz, An infinite dimensional affine nil algebra with finite Gelfand-Kirillov dimension, J.Amer. Math. Soc. 20 (2007), no. 4, 989-1001. [14] T. H. Lenagan, A. Smoktunowicz, and A. Young, Nil algebras with restricted growth, submitted. [15] V. T. Markov, Some examples of finitely generated algebras, Uspiekhi Mat. Nauk 221 (1981), 185 -- 186. [16] L. W. Small, J. T. Stafford, R. Warfield Jr., Affine algebras of Gelfand-Kirillov dimension one are PI, Math. Proc. Cambridge Philos. Soc. 97 (1985), no. 3, 407 -- 414. [17] Agata Smoktunowicz, Graded algebras associated to algebraic algebras need not be algebraic. Euro- pean Congress of Mathematics, Eur. Math. Soc. Z urich, 2010, 441 -- 449. [18] Agata Smoktunowicz, Makar-Limanov's conjecture on free subalgebras, Adv. Math. 222 (2009), no. 6, 2107 -- 2116. [19] Agata Smoktunowicz, A simple nil ring exists, Comm. Algebra 30 (2002), no. 1, 27 -- 59. Jason Bell, Department of Mathematics, Simon Fraser University, Burnaby, BC, V5A 1S6, CANADA E-mail address: [email protected] Lance Small, Department of Mathematics, University of California, San Diego, La Jolla, CA, 92093-0112, USA E-mail address: [email protected] Agata Smoktunowicz, Maxwell Institute for Mathematical Sciences, School of Math- ematics, University of Edinburgh, James Clerk Maxwell Building, King's Buildings, May- field Road, Edinburgh EH9 3JZ, Scotland, UK E-mail address: [email protected]
1711.03042	2	1711	2018-04-18T10:14:27	Hermitian Morita Theory: a Matrix Approach	[ "math.RA" ]	This bundle contains the paper "Hermitian Morita Theory: a Matrix Approach", published in Irish Math. Soc. Bulletin 62 (2008), pp. 37-41 (reproduced with permission of the Irish Mathematical Society), and with abstract "In this note an explicit matrix description of hermitian Morita theory is presented." and a short erratum.	math.RA	math	Irish Math. Soc. Bulletin 62 (2008), 37 -- 41 37 Hermitian Morita Theory: a Matrix Approach DAVID W. LEWIS AND THOMAS UNGER Abstract. In this note an explicit matrix description of her- mitian Morita theory is presented. 1. Introduction Let K be a field of characteristic different from two and let A be a central simple K-algebra equipped with an involution ∗. By a well- known theorem of Wedderburn, A is of the form Mn(D), a full matrix algebra over a division K-algebra D. Furthermore, there exists an involution − on D of the same kind as ∗ such that ∗ and − have the same restriction to K. Then ∗ is the adjoint involution adh0 of some nonsingular ε0-hermitian form h0 over (D, −), h0 : Dn × Dn −→ D, with ε0 = ±1. Thus X ∗ = adh0(X) = SX t S−1, ∀X ∈ Mn(D), where S ∈ GLn(D) is the matrix of h0, so that S t = ε0S. Let Grε(A, ∗) and Wε(A, ∗) denote the Grothendieck group and Witt group of ε-hermitian forms over (A, ∗), respectively. Hermitian Morita theory furnishes us with isomorphisms Grε(A, ∗) ∼= Grε0ε(D, −) and Wε(A, ∗) ∼= Wε0ε(D, −). These isomorphisms are the result of the following equivalences of categories   ε-hermitian forms over (Mn(D), ∗)   scaling / ε0ε-hermitian forms over (Mn(D), −t)    Morita equivalence / ε0ε-hermitian forms over (D, −)    8 1 0 2 r p A 8 1 ] . A R h t a m [ 2 v 2 4 0 3 0 . 1 1 7 1 : v i X r a o o / o o / 38 David W. Lewis and Thomas Unger (all forms are assumed to be nonsingular) which respect isometries, orthogonal sums and hyperbolic forms. In this note we describe these correspondences explicitly. In par- ticular we give a matrix description of Morita equivalence which does not seem to be generally known. Other explicit descriptions can be found in [3, 4, 5]. The subject is often treated in a more abstract manner, such as in [1] and [2, Chap. I, §9]. 2. Scaling Let M be a right Mn(D)-module and let h : M × M −→ Mn(D) be an ε-hermitian form with respect to ∗, i.e. t h(y, x) = εh(x, y)∗ = εSh(x, y) S−1. Proposition 2.1. The form S−1h : M × M −→ Mn(D), (x, y) 7−→ S−1h(x, y) is ε0ε-hermitian over (Mn(D), −t). Proof. Sesquilinearity of S−1h with respect to −t follows easily from sesquilinearity of h with respect to ∗: (S−1h)(xα, y) = S−1h(xα, y) = S−1α∗h(x, y) = S−1SαtS−1h(x, y) = αtS−1h(x, y) for any α ∈ Mn(D) and any x, y ∈ M . t Furthermore, using the fact that S = ε0S, we get (S−1h)(y, x) = S−1h(y, x) t = S−1εSh(x, y) S−1 t = εh(x, y) S−1 = εε0h(x, y) t t (S−1) t = εε0(S−1h)(x, y) for any x, y ∈ M . Remark 2.2. By the first part of the proof, scaling of a sesquilinear form h (rather than an ε-hermitian form h) with respect to ∗ results in a sesquilinear form S−1h with respect to −t. Hermitian Morita Theory: a Matrix Approach 39 Remark 2.3. The matrix S is not determined uniquely, but only up to scalar multiplication by λ ∈ K, since λS and S give the same involution adh0 . Hence the scaling correspondence is not canonical. 3. Morita Equivalence Every module over Mn(D) ∼= EndD(Dn) is a direct sum of simple modules, namely copies of Dn. Let (Dn)k be such a module. We identify (Dn)k with Dk×n, the k × n-matrices over D. We view each row of a k × n-matrix over D as an element of Dn. Note that Mn(D) acts on Dk×n on the right. Now let h : Dk×n × Dk×n −→ Mn(D) be an ε-hermitian form over (Mn(D), −t). Proposition 3.1. There exists an ε-hermitian k × k-matrix B ∈ Mk(D) such that h(x, y) = xtBy, ∀x, y ∈ Dk×n. (1) Proof. Let B = (bij). We will determine the entries bij. Let eij ∈ Dk×n, e′ ij ∈ Dn×k and Eij ∈ Mn(D) respectively denote the k × n- matrix, the n × k-matrix and the n × n-matrix with 1 in the (i, j)-th position and zeroes everywhere else. One can easily verify that eif Ef ℓ = eiℓ, (2) where 1 ≤ i ≤ k and 1 ≤ f, ℓ ≤ n. Also note that if C ∈ Mn(D), then computing the product Eij C picks the j-th row of C and puts it in row i while making all other entries zero. Similarly, computing the product CEij picks the i-th column of C and puts it in column j while making all other entries zero. The matrices eij and e′ ij behave in a similar fashion. The matrices {eij 1 ≤ i ≤ k, 1 ≤ j ≤ n} generate Dk×n as a right Mn(D)-module. Thus it suffices to compute h(eif , ejg) where 1 ≤ i, j ≤ k and 1 ≤ f, g ≤ n. Let us first compute h(eii, ejj): h(eii, ejj ) = h(eiiEii, ejj Ejj ) = Eiih(eii, ejj )Ejj = mijEij , where mij is the (i, j)-th entry of h(eii, ejj) ∈ Mn(D). In other words, the matrix h(eii, ejj) has only one non-zero entry, namely mij in position (i, j). 40 David W. Lewis and Thomas Unger Next, let us compute h(eif , ejg). We will use the fact that eif = eiiEif , where 1 ≤ i ≤ k and 1 ≤ f ≤ n, which follows from (2). We get h(eif , ejg) = h(eiiEif , ejj Ejg) = Ef ih(eii, ejj )Ejg = (cid:0)h(eii, ejj )(cid:1)ij = mij Ef g. Ef g Let bij = mij where 1 ≤ i, j ≤ k. We have eif tBejg = e′ f iBejg = bijEf g = mijEf g. Therefore, h(eif , ejg) = eif n, which establishes (1). Finally, tBejg where 1 ≤ i, j ≤ k and 1 ≤ f, g ≤ mjiEji = h(ejj , eii) = εh(eii, ejj) t = εmijEji, for 1 ≤ i, j ≤ k, which implies mji = εmij, for 1 ≤ i, j ≤ k. In other words, mji = εmij, for 1 ≤ i, j ≤ k, so that B = εB, which finishes the proof. t So, given an ε-hermitian form h over (Mn(D), −t), we have ob- tained an ε-hermitian form over (D, −) with matrix B as in Propo- sition 3.1. Conversely, given an ε-hermitian form ϕ : Dk × Dk −→ D, represented by the matrix B (i.e. B = (cid:0)ϕ(ei, ej)(cid:1) for a D-basis {ei} of Dk), we define h : Dk×n × Dk×n −→ Mn(D) by h(x, y) := xtBy, ∀x, y ∈ Dk×n, which gives an ε-hermitian form over (Mn(D), −t). Remark 3.2. The correspondence h ↔ ϕ already works for forms that are just sesquilinear, without assuming any hermitian symme- try. Since scaling also preserves sesquilinearity, as remarked earlier, we conclude that the category equivalences of §1 already hold for Hermitian Morita Theory: a Matrix Approach 41 sesquilinear forms over (Mn(D), ∗), (Mn(D), −t) and (D, −), respec- tively. References [1] Frohlich, A. and McEvett, A.M., Forms over rings with involution. J. Algebra 12 (1969), 79 -- 104. [2] Knus, M.-A., Quadratic and Hermitian forms over rings. Grundlehren der Mathematischen Wissenschaften 294, Springer-Verlag, Berlin, 1991. [3] Lewis, D.W., Forms over real algebras and the multisignature of a manifold. Advances in Math. 23 (1977), no. 3, 272 -- 284. [4] Riehm, C., Effective equivalence of orthogonal representations of finite groups. J. Algebra 196 (1997), no. 1, 196 -- 210. [5] Riehm, C., Orthogonal, symplectic and unitary representations of finite groups. Trans. Amer. Math. Soc. 353 (2001), no. 12, 4687 -- 4727. David Lewis and Thomas Unger, School of Mathematical Sciences, University College Dublin, Belfield, Dublin 4, Ireland, [email protected], [email protected] Received on 19 December 2008. Correction to Hermitian Morita Theory: a Matrix Approach DAVID W. LEWIS AND THOMAS UNGER Our description of the adjoint involution adh0 in §1 should be cor- rected as follows: adh0 (X) = S−1X t S, ∀X ∈ Mn(D). Consequently, S should be replaced by S−1 and vice versa every- where in §2. The proof of [1, Prop. 3.1] contains an error: the matrices eii are not defined for all values of i when k > n. We are very grateful to Bhanumati Dasgupta for pointing this out to us and for contributing to a correct proof which is presented below. Proposition 3.1. There exists an ε-hermitian k × k-matrix B ∈ Mk(D) such that h(x, y) = xtBy, ∀x, y ∈ Dk×n. (1) Proof. Let B = (bij). We will determine the entries bij. Let eij ∈ Dk×n, e′ ij ∈ Dn×k and Eij ∈ Mn(D) respectively denote the k × n- matrix, the n × k-matrix and the n × n-matrix with 1 in the (i, j)-th position and zeroes everywhere else. One can easily verify that eif Ef ℓ = eiℓ, (2) for all 1 ≤ i ≤ k and all 1 ≤ f, ℓ ≤ n. Also note that if C ∈ Mn(D), then computing the product Eij C picks the j-th row of C and puts it in row i while making all other entries zero. Similarly, computing the product CEij picks the i-th column of C and puts it in column j while making all other entries zero. The matrices eij and e′ ij behave in a similar fashion. The matrices {eij 1 ≤ i ≤ k, 1 ≤ j ≤ n} generate Dk×n as a right Mn(D)-module. Thus it suffices to compute h(eif , ejg) for all 1 8 1 0 2 r p A 8 1 ] . A R h t a m [ 2 v 2 4 0 3 0 . 1 1 7 1 : v i X r a 2 David W. Lewis and Thomas Unger 1 ≤ i, j ≤ k and all 1 ≤ f, g ≤ n. Using (2) we have for arbitrary 1 ≤ ℓ, r ≤ n that h(eif , ejg) = h(eiℓEℓf , ejrErg) = Ef ℓh(eiℓ, ejr)Erg = (cid:0)h(eiℓ, ejr)(cid:1)ℓr Ef g, where (cid:0)h(eiℓ, ejr)(cid:1)ℓr denotes the (ℓ, r)-th entry of the n × n matrix h(eiℓ, ejr). It follows that (cid:0)h(eiℓ, ejr)(cid:1)ℓr is independent of the choice of ℓ and r. For all 1 ≤ i, j ≤ k we define bij := (cid:0)h(eiℓ, ejr)(cid:1)ℓr . Thus h(eif , ejg) = bij Ef g for all 1 ≤ i, j ≤ k and all 1 ≤ f, g ≤ n. We also have eif tBejg = e′ f iBejg = bijEf g for all 1 ≤ i, j ≤ k and all 1 ≤ f, g ≤ n. Therefore, h(eif , ejg) = eif tBejg for all 1 ≤ i, j ≤ k and all 1 ≤ f, g ≤ n, which establishes (1). Finally, t bjiEgf = h(ejg, eif ) = εh(eif , ejg) = εbij Ef g t = εbijEgf , for all 1 ≤ i, j ≤ k and all 1 ≤ f, g ≤ n, which implies bji = εbij, for all 1 ≤ i, j ≤ k. In other words, bji = εbij, for 1 ≤ i, j ≤ k, so that B = εB, which finishes the proof. t References [1] Lewis, D.W. and Unger, T., Hermitian Morita Theory: a Matrix Approach. Irish Math. Soc. Bulletin 62 (2008), 37 -- 41. David Lewis and Thomas Unger, School of Mathematical Sciences, University College Dublin, Belfield, Dublin 4, Ireland, [email protected], [email protected] 16 April 2018
1502.06478	1	1502	2015-02-23T16:01:57	Meet-completions and ordered domain algebras	[ "math.RA" ]	Using the well-known equivalence between meet-completions of posets and standard closure operators we show a general method for constructing meet-completions for isotone poset expansions. With this method we find a meet-completion for ordered domain algebras which simultaneously serves as the base of a representation for such algebras, thereby proving that ordered domain algebras have the finite representation property. We show that many of the equations defining ordered domain algebras are preserved in this completion but associativity, (D2) and (D6) can fail.	math.RA	math	Meet-completions and ordered domain algebras R Egrot and R Hirsch May 10, 2019 Abstract Using the well-known equivalence between meet-completions of posets and standard closure operators we show a general method for con- structing meet-completions for isotone poset expansions. With this method we find a meet-completion for ordered domain algebras which simultaneously serves as the base of a representation for such algebras, thereby proving that ordered domain algebras have the finite represen- tation property. We show that many of the equations defining ordered domain algebras are preserved in this completion but associativity, (D2) and (D6) can fail. Keywords: Finite representation property, completion, partially or- dered set, ordered domain algebra. 1 Introduction When considering algebras of binary relations, it is generally not the case that a finite representable algebra has a representation on a finite base. Indeed, in any signature which includes the identity, intersection and com- position operators, any representation of the Point Algebra [22] interprets each diversity atom as a dense linear order and so the representation is necessarily infinite. There are two well-known cases where we do have the finite represen- tation property. For the signature with identity, converse and composition only, the Cayley representation maps an algebra element a to the binary relation {(x, x; a) : x ∈ A} over the algebra itself. At the other extreme, for the signature consisting of Boolean operators only we may modify the stan- dard Stone representation (which represents elements as unary relations) Corresponding author: Robin Hirsch, Department of Computer Science, University College, Gower Street, London WC1E 6BT, email: [email protected] 1 and represent an element a as the identity relation over the ultrafilters con- taining a. Similarly, for a signature with solely an order relation ≤ (in other words a poset, P say) we may construct a representation whose base P ∗ consists of the upward closed subsets of P . An element p of P is represented as the identity restricted to p = {u ∈ P ∗ : p ∈ u}. Clearly P ∗ is finite if P is. Note that as well as providing the base of a representation, (P ∗, ∪, ∩) forms a complete distributive lattice and is a completion of P (see section 2). An interesting case, then, is where the signature includes both compo- sition and an order relation. For the signature with composition, converse, the domain operator and an ordering (Ordered Domain Algebra) a construc- tion in [11] had aspects of the Cayley representation but also aspects of the upward closed set representation for a poset. Each element of an ordered domain algebra (ODA) is represented as a set of pairs of upward closed sub- sets of the algebra, but in order to make the representation work for the non-Boolean operators, these upward closed subsets are required to have certain other closure properties. [1] had already provided a complete, finite set of axioms defining the class of representable ordered domain algebras, the construction in [11] also showed that a finite representable ODA has a representation on a finite base. Here we show, further, that the opera- tors may be lifted from an ordered domain algebra A to an algebra ΓD[A] whose universe is the set of all closed subsets of A, where the ordering ≤ is reverse inclusion ⊇ and where the other operators are lifted from A. We show that ΓD[A] is a completion of A and it obeys many of the equations defining ODAs. On the other hand, rather important properties, like the associativity of composition, are shown to be fallible in ΓD[A]. Our main results are the following. We show how to lift the isotone op- erators of a poset expansion to operators on a meet-completion of the poset in definition 5.1. We prove that certain inequalities are preserved when passing to this completion (given certain conditions on the inequalities and the completion) in corollary 5.7. Focusing on ODAs, we define a particular completion ΓD in definitions 7.1, 7.4 and restate the result that this comple- tion can act as the base of a representation of an ODA in theorem 7.5. In proposition 7.7 we show that all but three of the equations defining ODAs are preserved in this completion, and in examples 7.11, 7.12 and 7.13 we show that those three equations can fail in the completion. The remainder of this paper is divided as follows. In the next section we give the basic definitions for meet-completions and standard closure opera- tors and provide a proof of the well-known correspondence between them. In section 5 we explain how to extend isotone operators on a poset to a 2 meet-completion of that poset. This provides a method for extending poset meet-completions to meet-completions of isotone poset expansions. We in- vestigate some general rules governing the preservation of inequalities by meet-completions of isotone poset expansions using this method. In section 6 we define ordered domain algebras, and in section 7 we apply the consider- ations of sections 2 and 5 to construct a completion for ODAs and determine which ODA equations it preserves. We show how this completion can be used as the base of a representation for that algebra. Finally in section 8 we draw some conclusions from these results and offer suggestions for further work in this area. 2 Meet-completions and closure operators The material in this section is well-known, dating back to the pioneering work of Ore [19, 18, 20]. The aim here is to provide formulations best suited for the work we undertake in later sections. Throughout, for any unary function f and subset S of the domain of f we write f [S] for {f (s) : s ∈ S}, more generally for an n-ary func- tion g (written prefix) we may apply g pointwise to a sequence of sub- sets Si of the domain of g (i = 1, . . . , n) and write g[S1 × . . . × Sn] to denote {g(s1, . . . , sn) : si ∈ Si for 1 ≤ i ≤ n}. Two exceptions to this no- tational convention, where the functions are not written prefix, are the unary operator ⌣ and the binary ; and we simply write S⌣, S; T for {s⌣ : s ∈ S}, {s; t : s ∈ S, t ∈ T } respectively. For any subset S of a poset P we write S↑ for {p ∈ P : ∃s ∈ S, s ≤ p}. For p ∈ P we write p↑ as short- hand for {p}↑. Definition 2.1 (Completion). Given a poset P we define a completion of P to be a complete lattice Q and an order embedding e : P → Q. We say e : P → Q is a meet-completion when e[P ] is meet-dense in Q. That is, when q =V{e(p) : p ∈ P and e(p) ≥ q} for all q ∈ Q. Definition 2.2 (P ). If P is a poset define P to be the complete lattice of up-sets (including ∅) of P ordered by reverse inclusion (S1 ≤ S2 ⇐⇒ S1 ⊇ S2). The order dual P δ is the lattice of up-sets ordered by inclusion with bottom element ∅. It's easy to see that the map ι : P → P defined by ι(p) = p↑ defines a meet-completion of P (note though that ι will not map the top element of P (if it exists) to the top element of P , as the top element of P will 3 be ∅). This particular completion plays an important role in the theory of meet-completions. Definition 2.3 (Closure operator). Given a poset P , a closure operator on P is a map Γ : P → P such that 1. p ≤ Γ(p) for all p ∈ P , 2. p ≤ q =⇒ Γ(p) ≤ Γ(q) for all p, q ∈ P , and 3. Γ(Γ(p)) = Γ(p) for all p ∈ P . Following [4] we say a closure operator Γ on P * or P δ is standard when Γ(p↑) = p↑ for all p ∈ P . It is well-known that a meet-completion e : P → Q defines a standard closure operator Γe : P δ → P δ by Γe(S) = {p ∈ P : e(p) ≥ V e[S]} (we take the dual of P as otherwise condition 1 of Definition 2.3 fails). In this case Q is isomorphic to the lattice Γe[P ] of Γe-closed subsets of P . This isomorphism is given by the map he : Q → Γe[P ] defined by he(q) = {p ∈ P : e(p) ≥ q}. Note that we are purposefully taking P rather than P δ here as we want to order by reverse inclusion. This is technically an abuse of notation as Γe is originally defined on P δ , but as these structures have the same carrier hopefully our meaning is clear. Conversely, whenever Γ is a standard closure operation on P δ it induces a meet-completion eΓ : P → Γ[P ] defined by eΓ(p) = p↑. For S ∈ P * we have ΓeΓ(S) = {p ∈ P : p↑ ≥ V{p↑ : p ∈ S}} = {p : p↑ ⊆ Γ(S)} = Γ(S), so ΓeΓ = Γ. Moreover, for all p ∈ P we have eΓe(p) = p↑ = he ◦ e(p) so the diagram in figure 1 commutes. For convenience we summarize the preceding discussion below: Notation 2.4. For meet-completion e : P → Q, and standard closure oper- ator Γ : P δ → P δ we define: Γe : P δ → P δ S 7→ {p ∈ P : e(p) ≥^ e[S]} he : Q → Γe[P ] q 7→ {p ∈ P : e(p) ≥ q} h−1 e : Γe[P ] → Q S 7→^ e[S] 4 P e Q ③③③③③③③③ he eΓe Γe[P ] Figure 1: The equivalence between meet-completions and standard closure operators. eΓ : P → Γ[P ] p 7→ p↑ We have seen, with notation from 2.4: Lemma 2.5. For any meet completion e : P → Q and any standard closure operator Γ : P δ → P δ 1. ΓeΓ = Γ 2. eΓe = he ◦ e Lemma 2.6. If e1 : P → Q1 and e2 : P → Q2 are meet-completions of P and g : Q1 → Q2 is an isomorphism such that g ◦ e1 = e2, then g is unique with this property. Proof. Suppose h is another such isomorphism. Then for all p ∈ P , and for all q ∈ Q1, we have e2(p) ≥ g(q) ⇐⇒ g ◦ e1(p) ≥ g(q) ⇐⇒ e1(p) ≥ q ⇐⇒ h ◦ e1(p) ≥ h(q) ⇐⇒ e2(p) ≥ h(q), so {p ∈ P : e2(p) ≥ g(q)} = {p ∈ P : e2(p) ≥ h(q)} and thus by meet-density we are done. Theorem 2.7. If e : P → Q is a meet-completion then there is a unique isomorphism he between Q and Γe[P ] such that the diagram in figure 1 commutes. Moreover, if e1 : P → Q1 and e2 : P → Q2 are meet-completions such that there is an isomorphism h : Q1 → Q2 with h ◦ e1 = e2 then Γe1 = Γe2. Proof. The isomorphism required has been given as he : q 7→ {p ∈ P : e(p) ≥ q}. That this is an isomorphism is easy to show using the fact that e : P → Q is a meet-completion. Uniqueness follows from Lemma 2.6. 5 / / Finally, if h : Q1 → Q2 with h ◦ e1 = e2 then Γe2(S) = {p ∈ P : e2(p) ≥^ e2[S]} = {p ∈ P : h ◦ e1(p) ≥^ h ◦ e1[S]} = {p ∈ P : h ◦ e1(p) ≥ h(^ e1[S])} = {p ∈ P : e1(p) ≥^ e1[S]} = Γe1(S) 3 Meet-completions and Cartesian products Definition 3.1 (en). If P is a poset then given a map e : P → Q we can define a map en : P n → Qn by en((p1, ..., pn)) = (e(p1), ..., e(pn)). Lemma 3.2. If e : P → Q is a meet-completion and n ≥ 2 then en : P n → Qn will be a meet-completion if and only if P has a top element ⊤ and e(⊤) is the top element of Q. Proof. Suppose first that P has top ⊤ and e(⊤) is the top element of Q. Since a finite product of complete lattices is again a complete lattice it remains only to check that en[P n] is meet-dense in Qn. Given (q1, ..., qn) ∈ Qn we claim that (q1, ..., qn) = V{en((p1, ..., pn)) : e(pi) ≥ qi for all i ∈ {1, ..., n}}. Since e(⊤) is the top element of Q we can be sure that this infimum is of a non-empty set. Now, (q1, ..., qn) is clearly a lower bound, so suppose (q′ n) is another such lower bound. Then, for i ∈ {1, ..., n}, we have q′ i ≤ e(pi) for all pi ∈ P with qi ≤ e(pi), so by meet-density of e[P ] in Q we have q′ n) ≤ (q1, ..., qn) as required. 1, ..., q′ i ≤ qi, and so (q′ 1, ..., q′ For the converse note that Q has a top element ⊤Q since it is complete, and if e(p) < ⊤Q for all p ∈ P then Qn will contain elements of form (e(p1), ⊤Q, e(p3), ..., e(pn)) which are not the infimum of any subset of e[P ]. Example 3.3 illustrates this issue. Example 3.3. Let P = {p}, let Q = {q, ⊤} with q < ⊤ and let e(p) = q. 6 Then e : P → Q is represented in the following diagram. ⊤ / q p ✤ e Now, P 2 = P and e2 : P 2 → Q2 is as follows: (⊤, ⊤) (q, ⊤) (⊤, q) ✉✉✉✉✉✉✉✉✉ ■■■■■■■■■ (p, p) ✤ e2 / (q, q) ■■■■■■■■■ ✉✉✉✉✉✉✉✉✉ Clearly e2 : P 2 → Q2 is not a meet-completion. Although en : P n → Qn is not, in general, a meet-completion we note that the problem elements cannot be below any member of en[P n], and so a true meet-completion can be obtained by simply removing them. We make a formal definition below: Definition 3.4 (❅❅Q n ). If ⊤Qn is the top element of Qn we obtain ❅❅Q n from Qn by removing all elements (q1, ..., qn) of Qn such that {(p1, ..., pn) ∈ P n : (e(p1), ..., e(pn)) ≥ (q1, ..., qn)} = ∅ and (q1, ..., qn) < ⊤Qn. The ordering of the remaining elements is left unchanged. Lemma 3.5. If e : P → Q is a meet-completion then the map en : P n → ❅❅Q n is a meet-completion (en is defined as in definition 3.1). Proof. ❅❅Q n remains a complete lattice as aside from ⊤Qn every element of ❅❅Q n is below an element of en[P n]. This also means that unless (q1, ..., qn) = ⊤Qn we have {en((p1, ..., pn)) : e(pi) ≥ qi for all i ∈ {1, ..., n}} 6= ∅, and so we can adapt the first part of the proof of lemma 3.2 to get the result. Note that ❅❅Q n embeds into Q as a subalgebra. Lemma 3.6. Given meet-completion e : P → Q and n ≥ 2, the map en : P n → Qn is a meet-completion if and only if Qn = ❅❅Q n . Proof. Qn = ❅❅Q n if and only if P has a top element ⊤ and e(⊤) is the top element of Q, so the result follows from lemma 3.2. 7 / / 4 Meet-completions of isotone poset expansions Definition 4.1 (poset expansion). A poset expansion is a structure P = (P, ≤, fi : i ∈ I) such that (P, ≤) is a poset and for each i ∈ I there is ni ∈ N with fi : P ni → P . Note that we define 0 to be an element of N. We say a poset expansion is isotone when fi is isotone for all i ∈ I. If e1 : P1 → Q1 is a meet-completion, e2 : P2 → Q2 is any completion of P2, and f : P1 → P2 is an isotone map, there is an intuitive method (introduced in [17]) for lifting f to an isotone map f : Q1 → Q2, given by This lifting is illustrated by figure 2 below. f (q) =^{e2 ◦ f (p) : e1(p) ≥ q} (1) P1 f e1 / Q1 f P2 / Q2 e2 Figure 2: An intuitive lift for an isotone map. Applying this to the special case of the meet-completions e : P → Q and en : P n → ❅❅Q n , and an isotone map f : P n → P we obtain: f : ❅❅Q n → Q The corresponding commuting square is shown in figure 3. q 7→^{e(f (p1, .., pn) : en(p1, ..., pn) ≥ q} P n en / ❅❅Q n f P f / Q e Figure 3: Lifting isotone operations to ❅❅Q n . We can extend f to an order preserving map f + : Qn → Q by defining: f + : Qn → Q q 7→(f (q) when q ∈ ❅❅Q n ⊤Q otherwise 8 / / / / Note that when Qn 6= ❅❅Q n we have f (⊤Qn) = f (⊤❅Q n ) = ⊤Q. The situation can be summarized by the commuting diagram in figure 4 (here ι stands in both cases for the appropriate inclusion function). In this diagram the maps γ and γ + are induced by the other maps. Lemma 4.2 gives an explicit definition for each. eΓen ∼=(hen ) P n f P en e ❅❅Q n / Qn o ι (Γe[P ])n ∼=(he)n f f + γ+ Q = Q o ∼=(he) / Γe[P ] ι = Γen[(P n)∗] γ Γe[P ] eΓe Figure 4: Lifting isotone operations Lemma 4.2. The maps γ : Γen[(P n)∗] → Γe[P ] and γ + : (Γe[P ])n → Γe[P ] can be defined as follows: 1. γ +(C1, ..., Cn) =(Γe(f [C1 × . . . × Cn]↑) when Ci 6= ∅ for all i ∈ {1, ..., n} 2. γ(C) =(Γe(f [C1 × . . . × Cn]↑) when C = (C1, ..., Cn) 6= ∅ Γe(∅) otherwise Γe(∅) otherwise Proof. First we check that the notation makes sense. Elements of (P n)∗ are the up-closed subsets of P n, so are either empty or of form S1 × . . . × Sn where the Si are non-empty up-closed subsets of P . Elements of (P )n are of form (S1, ..., Sn) where the Si are up-closed subsets of P which may be empty. The functions γ and γ + are defined by composing maps from figure 4. The relevant maps are as follows: (he)n : Qn → (Γe[P ∗])n (q1, ..., qn) 7→ ({p ∈ P : e(p) ≥ q1}, ..., {p ∈ P : e(p) ≥ qn}) 9 # # / / w w ( ( / o / / o o / / 5 5 o / ((he)n)−1 : (Γe[P ∗])n → Qn (C1, ..., Cn) 7→ (^ e[C1], ...,^ e[Cn]) f + : Qn → Q (q1, ..., qn) 7→(V{e(f (p1, ..., pn)) : e(p1, ..., pn) ≥ (q1, ..., qn)} if ⊤Q otherwise. 6= ∅ he : Q → Γe[P ∗] q 7→ {p ∈ P : e(p) ≥ q} 1. γ + is defined to be he ◦ f + ◦ ((he)n)−1. We have two cases. Suppose first that Ci ∈ Γe[P ∗] is non-empty for all i ∈ {1, ..., n}. Then (C1, ..., Cn) 7→ (^ e[C1], ...,^ e[C2]) 7→^{e(f (p1, ..., pn) : pi ∈ Ci for all i)} =^ e[f [C1, ..., Cn]] 7→ {p ∈ P : e(p) ≥^ e[f [C1, ..., Cn]]} = Γe(f [C1 × . . . × Cn]↑) Alternatively, suppose Ci = ∅ for some i. Then (C1, ..., Cn) 7→ (^ e[C1], ...,^ e[C2]) 7→ ⊤Q 7→ Γe(∅) 2. This follows from the fact that γ is ι ◦ γ +. 5 Preserving inequalities in meet-completions of isotone poset expansions Given any standard closure operator Γ : P δ f : P n → P , there is a map γ + : (Γ[P ])n → Γ[P ] defined by → P δ and n-ary function γ +(C1, ..., Cn) = Γ(f [C1 × . . . × Cn]↑) 10 such that the diagram in figure 5 commutes. Note that this γ + is the map that was introduced in lemma 4.2. The cases in the original definition are redundant so long as we set f [∅] = ∅. A special case of this definition is when f is a constant. For this case γ + = Γ({f }↑) = {f }↑. P n f P (eΓ)n (Γ[P ])n γ+ / Γ[P ] eΓ Figure 5: Lifting operations to meet-completions using closure operators Definition 5.1 (Γ(P ∗)). Let P = (P, ≤, fi : i ∈ I) be a poset with isotone operations fi : P ni → P of arities ni (for i ∈ I) and let Γ : P δ be a standard closure operator. We lift all the operations as in the preceding discussion to define → P δ Γ(P ∗) = (Γ[P ∗], ⊇, γ + i : i ∈ I) We note that frequently inequalities that hold with respect to the op- erations of P will fail in this completion. The remainder of this section is devoted to an examination of some conditions which guarantee inequality preservation. Definition 5.2 (Γι). Define Γι to be the identity on P δ . Definition 5.3 (LP ). If P = (P, ≤, fi : i ∈ I) is an isotone poset expansion then LP is the formal language composed of the standard logical symbols of first order logic along with the signature {≤} ∪ {fi : i ∈ I} that corresponds to the operations of P along with the binary relation ≤. Given an isotone poset expansion P = (P, ≤, fi : i ∈ I) we shall talk about terms in the language of P. These are the terms of the language LP constructed as per the usual rules of first order term construction. For example, if P has only the single binary operation f , then f (x, y) would be a term, as would f (x, x), and f (f (x, y), z) etc. Lemma 5.4. Let P = (P, ≤, fi : i ∈ I) be an isotone poset expansion, let x1, ..., xn be variables, and let φ(x1, ..., xn) be a term in the language of P. Let (C1, ..., Cn) ∈ (Γι(P ∗))n and consider Γι(P ∗) as a model for LP 11 / / / by interpreting fi as γ+ for all i ∈ I. Suppose we assign xi = Ci for all i i = 1, ..., n. Then the interpretation of φ(x1, ..., xn) in Γι(P ∗) under this assignment is {φ(y1, ..., yn) : (y1, ..., yn) ∈ C1 × ... × Cn}↑ = φ[C1 × ... × Cn]↑. Proof. We induct on the construction of φ. In the base case φ = fi for some i ∈ I. In this case the interpretation of fi(x1, ..., xn) is just γ + i (C1, ..., Cn) and the result follows from the definitions of γ + i and Γι. For the inductive step we are interested in the case where φ(x1, ..., xn) = f (φ1(¯x1), ..., φn(¯xn)), where f is an n-ary function from LP , ¯xi is a vector of variables from {x1, ..., xn}, and φi(¯xi)) is a term in LP for each i = 1, ..., n. To illustrate the proof we will use the special case where φ(x1, x2, x3) = f (φ1(x1, x2), φ2(x3)) for some φ1 and φ2. The general proof is similar but has a tedious notational burden so we choose to omit it. Now, by definition φ[C1, C2, C3]↑ = {φ(x1, x2, x3) : xi ∈ Ci for i = 1, 2, 3}↑ = {f (φ1(x1, x2), φ2(x3)) : xi ∈ Ci for i = 1, 2, 3}↑ (1) Also, using the inductive hypothesis and the definitions of γ + and Γι we have that the interpretation of φ(x1, x2, x3) in Γι(P ∗) is f [φ1[C1×C2]↑, φ2[C3]↑]↑. Now, f [φ1[C1 × C2]↑, φ2[C3]↑]↑ = {f (a, b) : a ∈ φ1[C1 × C2]↑, b ∈ φ2[C3]↑}↑ (2) where a ≥ φ1(c1, c2) for some (c1, c2) ∈ C1 × C2, and b ≥ φ2(c3) for some c3 ∈ C3. Clearly (1) ⊆ (2), and that (2) ⊆ (1) follows from the fact that f is isotone. Definition 5.5 (P ). Given an isotone poset expansion P = (P, ≤, fi : i ∈ I) define P * = Γι(P ) = (P , ⊇, γ + i : i ∈ I). Proposition 5.6. Let P = (P, ≤, fi : i ∈ I) be an isotone poset expansion, let φ(x1, ..., xn) and ψ(x1, ..., xn) be terms of LP . Then P = φ(x1, ..., xn) ≤ ψ(x1, ..., xn) ⇐⇒ P * = φ(x1, ..., xn) ≤ ψ(x1, ..., xn) assuming P and P * are interpreted as LP -structures in the natural way. 12 Proof. Using lemma 5.4 and the definition of P * P * = φ(x1, ..., xn) ≤ ψ(x1, ..., xn) ⇐⇒ φ[C1 × ... × Cn]↑ ⊇ ψ[C1 × ... × Cn]↑ for all C1, ..., Cn ∈ P * =⇒ φ[x↑ 1 × ... × x↑ ⇐⇒ P = φ(x1, ..., xn) ≤ ψ(x1, ..., xn) n]↑ for all x1, ..., xn ∈ P 1 × ... × x↑ n]↑ ⊇ ψ[x↑ This proves the right to left implication. To prove the other direction let C1, ..., Cn ∈ P . We must show that φ[C1 × ... × Cn]↑ ⊇ ψ[C1 × ... × Cn]↑ for all C1, ..., Cn ∈ P There are two cases. In the trivial case there is i ∈ {1, ..., n} with Ci = ∅. In this case ψ[C1×...×Cn]↑ = ∅ = φ[C1×...×Cn]↑. Suppose instead that Ci 6= ∅ for all i = 1, ..., n and let a′ ∈ ψ[C1 × ... × Cn]↑. Then a′ ≥ a ∈ ψ[C1 × ... × Cn] for some a ∈ P, and thus a = ψ(a1, ..., an) for some (a1, ..., an) ∈ P n. Now, by assumption φ(a1, ..., an) ≤ ψ(a1, ..., an) and so a′ ∈ φ[C1 × ... × Cn]↑ and we are done. The following corollary to this result provides a condition on the rela- tionship between a combination of operations of P and a standard closure operator Γ on P δ sufficient to guarantee the preservation of an inequality in the meet-completion induced by Γ. Somewhat surprisingly it turns out that only the 'larger' term is important here. Corollary 5.7. Let P = (P, ≤, fi : i ∈ I) be an isotone poset expansion, let φ(x1, ..., xn) and ψ(x1, ..., xn) be terms of LP , and let Γ be a standard closure operator on P δ : i ∈ I) as in definition 5.1. Suppose that ψ(C1, ..., Cn) = Γ(ψ[C1×...×Cn]↑) for all (C1, ..., Cn) ∈ Γ[P ]n. Then . Define Γ[P] = (Γ[P ], ⊇, γ + i P = φ(x1, ..., xn) ≤ ψ(x1, ..., xn) =⇒ Γ[P] = φ(x1, ..., xn) ≤ ψ(x1, ..., xn). Proof. By proposition 5.6 we have P * = φ(x1, ..., xn) ≤ ψ(x1, ..., xn), so in particular ψ[C1 × ... × Cn]↑ ⊆ φ[C1 × ... × Cn]↑ for all (C1, ..., Cn) ∈ Γ[P ]n. Note that in Γ(P) we must have Γ(φ[C1 × ... × Cn]↑) ⊆ φ(C1, ..., Cn), and similar for ψ. So Γ(ψ[C1 × ... × Cn]↑) ⊆ Γ(φ[C1 × ... × Cn]↑) ⊆ φ(C1, ..., Cn) and since by assumption ψ(C1, ..., Cn) = Γ(ψ[C1 × ... × Cn]↑) this gives ψ(C1, ..., Cn) ⊆ φ(C1, ..., Cn), and thus Γ[P] = φ(x1, ..., xn) ≤ ψ(x1, ..., xn) as required. 13 Corollary 5.8. With all notation as in corollary 5.7 suppose P = φ(x1, ..., xn) = ψ(x1, ..., xn). Then if 1. ψ(C1, ..., Cn) = Γ(ψ[C1 × ... × Cn]↑), and 2. φ(C1, ..., Cn) = Γ(φ[C1 × ... × Cn]↑) in Γ[P] for all (C1, ..., Cn) ∈ Γ[P ]n, then Γ[P] = φ(x1, ..., xn) = ψ(x1, ..., xn) Proof. Immediate, from corollary 5.7. 6 Ordered domain algebras The axioms in this section originate with Bredikhin, and the presentation here is that used in [11]. Definition 6.1. The class R(;, dom, ran, ⌣, 0, id, ≤) is defined as the iso- morphs of A = (A, ;, dom, ran, ⌣, ∅, id, ⊆) where A ⊆ ℘(U × U ) for some base set U and x ; y = {(u, v) ∈ U × U : (u, w) ∈ x and (w, v) ∈ y for some w ∈ U } dom(x) = {(u, u) ∈ U × U : (u, v) ∈ x for some v ∈ U } ran(x) = {(v, v) ∈ U × U : (u, v) ∈ x for some u ∈ U } x⌣ = {(v, u) ∈ U × U : (u, v) ∈ x} id = {(u, v) ∈ U × U : u = v} for every x, y ∈ A. Let Ax denote the following formulas: Partial order ≤ is reflexive, transitive and antisymmetric, with lower bound 0. Isotonicity and normality the operators ⌣, ;, dom, ran are isotonic, e.g. a ≤ b → a ; c ≤ b ; c etc. and normal 0⌣ = 0 ; a = a ; 0 = dom(0) = ran(0) = 0. Involuted monoid ; is associative, id is left and right identity for ;, id⌣ = id and ⌣ is an involution: (a⌣)⌣ = a, (a ; b)⌣ = b⌣ ; a⌣. Domain/range axioms 14 (D1) dom(a) = (dom(a))⌣ ≤ id = dom(id) (D2) dom(a) ≤ a ; a⌣ (D3) dom(a⌣) = ran(a) (D4) dom(dom(a)) = dom(a) = ran(dom(a)) (D5) dom(a) ; a = a (D6) dom(a ; b) = dom(a ; dom(b)) (D7) dom(dom(a) ; dom(b)) = dom(a) ; dom(b) = dom(b) ; dom(a) Two consequences of these axioms (use (D6), (D7) for the first, use (D4), (D5) for the second) are (D8) dom(dom(a) ; b) = dom(a) ; dom(b) (D9) dom(a) ; dom(a) = dom(a) A model of these axioms is called an ordered domain algebra (ODA). Each of the axioms (D1) -- (D8) has a dual axiom, obtained by swapping domain and range and reversing the order of compositions, and we denote the dual axiom by a ∂ superscript, thus for example, (D6)∂ is ran(b ; a) = ran(ran(b) ; a). The dual axioms can be obtained from the axioms above, using the involution axioms and (D3). Another consequence of the ODA axioms is the following lemma, which we shall use later. Lemma 6.2. Let B be any ODA and let b, c ∈ B. Then dom(b ; c) ; b ≥ b ; dom(c) b ; ran(c ; b) ≥ ran(c); b and Proof. dom(b ; c) ; b = dom(b ; dom(c)) ; b ≥ dom(b ; dom(c)) ; b ; dom(c) = b ; dom(c) by (D6) (D1) (D5) The other part is similar. 15 7 A completion Definition 7.1 (ΓD). Given an ODA A with underlying poset P , define ΓD : P δ by defining the closed sets of P to be those X ∈ P * such that {dom(x) ; y ; ran(z) : x, y, z ∈ X}↑ = X. → P δ Lemma 7.2. ΓD is a standard closure operator on P δ . Proof. Routine. Lemma 7.3. Given X ∈ P , if we define X0 = X, and Xn+1 = {dom(x) ; y ; ran(z) : x, y, z ∈ Xn}↑ for all n ∈ ω Proof. It's easy to show that Xn ⊆ Xn+1 for all n ∈ ω, so given x, y, z ∈ then ΓD(X) =Sω Xn. Sω Xn there is a k ∈ ω with x, y, z ∈ Xk. Thus (dom(x) ; y ; ran(z))↑ ⊆ Xk+1 ⊆Sω Xn, henceSω Xn is ΓD-closed. Clearly any closed set containing X must contain Sω Xn, so we must have ΓD(X) =Sω Xn as required. The next definition is a special case of definition 5.1. Definition 7.4 (ΓD[A]). Given an ODA A with underlying poset P , we define ΓD[A] = (ΓD[P ], ⊇, γ + : fi ∈ {;, dom, ran, ⌣, 0, id}). For clarity we i will write ;γ, domγ, ranγ, ⌣γ , 0γ, idγ for the operations γ + i . Theorem 7.5 (Hirsch, Mikulas). Let A be an ordered domain algebra. The map h : A → ℘(ΓD[A] × ΓD[A]) defined by (X, Y ) ∈ h(a) ⇐⇒ X ;γ a ↑ ⊆ Y and Y ;γ(a⌣)↑ ⊆ X is a representation of A over the base ΓD[A]. This theorem is proved, with minor notational variations, in the proof of [11, theorem 2.2]. Lemma 7.6. Let A be an ODA with underlying poset P and consider the closure operator ΓD. Then for f ∈ {dom, ran,⌣ } and C ∈ ΓD[P ] we have γ +(C) = f [C]↑. Moreover, 0γ = {0}↑ and idγ = {id}↑. 16 Proof. That 0γ and idγ are just {0}↑ and {id}↑ is direct from the definition of γ +. For dom let C ∈ ΓD[P ] and let x, y, z ∈ C. We are required to show that dom[C]↑ is ΓD-closed. Now, dom(dom(x)) ; dom(y) ; ran(dom(z)) = dom(x) ; dom(y) ; dom(z) = dom(dom(x) ; y) ; dom(z) by ODA axioms (D4), (D7), and (D8). As C is ΓD-closed we must have dom(x) ; y ∈ C, so we have something of form dom(x′) ; dom(z) for x′, z ∈ C. Another application of (D8) gives dom(x′) ; dom(z) = dom(dom(x′) ; z), and thus as C is closed we have something of form dom(y′) for y′ ∈ C, which is in dom[C]. The ran case is dual, and the ⌣ case follows from axiom (D3) and the fact that ⌣ is an involution. We ask how close ΓD[A] is to being an ODA. It turns out that most of the axioms (D1)-(D8) hold (proposition 7.7), with the exceptions being (D2) and (D6) (Examples 7.11 and 7.12), the operations on ΓD[A] remain isotone and normal, idγ remains a left and right identity for composition and ⌣γ is still an involution (Lemma 7.9). The dramatic deviation is that ;γ is not necessarily associative (Example 7.13). The remainder of this section will be taken up with proving the claims in this paragraph. Proposition 7.7. Given ODA A, axioms (D1), (D3), (D4), (D5), and (D7) hold in ΓD[A]. Proof. That ΓD[A] = {(D1), (D3), (D4)} follows easily from corollary 5.7 and Lemma 7.6. Since domγ(C1) ;γ domγ(C2) = ΓD(dom[C1] ; dom[C2]↑) for all C1, C2 ∈ ΓD[A], by corollary 5.7 it is a necessary and sufficient con- dition for ΓD[A] = (D7) that ΓD(dom[dom[C1]; dom[C2]]↑) = dom[ΓD(dom[C1]; dom[C2])]↑ for all C1, C2 ∈ ΓD[A]. We shall show that (dom[C1]; dom[C2])↑ is ΓD- closed, as in that case the required equality follows from lemma 7.6: Let x1, x2, x3 ∈ C1, and let y1, y2, y3 ∈ C2. Then dom(dom(x1) ; dom(y1)) ; dom(x2) ; dom(y2) ; ran(dom(x3) ; dom(y3)) = dom(x1) ; dom(x2) ; dom(x3) ; dom(y1) ; dom(y2) ; dom(y3) by axioms (D4) and (D7). Since dom[C1]↑ and dom[C2]↑ are closed by lemma 7.6 it is easy to show that dom(x1) ; dom(x2) ; dom(x3) ∈ dom[C1]↑ and dom(y1) ; dom(y2) ; dom(y3) ∈ dom[C2]↑ and thus ΓD[A] = (D7) as required. That ΓD[A] = (D5) follows easily from lemma 7.6. Lemma 7.8. For all S ∈ P *, ΓD(S)⌣ = ΓD(S⌣). 17 Proof. Since S⌣ ⊆ ΓD(S)⌣ and ΓD(S)⌣ is ΓD-closed by lemma 7.6, ⊇ follows from the isotonicity of closure operators. Define X0 = S and Xn 0 = S⌣ ⊆ ΓD(S⌣), and for all as in lemma 7.3 for all n ∈ ω. Then X⌣ k ∈ ω and every a ∈ Xk we have a ≥ b = dom(b1) ; b2 ; ran(b3) for some b1, b2, b3 ∈ Xk−1. So b⌣ = dom(b⌣ 1 ) by involution and axioms k ⊆ ΓD(S⌣). Since (D1) and (D3), and thus if X⌣ n we are done. k−1 ⊆ ΓD(S⌣) then X⌣ Lemma 7.9. For all f ∈ {;, dom, ran, ⌣, 0, id} the extension γ + is isotone and normal, moreover 3 ) ; b⌣ 2 ; ran(b⌣ ΓD(S)⌣ =Sn∈ω X⌣ 1. idγ is a left and right identity for ;γ, and 2. ⌣γ is an involution. Proof. Isotonicity of the operations is automatic from the lifting process, and normality follows from the fact that 0γ = {0}↑. That idγ is a left and right identity for ;γ follows easily from the definition of ;γ and the fact that idγ = {id}↑. To see that (a ; b)⌣ = b⌣ ; a⌣ holds in ΓD[A] define terms φ(x, y) = (x ; y)⌣ and ψ(x, y) = x⌣ ; y⌣ in LP . Then using lemma 7.6 it's easy to see that in ΓD[A] we have ψ(C, D) = ΓD(ψ[C × D]↑) for all C, D ∈ ΓD[A], and that φ(C, D) = ΓD(φ[C × D]↑) follows from lemma 7.8. The result then follows from corollary 5.8. Lemma 7.10. Let X, Y ∈ ΓD[A]. If domγ(X) = domγ(Y ) and ranγ(X) = ranγ(Y ) then X ∪ Y ∈ ΓD[A]. Proof. Let z1, z2, z3 ∈ X ∪ Y . We are required to prove that dom(z1); z2; ran(z3) ∈ X ∪ Y Without loss of generality, let z2 ∈ X. Since domγ(X) = domγ(Y ) we know that dom(z1) ∈ domγ(X) and similarly ran(z3) ∈ ranγ(X), hence dom(z1); z2; ran(z3) ∈ X, by the closure of X. Example 7.11. (D2) can fail in ΓD[A]. Let A be the full proper ODA over a base of four elements {a, b, c, d}. Define x, y ∈ A by x = {(a, b), (c, d)}, and y = {(a, d), (c, b)}. Let A = {x, y}↑. Then dom(x) = dom(y) and ran(x) = ran(y), and consequently A is ΓD-closed. We aim to show that A ;γ A⌣γ 6⊆ domγ(A). For this claim, x ; y⌣ ∈ ΓD(A ; A⌣)↑, x ; y⌣ = {(a, c), (c, a)}, and domγ(A) = dom(x)↑ = dom(y)↑ = {(a, a), (c, c)}↑, so x ; y⌣ 6∈ domγ(A) hence A ;γ A⌣γ 6⊆ domγ A and thus ΓD[A] 6= (D2). 18 Example 7.12. (D6) can fail in ΓD[A]. Let A be the full proper ODA over the two element base {a, b}. Define x = {(a, b), (b, a)} and let id = {(a, a), (b, b)} be the identity as usual. Let B = {x, id}↑. Then, as dom(x) = dom(id) = ran(x) = ran(id) = id, B is ΓD-closed. Define A = {{(a, a)}}↑. Then A ;γ B = {{(a, a)}}↑ ;γ {id, x}↑ = ΓD({{(a, a)}, {(a, b)}}↑) = ∅↑ because {(a, b)}; ran {(a, a)} = ∅. Hence domγ(A ;γ B) = domγ(∅↑) = ∅↑. However, domγ(B) = idγ so domγ(A ;γ domγ B) = domγ A = {(a, a)}↑ 6= dom(A ;γ B) and thus ΓD[A] 6= (D6). Example 7.13. Associativity can fail in ΓD[A]. Let A be the full proper ODA over a base of five elements {a, b, c, d, e}, let x = {(a, a)}, let y = {(a, b), (c, d)}, let z = {(a, d), (c, b)}, and let u = {(b, e), (d, e)}. Define A = {x}↑, B = {y, z}↑, and C = {u}↑. Then A and C are principal and hence ΓD-closed, and dom(z) = dom(y) and ran(z) = ran(y) so B is also ΓD-closed. Now, A ;γ B = ΓD({x ; y, x ; z}↑) = ΓD({{(a, b)}, {(a, d)}}↑) = ∅↑, as {(a, b)} ; ran({(a, d)}) = ∅, so (A ;γ B) ;γ C = ∅↑. However, B ;γ C = ΓD{y ; u, z ; u}↑ = {{(a, e), (c, e)}}↑, which is principal and hence ΓD-closed. Thus A ;γ(B ;γ C) = ΓD({x ; y ; u, x ; z ; u}↑) = ΓD({{(a, e)}}↑) = {{(a, e)}}↑ 6= ∅↑, and so (A ;γ B) ;γ C 6= A ;γ(B ;γ C). This example also shows that the weak associativity law, where associativity is only required for A ≤ idγ, can fail in ΓD[A]. Problem 7.14. Consider the partial binary relation ∗ on the base of ΓD[A] where B ∗ C is only defined (for B, C ∈ ΓD[A]) if ranγ(B) = domγ(C) and then B ∗ C = B ;γ C. Is ∗ associative, i.e. is A ∗ (B ∗ C) = (A ∗ B) ∗ C whenever either side is defined? 8 Conclusions and further work We have shown how to lift the operators of an isotone poset expansion to a meet-completion. We have identified a family of equations preserved in certain meet-completions. For the particular case of ordered domain algebras we have defined a meet-completion ΓD[A] of an ordered domain algebra A and shown that some, but not all, of the equations defining ordered domain 19 algebras are inherited by ΓD[A]. Furthermore we have seen that ΓD[A] may be used as the base of a representation of A. It may be of interest to compare all this with similar research in the field of Boolean algebras with operators (BAOs), in particular relation al- gebras, where the signature is more expressive. There are many possible completions of a BAO ranging from the MacNeille completion (in some of the references below this is simply called the completion) up to the canonical extension. Much work has been done to identify a large class of formulas preserved by completions, e.g. [21, 2, 7]. More generally, canonical exten- sions, MacNeille completions, and other completions for lattice and poset expansions have received considerable attention in recent years, partly due to their connections to non-classical logics (see e.g. [3, 5, 6]). The canonical extension of a Boolean algebra with operators can be defined by the second dual of the Boolean algebra, with operators lifted from the algebra [13]. An equation is canonical if it holds in the canonical extension of an algebra whenever it holds in the algebra itself. Not all equations are canonical, but all the equations defining the class of relation algebras are canonical, [14] or [9, theorem 3.16]. By a theorem due to J. Monk, the class of all representable relation algebras is a canonical class (an algebra is representable if and only if its canonical extension is, for a proof see [16] or [9, theorem 3.36]) but any equational axiomatization of this representation class must involve infinitely many non-canonical equations [12]. The MacNeille completion of a BAO preserves essentially infinite meets and joins. Again the operators are lifted from the algebra to its comple- tion. It has been shown that a representable relation algebra can have an unrepresentable MacNeille completion [10], hence some equations fail to be preserved when passing to the MacNeille completion. Bearing this in mind there are two primary directions the work here can be extended. First, by building a deeper understanding of the preservation of inequalities by meet-completions, in the spirit of [21] and in particular its numerous algebraic descendents (e.g. [6, 7, 8]). Section 5 makes a small step in this direction, but it seems likely that the results here could be refined and extended considerably. Second, the inspiration for this paper was the implicit appearance of the meet-completion structure described in section 7 in the results of [11]. It remains to be seen whether this is an isolated event or whether the role of the completion in the representation process hints at some deeper structure which could possibly be used for further representation results. With these thoughts in mind we propose some problems we believe are of particular interest: 20 Problem 8.1. Let A be an ODA and let Γ be a standard closure operator defining a completion of A. Under what conditions is the completion ΓD[A] an ODA, or at least when is the completion associative? More generally, under what conditions is Γ[A] associative? Problem 8.2. Is it possible to generalize the definition of a representation of an ordered domain algebra in such a way that the completion ΓD[A] of an ordered domain algebra does possess a weak representation. C.f. For rela- tion algebras we can generalize the notion of a representation to a relativized representation where all operators are restricted to some reflexive and sym- metric (but not necessarily transitive) maximal relation. When evaluated in a relativized representation, composition need not be associative. A weakly associative algebra is a relation-type algebra obeying all the relation algebra axioms except perhaps associativity and satisfying the weak associativity axiom (x; 1); 1 = x; (1; 1) instead. Maddux proved that the class of relation- type algebras isomorphic to algebras of binary relations with relativized operators is the class of weakly associative algebras [15]. Problem 8.3. All of the ODA axioms except associativity, (D2) and (D6) are valid over ΓD(ODA) = {ΓD[A] : A ∈ ODA}. Can this set of axioms be extended to a (finite) set of formulas (or equations) so as to define the closure under isomorphism of ΓD(ODA)? References [1] D. Bredikhin. An abstract characterization of some classes of algebras of binary relations. In Algebra and Number theory, number 2, Kabardino- Balkarsk. Gos. Univ., Nalchik, 1977. (In Russian). [2] M. de Rijke and Y. Venema. Sahlqvist's theorem for Boolean alebras with operators with an application to cylindric algebras. Studia Logica, 54:61 -- 78, 1995. [3] J.M. Dunn, M. Gehrke, and A. Palmigiano. Canonical extensions and relational completeness of some substructural logics. J. Symb. Logic, 70:713 -- 740, 2005. [4] M. Ern´e. Adjunctions and standard constructions for partially ordered sets. In Holder, Pichler, and Tempsky, editors, Contributions to General Algebra, volume 2, pages 77 -- 106, 1983. 21 [5] M. Gehrke, J. Harding, and Y. Venema. MacNeille completions and canonical extensions. Trans. Am. Math. Soc., 358:573 -- 590, 2006. [6] M. Gehrke, H. Nagahashi, and Y. Venema. A Sahlqvist theorem for distributive modal logic. Ann. Pure and Appl. Logic, 131:65 -- 102, 2005. [7] S. Givant and Y. Venema. The preservation of Sahlqvist equations in completions of Boolean algebras with operators. Algebra Universalis, 41:47 -- 84, 1999. [8] R. Goldblatt. Varieties of complex algebras. Ann. Pure and Appl. Logic, 44:173 -- 242, 1989. [9] R. Hirsch and I. Hodkinson. Relation Algebras by Games. North- Holland, Amsterdam, NL, 2002. [10] R Hirsch and I. Hodkinson. Strongly representable atom structures of relation algebras. Proc. Amer. Math. Soc., 130:1819 -- 1831, 2002. [11] R Hirsch and S. Mikul´as. Ordered domain algebras. Journal of Applied Logic, 11(3):253 -- 271, 2013. [12] I. Hodkinson and Y. Venema. Canonical varieties with no canonical axiomatisation. Trans. Am. Math. Soc., 357:4579 -- 4605, 2005. [13] B. J´onsson and A. Tarski. Boolean algebras with operators I. Am. J. Math., 73:891 -- 939, 1951. [14] B. J´onsson and A. Tarski. Boolean algebras with operators II. Am. J. Math., 74:127 -- 162, 1952. [15] R. Maddux. Some varieties containing relation algebras. Trans. Amer. Math. Soc., 272(2):501 -- 526, 1982. [16] R. Maddux. A sequent calculus for relation algebras. Ann. Pure and Appl. Logic, 25:73 -- 101, 1983. [17] J. D. Monk. Completions of Boolean algebras with operators. Math. Nachrichten, 46:47 -- 55, 1970. [18] O. Ore. Combinations of closure relations. Annals of Math., 44:514 -- 533, 1943. [19] O. Ore. Some studies on closure relations. Duke Math. J., 10:573 -- 627, 1943. 22 [20] O. Ore. Galois connexions. Trans. Amer. Math., 55:493 -- 513, 1944. [21] H. Sahlqvist. Completeness and correspondence in the first and second order semantics for modal logic. In S. Kanger, editor, Proceedings of the Third Scandinavian Logic Symposium 1973, volume 82 of Stud. Logic Found. Math., pages 110 -- 143, Amsterdam, 1975. North-Holland. [22] M Vilain and H. Kautz. Constraint propagation algorithms for temporal reasoning. In Proc. fifth national conference on artificial intelligence, 1986. 23
1810.03152	1	1810	2018-10-07T14:31:29	Third-order Jacobsthal Generalized Quaternions	[ "math.RA" ]	In this paper, the third-order Jacobsthal generalized quaternions are introduced. We use the well-known identities related to the third-order Jacobsthal and third-order Jacobsthal-Lucas numbers to obtain the relations regarding these quaternions. Furthermore, the third-order Jacobsthal generalized quaternions are classified by considering the special cases of quaternionic units. We derive the relations between third-order Jacobsthal and third-order Jacobsthal-Lucas generalized quaternions.	math.RA	math	THIRD-ORDER JACOBSTHAL GENERALIZED QUATERNIONS GAMALIEL CERDA-MORALES1 1Instituto de Matem´aticas, Pontificia Universidad Cat´olica de Valpara´ıso, Blanco Viel 596, Valpara´ıso, Chile. E-mails: [email protected] / [email protected] Abstract. In this paper, the third-order Jacobsthal generalized quater- nions are introduced. We use the well-known identities related to the third- order Jacobsthal and third-order Jacobsthal-Lucas numbers to obtain the relations regarding these quaternions. Furthermore, the third-order Ja- cobsthal generalized quaternions are classified by considering the special cases of quaternionic units. We derive the relations between third-order Jacobsthal and third-order Jacobsthal-Lucas generalized quaternions. Mathematical subject classification: Primary: 11B37; Secondary: 11R52, 11Y55. Key words: Third-order Jacobsthal number, generalized quaternion, split quaternion, semi- quaternion, third-order Jacobsthal and third-order Jacobsthal-Lucas generalized quaternion. 1. Introduction and Preliminaries Recently, the topic of number sequences in real normed division algebras has attracted the attention of several researchers. It is worth noticing that there are exactly four real normed division algebras: real numbers (R), complex numbers (C), quaternions (H) and octonions (O). In [Bae] Baez gives a comprehensive discussion of these algebras. The real quaternion algebra H = {q = qr + qii + qj j + qkk : qr, qs ∈ R, s = i, j, k} is a 4-dimensional R-vector space with basis {1 ≃ e0, i ≃ e1, j ≃ e2, k ≃ e3} satisfying multiplication rules qr1 = qr, e1e2 = −e2e1 = e3, e2e3 = −e3e2 = e1 and e3e1 = −e1e3 = e2. There has been an increasing interest on quaternions and octonions that play an important role in various areas such as computer sciences, physics, differential geometry, quantum physics, signal, color image processing and geostatics (for more details, see [Ad, Car, Go1, Go2, Ko1, Ko2]). A variety of new results on Fibonacci-like quaternion and octonion numbers can be found in several papers [Ce, Ci1, Ci2, Ha1, Ha2, Ho1, Ho2, Iy, Ke-Ak, 1 2 G. CERDA-MORALES Szy-Wl]. The origin of the topic of number sequences in division algebra can be traced back to the works by Horadam in [Ho1] and by Iyer in [Iy]. In this sense, Horadam [Ho1] defined the quaternions with the classic Fibonacci and Lucas number components as QFn = Fn + Fn+1i + Fn+2j + Fn+3k (Fn1 = Fn) and QLn = Ln + Ln+1i + Ln+2j + Ln+3k (Ln1 = Ln), respectively, where Fn and Ln are the n-th classic Fibonacci and Lucas numbers, respectively, and the author studied the properties of these quaternions. Several interesting and useful extensions of many of the familiar quaternion numbers (such as the Fibonacci and Lucas quaternions [Ak, Ha1, Ho1], Pell quaternion [Ca, Ci1] and Jacobsthal quaternions [Szy-Wl] have been considered by several authors. After the work of Hamilton, James Cockle introduced the set of split quater- nions which can be represented as 2 = e2 H(1,−1) = {q = qr + qie1 + qje2 + qke3 : qr, qs ∈ R, s = i, j, k} 1 = −1, e2 where e2 3 = 1 and e1e2e3 = 1. Note that e1e2 = e3 = −e2e1, e2e3 = −e1 = −e3e2 and e3e1 = e2 = −e1e3. The set of split quaternions is also noncommutative. Unlike quaternion algebra, the set of split quaternions contains zero divisors, nilpotent and nontrivial idempotent elements, [Ku]. For more properties of the split quaternions the reader is refereed to [Oz]. The set of generalized quaternions which can be represented as H(α,β) = {q = qr + qie1 + qje2 + qke3 : qr, qs ∈ R, s = i, j, k}, where e1, e2 and e3 are quaternionic units which satisfy the equalities (1.1) ( e2 1 = −α, e2 2 = −β, e2 3 = −αβ, e1e2 = e3 = −e2e1, e2e3 = βe1 = −e3e2, e3e1 = αe2 = −e1e3, where α, β ∈ R. By choosing α and β, there are following special cases: • α = β = 1 is considered, then H(1,1) is the algebra of real quaternions. • α = 1, β = −1 is considered, then H(1,−1) is the algebra of split quater- nions. • α = 1, β = 0 is considered, then H(1,0) is the algebra of semi-quaternions. • α = −1, β = 0 is considered, then H(−1,0) is the algebra of split semi- • α = β = 0 is considered, then H(0,0) is the algebra of 1 4 -quaternions. quaternions. THIRD-ORDER JACOBSTHAL GENERALIZED QUATERNIONS 3 Pottman and Wallner provided a brief introduction of the generalized quater- nions in [Po-Wa]. Furthermore, in [Ja-Ya], Jafari and Yaylı studied some alge- braic properties of generalized quaternions and operations over them. A gener- alized quaternion q is a sum of a scalar and a vector, called scalar part, Sq = qr, and vector part Vq = qie1 + qj e2 + qke3 ∈ R3 αβ. Therefore, H(α,β) forms a 4- dimensional real space which contains the real axis R and a 3-dimensional real linear space E3 αβ (for more details, see [Ja-Ya]). αβ, so that, H(α,β) = RL E3 2. Third-order Jacobsthal Quaternions The Jacobsthal numbers have many interesting properties and applications in many fields of science (see, e.g., [Ba, Ho3]). The Jacobsthal numbers Jn are defined by the recurrence relation (2.1) J0 = 0, J1 = 1, Jn+1 = Jn + 2Jn−1, n ≥ 1. Another important sequence is the Jacobsthal-Lucas sequence. This sequence is defined by the recurrence relation (2.2) (see, [Ho3]). j0 = 2, j1 = 1, jn+1 = jn + 2jn−1, n ≥ 1. In [Cook-Bac] the Jacobsthal recurrence relation is extended to higher order recurrence relations and the basic list of identities provided by A. F. Horadam [Ho3] is expanded and extended to several identities for some of the higher order n }n≥0, and third cases. For example, the third order Jacobsthal numbers, {J (3) order Jacobsthal-Lucas numbers, {j(3) n+1 + 2J (3) n }n≥0, are defined by n , J (3) 0 = 0, J (3) n+3 = J (3) J (3) n+2 + J (3) 1 = J (3) (2.3) 2 = 1, n ≥ 0, and (2.4) j(3) n+3 = j(3) n+2 + j(3) n+1 + 2j(3) n , j(3) 0 = 2, j(3) 1 = 1, j(3) 2 = 5, n ≥ 0, respectively. The following properties given for third order Jacobsthal numbers and third order Jacobsthal-Lucas numbers play important roles in this paper (see [Ce, Cook-Bac]). (2.5) (2.6) (2.7) 3J (3) n + j(3) n = 2n+1, n = 2j(3) n−3, n − 3J (3) j(3) n =( −2 if n ≡ 1 (mod 3) if n 6≡ 1 (mod 3) 1 J (3) n+2 − 4J (3) , 4 (2.8) (2.9) (2.10) (2.11) (2.12) (2.13) and (2.14) n − 4J (3) j(3) n = n+2 = n − J (3) j(3) n−3(cid:17)2 (cid:16)j(3) k =( J (3) k =( j(3) n (cid:17)2 (cid:16)j(3) nXk=0 nXk=0 J (3) j(3) G. CERDA-MORALES if n ≡ 0 (mod 3) 2 −3 if n ≡ 1 (mod 3) 1 if n ≡ 2 (mod 3) j(3) n+1 + j(3) n = 3J (3) n+2, 1 if n ≡ 0 (mod 3) −1 if n ≡ 1 (mod 3) if n ≡ 2 (mod 3) 0 + 3J (3) n j(3) n = 4n, , , n+1 J (3) n+1 − 1 n+1 − 2 j(3) n+1 + 1 if n 6≡ 0 (mod 3) if n ≡ 0 (mod 3) , if n 6≡ 0 (mod 3) if n ≡ 0 (mod 3) n (cid:17)2 − 9(cid:16)J (3) = 2n+2j(3) n−3. Using standard techniques for solving recurrence relations, the auxiliary equa- tion, and its roots are given by x3 − x2 − x − 2 = 0; x = 2, and x = −1 ± i√3 2 . Note that the latter two are the complex conjugate cube roots of unity. Call them ω1 and ω2, respectively. Thus the Binet formulas can be written as (2.15) and J (3) n = 2 7 2n − (2.16) 7 respectively. Now, we use the notation j(3) n = 2n + 8 7 3 + 2i√3 21 3 + 2i√3 3 − 2i√3 21 ωn 2 ωn 1 − ωn 1 + 3 − 2i√3 7 ωn 2 , (2.17) V (3) n = Aωn 2 1 − Bωn ω1 − ω2 2 if n ≡ 0 (mod 3) −3 if n ≡ 1 (mod 3) 1 if n ≡ 2 (mod 3) , where A = −3 − 2ω2 and B = −3 − 2ω1. Furthermore, note that for all n ≥ 0 we have = (2.18) n+2 = −V (3) V (3) n+1 − V (3) n , V (3) 0 = 2 and V (3) 1 = −3. THIRD-ORDER JACOBSTHAL GENERALIZED QUATERNIONS 5 From the Binet formulas (2.15), (2.16) and Eq. (2.17), we have (2.19) J (3) n = 1 7(cid:16)2n+1 − V (3) n (cid:17) and j(3) n = 1 7(cid:16)2n+3 + 3V (3) n (cid:17) . In [Ce], the author introduced the so-called third order Jacobsthal quater- nions, which are a new class of quaternion sequences. They are defined by (2.20) J Q(3) n = J (3) n+ses = J (3) n + 3Xs=0 3Xs=1 J (3) n+ses, (J (3) n 1 = J (3) n ) is the n-th third order Jacobsthal number, e2 1 = e2 2 = e2 3 = −1 and n where J (3) e1e2e3 = −1. The main objective of this paper is to define third-order Jacobsthal general- ized quaternions and obtain the relations related to these quaternions. (i.e., for split third-order Jacobsthal quaternions, third-order Jacobsthal semi-quaternions and split third-order Jacobsthal semi-quaternions). 3. Third-order Jacobsthal Generalized Quaternions The third-order Jacobsthal and third-order Jacobsthal-Lucas generalized quater- nions have respectively the expressions of following forms (3.1) and (3.2) J Q(3) α,β,n = jQ(3) α,β,n = 3Xs=0 3Xs=0 J (3) n+ses = J (3) n + j(3) n+ses = j(3) n + J (3) n+ses, (J (3) n 1 = J (3) n ) j(3) n+ses, (j(3) n 1 = j(3) n ), 3Xs=1 3Xs=1 n is the n-th third-order Jacobsthal number, j(3) where J (3) n is the n-th third-order Jacobsthal-Lucas number and e1, e2 and e3 are quaternionic units which satisfy the equalities ( e2 1 = −α, e2 2 = −β, e2 3 = −αβ, e1e2 = e3 = −e2e1, e2e3 = βe1 = −e3e2, e3e1 = αe2 = −e1e3. Let us denote the sets of the third-order Jacobsthal and third-order Jacobsthal- Lucas generalized quaternions by J Q(3) α,β respectively and their natural basis by choosing α and β: α,β and jQ(3) • For α = β = 1, J Q(3) and jQ(3) [Ce]. 1,1 is the set of third-order Jacobsthal real quaternions 1,1 is the set of third-order Jacobsthal-Lucas real quaternions, • For α = 1, β = −1, J Q(3) quaternions and jQ(3) quaternions. 1,−1 is the set of split third-order Jacobsthal 1,−1 is the set of split third-order Jacobsthal-Lucas 6 G. CERDA-MORALES quaternions. • For α = 1, β = 0, J Q(3) • For α = −1, β = 0, J Q(3) • For α = β = 0, J Q(3) semi-quaternions. 1,0 is the set of third-order Jacobsthal semi- −1,0 is the set of split third-order Jacobsthal 0,0 is the set of third-order Jacobsthal 1 4 -quaternions. Throughout this paper, we study on third-order Jacobsthal generalized quater- nions J Q(3) α,β . Similar relations hold for third-order Jacobsthal-Lucas generalized quaternions jQ(3) α,β. In the following we will study the important properties of the third-order Jacobsthal generalized quaternions and third-order Jacobsthal-Lucas generalized quaternions: • The sum and subtract of J Q(3) α,β,n and jQ(3) α,β,n is defined as J Q(3) α,β,n ± jQ(3) α,β,n = 7Xs=0 (J (3) n+s ± j(3) n+s)es, α,β,n = S JQ(3) α,β,n + V JQ(3) α,β,n , where s=1 J (3) n+ses are called the scalar and where J Q(3) Furthermore, we can be written as J Q(3) α,β,n ∈ H(α,β). α,β,n, jQ(3) S JQ(3) α,β,n = J (3) n and V JQ(3) vector parts, respectively. α,β,n =P7 n − h(cid:16)V jQ(3) α,β,n + J (3) n V • The multiplication of these quaternions are defined by J Q(3) α,β,n = J (3) n j(3) , V jQ(3) α,β,n · jQ(3) JQ(3) α,β,n α,β,n(cid:17) + V + j(3) n V JQ(3) α,β,n JQ(3) α,β,n × V jQ(3) α,β,n , where h(cid:16)V JQ (3) α,β,n and , V jQ (3) α,β,n(cid:17) = αJ Q(3) α,β,n+1 · jQ(3) α,β,n+1 + βJ Q(3) α,β,n+2 · jQ(3) α,β,n+2 + αβJ Q(3) α,β,n+3 · jQ(3) α,β,n+3 (3.3) (3.4) V JQ(3) α,β,n × V jQ(3) α,β,n = βi(cid:16)J Q(3) + αj(cid:16)J Q(3) + k(cid:16)J Q(3) α,β,2 · jQ(3) α,β,3 · jQ(3) α,β,1 · jQ(3) α,β,n is defined by α,β,3 − J Q(3) α,β,1 − J Q(3) α,β,2 − J Q(3) α,β,2(cid:17) α,β,3 · jQ(3) α,β,3(cid:17) α,β,1 · jQ(3) α,β,1(cid:17) . α,β,2 · jQ(3) • The conjugate of J Q(3) (3.5) J Q(3) α,β,n = S JQ(3) α,β,n − V JQ(3) α,β,n = J (3) n − J (3) n+ses 7Xs=1 THIRD-ORDER JACOBSTHAL GENERALIZED QUATERNIONS 7 and this operation satisfies (3.6) (3.7) (3.8) J Q(3) α,β,n = J Q(3) α,β,n = J Q(3) α,β,n = jQ(3) α,β,n, α,β,n + jQ(3) α,β,n · J Q(3) α,β,n, α,β,n, J Q(3) J Q(3) α,β,n + jQ(3) α,β,n · jQ(3) α,β,n ∈ H(α,β). for all J Q(3) α,β,n, jQ(3) • The norm of an third-order Jacobsthal generalized quaternion, which agrees with the standard Euclidean norm on R4 is defined as N r(J Q(3) • The inverse of J Q(3) α,β,n(cid:12)(cid:12)(cid:12). α,β,n(cid:12)(cid:12)(cid:12) =(cid:12)(cid:12)(cid:12)J Q(3) α,β,n) =(cid:12)(cid:12)(cid:12)J Q(3) α,β,n · J Q(3) α,β,n · J Q(3) α,β,n(cid:17)−1 α,β,n 6= 0 is given by (cid:16)J Q(3) α,β,n(cid:17)−1 (cid:16)J Q(3) α,β,n(cid:17)−1 =(cid:16)jQ(3) α,β,n(cid:17)−1 ·(cid:16)J Q(3) α,β,n · jQ(3) = . From the above two definitions it is deduced that JQ(3) α,β,n N r(JQ (3) α,β,n) . Now, by the addition, subtraction and multiplication we can give the following theorems. Theorem 3.1. Let J Q(3) For n ≥ 1, the following relations hold (3.9) α,β,n + J Q(3) 2J Q(3) α,β,n+1 + J Q(3) α,β,n+2 = J Q(3) α,β,n+3, α,β,n be the third-order Jacobsthal generalized quaternion. (3.10) J Q(3) α,β,n − e1J Q(3) α,β,n+1 − e2J Q(3) α,β,n+2 − e3J Q(3) =( (1 + 2β + 10αβ)J (3) +(α + 2β + 9αβ)J (3) n+2 α,β,n+3 n + (3β + 9αβ)J (3) n+1 ) , where α and β are real numbers and J (3) ber. n is the n-th third-order Jacobsthal num- Proof. (3.9): By the using the equations (3.1) and (3.3), we have 2J Q(3) α,β,n + J Q(3) α,β,n+1 + J Q(3) α,β,n+2 n + 2J (3) 3Xs=1 n + J (3) n+1 + J (3) n+1 + n+ses! + J (3) 3Xs=1 3Xs=1(cid:16)2J (3) n+2(cid:17) + J (3) n+2 + n+s+1es! + J (3) n+s+2(cid:17) es n+s+1 + J (3) n+s + J (3) n+s+2es! J (3) 3Xs=1 = J (3) n+3 + J (3) n+s+3es = J Q(3) α,β,n+3. = 2J (3) =(cid:16)2J (3) 3Xs=1 8 G. CERDA-MORALES Using the identity of third-order Jacobsthal numbers n+2 + J (3) in (2.3), the last equation becomes 2J Q(3) n+3 = J (3) J (3) n+1 + 2J (3) α,β,n+J Q(3) n , (n ≥ 0) α,β,n+1+J Q(3) α,β,n+2 = J Q(3) α,β,n+3. (3.10): From equations (3.1) and (1.1), we conclude that J Q(3) α,β,n+3 n + α,β,n+1 − e2J Q(3) α,β,n − e1J Q(3) = J (3) 3Xs=1 − e2 J (3) α,β,n+2 − e3J Q(3) 3Xs=1 n+s+2es! − e3 J (3) n+ses! − e1 J (3) 3Xs=1 n+2 + n+1 + J (3) J (3) = J (3) n + αJ (3) n+2 + βJ (3) n+4 + αβJ (3) n+6. J (3) n+s+1es! 3Xs=1 J (3) n+3 + n+s+3es! Substituting the identities of third-order Jacobsthal numbers J (3) 3J (3) relation (3.3) into the last equation and after simplifying we can assert that n+2 + n which are well-known using n+2 + 10J (3) n+4 = 2J (3) n+6 = 9J (3) n+1 + 9J (3) n+1 + 2J (3) and J (3) n J Q(3) α,β,n − e1J Q(3) α,β,n+1 − e2J Q(3) =( (1 + 2β + 10αβ)J (3) +(α + 2β + 9αβ)J (3) n+2 α,β,n+2 − e3J Q(3) α,β,n+3 n + (3β + 9αβ)J (3) n+1 ) . Special Cases: • For α = β = 1, the equation (3.10) is equivalent to ( J Q(3) −e2J Q(3) 1,1,n − e1J Q(3) 1,1,n+2 − e3J Q(3) 1,1,n+1 1,1,n+3 ) = 37J (3) n + 12j(3) n , wich was given by Cerda-Morales in [Ce]. • For the case α = 1 and β = −1, the equation (3.10) becomes n+1 + 5j(3) n + 17J (3) 1,−1,n+1 J Q(3) −e2J Q(3) 1,−1,n − e1J Q(3) 1,−1,n+2 − e3J Q(3) 1,−1,n+3 ) = −(cid:16)J (3) ( n+1(cid:17) . • Let β = 0. For α = 1, α = −1 and α = 0, there are following relations J Q(3) 1,0,n − e1J Q(3) −1,0,n − e1J Q(3) 1,0,n+1 − e2J Q(3) −1,0,n+1 − e2J Q(3) 1,0,n+2 − e3J Q(3) −1,0,n+2 − e3J Q(3) 1,0,n+3 = J (3) −1,0,n+3 = J (3) n + J (3) n+2, n − J (3) n+2 J Q(3) and J Q(3) 0,0,n − e1J Q(3) 0,0,n+1 − e2J Q(3) 0,0,n+2 − e3J Q(3) 0,0,n+3 = J (3) n , respectively. (cid:3) THIRD-ORDER JACOBSTHAL GENERALIZED QUATERNIONS 9 n and j(3) Theorem 3.2. Let J (3) be the third-order Jacobsthal and third-order Jacobsthal-Lucas numbers, J Q(3) α,β,n be the third-order Jacobsthal and third-order Jacobsthal-Lucas generalized quaternions, respectively. In this case, the following equations can be given α,β,n and jQ(3) n α,β,n+2(cid:17)2 +(cid:16)J Q(3) n+1 · J Q(3) α,β,n + J (3) −3 · 22(n+1)(1 + 4α + 16β + 64αβ) α,β,n+1 + J (3) n + 2αU (3) n+1 + 4βU (3) −2(1 + α + β + αβ) α,β,n+2(cid:17) n+2 · J Q(3) n+3(cid:17) n+2 + 8αβU (3) (cid:16)J Q(3) α,β,n(cid:17)2 14(cid:16)J (3) (3.11) = 1 7 (cid:16)jQ(3) α,β,n(cid:17)2 (3.12)  α,β,n+1(cid:17)2 +(cid:16)J Q(3) n · J Q(3) − · 2n+2(cid:16)U (3) α,β,n(cid:17)2 − 9(cid:16)J Q(3) =( −2n+2(cid:16)j(3) .  n (cid:17) ) , α,β,n − 18J (3) n · J Q(3) α,β,n 2j(3) n · jQ(3) n−3 + 2αj(3) n = j(3) n−2 + 4βj(3) n−1 − J (3) n+1. n−1 + 8αβj(3) where α and β are real numbers, and U (3) Proof. (3.11): From equation (3.1), we get (cid:16)J Q(3) α,β,n(cid:17)2 n + = J (3) 3Xs=1 = −(cid:18)(cid:16)J (3) n (cid:17)2 n · J Q(3) + 2J (3) α,β,n. J (3) J (3) n + n+ses! · J (3) n+1(cid:17)2 + α(cid:16)J (3) n+ses! 3Xs=1 n+2(cid:17)2 + αβ(cid:16)J (3) + β(cid:16)J (3) n+3(cid:17)2(cid:19) Combining the equation (3.3) with the last equation gives α,β,n(cid:17)2 (cid:16)J Q(3) +(cid:16)J Q(3) α,β,n+1(cid:17)2 +(cid:16)J Q(3) α,β,n+2(cid:17)2 n+2(cid:17)2(cid:19) = −(cid:18)(cid:16)J (3) +(cid:16)J (3) +(cid:16)J (3) n+1(cid:17)2 n (cid:17)2 n+3(cid:17)2(cid:19) − α(cid:18)(cid:16)J (3) n+2(cid:17)2 n+1(cid:17)2 +(cid:16)J (3) +(cid:16)J (3) n+4(cid:17)2(cid:19) − β(cid:18)(cid:16)J (3) n+3(cid:17)2 n+2(cid:17)2 +(cid:16)J (3) +(cid:16)J (3) n+5(cid:17)2(cid:19) − αβ(cid:18)(cid:16)J (3) n+4(cid:17)2 n+3(cid:17)2 +(cid:16)J (3) +(cid:16)J (3) + 2(cid:16)J (3) n+1 · J Q(3) n · J Q(3) α,β,n + J (3) α,β,n+1 + J (3) n+2 · J Q(3) α,β,n+2(cid:17) . 10 G. CERDA-MORALES + = 1 after simplifying we obtain that Thus, using the identity of the third-order Jacobsthal numbers(cid:16)J (3) n (cid:17)2 n+2(cid:17)2 (cid:16)J (3) (cid:16)J Q(3) n+1(cid:17)2 +(cid:16)J (3) n + 2(cid:17) by (2.15) into the last equation and +(cid:16)J Q(3) n+1 · J Q(3) α,β,n+2(cid:17)2 7(cid:16)3 · 22(n+1) − 2n+2U (3) α,β,n(cid:17)2 α,β,n+1(cid:17)2 +(cid:16)J Q(3) 2(cid:16)J (3) α,β,n + J (3) n · J Q(3) − 3 7 · 22(n+1)(1 + 4α + 16β + 64αβ) 7 · 2n+2(cid:16)U (3) n+1 + 4βU (3) n + 2αU (3) − 1 − 2 7 (1 + α + β + αβ) n−1 − J (3) n+1. α,β,n+2(cid:17) n+2 · J Q(3) n+3(cid:17) (3.12): In the same manner, from the equations (3.1), (3.2) and (3.3) we can n+2 + 8αβU (3) α,β,n+1 + J (3)   where U (3) n = j(3) = , see that α,β,n = 2j(3) (cid:16)jQ(3) α,β,n(cid:17)2 α,β,n(cid:17)2 − 9(cid:16)J Q(3) n · J Q(3) n · jQ(3) α,β,n − 18J (3) −(cid:18)(cid:16)j(3) n (cid:17)2(cid:19) − α(cid:18)(cid:16)j(3) n+1(cid:17)2 n (cid:17)2 − 9(cid:16)J (3) − 9(cid:16)J (3) n+2(cid:17)2(cid:19) − αβ(cid:18)(cid:16)j(3) − β(cid:18)(cid:16)j(3) n+3(cid:17)2 n+2(cid:17)2 − 9(cid:16)J (3) Putting the identity(cid:16)j(3) n (cid:17)2 −9(cid:16)J (3) n (cid:17)2 α,β,n(cid:17)2 − 9(cid:16)J Q(3) (cid:16)jQ(3) α,β,n(cid:17)2 = 2n+2j(3) = 2j(3) n−3 of the third-order Jacobsthal numbers (see [Cook-Bac]) into the last equations and after an easy computation we obtain n+1(cid:17)2(cid:19) − 9(cid:16)J (3) n+3(cid:17)2(cid:19) . n · jQ(3) − 2n+2(cid:16)j(3) α,β,n − 18J (3) n−3 + 2αj(3) n · J Q(3) n−2 + 4βj(3) α,β,n n−1 + 8αβj(3) n (cid:17) . Special cases: • For α = β = 1, we have the following relations for the third-order Jacobsthal quaternions which were given by Cerda-Morales in [Ce]. 1,1,n(cid:17)2 (cid:16)jQ(3) 1,1,n(cid:17)2 − 9(cid:16)J Q(3) = 2( j(3) n · jQ(3) −2n(cid:16)17j(3) 1,1,n − 9J (3) n + 7j(3) n · J Q(3) n−1 + 3j(3) • Let α = 1 and β = −1. In this case, we have the following relations for the split third-order Jacobsthal quaternions: 1,−1,n(cid:17)2 (cid:16)jQ(3) − 9(cid:16)J Q(3) 1,−1,n(cid:17)2 = 2( j(3) 1,−1,n − 9J (3) n + 3j(3) n · jQ(3) +3 · 2n(cid:16)5j(3) n · J Q(3) n−1 − j(3) 1,1,n n−2(cid:17) ) . n−2(cid:17) ) . 1,−1,n THIRD-ORDER JACOBSTHAL GENERALIZED QUATERNIONS 11 • Let β = 0. For α = 1, α = −1 and α = 0, the equation (3.12) becomes 1,0,n(cid:17)2 1,0,n(cid:17)2 − 9(cid:16)J Q(3) (cid:16)jQ(3) −1,0,n(cid:17)2 −1,0,n(cid:17)2 − 9(cid:16)J Q(3) (cid:16)jQ(3) 0,0,n(cid:17)2 0,0,n(cid:17)2 = 2n j(3) − 9(cid:16)J Q(3) (cid:16)jQ(3) and respectively. = 2( j(3) = 2( j(3) ) , 1,0,n n · jQ(3) 1,0,n − 9J (3) n · J Q(3) n−1(cid:17) −2n+1(cid:16)j(3) n − j(3) n · J Q(3) n · jQ(3) −1,0,n − 9J (3) −2n+1(cid:16)j(3) n−2(cid:17) n−3 − 2j(3) 0,0,n − 2n+1j(3) 0,0,n − 9J (3) n · J Q(3) −1,0,n n · jQ(3) ) n−3 o , In Theorem 3.3, the first identity of norm for α = β = 1 is analogous to the (cid:3) ordinary third-order Jacobsthal quaternions (3.13) · N r(J Q(3) n ) = (for more details, see [Ce]). 1 49 340 · 22n − 64 · 2n + 18 if n ≡ 0 (mod 3) 340 · 22n + 68 · 2n + 23 if n ≡ 1 (mod 3) 340 · 22n − 4 · 2n + 15 if n ≡ 2 (mod 3) , be the third-order Jacobsthal number, J Q(3) n Theorem 3.3. Let J (3) third-order Jacobsthal generalized quaternion and J Q(3) J Q(3) (3.14) α,β,n. Then, the following equation hold α,β,n be the α,β,n be the conjugate of N r(J Q(3) α,β,n) = 1 49 and V (3) n as in Eq. (2.17).  22(n+1)(1 + 4α + 16β + 64αβ) n+1(cid:17) −2n+2(cid:16)(1 + 8αβ − 4β)V (3) n + (2α − 4β)V (3) n+2(cid:17)2 n+1(cid:17)2 + β(cid:16)V (3) n (cid:17)2 +(1 + αβ)(cid:16)V (3) + α(cid:16)V (3) ,  Proof. By multiplication of two third-order Jacobsthal generalized quaternions, and by using the identity of the third-order Jacobsthal numbers J (3) n+2 + J (3) n+1 + 2J (3) it may be concluded that n+3 = J (3) n J (3) J (3) n + n − n+ses! · J (3) α,β,n) = J (3) 3Xs=1 3Xs=1 n+2(cid:17)2 n+1(cid:17)2 n (cid:17)2 + β(cid:16)J (3) + α(cid:16)J (3) =(cid:16)J (3) 49(cid:18)22(n+1) − 2n+2V (3) 49(cid:16)2n+1 − V (3) n (cid:17)2 n+ses! n+3(cid:17)2 + αβ(cid:16)J (3) n (cid:17)2(cid:19) . n +(cid:16)V (3) it is obvious that = 1 1 . N r(J Q(3) (cid:16)J (3) n (cid:17)2 = Finally, from the Binet formula (2.19) of J (3) n 12 G. CERDA-MORALES   n+1 + 4βV (3) n+2 + 8αβV (3) 22(n+1)(1 + 4α + 16β + 64αβ) n + 2αV (3) n+3(cid:17) −2n+2(cid:16)V (3) n+3(cid:17)2 n+2(cid:17)2 n+1(cid:17)2 + αβ(cid:16)V (3) n (cid:17)2 + β(cid:16)V (3) +(cid:16)V (3) + α(cid:16)V (3) n+1(cid:17) −2n+2(cid:16)(1 + 8αβ − 4β)V (3) n + (2α − 4β)V (3) n+2(cid:17)2 n+1(cid:17)2 +(1 + αβ)(cid:16)V (3) n (cid:17)2 + β(cid:16)V (3) + α(cid:16)V (3) 22(n+1)(1 + 4α + 16β + 64αβ) ,   using the relations V (3) n + V (3) n+1 + V (3) n+2 = 0 and V (3) n = V (3) n+3 for n ≥ 0. By scrutinizing α and β, the equation (3.14) becomes as follow: Then, we have N r(J Q(3) α,β,n) = 1 49 = 1 49 Special Cases: • N r(J Q(3) 1,1,n) = • 1 1 1 1,0,n) = n+1(cid:17)2 +(cid:16)V (3) 49(cid:26) 85 · 22(n+1) − 2n+2(cid:16)5V (3) n+1(cid:17) +(cid:16)V (3) n (cid:17)2 n − 2V (3) n (cid:17) −75 · 22(n+1) − 3 · 2n+2(cid:16)2V (3) 49 n+1 − V (3) n+2(cid:17)2 −(cid:16)V (3) 5 · 22(n+1) − 2n+2(cid:16)V (3) 49  +(cid:16)V (3) n+1(cid:17)2 n+1(cid:17) −3 · 22(n+1) − 2n+2(cid:16)V (3) 49  n − 2V (3) n+1(cid:17)2 +(cid:16)V (3) n (cid:17)2 −(cid:16)V (3) 49(cid:26) 22(n+1) − 2n+2V (3) n (cid:17)2 (cid:27) , n +(cid:16)V (3) +(cid:16)V (3) n (cid:17)2 n+1(cid:17) n + 2V (3) 0,0,n) = 1 1 , + 14 (cid:27) ,  , , (cid:3) N r(J Q(3) 1,−1,n) = N r(J Q(3) N r(J Q(3) −1,0,n) = N r(J Q(3) • • • In the following theorem, the first and second formulas are analogous to the Theorem 3.3 and 3.4 in [Ce]. Theorem 3.4 (Binet's Formulas). Letb2 = 1 + 2e1 + 4e2 + 8e3,cω1 = 1 + ω1e1 + 1e2 +e3 andcω2 = 1+ω2e1 +ω2 α,β,n and α,β,n be the third-orderJacobsthal and third-order Jacobsthal-Lucas generalized 2e2 +e3 generalized quaternions. Let J Q(3) ω2 jQ(3) THIRD-ORDER JACOBSTHAL GENERALIZED QUATERNIONS 13 quaternions, respectively. For n ≥ 0, the Binet formulas for these quaternions are given as: (3.15) and (3.16) J Q(3) α,β,n = jQ(3) α,β,n = 1 n (cid:17) 7(cid:16)2n+1b2 − V Q(3) n (cid:17) , 7(cid:16)2n+3b2 + 3V Q(3) n is defined by 1 respectively. Here, the sequence V Q(3) (3.17) V Q(3) n = = Aωn ω1 − ω2 1cω1 − Bωn 2cω2 n+1 − V Q(3) n+2 = −V Q(3) n . 2 − 3e1 + e2 + 2e3 −3 + e1 + 2e2 − 3e3 1 + 2e1 − 3e2 + e3 if n ≡ 0 (mod 3) if n ≡ 1 (mod 3) if n ≡ 2 (mod 3) , where A = −3 − 2ω2 and B = −3 − 2ω1. Furthermore, note that for all n ≥ 0 we have V Q(3) The following theorem gives d'Ocagne's identities for third-order Jacobsthal generalized quaternion. Theorem 3.5. If J Q(3) nion. Then, for any integers n and m, we have (3.18) α,β,n be the n-th third-order Jacobsthal generalized quater- J Q(3) α,β,n = α,β,m+1·J Q(3) α,β,n+1−J Q(3) α,β,m·J Q(3) whereb2 = 1+2e1 +4e2 +8e3,cω1 = 1+ω1e1 +ω2 α,β,n−1 − J Q(3) n = jQ(3) and U Q(3) α,β,n+1. 1 √3 − 7( 2m+1b2U Q(3) 2 cω2cω1(cid:1) ) n+1 − 2n+1U Q(3) m+1b2 3 i(cid:0)ωm−n 1 cω1cω2 − ωm−n 1e2 +e3,cω2 = 1+ω2e1 +ω2 2e2 +e3 Proof. Using the Binet formula for the third-order Jacobsthal generalized quater- nions and V Q(3) (3.19) n in (3.17) gives J Q(3) = = = 1 1 α,β,n α,β,m · J Q(3) α,β,n+1 − J Q(3) α,β,m+1 · J Q(3) 49 (cid:16)2m+1b2 − V Q(3) −(cid:16)2m+2b2 − V Q(3) 49( −2m+1b2V Q(3) 7 2m+1b2U Q(3) n (cid:17)  n+1(cid:17) m(cid:17)(cid:16)2n+2b2 − V Q(3) m+1(cid:17)(cid:16)2n+1b2 − V Q(3) mb2 + 2m+2b2V Q(3) n+1 − V Q(3) i(cid:0)ωm−n m+1b2 − 1 cω1cω2 − ωm−n n+1 − 2n+2V Q(3) m V Q(3) n+1 − 2n+1U Q(3) m+1V Q(3) +V Q(3) √3 3 1 n α,β,n−1 − J Q(3) α,β,n+1 for all n ≥ 1. ) m+1b2 2 cω2cω1(cid:1)! , n + 2n+1V Q(3) where U Q(3) n = jQ(3) (cid:3) 14 G. CERDA-MORALES Taking m = n + 1 in the Theorem 3.5 and using the identity ω1cω1cω2 − ω2cω2cω1 = (ω1 − ω2)( (2 − e1 − e2 + 2e3) − (1 + α + β + αβ) +(βe1 + αe2 + e3) ) . we obtain a type of Cassini-like identity for third-order Jacobsthal generalized quaternions. Corollary 3.6. For any integer n ≥ 0, we have (3.20) (cid:16)J Q(3) α,β,n+1(cid:17)2 − J Q(3) α,β,n+2 · J Q(3) α,β,n = 1 7 Special Cases: In particular, for the third-order Jacobsthal and third-order Jacobsthal-Lucas generalized quaternions by considering the special cases of α and β, respectively, the Cassini-like Identities are as follows: n+2b2(cid:17) n+1 − U Q(3) +(2 − e1 − e2 + 2e3) −(1 + α + β + αβ) +(βe1 + αe2 + e3) 2n+1(cid:16)2b2U Q(3)  7( 2n+1(cid:16)2b2U Q(3) 7( 2n+1(cid:16)2b2U Q(3) 7( 2n+1(cid:16)2b2U Q(3) 7( 2n+1(cid:16)2b2U Q(3) 7( 2n+1(cid:16)2b2U Q(3) −e1 + 3e3 −1 + 3e3 −2 + 3e3 .  n+2b2(cid:17) ) , n+2b2(cid:17) n+2b2(cid:17) ) , n+2b2(cid:17) n+2b2(cid:17) ) . n+1 − U Q(3) 2 − e1 − 2e2 + 3e3 n+1 − U Q(3) n+1 − U Q(3) n+1 − U Q(3) +2 − 2e2 + 3e3 n+1 − U Q(3) ) , ) , • (cid:16)J Q(3) • (cid:16)J Q(3) • (cid:16)J Q(3) • (cid:16)J Q(3) • (cid:16)J Q(3) 1,1,n+1(cid:17)2 1,−1,n+1(cid:17)2 1,0,n+1(cid:17)2 −1,0,n+1(cid:17)2 0,0,n+1(cid:17)2 −J Q(3) 1,1,n+2·J Q(3) 1,1,n = 1 −J Q(3) 1,−1,n+2·J Q(3) 1,−1,n = 1 −J Q(3) 1,0,n+2·J Q(3) 1,0,n = 1 −J Q(3) −1,0,n+2·J Q(3) −1,0,n = 1 −J Q(3) 0,0,n+2·J Q(3) 0,0,n = 1 The third-order Jacobsthal generalized quaternions are given by 4. Conclusions J Q(3) α,β,n = J (3) n + e1J (3) n+1 + e2J (3) n+2 + e3J (3) n+3, where J (3) quaternionic units which satisfy the equalities n is the n-th third-order Jacobsthal number and e1, e2 and e3 are 1 = −α, e2 e2 2 = −β, e2 3 = −αβ, e1e2 = e3 = −e2e1, e2e3 = βe1 = −e3e2, e3e1 = αe2 = −e1e3, THIRD-ORDER JACOBSTHAL GENERALIZED QUATERNIONS 15 For α = β = 1, the third-order Jacobsthal generalized quaternion J Q(3) 1,1,n which was given by [Ce] becomes the real third-order Jacobsthal quaternions. For α = 1 and β = −1, the third-order Jacobsthal generalized quaternion J Q(3) 1,−1,n becomes the split third-order Jacobsthal quaternion. Starting from ideas given by Horadam [Ho1], Pottman and Wallner [Po-Wa], the third-order Jacobsthal generalized quaternions are studied and the relations related to these quaternions are obtained (i.e., for third-order Jacobsthal semi-quaternions, split third-order Jacobsthal semi-quaternions and third-order Jacobsthal 1 4 -quaternions). References [Ad] S.L. Adler, Quaternionic quantum mechanics and quantum fields, New York: Oxford University Press, 1994. [Ak-Ke] I. Akkus and O. Ke¸cilioglu, Split Fibonacci and Lucas octonions, Adv. Appl. Clifford Algebras, 25(3), (2015), 517 -- 525. [Ak] M. Akyigit, H.H. Kosal and M. Tosun, Split Fibonacci quaternions, Adv. Appl. Clifford Algebras, 23, (2013), 535 -- 545. [Bae] J.C. Baez, The octonions, Bull. Am. Math. Soc., 39, (2002), 145 -- 205. [Ba] P. Barry, Triangle geometry and Jacobsthal numbers, Irish Math. Soc. Bull., 51, (2003), 45 -- 57. [Car] K. Carmody, Circular and Hyperbolic Quaternions, Octonions and Sedenions, Appl. Math. Comput., 28, (1988), 47 -- 72. [Ca] P. Catarino, The modified Pell and the modified k-Pell quaternions and octonions, Adv. Appl. Clifford Algebras, 26, (2016), 577 -- 590. [Ce] G. Cerda-Morales, Identities for Third Order Jacobsthal Quaternions, Advances in Ap- plied Clifford Algebras, 27(2), (2017), 1043 -- 1053. [Ce1] G. Cerda-Morales, On a Generalization of Tribonacci Quaternions, Mediterranean Jour- nal of Mathematics, 14:239 (2017), 1 -- 12. [Cook-Bac] Ch. K. Cook and M. R. Bacon, Some identities for Jacobsthal and Jacobsthal- Lucas numbers satisfying higher order recurrence relations, Annales Mathematicae et Informaticae, 41, (2013), 27 -- 39. [Che-Lou] W. Y. C. Chen and J. D. Louck, The combinatorial power of the companion matrix, Linear Algebra Appl., 232, (1996), 261 -- 278. [Ci1] C.B. C¸ imen and A. Ipek, On Pell quaternions and Pell-Lucas quaternions, Adv. Appl. Clifford Algebras, 26(1), (2016), 39 -- 51. [Ci2] C.B. C¸ imen and A. Ipek, On Jacobsthal and Jacobsthal-Lucas Octonions, Mediterranean Journal of Mathematics, 14:37, (2017), 1 -- 13. [Go1] M. Gogberashvili, Octonionic Geometry, Adv. Appl. Clifford Algebras, 15, (2005), 55 -- 66. [Go2] M. Gogberashvili, Octonionic electrodynamics, J. Phys. A: Math. Gen., 39, (2006), 7099 -- 7104. [Ha1] S. Halici, On Fibonacci quaternions, Adv. Appl. Clifford Algebras, 22, (2012), 321 -- 327. [Ha2] S. Halici, On complex Fibonacci quaternions, Adv. Appl. Clifford Algebras, 23, (2013), 105 -- 112. [Ho1] A. F. Horadam, Complex Fibonacci numbers and Fibonacci quaternions, Am. Math. Month., 70, (1963), 289 -- 291. [Ho2] A. F. Horadam, Quaternion recurrence relations, Ulam Quarterly, 2, (1993), 23 -- 33 . 16 G. CERDA-MORALES [Ho3] A. F. Horadam, Jacobsthal representation numbers, Fibonacci Quarterly, 34, (1996), 40 -- 54. [Iy] M.R. Iyer, A note on Fibonacci quaternions, Fibonacci Quaterly, 7(3), (1969), 225 -- 229. [Ja-Ya] M. Jafari, Y. Yaylı, Generalized Quaternion and Rotation in 3-space E 3 αβ , TWMS J. Pure Appl. Math., 6(2), (2015), 224 -- 232. [Ke-Ak] O. Ke¸cilioglu and I. Akkus, The Fibonacci Octonions, Adv. Appl. Clifford Algebras, 25(1), (2015), 151 -- 158. [Ko1] J. Koplinger, Signature of gravity in conic sedenions, Appl. Math. Computation, 188, (2007), 942 -- 947. [Ko2] J. Koplinger, Hypernumbers and relativity, Appl. Math. Computation, 188, (2007), 954 -- 969. [Ku] L. Kula, Y. Yaylı, Split Quaternions and Rotations in Semi Euclidean Space, J. Korean Math. Soc., 144(6), (2007), 1313 -- 1327. [Oz] M. Ozdemir, The Roots of a Split Quaternion, Appl. Math. Lett., 22(2), (2009), 258 -- 263. [Po-Wa] H. Pottman and J. Wallner, Computational Line Geometry, Springer-Verlag, New York, 2000. [Szy-Wl] A. Szynal-Liana and I. W loch, A Note on Jacobsthal Quaternions, Adv. Appl. Clifford Algebras, 26, (2016), 441 -- 447. [Ta] Y. Tian, Matrix representations of octonions and their applications, Adv. Appl. Clifford Algebras, 10(1), (2000), 61 -- 90.
1404.4349	3	1404	2018-05-10T17:44:47	Refinements to patching and applications to field invariants	[ "math.RA", "math.AG" ]	We introduce a notion of refinements in the context of patching, in order to obtain new results about local-global principles and field invariants in the context of quadratic forms and central simple algebras. The fields we consider are finite extensions of the fraction fields of two-dimensional complete domains that need not be local. Our results in particular give the u-invariant and period-index bound for these fields, as consequences of more general abstract results.	math.RA	math	Refinements to patching and applications to field invariants David Harbater, Julia Hartmann, and Daniel Krashen Abstract We introduce a notion of refinements in the context of patching, in order to obtain new results about local-global principles and field invariants in the context of quadratic forms and central simple algebras. The fields we consider are finite extensions of the fraction fields of two-dimensional complete domains that need not be local. Our results in particular give the u-invariant and period-index bound for these fields, as conse- quences of more general abstract results. 1 Introduction In this manuscript we introduce the notion of refinements in the context of patching, and use this to obtain results about quadratic forms and central simple algebras over fraction fields of two-dimensional complete domains. These provide strengthenings and analogs of results in earlier papers. Among our results here are local-global principles, which in the case of quadratic forms concern isotropy, the Witt group, the Witt index, and the u-invariant. In the case of central simple algebras they concern Brauer equivalence and the index. In addition, we obtain explicit results about the values of the u-invariant and the period-index bounds for these fraction fields. Classically, one relates the u-invariant and period-index bound for complete discretely valued fields to those of their residue fields. Here, we consider the analogous situation of fraction fields of two-dimensional complete domains, which need not be local. We focus on these two situations: (i) the fraction field of a two-dimensional Noetherian complete local domain R (e.g. k((x, t))); (ii) a finite separable extension of the fraction field of the t-adic completion of T [x], where T is a complete discrete valuation ring with uniformizer t. In the context of central simple algebras, we obtain the following result. (Our use of the term "Brauer dimension" is explained before Theorem 4.21.) Theorem 1.1. In the above two situations, assume that the residue field k of R (resp. T ) has Brauer dimension d away from p := char(k). Then ind(α) divides per(α)d+1 for all α ∈ Br(E) whose period is not divisible by p. 1 See Theorem 4.23, which also treats Brauer classes α ∈ Br(E) of arbitrary period in the mixed characteristic case. Using that, we obtain a local analog of [PS14, Theorem 1]: Corollary 1.2. Let L be the fraction field of Zp[[x]] or of the p-adic completion of Zp[x], and let E be a finite extension of L. Then ind(α) divides per(α)2 for all α ∈ Br(E). Theorem 4.23 also shows that the same conclusion holds if instead L is the fraction field of the p-adic completion of Zur p [[x]] or Zur p [x], with per(α) = ind(α) if per(α) is prime to p. For quadratic forms, we prove an analog of Theorem 1.1; see Theorem 4.11. This yields results about values of the u-invariant in mixed characteristic (Corollaries 4.13 and 4.14), and also the following result in equicharacteristic: Theorem 1.3. Let k be a field of characteristic unequal to two, and let E be a finite separable extension of the fraction field of k[x][[t]] or of k[[x, t]]. Then • u(E) = 4 if k is algebraically closed. • u(E) = 8 if k is finite, or if k = k0((z)) with k0 algebraically closed. • u(E) = 16 if k = k0((z)) with k0 finite, or if k = Qp for some p (which can equal 2). • u(E) = 32 if k = Qp(z) or if k = Qp((z)) for some p. Note that the value of the u-invariant or the period-index bound for a given field does not in general give much information about the corresponding invariant for arbitrary finite separable extensions. So the above results would not follow simply from knowing the values of these invariants for the fraction fields of rings of the form T [x] or T [[x]]. Previous results about the u-invariant and period-index bounds for related fields ap- peared in such papers as [PS10], [HHK09], [Lee13], [Sal08], [deJ04], and [Lie11]. To obtain our present results, we build on the patching framework that was used in our previous manuscripts [HH10], [HHK09], [HHK15], and [HHK13]. As in those papers, the patching framework also enables us to obtain local-global princi- ples. In particular, our Corollary 4.7 proves a local-global principle for isotropy; Corollary 4.8 relates the u-invariants of fields to those of their completions; and Corollary 4.17 provides a local-global principle for the period-index bound of a field. See also related results in [CPS12], [PS14], and [Hu15]. The key new ingredient in this paper is a refinement principle for patching. As in the patching framework, we consider a projective normal curve over a complete discrete valuation ring, and we choose a finite partition of the closed fiber. Criteria for patching and local- global principles are given in terms of intersection and factorization properties for a certain quadruple of rings arising from the partition. By enlarging the given partition or modifying the model (e.g. by blowing up), one can refine a quadruple, obtaining a new one with one part expanded. The refinement principle that we state in this manuscript relates the intersection and factorization properties of a given quadruple to that of two other quadruples: the refined one, and the quadruple arising from the part that was expanded. This principle is first stated in an abstract context (Proposition 2.14), and then used in a geometric context to establish "patching on patches" (Proposition 3.9) and patching on exceptional divisors of a blow-up (Proposition 3.10, answering a question of Yong Hu). 2 The manuscript is organized as follows. In Section 2 we present general results about patching and local-global principles for quadruples of groups or rings, and then state our abstract refinement principle. In Section 3 we turn to the geometric situation, generalizing the patching setup of [HHK09] in Section 3.1 to allow more general open subsets of the closed fiber, and then obtaining consequences of the refinement principle, in Sections 3.2 and 3.3. We then turn to quadratic forms and central simple algebras in Section 4, first proving local- global results in an abstract context (Theorems 4.1 and 4.15), and then specializing to the geometric situation to obtain results about numerical invariants, including those above. We wish to thank Annette Maier, Yong Hu, and Jean-Louis Colliot-Thélène for helpful discussions concerning results and ideas in this manuscript. 2 Patching and refinement This section proves a refinement principle (Proposition 2.14) that will afterwards permit us to obtain results about patching and local-global principles over certain function fields, once such results are known for related fields. In this section the presentation is in a more general framework. We begin with a discussion of patching and local-global principles, and the related conditions of factorization and intersection for quadruples. That discussion draws heavily on prior papers of the present authors. 2.1 Diamonds of groups and rings Patching is a method that mimics constructions from complex geometry to obtain global objects from more local ones. In our algebraic version, we work with objects in a cate- gory C consisting of sets, possibly with additional structure (e.g. groups or rings), and we will consider quadruples of objects S• = (S, S1, S2, S0) together with morphisms forming a commutative diagram S1 S0 β1 : :✈✈✈ d■■■ Sα1 β2 d❍❍❍ S2 :✉✉✉ α2 (∗) As a motivating example, one can think of these as being the collection of functions or other objects on a global space X, on two subsets U1 and U2 that cover X, and on their intersection U0, as indicated in the following commutative diagram: U1 U0 X z✉✉✉ $❏❏❏ $■■■ zttt U2 For patching, the two key properties of a quadruple are factorization and intersection (see Theorem 2.8). More precisely: 3 d d : z $ $ z Definition 2.1. (a) A diamond S• in C consists of a commutative diagram (∗) as above such that (α1, α2) : S → S1 ×S0 S2 is a monomorphism. For short we will often write S• = (S, S1, S2, S0) as a quadruple if the maps αi, βi are understood. (b) A diamond S• as above has the intersection property if the map (α1, α2) : S → S1×S0 S2 is an isomorphism. It is injective if each of the maps αi, βi is injective. (c) Let G• = (G, G1, G2, G0) be a diamond of groups, together with maps αi, βi. We say that G• has the factorization property if G0 = β1(G1)β2(G2), i.e. every element of G0 is of the form β1(g1)β2(g2) with gi ∈ Gi. In the injective case, we often regard the maps αi, βi as inclusions; and to emphasize this, we write the diamond as (S ≤ S1, S2 ≤ S0). With these identifications, the intersection property asserts that S = S1 ∩ S2 in S0; and the factorization property for diamonds of groups then asserts that each element of G0 can be factored as g1g2 with gi ∈ Gi. By applying factorization to the inverse of each element of G0, we obtain: Lemma 2.2. A diamond of groups (G, G1, G2, G0) has the factorization property (respectively the intersection property) if and only if (G, G2, G1, G0) does. We will often consider injective diamonds of rings F• = (F ≤ F1, F2 ≤ F0) with F is a field and each Fi a direct product of finitely many fields. In particular, we have: Example 2.3. Let Γ be a bipartite connected (multi-)graph, with vertex set V = V1 ⊔ V2 and edge set E. Suppose that we are given a Γ-field in the sense of [HHK14, Section 2.1.1]; i.e. a field Fv for each v ∈ V and a field Fe for each e ∈ E, together with an inclusion Fv ֒→ Fe whenever v is a vertex of e. These fields and inclusions define an inverse system of fields; and if the inverse limit is a field F then this is called a "factorization inverse system" over F ([HHK15], Section 2), and the graph together with the associated fields is called a Fv for i = 1, 2 and set Γ/F -field ([HHK14], Section 2.1.1). In this situation, set Fi =Qv∈Vi F0 = Qe∈E Fe. Then the inclusions Fv ֒→ Fe induce inclusions Fi ֒→ F0 for i = 1, 2; and F• = (F ≤ F1, F2 ≤ F0) is an injective diamond of the above form. In fact, every such diamond arises in this way: Proposition 2.4. Let F• = (F ≤ F1, F2 ≤ F0) be an injective diamond of rings having the intersection property, with F a field and each Fi a finite product of fields. Then F• is induced from a factorization inverse system over F as in Example 2.3. Fλ with each Fλ a field, and let E = Λ0, V1 = Λ1, V2 = Λ2. To give the structure of a graph Γ, we will associate to every e ∈ E elements vi ∈ Vi for Fv → Proof. Write Fi = Qλ∈Λi i = 1, 2. To do this, choose e′ ∈ E and for i = 1, 2 consider the composition Qv∈Vi Qe∈E Fe → Fe′. Since Fe′ is a field, the image of this homomorphism is a domain. So the unique projectionQv∈Vi kernel must be a prime and hence maximal ideal. Thus this composition factors through a 2) gives a graph Γ, and the → Fe′ give the structure of a Γ-field. The inverse limit of the above homomorphisms Fv′ fields Fv, Fe is the intersection F1 ∩ F2, which is equal to F by the intersection property; and so these fields form a factorization inverse system, which induces the diamond. . The assignment of e to (v′ 1, v′ Fv → Fv′ i i 4 2.2 Patching and local-global principles To study patching and local-global principles in this framework, we will need to introduce the notion of diamonds of categories and tensor categories. Let C1, C2, C0 be categories, and suppose we are given functors Gi : Ci → C0. The (2-)fiber product C1 ×C0 C2 is defined to be the category whose objects are triples (C1, C2, φ) where Ci ∈ Ci and φ : G1(C1) → G2(C2) is an isomorphism. A morphism (C1, C2, φ) → (D1, D2, ψ) is defined to be a pair or morphisms fi : Ci → Di such that we have a commutative square G1(C1) f1 G1(D1) φ ψ / G2(C2) f2 / G2(D2) Definition 2.5 (Patching Problems). A diamond of (tensor) categories is a diagram C1 G1 >⑥⑥⑥ a❇❇❇❇ F1 C0 C G2 `❆❆❆ C2 =⑤⑤⑤⑤ F2 of (tensor) categories and functors, together with a natural isomorphism of functors α : G1F1 → G2F2, such that the functor Φ : C → C1 ×C0 C2, given by Φ(c) = (F1(c), F2(c), α(c)), is essentially injective. In this situation: (a) A patching problem is an object C• in the fiber product category C1 ×C0 (b) A solution to a patching problem C• is an object C ∈ C such that Φ(C) ∼= C•. (c) Patching holds for the diamond if Φ is an equivalence of categories. C2. Example 2.6 (Patching for torsors). If R• = (R, R1, R2, R0) is a diamond of rings, and G is an algebraic group over R, we obtain a diamond of categories of torsors Tors(GR•). In this case, we refer to patching problems (solutions, etc.) for Tors(GR•) as patching problems (solutions, etc.) for G-torsors. Example 2.7 (Patching for free modules). Similarly, if R• = (R, R1, R2, R0) is a diamond of rings, we obtain a diamond F(R•) of tensor categories of free modules of finite rank. If patching holds for F(R•), it follows that patching holds for other categories of structures which may be defined in terms of the category of vector spaces and its tensor structure (e.g. central simple algebras). The proof is as in [HH10], Theorems 7.1 and 7.5. Theorem 2.8. Let R• = (R, R1, R2, R0) be a diamond of rings. (a) The following conditions are equivalent: (i) For every n ≥ 1, the diamond of groups GLn(R•) satisfies the intersection and factorization properties. 5 / / > ` a = (ii) Patching holds for free modules; that is, for the diamond of categories F(R•). (b) Under these conditions, the inverse to the equivalence Φ : F(R) → F(R1) ×F(R0) F(R2) is given by taking the intersection of free modules. (c) Suppose that R is a field, that each Ri is a finite product of fields, and the diamond is injective. Then the above two conditions are also equivalent to: (iii) For every linear algebraic group G over R, patching holds for G-torsors, i.e. for Tors(GR•). Proof. First observe that if Φ is an equivalence of categories, then necessarily R is the fiber product R1 ×R0 R2. Namely, for any c ∈ R1 ×R0 R2, consider the endomorphism of the rank one object (R1, R2, id) given by multiplication by c. Since Φ is an equivalence, this morphism is induced by an endomorphism of the free rank one R-module R; i.e. by multiplication by some element of R, which is necessarily equal to c. Thus R1 ×R0 R2 = R. The first two parts of the assertion now follow from [Har84, Proposition 2.1], which says that if R = R1 ×R0 R2, then Φ is an equivalence of categories if and only if factorization holds; and moreover that in this case the inverse of Φ is given by taking the fiber product of objects. The third part follows from [HHK14, Theorem 2.1.4], which applies here by Proposition 2.4 above. Definition 2.9. Patching holds for the diamond of rings R• if either of the equivalent conditions of Theorem 2.8(a) holds. Lemma 2.10. Let R• = (R, R1, R2, R0) be a diamond of rings. Let R ⊆ S be a finite extension of rings such that S is a free R-module. Set Si = S ⊗R Ri. (a) If the intersection property holds for R•, then it also holds for S•. (b) If patching holds for R•, then it also holds for S•. Proof. For the first part, suppose that R• has the intersection property. We thus have a left exact sequence 0 → R → R1 × R2 → R0 of R-modules, where the map on the right is given by subtracting the image under R2 → R0 from the image under R1 → R0. Since S is free over R, the sequence 0 → S → S1 × S2 → S0 is also left exact. Consequently S• also has the intersection property. For the second statement, let (sj) be a basis for the free R-module S and suppose j,ksℓ. Then the category of finitely generated free Si-modules is equiv- alent to the category whose objects are finitely generated free Ri-modules together with that sjsk = P aℓ Ri-endomorphismsesj (corresponding to multiplication by sj) such thatesjesk =P aℓ j,kesℓ, and where the morphisms in the category are required to commute with eachesj. In particular, since patching holds for the diamond of categories F(R•), it follows that patching also holds for F(S•). That is, patching holds for S•. Definition 2.11. Let R• = (R, R1, R2, R0) be a diamond of rings. Let V be a class of R-varieties. We say that the local-global principle holds for V with respect to R• if for each V ∈ V, the condition V (Ri) 6= ∅ for i = 1, 2 implies that V (R) 6= ∅. 6 This definition applies in particular to the key case considered above, where we are given an injective diamond (F ≤ F1, F2 ≤ F0), with F is a field and each Fi a finite product of fields group G over F . The set of isomorphism classes of G-torsors over F is in natural bijection Qj Fij, and where we take V to be the class of G-torsors over F , for some linear algebraic with the pointed Galois cohomology set H 1(F, G), and we write H 1(Fi, G) forQj H 1(Fij, G). The local-global principle holds for (the class of) G-torsors if and only if the natural map φ : H 1(F, G) → H 1(F1, G) × H 1(F2, G) has trivial kernel. Given a diamond R• = (R, R1, R2, R0) of rings, and a linear algebraic group G over R, there is an associated diamond of groups G(R•) = (G(R), G(F1), G(F2), G(F0)) of rational points. By embedding G in GLn,R ⊂ AN R and considering coordinates, we immediately obtain the following lemma, which was implicitly used in [HHK09], Section 3. Lemma 2.12. Suppose that R• is a diamond of rings with the intersection property and that G is a linear algebraic group over R. Then G(R•) also has the intersection property. Theorem 2.13. Suppose that F• = (F ≤ F1, F2 ≤ F0) is an injective diamond of rings with F a field and each Fi a finite direct product of fields. Assume moreover that patching holds for F•. Then the following statements are equivalent for a linear algebraic group G over F : (i) G(F•) satisfies factorization. (ii) The local-global principle holds for G-torsors with respect to F•. (iii) The local-global principle holds, with respect to F•, for the class of F -varieties V equipped with a G-action such that G(F0) acts transitively on V (F0). Proof. By Proposition 2.4, the diamond F• arises from a factorization inverse system; and so [HHK15, Theorem 2.4] applies. Thus there is the following exact sequence of pointed sets: H 0(F1, G) × H 0(F2, G) → H 0(F0, G) → H 1(F, G) → H 1(F1, G) × H 1(F2, G). Condition (i) asserts surjectivity of the first arrow, which is equivalent to the third arrow having trivial kernel. That latter property is the same as condition (ii), as discussed above. Moreover the triviality of that kernel implies condition (iii) by [HHK15, Corollary 2.8]. Finally, condition (iii) trivially implies condition (ii), since G-torsors over F satisfy the transitivity hypothesis of condition (iii). 2.3 A refinement principle In studying factorization and intersection for a diamond G• = (G, E, H, J) of groups, it will prove useful to consider the situation when H is itself the base for another diamond of groups H• = (H, H1, H2, H0). We would then like to combine the two diamonds into one new diamond, thereby refining the original situation. The natural question is to what extent such refinements preserve factorization and intersection. Below, given maps f : A → B, g : A → B′, and f ′ : A′ → B′, we write (f, g) : A → B ×B′ for the map a 7→ (f (a), g(a)), and write f × f ′ : A × A′ → B × B′ for (a, a′) 7→ (f (a), f ′(a′)). 7 Proposition 2.14 (Refinement Principle). Suppose we are given a commutative diagram of groups and homomorphisms R = E J β2 β1 b❊❊❊❊❊❊ =④④④④④④ b❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊❊ δ2 γ1 ;✇✇✇✇✇ c❍❍❍❍❍❍ :✈✈✈✈✈✈✈ ε1 δ1 H1 G H0 H γ2 c●●●●● ;✈✈✈✈✈✈ ε2 such that the following diagrams are diamonds of groups: G• = Then G′ • = β1ε1 ]❀❀❀❀ H A✄✄✄✄ δ1 β2 B☎☎☎☎ \✿✿✿✿ δ2 E J G (β1,γ1) β2×γ2 J × H0 ;✇✇✇✇✇ c❍❍❍❍❍❍ e❑❑❑❑❑❑ 9rrrrrrr ε1δ1 G H1 (δ2,ε2δ1) E × H2 H2 H• = H0 γ1 @ _❄❄❄❄ ε1 H1 H γ2 ^❃❃❃❃ ?⑧⑧⑧⑧ ε2 H2 1. If G′ 2. If G′ • has the factorization property then so does H•. • has the factorization property and H• has the intersection property then G• has the factorization property. 3. If G′ 4. If G′ • has the intersection property then so does G•. • has the intersection property and G• has the factorization property then H• has the intersection property. 5. If G• and H• have the intersection property then so does G′ •. 6. If G• and H• have the factorization property then so does G′ •. Proof. Part 1: Assume that G′ • has the factorization property. Let h0 ∈ H0. By hypothesis, we may write (1, h0) ∈ J × H0 as (β1, γ1)(h1) · (β2(e), γ2(h2)) for some elements h1 ∈ H1 and (e, h2) ∈ E × H2. In particular, γ1(h1)γ2(h2) = h0, hence H• has the factorization property. Part 2: We now additionally assume that H• has the intersection property. To see that G• has the factorization property, suppose that j ∈ J and consider (j, 1) ∈ J × H0. Using factorization for G′ •, we may find h1 ∈ H1 and (e, h2) ∈ E × H2 such that j = β1(h1)β2(e) and 1 = γ1(h1)γ2(h2). Since by the latter equality, h1 and h−1 2 have the same image in H0, the intersection hypothesis for H• implies that there exists h ∈ H with ε1(h) = h1 and ε2(h) = h−1 2 . Hence j = β1ε1(h)β2(e), which proves factorization for G• (see Lemma 2.2). 8 = b ; c c ; b : B ] \ A ; e 9 c @ ^ _ ? Part 3: Suppose that G′ • has the intersection property, and let (e, h) ∈ E × H satisfy β2(e) = β1ε1(h). We wish to show that e and h are the images of a single element g ∈ G. To see this, we first note that the elements ε1(h) ∈ H1 and (e, ε2(h)) ∈ E × H2 have the same image in J × H0. Since G′ • has the intersection property, we may find g ∈ G so that ε1δ1(g) = ε1(h), δ2(g) = e, and ε2δ1(g) = ε2(h). But (ε1, ε2) is injective, since H• is assumed to be a diamond. It follows that δ1(g) = h, and g is as desired. Part 4: Suppose that G′ • has the intersection property and G• has the factorization prop- erty. Assume that h0 = γ1(h1) = γ2(h2) with hi ∈ Hi. We would like to show that h1, h2 are the images in H1, H2 of some element of H. Using the factorization property for G•, we may write β1(h1) ∈ J as β2(e) · β1ε1(h) with e ∈ E, h ∈ H. Let h′ 1 = h1ε1(h)−1 ∈ H1 and 2 = h2ε2(h)−1 ∈ H2. Thus h′ h′ 2 have the same image in H0 under γ1, γ2 respectively, viz. the element h′ 1) = β2(e). Consider the im- ages of h′ 2)), which by the previous considerations are equal. Since G′ and ε2δ1(g) = h′ and h2 ∈ H2 are the images of the common element δ1(g)h ∈ H. • has the intersection property, there exists an element g ∈ G for which ε1δ1(g) = h′ 1 2 ∈ H2 under ε1, ε2. Thus h1 ∈ H1 0 := h0 γ1ε1(h)−1 = h0 γ2ε2(h)−1. Moreover β1(h′ 2) ∈ E × H2 in J × H0. These are (β1, γ1)(h′ 2. So δ1(g) ∈ H maps to h′ 1 ∈ H1 and of (e, h′ 1) and (β2(e), γ2(h′ 1, h′ 1 ∈ H1 and h′ Part 5: Suppose h1 ∈ H1 and (e, h2) ∈ E × H2 satisfy (β1(h1), γ1(h1)) = (β2(e), γ2(h2)). The intersection property for H• yields an element h ∈ H such that εi(h) = hi for i = 1, 2. The intersection property for G• then yields an element g ∈ G such that δ2(g) = e and δ1(g) = h. Thus ε1δ1(g) = ε1(h) = h1 and (δ2, ε2δ1)(g) = (e, h2); i.e. h1 and (e, h2) are the images of the common element g ∈ G. Part 6: Let (j, h0) ∈ J × H0. By factorization for H•, there exist hi ∈ Hi for i = 1, 2, such that h0 = γ1(h1)γ2(h2). By factorization for G• and Lemma 2.2, there exist h ∈ H and e ∈ E such that β1(h1)−1j ∈ J equals β1ε1(h) · β2(e); i.e. j = β1(h1ε1(h)) · β2(e) ∈ J. Moreover h0 = γ1(h1)γ2(h2) = γ1(h1)γ1ε1(h)γ2ε2(h)−1γ2(h2) = γ1(h1ε1(h))γ2(ε2(h)−1h2). Thus the elements h1ε1(h) ∈ H1 and (e, ε2(h)−1h2) ∈ E × H2 provide a factorization of (j, h0) ∈ J × H0. The following result will be useful in conjunction with the above proposition. Lemma 2.15. (a) Let H (j) • = (H (j), H (j) Write H = Q H (j) and Hi = Q H (j) 2 , H (j) 0 ) be a diamond of groups for each j. for i = 0, 1, 2, and let H• = (H, H1, H2, H0), together with the products of the maps defining the diamonds H (j) • . Then H• is a diamond, and it satisfies intersection (resp. factorization) if and only if each H (j) • does. 1 , H (j) i (b) Let eH• = (eH, eH1, eH2, eH0) be a diamond of groups, with associated maps αi : eH → eHi and βi : eHi → eH0 for i = 1, 2. Let A be a group and write H = eH × A, H1 = eH1 × A, and Hi = eHi for i = 0, 2. Then H• := (H, H1, H2, H0) is a diamond with respect to the satisfies intersection (resp. factorization) if and only if eH• does. maps α1 × id, α2 ◦ pr1, β1 ◦ pr1, β2, where pr1 is the first projection map. Moreover H• 9 Here part (a) is immediate, and part (b) then follows by applying part (a) to the two diamonds eH• and (A, A, 1, 1). 3 Patches and their fields In this section, we apply the refinement principle, Proposition 2.14, to fields arising from curves over complete discretely valued fields. In that situation, it was shown in [HHK09] that patching holds for diamonds arising from a partition of the closed fiber of a normal projective model of the curve into finitely many closed points and irreducible open sets. Here we show the same holds for more general partitions of the closed fiber (Proposition 3.7); for partitions of a connected open subset of the closed fiber (Proposition 3.9); and for partitions of the exceptional divisor of a blow-up (Proposition 3.10). The second of these can be regarded as an assertion about "patching on patches." Related results for factorization with respect to algebraic groups, which will be used in Section 4, appear in Section 3.3. For the sake of the applications in Section 4, we will need to consider reducible open sets in our partitions, in order to be able to treat branched covers that have reducible closed fibers over a given open subset of the base. This will require us first to generalize somewhat the framework that was considered in [HH10] and [HHK09], where the open sets had been assumed to be irreducible. (See also [Cuo13], where a similar generalization is considered.) 3.1 Setup Consider a complete discrete valuation ring T with uniformizer t, residue field k, and fraction a normal connected projective T -curve with function field F . Let X be the reduced closed field K. Let F be a one-variable function field over K, and let bX be a normal model of F , i.e. fiber of bX. The following definition generalizes the notation in [HH10, Section 6] and [HHK09], where the open sets of X were required to be irreducible. FW to be its fraction field. Definition 3.1. For a point P ∈ X we let O bX,P be its local ring, consisting of the elements of F that are regular at P . For a nonempty strict open subset W ⊂ X, we define RW = at each point of W . The above definition agrees with those in [HH10] and [HHK09], which \P ∈W O bX,P , and we let bRW be the t-adic completion of RW . If bRW is a domain we also define Thus RW is the subring of F consisting of the rational functions on bX that are regular considered RW and bRW only when W meets just one irreducible component of X. For W an affine open subset of X, we view Spec(bRW ) as a "thickening" of W , just as bX is a "thickening" we have a second curve bX ′ with function field F ′, we will write R′ X ′ of bX ′). of its closed fiber X. See Lemma 3.3(a) below. By convention, we also write FX := F . If W for the analogously defined rings (where W is now a nonempty strict open subset of the closed fiber 10 W , and F ′ W , bR′ Remark 3.2. (a) For W a non-empty open subset of X as above, and for any point 34.A); this uses that T and hence RW is excellent (by [Mat80], 34.B, 34.A). attention to affine open sets W . This is for the following reason: Given a collection of some but not all irreducible components of X, by [BLR90, Sect. 6.7, Proposition 4] P ∈ W , the completion of bRW at the ideal defined by P ∈ bX is the complete local ring of bX at P , which is a normal domain. Moreover, since RW is normal and reduced (i.e. has no nilpotents), the same conditions hold for its completion bRW (by [Mat80], 33.I, (b) When studying rings of the form bRW on normal models of F , it suffices to restrict there is an associated contraction π : bX → bY . This is a projective birational T - morphism π to a normal model bY of F over K that is an isomorphism away from these of bY given in [BLR90, Sect. 6.7, Theorem 1], if U is an open set strictly contained in the closed fiber Y of bY , and W = π−1(U), then the T -morphism π induces a T -algebra isomorphism between RW (taken on bX) and RU (taken on bY ). This in turn induces an isomorphism between the t-adic completions bRW and bRU . In particular, let W be any those components of X that are contained in W . Then bRW = bRU for U = π(W ), and connected open set strictly contained in X, and let π be chosen to contract precisely components, and which sends each of these components to a point. By the description moreover U is affine. With W ⊂ X as above, the reduced closed fiber of Spec(RW ) is Spec(RW /J), where J is the Jacobson radical of RW (i.e. the radical of the ideal tRW ). The corresponding statement holds for Spec(bRW ). Lemma 3.3. In the above situation, let W be a non-empty open subset of X. (a) If W is an affine open subset of X, then the reduced closed fibers of Spec(RW ) and irreducible components of X that are contained in W . Spec(bRW ) are each isomorphic to W . (b) More generally, if W 6= X, then the reduced closed fibers of Spec(RW ) and Spec(bRW ) are each isomorphic to π(W ), where π is the contraction of bX with respect to the Proof. (a) The assertion for RW is clear, and it then follows for bRW because RW and bRW (b) Note that π( W ) is open because the only components of W that are contracted by π are those contained in W . Also RW = Rπ(W ), as observed in Remark 3.2(b). So the assertion follows. have the same reduction modulo (t). W1, . . . , Wn be the connected components of W . Proposition 3.4. Let W be a nonempty proper open subset of X, the closed fiber of bX. Let (a) The ring bRW is a domain (and so FW is defined) if and only if W is connected. (b) The ring bRW is isomorphic toQn i=1 bRWi. 11 i=1 RW /J n i=1 J n i i = Qr (c) Let F ′ be a finite extension of F , and let bX ′ be the normalization of bX in F ′, with associated morphism π : bX ′ → bX and closed fiber X ′. Let W ′ = π−1(W ) ⊂ X ′. Then bRW ⊗RW R′ components of W ′. If W is connected, FW ⊗F F ′ is isomorphic toQn W ′ is isomorphic to bR′ W ′ =Qn n are the connected j=1 bR′ , where W ′ 1, . . . , W ′ j=1 F ′ W ′ j W ′ j . The forward direction of part (a) is now immediate. For the reverse implication of (a), reduced and normal. Moreover a normal scheme is irreducible if and only if it is connected. Proof. We begin with part (b). By Lemma 3.3(b), the reduced closed fiber of Spec(RW ) is the disjoint union of the reduced closed fibers of Spec(RWi), for i = 1, . . . , r. Thus the ideal J ⊂ RW defining the former closed fiber is the product of the relatively prime ideals for all n. The Chinese i=1 RWi/J n i RWi for all n. Since J is the radical of the ideal (t) ⊂ RW , and similarly for Ji and RWi, the asserted isomorphism follows by passing to the inverse limit. Ji ⊂ RW defining the latter; and more generally J n = Qr Remainder Theorem implies that RW /J n = Qr note that the condition that bRW is a domain is equivalent to its spectrum being reduced and irreducible. It was observed in Remark 3.2(a) that bRW (or equivalently its spectrum) is Thus it suffices to show that Spec(bRW ) is connected. So suppose that Spec(bRW ) is the disjoint union of two Zariski open subsets Y1 and Y2. We wish to show that Y1 or Y2 is empty. First note that Spec(bRW /(t)) is the disjoint union of the two Zariski open subsets ¯Y1 and ¯Y2, the restrictions of Y1 and Y2 to Spec(bRW /(t)). Also, Spec(bRW /(t)) is connected since W and hence π(W ) is, using Lemma 3.3(b). So either ¯Y1 or ¯Y2 is empty. But any maximal ideal of bRW contains t, and so any closed point of Spec(bRW ) (including any closed point of Yi) lies on Spec(bRW /(t)). Hence if Yi is non-empty, then so is W ′ → bR′ Finally, we prove part (c). The map bRW ⊗RW R′ III, §3.4, Theorem 3(ii)]. For the second assertion, where W is connected, write S = bR× Then the localization S−1bR′ W ′ = FW ⊗ bRW QbR′ FW ⊗ bRW bRW ⊗RW R′ The above definition of bRW requires that W is non-empty. But if the closed fiber X of bX is irreducible with generic point η, then we define bR∅ to be bRη. In this situation note that The fields FW arise in particular when considering a finite morphism f : bX → bX ′ of empty connected open subset of the closed fiber X ′ of bX ′, then by Proposition 3.4(c) the W ′ is an isomorphism by [Bou72, W . is a domain that is a finite extension of FW . W ′ = tensor product F ′ U ⊗F ′ F is a product of finitely many fields, each of them of the form FW for some open subset W ⊆ X. Namely, these sets W are the connected components of f −1(U) ⊆ X. Here the sets W can each meet more than one irreducible component of X, even if U meets just one irreducible component of X ′. projective normal T -curves, corresponding to a finite field extension F/F ′. If U is a non- ¯Yi. Thus either Y1 or Y2 is empty, concluding the proof of (a). W ′ = FW ⊗ bRW bR′ Thus it is a field, and is equal to its fraction field F ′ = FW ⊗ bRW bR′ =Q S−1bR′ the equivalence in Proposition 3.4 still holds. . So FW ⊗F F ′ = FW ⊗RW R′ W ′ j =Q F ′ . W ′ j W ′ j W ′ j W ′ j W ′ j 12 U , where k; and where In the other direction, consider a finite separable extension EU of the field F ′ U is F ′ and FW is F ′ T ; where U = A1 U -isomorphic to the given field EU . one-variable function field F over the complete discretely valued field K, and X the closed F ′ second part of the next result, there is a finite field extension F/F ′, corresponding to a finite U -isomorphic F ′ = K(x) is the function field of the projective line bX ′ = P1 U is the patching field associated to U on the closed fiber X ′ of bX ′. Then according to the morphism f : bX → bX ′ of projective normal T -curves, such that F ⊗F ′ F ′ to EU . Hence W := f −1(U) is a connected affine open subset of the closed fiber X of bX, Before stating the proposition, recall some notation. Let bX be a normal model of a fiber of bX. For each point P ∈ X, let RP be the local ring of bX at P . Its completion bRP ℘ in bRP that contains the uniformizer t of K defines a branch of X at P , lying on some irreducible component of X. The t-adic completion bR℘ of the local ring R℘ of bRP at ℘ is a ℘, bR′ such that P ∈ ¯U r U. If bX ′ is another curve with function field F ′ we will write bR′ complete discrete valuation ring; its fraction field is denoted by F℘. Hence F℘ contains FP , and is its completion. The field F℘ also contains FU if U is an irreducible open subset of X P , F ′ ℘ is a domain ([HH10, page 88]), with fraction field denoted by FP . Each height one prime etc. for the analogously defined objects. k, let P be the point (x = ∞) on P1 k, and let ℘ be the unique Proposition 3.5. Let U = A1 branch of P1 k at P , where X = P1 (a) For every finite separable field extension E℘ of F℘, there is a finite separable field extension EP of FP such that EP ⊗FP F℘ ∼= E℘ over F℘. k is viewed as the closed fiber of bX = P1 T . (b) For every finite separable field extension EU of FU , there is a finite separable field ∼= F ′ then F ′ ⊗F FU extension F ′ of F := K(x) such that F ′ ⊗F FU U ′ over FU , where U ′ = π−1(U) ⊂ X ′ is connected. ∼= EU over FU . Moreover if bX ′ is the normalization of bX in F ′, with closed fiber X ′ and associated morphism π : bX ′ → bX, (b) The tensor product EU ⊗FU F℘ is a finite direct product Qi E℘,i of finite separable Proof. (a) Since F℘ is the ℘-adic completion of FP , this follows from Krasner's Lemma ([Lan70], Prop. II.2.3). field extensions E℘,i of F℘. By part (a), for each i there is a finite separable field extension EP,i of FP such that EP,i ⊗FP F℘ is isomorphic to E℘,i over F℘. We thus have an isomorphism of separable F℘-algebras EU ⊗FU F℘ →(cid:0)Yi EP,i(cid:1) ⊗FP F℘. Applying the patching assertion Theorem 7.1 of [HH10] (in the context of Theorem 5.9 of that paper), we obtain a finite separable F -algebra F ′ that induces EU over FU and induces Qi EP,i over FP , compatibly with this isomorphism. Since EU is a field, so is F ′. by Proposition 3.4(c), where U ′ For the last part, F ′ ⊗F FU n are the ∼= EU is a field. So n = 1 and the assertion 1, . . . , U ′ j=1 F ′ U ′ j connected components of U ′. But F ′ ⊗F FU follows. ∼=Qn 13 As the above proof shows, Proposition 3.5 holds more generally for non-empty affine open subsets U of the closed fiber X of a smooth projective T -curve bX, together with the set of points P ∈ X in the complement of U. For this, one cites Theorem 7.1 of [HH10] in the context of Theorem 5.10 of that paper. In the case that T is an equal characteristic complete discrete valuation ring, an analog of Proposition 3.5(b) for a finite extension of FP appeared in [HHK13, Lemma 3.8]. For a more general choice of T , there is the following weaker result, which nevertheless will suffice for our purposes below (in Corollary 3.12): closed fiber X, and let E be a finite separable extension of FP . Let S be the integral closure Proposition 3.6. Let bX be a projective normal T -curve, let P be a closed point on the of bRP in E, and let bV ∗ → Spec(S) be a birational projective morphism, with bV ∗ normal. Then there exist normal schemes bV , bZ, and bY and a commutative diagram / Spec(S) bV ∗ Spec(bRP ) / bX bV bZ bY bV bZ of T -schemes, where the horizontal maps are birational projective morphisms that are iso- morphisms away from (the inverse image of ) P ; the bottom half of the diagram is a pullback Proof. Let L be the Galois closure of E over FP , let G = Gal(L/FP ), and let R be the The Galois group G = Gal(E/FP ) acts on the (isomorphism classes of) birational pro- jective morphisms to Spec(S). Consider the fiber product of the morphisms in the orbit of square; and the morphism bV → bZ is finite. integral closure of bRP in L. Let H = Gal(L/E) and letcW ∗ be the normalization of bV ∗ ×S R. It suffices to prove the assertion with E, S, and bV replaced by L, R, and cW ∗, provided we also show that the G-action on Spec(R) lifts to a G-action on the asserted spacecW . Namely, we can then take bV =cW /H. So we now assume that E is Galois over FP . bV ∗ → Spec(S), and let bV be the irreducible component that dominates Spec(S). Then bV is normal since each G-conjugate of bV ∗ is; and bV → Spec(S) is a G-stable birational projective morphism that factors through bV ∗ → Spec(S). Thus the action of G on Spec(S) lifts to an action of G on bV . Let bZ be the quotient of bV by G. Then the birational projective morphism bV → Spec(S) descends to a birational projective morphism bZ → Spec(bRP ). That is, we obtain a commutative diagram bV ∗ / Spec(S) / Spec(bRP ) 14 / / / / / / / / / / whose vertical arrows are each finite and G-Galois, with generic fiber corresponding to the G-Galois field extension E/FP . The bottom horizontal morphism is a composition of blowups and blowdowns, centered at P and at points on exceptional divisors lying over P . We may perform the corresponding blowups and blowdowns on bX, observing inductively that at each step the spaces mapping to Spec(bRP ) and to bX have the same exceptional divisors (fibers over P ), and that generators of the local ring at a closed point over P ∈ bX also generate the local ring at the corresponding closed point over P ∈ Spec(bRP ). This process yields a pullback diagram bZ bY Spec(bRP ) / bX where the bottom horizontal map is a birational projective morphism which is an isomor- 3.2 Patching phism away from P ∈ bX. This gives the desired conclusion. Below, bX is a (projective) normal model of a one-variable function field F over the complete discretely valued field K, and X is the closed fiber of bX. As in Section 3.1, for each point P on the closed fiber X of bX we have an associated complete local domain bRP with fraction field the domain bRU and its fraction field FU . For P ∈ P and ℘ a branch of X at P , we also have the complete discrete valuation ring bR℘ and its fraction field F℘. Consider a non-empty finite subset P ⊂ X. Let W be the set of connected components of the complement of P; each of these connected components U ∈ W is a strict open subset of X. The set of all the branches of X at points of P is denoted by B. FP ; and for each non-empty connected Zariski strict open subset U of X we may consider If U ⊂ U ′ are connected strict open subsets of X, then FU ′ ⊂ FU . For P ∈ U, there is also an inclusion FU ⊂ FP ; and if ℘ is a branch at P lying on the closure of U, then there are inclusions FP , FU ⊂ F℘. These containments are compatible, as U, U ′ vary. The next result generalizes [HH10, Theorem 6.4] and [HHK13, Proposition 2.3(a)]. following injective diamonds. connected components of the complement V of P in the closed fiber X, and let B be the set Q be the subring of FQ that consists Proposition 3.7. Let P be a non-empty finite set of closed points of bX, let W be the set of of branches of X at the points of P. For Q ∈ P, let bR◦ of the elements that lie in bR℘ for each branch ℘ of X at Q. Then patching holds for the (a) F• = (F ≤ QU ∈W FU , QQ∈P FQ ≤ Q℘∈B F℘). (b) R• = (RV ≤ QU ∈W bRU , QQ∈P bR◦ Q ≤ Q℘∈B bR℘). 15 / / / R◦ R1, R◦ 2 ≤ R0). The proofs for the two diamonds are similar. We begin with F•. Proof. For short write F1 = QU ∈W FU , F2 = QQ∈P FQ, F0 = Q℘∈B F℘, R1 = QU ∈W bRU , 2 = QQ∈P bR◦ Q, and R0 = Q℘∈B bR℘. Thus F• = (F ≤ F1, F2 ≤ F0) and R• = (RV ≤ trivially. By [HHK15, Proposition 3.3], there is then a finite morphism f : bX → P1 First consider the special case that the set P meets each irreducible component of X non- T such that T . Thus W is the set of connected T , so that the map f 0) be defined analogously as T , with P′ = {∞} and W′ = {U ′}. By [HH10, Theorem 5.9], patching •. Since F/F ′ is a finite field extension, Lemma 2.10 implies that patching holds 1 ⊗F ′ F, F ′ 0 ⊗F ′ F ). The proposition now follows from the assertion that i ⊗F ′ F , which holds for i = 1 by Proposition 3.4(c) and for i = 0, 2 by [HH10, P is the fiber over the point ∞ on the closed fiber P1 components of f −1(U ′), where U ′ = A1 gives a finite field extension F/F ′. Let F ′ above for the curve P1 holds for F ′ for (F, F ′ Fi = F ′ Lemma 6.2] (enlarging the set S′ there if necessary). k. Let F ′ be the function field of P1 • = (F ′ ≤ F ′ 2 ⊗F ′ F, F ′ 2 ≤ F ′ k of P1 1, F ′ Now consider the case that P does not meet each irreducible component of X. Since P is non-empty, not every irreducible component of X is disjoint from P. So by Remark 3.2(b), we may contract the components that are disjoint from P, via a proper birational morphism proof of patching for F•. isomorphism between F℘ and Fπ(℘) for ℘ ∈ B. Moreover for U ∈ W, the morphism π induces an isomorphism between FU and Fπ(U ), by Remark 3.2(b), since these are the fraction fields π : bX → bY . The set P maps bijectively to its image in bY , with π inducing an isomorphism between FQ (taken on bX) and Fπ(Q) (taken on bY ), for Q ∈ P. Similarly, π induces an of bRU and bRπ(U ) (where the rings are taken on bX and bY respectively). Thus the assertion for bX is equivalent to the assertion for bY , which holds by the first case. This completes the morphism f : bX → P1 1 = bR′ Next we turn to patching for the diamond R•. As above, we are reduced to the case that the set P meets each irreducible component of X non-trivially, so that there is a finite T such that P = f −1(∞). With notation as above, consider the 2 ≤ R′ 0) taken with respect to P′ = {∞}, where • = (R′ U ′ RV by Proposition 3.4(b,c). Together with [HH10, • ⊗R′ analogous diamond R′ R′ Lemma 6.2], this implies that R• = R′ U ′. Note that R1 = bRV = R′ Also, RV is a finitely generated free module over R′ using that the finitely generated module RV /tRV over the principal ideal domain R′ k[x] is torsion-free and hence free and also that (t) is the Jacobson radical of R′ intersection holds for R′ 1 ∩ F ′ = R′ factorization holds for GLn(R′ R′ U ′ by [Bou72, Proposition II.3.2.5(ii)], U ′/(t) = U ′. Now U ′, and •) by [HHK13, Proposition 2.3(a)]. That is, patching holds for •. Hence it also holds for R• = R′ • because R′ U ′ RV , by Lemma 2.10. U ′ ≤ R′ U ′ ⊆ R′ 1 ∩ R′◦ 2 ⊆ R′ U ′ RV . 1 ∩ F ′ 1 ∩ F ′ 2 = R′ 1, R′◦ 1 ⊗R′ • ⊗R′ The next result generalizes [HHK13, Corollary 2.4, Theorem 3.1(a), and Corollary 3.3(a)], the second of which is a form of the Weierstrass Preparation Theorem. It follows easily from Proposition 3.7(b) in the same way that those three previous results followed from [HHK13, Proposition 2.3(a)]. See also [Cuo13, Theorem 3.6]. Corollary 3.8. Let W be as in Proposition 3.7. 16 (a) Suppose that for every U ∈ W we are given an element aU ∈ F × U . Then there exist U such that aU = bcU ∈ F × b ∈ F and elements cU ∈ bR× (b) If U ∈ W and a ∈ FU then there exist b ∈ F and c ∈ bR× of the residue field k of T , then there exist b ∈ F and c ∈ bR× U such that a = bc ∈ FU . More generally, if a ∈ FU and n is a positive integer that is not divisible by the characteristic U such that a = bcn ∈ FU . The next result is analogous to Proposition 3.7(a), with F replaced by FW , for W ⊂ X. U for all U ∈ W. Proposition 3.9. Let W ⊆ X be a connected open subset of X. Let P be a non-empty finite set of closed points of W ; let W be the set of connected components of the complement of P in W ; and let B be the set of branches of W at the points of P. Then patching holds for the injective diamond FW • = FW ≤ YU ∈W FU ,YQ∈P FQ ≤Y℘∈B F℘! . Proof. If W = X then the assertion is given by Proposition 3.7(a). So assume that W is strictly contained in X. After blowing down all irreducible components of X that do not intersect W as in Remark 3.2(b), we may assume that the closure of W is X. Let eP be the complement of W in its closure X. Also let eB be the set of branches at the points of eP. By Proposition 3.7(a), patching holds for the diamond eF• = (F ≤ FW ,QQ∈eP FQ ≤Q℘∈ eB F℘). Let bP be the disjoint union eP ⊔ P and let bB be the set of branches at the points of bP. Thus bB = eB ⊔ B. Notice that the set of connected components of the complement ofbP in X holds for the diamond bF• = (F ≤QU ∈W FU ,QQ∈bP FQ ≤Q℘∈ bB F℘) by Proposition 3.7(a). is the set of connected components of the complement of P in W , i.e., W. Again, patching The general linear groups for the various products of fields form the following diagram: GLn(F℘) GLn(F℘) GLn(FQ) GLn(FU ) GLn(FQ) YQ∈eP Y℘∈ eB <②②②②②② b❊❊❊❊❊❊ YU ∈W f▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼ Y℘∈B <②②②②②②② c●●●●●●● YQ∈P e❑❑❑❑❑❑ 9rrrrrr GLn(FW ) 7♥♥♥♥♥♥♥♥ GLn(F ) As noted above, patching holds for the diamonds eF• and bF•; and so by Theorem 2.8(a), factorization and intersection hold for the diamonds of groups G• := GLn(eF•) =(cid:0)GLn(F ) ≤ GLn(FW ),YQ∈eP GLn(FU ),YQ∈bP • := GLn(bF•) =(cid:0)GLn(F ) ≤ YU ∈W GLn(FQ) ≤Y℘∈ eB GLn(FQ) ≤Y℘∈ bB GLn(F℘)(cid:1), GLn(F℘)(cid:1). G′ 17 < b < c e 9 7 f Proposition 2.14 (parts 1 and 4) and Lemma 2.2 yield factorization and intersection for GLn(FW •) =(cid:0)GLn(FW ) ≤ YU ∈W GLn(FU ),YQ∈P GLn(FQ) ≤Y℘∈B GLn(F℘)(cid:1). That is, patching holds for the diamond FW •, as desired. The next result, which answers a question posed by Yong Hu, permits patching on the in which a non-empty connected union V of some but not all of the irreducible components T -curves, having closed fibers X, Y respectively. Let P ∈ Y be a closed point, let V ⊂ X FQ,YU ∈W FU ≤Y℘∈B F℘! , with respect to the natural inclusions. Proof. First observe that there are natural inclusions of FP into FU and FQ, for U ∈ W the set of connected components of V r P, and let B be the set of branches at the points in P along the components of V . Then patching holds for the injective diamond FP • = exceptional divisor of a blow-up f : bX → bY . Alternatively, we can view f as a blow-down, of the closed fiber X ⊂ bX are contracted to a point P ∈ Y ⊂ bY (cf. Remark 3.2(b)). Proposition 3.10. Let f : bX → bY be a proper birational morphism of projective normal be the inverse image of P in X, and let eY ⊆ X be the proper transform of Y . Suppose that dim(V ) = 1, and that f restricts to an isomorphism bX r V → bY r {P }. Choose a finite collection of closed points P in V that includes all the points of V ∩ eY . Let W be FP ≤ YQ∈P and Q ∈ P. Namely, the natural morphism Spec(bRU ) → bX factors through bXP := f −1(Spec(bRP )), the pullback of bX → bY via Spec(bRP ) → bY . Since bX → bY is birational, so is bXP → Spec(bRP ), and hence the function field of bXP is FP . The morphism Spec(bRU ) → bXP Let eW be the set of connected components of Y r {P }. Let eB be the set of branches of regard the elements of eW as open subsets of X, and the elements of eB as branches of eY . Write bW for the disjoint union eW ⊔ W. The set of points of X that lie in no element of bW is exactly P; let bB be the set of branches of X at points of P. Thus bB equals the disjoint union eB ⊔ B. Note that at the points of P, some of the branches of X are elements of B and some are in eB, depending on whether the branches lie on a component of V or of eY . induces a homomorphism of function fields in the other direction, FP → FU , which is neces- sarily an inclusion. The case of FP → FQ is similar. Y at P . Via f , we may identify X r V with its isomorphic image Y r {P }. We may thus 18 We may now consider the associated diagram of groups: YU ∈ eW Y℘∈ eB <①①①①①① b❊❊❊❊❊❊ YQ∈P f▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼ GLn(FU ) GLn(FQ) GLn(FU ) GLn(F℘) GLn(F℘) Y℘∈B <②②②②②②② c❍❍❍❍❍❍❍ YU ∈W e❏❏❏❏❏❏ 8rrrrrr GLn(FP ) 7♦♦♦♦♦♦♦♦ GLn(F ) By Proposition 3.7(a) and Lemma 2.2, patching holds for the diamonds eF• = (F ≤ YU ∈ eW FU , FP ≤Y℘∈ eB F℘), bF• = (F ≤ YQ∈P FQ,YU ∈ bW FU ≤Y℘∈ bB F℘). Y = P1 G′ satisfies factorization and intersection; i.e. patching holds for FP •, as desired. That is, factorization and intersection hold for the diamonds of groups G• := GLn(eF•) and • := GLn(bF•). Parts (1) and (4) of Proposition 2.14 imply that the diamond GLn(FP •) Example 3.11. Let T = k[[t]] and let bY be the projective y-line P1 k. Let P be the point y = 0 on Y , with complete local ring bRP = k[[y, t]]. Consider the blow-up bX → bY of bY at P . The exceptional divisor V is a copy of the x-line over k, with x = 0 at the point P ′ of bX where V meets the proper transform of Y . The complete local ring bRP ′ is k[[y, t, x]]/(t − xy) = k[[x, y]]. Writing W for the complement of P ′ in V , the ring bRW (with y = x−1t). The unique branch ℘ of V at P ′ has associated ring bR℘ = k((x))[[y]], which contains bRP ′ and bRW . The intersection of these two rings in bR℘ is bRP . The respective fraction fields are FP = k((y, t)), FP ′ = k((x, y)), FW = frac(cid:0)k[x−1][[t]](cid:1), and F℘ = k((x))((y)). They satisfy the intersection condition FP = FP ′ ∩ FW ⊂ F℘, and they also satisfy factorization for GLn. This example is a twisted form of the example given in [HH10] after Theorem 5.9 there. It is also related to the situations discussed in [PN10, Section 1] and [BT13]. is the t-adic completion of k[[y, t]][x−1]/(x−1t−y), which is naturally isomorphic to k[x−1][[t]] T , with closed fiber The next corollary, which will be useful in Section 4.2, is a variant of the previous proposition. Here we blow up Spec(S) for some two-dimensional complete ring S that need not be of the form bRP , but instead can be a finite extension of some bRP or some bRW . In this situation, we can again consider fields of the form FQ, FU , and F℘, associated to this blowup; the previous definitions carry over mutatis mutandis to the case of any two- dimensional normal scheme whose closed points all lie on a connected curve. either a closed point P ∈ X or a connected affine open subset W ⊂ X. Let E be a finite Corollary 3.12. Let bX be a projective normal T -curve with closed fiber X, and let ξ be separable extension of Fξ, let S be the integral closure of bRξ in E, and let eξ be the inverse 19 < b < c e 8 7 f image of ξ under Spec(S) → Spec(bRξ). Let D be a divisor on Spec(S). Then there exist a birational projective morphism π : bV → Spec(S) and a finite set P of closed points of V := π−1(eξ) such that the following hold: (i) bV is a normal scheme. (ii) D′ := π−1(D) is a normal crossing divisor on bV . (iii) P contains all the points of V where V ∪D′ is not regular, and it meets every connected component of the exceptional locus of π. (iv) Let W be the set of connected components of V r P, and let B be the collection of branches of V at the points of P. If P, W are non-empty, then patching holds for the injective diamond E• = (E ≤YQ∈P FQ,YU ∈W FU ≤Y℘∈B F℘). Proof. Case I: ξ = P ∈ X. Since Spec(S) is two-dimensional, excellent, and normal, by [Abh69] and [Lip75] there is a birational projective morphism (viz. a composition of blowups) π∗ : bV ∗ → Spec(S) such that bV ∗ is regular and D∗ := (π∗)−1(D) ⊂ bV ∗ is a normal crossing divisor. By Proposition 3.6, we obtain a diagram π / Spec(S) bV ∗ bV bZ bY σ ω α Spec(bRP ) / bX with the properties asserted there. Here D′ := π−1(D) is a normal crossing divisor on bV because D∗ is, and since the map bV → bV ∗ is a blow-up. Recall that V := π−1(eP ), where eP ∈ Spec(S) is the inverse image of P ∈ Spec(bRP ). Let Y = ω−1(X) be the closed fiber of bY , and V0 := α(V ) = ω−1(P ) ⊂ Y . Choose a non-empty finite set P0 of closed points of V0 that contains the image under α of the locus where V ∪ D′ is not regular, and also contains the points where V0 meets the proper transform of X in Y . Let P = α−1(P0) ⊆ V . Thus the above properties (i), (ii), (iii) hold. Let W be as in (iv). If π is an isomorphism, then W is empty and we are done. Otherwise, V is a curve, W is non-empty, and it remains to show that patching holds for the diamond E•. Let W0 be the set of connected components of V0 r P0 and let B0 be the set of branches of V0 at the points of P0. Thus the elements of W are the inverse images of the elements of W0, and similarly for B and B0. By Proposition 3.10, patching holds for the diamond FU ≤ Y℘∈B0 F℘(cid:1), taking FP with respect to bX and taking the FP • = (cid:0)FP ≤ YQ∈P0 FQ, YU ∈W0 other fields with respect to bY . 20 / / ) ) / / / / By Proposition 3.6, the bottom half of the above diagram is a pullback square, with Z := σ−1(P ) → V0 an isomorphism. For each U ∈ W0, with inverse image U ′ ⊆ Z, the natural map FU → FU ′ is an isomorphism; and similarly for P0 and B0. The diamond FP • may thus be considered to be taken with respect to bZ. The morphism bV → bZ in Proposition 3.6 is finite. ThusQU ∈W FU =QU ∈W0 FU ⊗FP E, where FU on the left is taken with respect tobV and FU on the right is taken with respect to bZ (or bY ); and the analogous isomorphisms hold for the fields FQ and F℘. (These isomorphisms are as in Proposition 3.4(c) and [HH10, Lemma 6.2], whose statements and proofs carry over to this somewhat more general situation.) Applying Lemma 2.10(b) with respect to the field extension E/FP , we obtain the desired conclusion. Case II: ξ is a connected affine open subset W ⊂ X. Definition 2.9). FQ ≤ Y℘∈B∗ and similarly for the fields FQ and F℘, as at the end of the proof of Case I. As in that Let W∗ be the set of connected components of W r P∗ and let B∗ be the set of branches of Recall that fW ⊂ Spec(S) is the inverse image of W ⊂ Spec(bRW ). Choose a non-empty finite subset P∗ of closed points of W that contains the image under f : Spec(S) → Spec(bRW ) of the points wherefW ∪ D is not regular, and leteP = f −1(P∗) ⊂fW ⊂ Spec(S). Write eW for the set of connected components of the complement ofeP infW , and eB for the set of branches offW at the points of eP. W at the points of P∗. Thus the elements of eW are the inverse images under f of the elements of W∗, and similarly for eB and B∗. By Proposition 3.9, patching holds for the diamond FU , YQ∈P∗ FW • =(cid:0)FW ≤ YU ∈W∗ F℘(cid:1). Since f is finite,QU ∈ eW FU =QU ∈W∗ FU ⊗FW E, proof, patching holds for the diamond eE• := (E ≤ QU ∈ eW FU ,QQ∈eP FQ ≤ Q℘∈ eB F℘), by Lemma 2.10(b). That is, factorization and intersection hold for G• := GLn(eE•) for all n (see LeteP′ ⊆eP consist of the closed points offW where D is not a normal crossing divisor, and write bP =eP reP′. Thus eP =eP′ ⊔bP. Our strategy will be to blow up Spec(S) at the points of eP′, obtaining a refined diamond E•; and then to use that patching holds for eE• and the For each Q ∈ eP′, consider the complete local ring bRQ of Spec(S) at the point Q, with fraction field FQ. Thus f (Q) ∈ P∗ ⊂ W ⊂ cW = Spec(bRW ), and FQ is a finite separable extension of Ff (Q), viz. a factor of Ff (Q) ⊗FW E. Let DQ be the restriction of fW ∪ D to Spec(bRQ). By Case I, for each Q ∈eP′, there is a birational projective morphism (viz. a composition Q : bV ′ Q → Spec(bRQ) for which conditions (i)-(iv) are satisfied, with respect to Q• := (FQ ≤ QQ′∈P′ FQ′,QU ∈W′ Q are non-empty since Q ∈eP′. That is, intersection and • =QQ∈eP′ F ′ Q•) for all n. By Lemma 2.15(a), these properties also hold for Q•, with the product of diamonds being taken entry by entry. of blowups) π′ the divisor DQ, some finite subset P′ the diamond F ′ Q• by (iv), where P′ holds for F ′ factorization hold for GLn(F ′ GLn(F ′ •), where F ′ Q (Q), the associated sets W′ Q, and F℘). In particular, patching Q and B′ diamond arising from the exceptional locus to obtain the same for E• via Proposition 2.14. FU ≤ Q℘∈B′ Q Q of V ′ Q := π′−1 Q Q Q, W′ 21 Since blowing up is local, we may take the corresponding blowups of Spec(S) at ideals be the disjoint union of the sets P′ Now properties (i), (ii), (iii) hold for π with respect to the divisor D and the set P, respectively supported at the points Q ∈ eP′. We thus obtain a projective birational mor- phism π : bV → Spec(S) which is an isomorphism away from eP′, and whose pullback under Q, for Q ∈ eP′. We may similarly regard Spec(bRQ) → Spec(S) may be identified with bV ′ fW reP′ as contained in bV . With respect to these identifications, let P′ ⊂ V := π−1(fW ) Q for Q ∈ eP′, and similarly define W′ and B′. Note that P := P′ ⊔bP contains all the points of V at which V ∪ D′ is not regular, where D′ = π−1(D). where (ii) uses that D ⊂ Spec(S) is a normal crossing divisor away from eP′. Let W be the of P. Thus W = eW ⊔ W′ and B = eB ⊔ B′. It remains to show that property (iv) is satisfied for the diamond E• := (E ≤QQ∈P FQ,QU ∈W FU ≤Q℘∈B F℘). GLn(FQ′), YU ∈W′ set of connected components of V r P, and let B be the set of branches of V at the points It was shown above that for any n, intersection and factorization hold for GLn(FQ) ≤ YQ′∈P′ GLn(FU ) ≤ Y℘∈B′ Applying Lemma 2.15(b) with A = GLn(QQ∈bP FQ), it follows that these two properties also hold for the (non-injective) diamond eH• := GLn(F ′ GLn(F℘)(cid:1). •) =(cid:0)YQ∈eP′ GLn(FQ), YU ∈W′ GLn(FU ),Y℘∈B′ GLn(F℘)(cid:1). GLn(FQ),YQ∈P H• :=(cid:0)YQ∈eP • := GLn(E•) =(cid:0)GLn(E) ≤ YQ∈P G′ (Here we use that eP =eP′ ⊔bP and that P = P′ ⊔bP.) Consider the injective diamond GLn(F℘)(cid:1). Since W = eW ⊔ W′ and B = eB ⊔ B′, Proposition 2.14 applies to the diamonds G•, G′ GLn(FU ) ≤Y℘∈B GLn(FQ), YU ∈W H•, with respect to the following diagram: •, and :✉✉✉✉ GLn(FU ) YU ∈ eW d❍❍❍❍ GLn(F℘) Y℘∈ eB YQ∈P e❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑❑ Y℘∈B′ :✉✉✉✉✉ GLn(FQ) d■■■■■ YQ∈eP 8♣♣♣♣ GLn(E) GLn(F℘) e❑❑❑❑❑❑ YU ∈W′ 9ssssss GLn(FQ) GLn(FU ) By parts (5) and (6) of that proposition, it follows that intersection and factorization hold for G′ • = GLn(E•) for all n; i.e. patching holds for E• = (E ≤ YQ∈P FU ≤Y℘∈B FQ,YU ∈W F℘). 22 : d : e d 9 8 e Remark 3.13. As the proof of Corollary 3.12 shows, if the conclusion holds for a given choice of bV and of P ⊂ V , and if P′ ⊂ V is any other finite subset of V , then one can enlarge P so as to contain P′ and still satisfy the conclusion of the corollary. 3.3 Factorization for diamonds of groups In the situation of Section 3.2, we can obtain results about factorization for diamonds that arise from algebraic groups other than just GLn. This is useful for obtaining local-global principles for torsors; see Theorem 2.13. Proposition 3.14. Let bX be a normal model of a one-variable function field F over K, and • = (F ≤QU ∈W′ FU , QQ∈P′ FQ ≤Q℘∈B ′ F℘). let G be an algebraic group over F . In the notation of Proposition 3.7, 3.9, or 3.10, let P′ be a finite set of closed points of X that contains P. In the context of Proposition 3.9, assume also that P′ contains W r W . Let W′ be the set of components of X r P′, and let B′ be the set of branches of X at the points of P′. Set F ′ If factorization holds for the diamond G(F ′ respectively. •) then it holds for G(F•), G(FW •), or G(FP •), Proof. For short write F ′ 0). We consider each of the three cases in turn. Case of Proposition 3.7. Let n be the number of points in P′ that are not in P. By induction we are reduced to the case that n = 1, since each set P′′ with P ⊂ P′′ ⊂ P′ is also an allowable finite subset of X in the notation of Proposition 3.7. Write P′ = P ⊔ {P }. • = (F ≤ F ′ 2 ≤ F ′ 1, F ′ Let W be the unique element of W that contains P , and let W ′ = W r {P }. Consider the associated diamond FW •, defined as in Proposition 3.9 with respect to the set {P } ⊂ F℘), where W′ be the set of connected components of W ′, and where BP is the set of branches of W at P . Patching holds for FW • by Proposition 3.9; and in particular the intersection property holds for FW • (see Definition 2.9). W . Thus FW • = (FW ≤ QU ∈W′ P Let By Lemma 2.12, intersection holds for eH• since it holds for FW •. Let A =QU ∈Wr{W } G(FU ). Let H• be the coordinate-wise product of diamonds eH• × (A, A, 1, 1). That is, G(F℘)). G(FU ), G(FP ) ≤ Y℘∈BP Then intersection holds for H• by Lemma 2.15(b). Using Lemma 2.2, the desired conclusion then follows from Proposition 2.14(2), with G•, G′ •, and H• as above, with respect to the following refinement diagram: 23 FU , FP ≤ Q℘∈BP G(FU ), YQ∈P G(FQ) ≤ Y℘∈B G(FU ), YQ∈P′ G(FQ) ≤ Y℘∈B′ G(FU ), G(FP ) ≤ Y℘∈BP G(F℘)), G(F℘)), G(F℘)). G′ • = G(F ′ G• = G(F•) = (G(F ) ≤ YU ∈W •) = (G(F ) ≤ YU ∈W′ eH• = G(FW •) = (G(FW ) ≤ YU ∈W′ G(FU ) ≤ YU ∈W′ H• = (YU ∈W QQ∈P G(FQ) 5❦❦❦❦❦❦❦❦ i❙❙❙❙❙❙❙❙❙ Q℘∈B G(F℘) i❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ QU ∈W′ G(FU ) 5❦❦❦❦❦❦❦❦❦ i❚❚❚❚❚❚❚❚❚ 5❥❥❥❥❥❥❥❥❥❥❥ G(F ) G(F℘) Q℘∈BP QU ∈W G(FU ) h◗◗◗◗◗◗◗◗ 6♠♠♠♠♠♠♠♠♠ G(FP ) Case of Proposition 3.9. Let P′′ be the subset of P′ obtained by deleting those points • . By the case of • ). So after replacing P′ by P′′, that lie in W but not in P, and consider the corresponding diamond F ′′ Proposition 3.7, it follows that factorization holds for G(F ′′ we may assume that P′ ∩ W = P. • = G(F ′ as in that case. Using Lemma 2.2, it follows from Proposition 2.14(1) that H• satisfies Write P′ as a disjoint union P ⊔ P. Thus P is disjoint from W . Let W be the set of connected components of the complement of P in X. Thus W ∈ W, using the hypothesis that P′ contains W r W . Let B be the set of branches of X at the points of P. Consider the diamond F• = (F ≤ F1, F2 ≤ F0), where F1, F2, F0 are defined analogously to F ′ 2, F ′ 0. Write G• = G( F•), G′ •), and eH• = G(FW •). Letting A = QU ∈ eWr{W } G(FU ), and setting H• = eH• × (A, A, 1, 1) as in the first case, we then obtain a refinement diagram factorization. By Lemma 2.15(b), so does eH• = G(FW •). Case of Proposition 3.10. We proceed analogously to the case of Proposition 3.9. As in that case, we may assume that P′ ∩V = P, via the case of Proposition 3.7. Write P′ = P⊔ P′′, so that P′′ ⊂ X is disjoint from V . Identifying X r V with Y r {P }, we may view P′′ as a subset of Y r {P }. Let P = P′′ ⊔ {P } ⊂ Y . Let W be the set of connected components of the complement of P in Y , and B the set of branches of Y at the points of P. Consider the diamond F• = (F ≤ F1, F2 ≤ F0), where F1, F2, F0 are defined as in the previous case 1, F ′ • = G(F ′ Write G• = G( F•), G′ case, using Lemma 2.2, factorization for H• follows from Proposition 2.14(1). Lemma 2.15(b) (though with respect to bY , rather than bX; note that F is also the function field of bY ). •), and eH• = G(FP •). Setting A = QQ∈ eP r{P } G(FQ) and H• = eH• × (A, A, 1, 1) as before, we again obtain a refinement diagram. As in the previous then implies that factorization also holds for eH• = G(FP •). Corollary 3.15. Under the hypothesis of Proposition 3.7, 3.9, or 3.10, let G be a rational connected linear algebraic group over F . Then factorization holds for the diamond G(F•), G(FW •), or G(FP •), respectively. Proof. Choose a finite set P′ of closed points of X that contains P and all the points where irreducible components of X meet. In the context of Proposition 3.9, assume also that P′ contains W r W . With notation as in the statement of Proposition 3.14, factorization holds for the diamond G(F ′ •) by [HHK09, Theorem 3.6] (using that the simultaneous factorization condition in [HHK09] is equivalent to the factorization assertion for diamonds; see [HHK14, Section 2.1.3]). Hence the conclusion follows by Proposition 3.14. 24 5 i 5 h i 6 i 5 In the situation of Corollary 3.15, it then follows from Theorem 2.13 that if V is the class then the local-global principle holds for V with respect to F•. The corresponding local-global of F -varieties V with a G-action such that G(Q℘∈B F℘) acts transitively on V (Q℘∈B F℘), assertions also follow for varieties over FW or FP (satisfying transitivity overQ℘∈B F℘), with respect to FW • or FP •, respectively. Remark 3.16. Proposition 3.14 similarly implies the conclusion of Corollary 3.15 for any linear algebraic group G over F such that factorization holds for G with respect to any choice of a non-empty finite subset P′ of X that includes all the points where distinct components meet. This includes all connected retract rational groups G, by [Kra10]. As shown in [HHK15, Corollary 6.5], if the reduction graph associated to F is a tree then this property also holds for all linear algebraic groups G over F that are rational but not necessarily connected (i.e. each connected component is F -rational). 4 Applications to local-global principles and field invari- ants We now apply the above results in order to obtain applications in the contexts of quadratic forms and central simple algebras. These applications, which concern local-global principles and invariants of fields, extend and build on results that appeared in [HHK09], [HHK15], and [HHK13], as well as in [Lee13], [Hu13], [Hu12], [Hu15], and [PS14]. The fields we consider will be finite extensions of fields FP and FU , for P a closed point and U a connected open subset of the closed fiber of a curve over a complete discrete valuation ring, in the notation of Section 3. 4.1 Applications to quadratic forms Here we present applications to quadratic forms, concerning local-global principles and in- variants of fields, especially the u-invariant. We focus on the fields arising in Section 3 and finite separable extensions of these, in particular proving results that generalize and extend assertions in [HHK15, Section 9.2], and [HHK13, Section 4.1] regarding the Witt ring, Witt index, and u-invariant. As a consequence, we obtain a local-global result for the value of the u-invariant (Corollary 4.8). Due to [PS14], this latter result also applies in the case of mixed characteristic (0, 2), which is often avoided in quadratic form theory. Afterwards we obtain a result (Theorem 4.11) concerning the value of the u-invariant for finite separable extensions of fields such as k((x, t)) and the fraction field of k[x][[t]], as well as mixed characteristic analogs of such fields. We begin with local-global principles, starting with a more general result in the abstract context of diamonds. The reader is referred to [Lam05] for basic notions such as isotropic and hyperbolic forms, the Witt ring and fundamental ideal, the Witt index, and the u-invariant. If Fv are fields (for v in some index set), we define W (Q Fv) :=Q W (Fv) and I(Q Fv) := Q I(Fv). Recall that for any field E and quadratic form q over E, H 1(E, SO(q)) classifies 25 quadratic forms of the same dimension and discriminant as q, with q corresponding to the distinguished element of the Galois cohomology set; see [KMRT98, 29.29]. (In part (b) below we write µ2 rather than Z/2Z as in [HHK15]; but these are equivalent since the characteristic is not two.) Theorem 4.1. Suppose that F• = (F ≤ F1, F2 ≤ F0) is a diamond of rings with F a field of characteristic unequal to two, and each Fi a finite direct product of fields. Write Fv where each Fv is a field. Assume moreover that patching holds for F• and that we have factorization for the diamond SO(nh)(F•) for each n > 0, where h is a hyperbolic plane h1, −1i. Then: Fi =Qv∈Vi (a) We have an exact sequence 0 → PF• → W(F ) → W(F1) × W(F2) → W(F0), where the map on the right is given by taking the difference of the restrictions of the two Witt classes, and where PF• is the subgroup of W(F ) consisting of classes of locally hyperbolic binary Pfister forms; i.e. forms h1, −di where d ∈ F × becomes a square in each Fi. (b) The group PF• is naturally isomorphic to the kernel of the local-global map H 1(F, µ2) →Yv H 1(Fv, µ2), i.e. the group of square classes in F × that become the trivial square class in each F × v . (c) For every quadratic form q over F , factorization holds for the diamond of groups SO(q)(F•). (d) The local-global principle for isotropy holds for quadratic forms over F that are of dimension unequal to two, and for binary forms that do not lie in PF•. That is, if such a form q becomes isotropic over Fv for each v ∈ V1 ∪ V2, then q is isotropic. (e) If q is a regular quadratic form over F then iW(q) = min{iW(qv) v ∈ V1 ∪ V2} − ε, where ε = 1 if q represents a non-trivial class in PF• and otherwise ε = 0. Proof. Proof of part (a): Let PF• be the kernel of the diagonal map W(F ) → W(F1)×W(F2), and let α ∈ PF• ⊆ W(F ). Thus αFv = 0 for each v. Here α is the class of a quadratic form q such that qFv is hyperbolic; hence q is of even dimension 2n. Let d be the discriminant of q and let b = h1, −di. Since qFv has trivial discriminant for each v, it follows that bFv is hyperbolic for each v; i.e. d ∈ (F × v )2. Hence the form q′ = q ⊥ −b has trivial discriminant and is hyperbolic for each v. Thus q′ corresponds to a class in H 1(F, SO((n + 1)h)). Let H be the SO((n + 1)h)-torsor corresponding to q′. Since q′ Fv is hyperbolic for each v, we see that H(Fi) 6= ∅ for i = 1, 2. But by Theorem 2.13, since factorization holds for SO((n + 1)h)(F•), it follows that H(F ) is non-empty. Hence the torsor H is split, and so q′ = q ⊥ −b is hyperbolic. But this implies that q is equivalent to the binary Pfister form b. 26 For exactness on the right, suppose that we have Witt classes αi ∈ W(Fi) such that (α1)F0 = (α2)F0. We wish to show that there is a class α ∈ W(F ) such that αFi = αi for i = 1, 2. To begin, we choose representative forms qi over Fi with class αi of the same dimension n. Since these forms become Witt equivalent over F0, and since they have the same dimension, they necessarily become isometric over F0. But the category of quadratic forms of dimension n under isometry is equivalent to the category of O(n h1i)-torsors; so by Theorem 2.8, there is a quadratic form q over F such that qFi ∼= qi, as desired. Proof of part (b): Since elements of PF• are represented by binary forms, they lie in the fundamental ideal I(F ). So the exact sequence in part (a) restricts to an exact sequence 0 → PF• → I(F ) → I(F1) × I(F2) → I(F0). We claim that the induced map PF• → I(F )/I 2(F ) is injective. To see this, observe that if q is a quadratic form whose Witt class is in PF• ∩ I 2(F ), then q has trivial discriminant and even dimension 2n, and thus corresponds to a class α ∈ H 1(F, SO(nh)) with αFv trivial for each v. But since factorization holds for SO(nh)(F•), it follows from Theorem 2.13 that α is split and hence q is hyperbolic. Consequently, the above map is injective. Consider the composition PF• → I(F ) → I(F )/I 2(F ) →∼ F ×/(F ×)2 →∼ H 1(F, µ2), which is thus also injective. Its image is contained in the kernel of the map H 1(F, µ2) → Qv H 1(Fv, µ2), by the definition of PF• and the functoriality the isomorphism I(F )/I 2(F ) →∼ H 1(F, µ2). The reverse containment follows from the description of PF• in part (a), together with the fact that the map I(F ) → F ×/(F ×)2 takes the class of the quadratic form h1, −di to the square class of d. Hence we obtain the asserted isomorphism. Proof of part (c): Let n = dim q. Suppose that q′ is a quadratic form class with [q′] ∈ H 1(F, SO(q)) such that [q′]Fv is trivial in H 1(Fv, SO(q)) for each v. Then q ⊥ −q′ is a quadratic form over F of even dimension and trivial discriminant that is trivial over each Fv; and hence its Witt class lies in PF•. Since none of the nontrivial elements of PF• have trivial discriminant, it follows that q ⊥ −q′ is hyperbolic, and that q and q′ are isometric (being of the same dimension). Hence [q′] is trivial in H 1(F, SO(q)). This shows that the local-global principle holds for SO(q)-torsors. By Theorem 2.13, it follows that factorization holds for SO(q). Proof of part (d): We prove the contrapositive; i.e. if q is anisotropic then so is some qFv. This is clear if q is binary but not in the kernel PF• of the local-global map on Witt rings, since a binary form is hyperbolic if and only if it is isotropic. So now assume that the anisotropic form q is not binary. The group O(q) acts on the projective quadric hypersurface Q defined by q, and the action of O(q)(F0) is transitive on Q(F0) by the Witt Extension Theorem (see the proof of [HHK09, Theorem 4.2]). Since dim(q) > 2, it follows that Q is connected; and hence SO(q)(F0) also acts transitively on Q(F0). By part (c) above, factorization holds for SO(q)(F•). Hence Theorem 2.13 implies that the local-global principle holds for Q. If q is anisotropic, then Q(F ) is empty and thus some Q(Fv) must be empty; i.e. qFv is anisotropic. 27 Proof of part (e): By Witt decomposition, we are reduced to the case that q is anisotropic. The desired assertion is clear if the class of q is in PF•, so assume otherwise. Then some qFv is anisotropic by part (d). Hence iW(q) = min{iW(qv)} = 0, and thus ε = 0 as asserted. This abstract result can in particular be applied to the concrete situation of Section 3. We recall the standing hypotheses: T is a complete discrete valuation ring with fraction field K and residue field k; and bX is a projective normal T -curve with fraction field F and closed fiber X. For the remainder of this section on quadratic forms, we additionally assume that K (or equivalently, F ) has characteristic unequal to two. Theorem 4.1 yields local-global principles for quadratic forms in this context: Example 4.2. In the situation of Proposition 3.7, 3.9, or 3.10, write L for the field F , FW , or FP , respectively. Recall that we assume char(L) 6= 2. Theorem 4.1 applies because those three propositions say that patching holds for the given diamond, and because factorization holds for SO(nh)(F•) by Corollary 3.15 (since SO(nh) is a rational connected linear algebraic F -group by the Cayley parametrization; e.g. see [HHK09, Remark 4.1]). The local-global principles given in Example 4.2 can be carried over to the situation of points on the closed fiber. First we prove a lemma, with notation as above. Lemma 4.3. Let X0 be an irreducible component of X, with generic point η. Let U0 be a non-empty affine open subset of X0 that meets no other irreducible component of X, and let q be a quadratic form over FU0. If q becomes isotropic over Fη then q is isotropic over FU for some non-empty affine open subset U ⊆ U0. Proof. We may assume that q is a diagonal form ha1, . . . , ani, with each ai ∈ FU0 ⊂ Fη. By [HHK13, Corollary 3.3(a)], there exist elements bi ∈ F and ci ∈ F × i . So q is isometric over FU0 to the form q′ := hb1, . . . , bni that is defined over F . The projective quadric hypersurface Q over F that is defined by q′ has an Fη-point, since q′ is isotropic over Fη. Hence by [HHK15, Proposition 5.8], Q has an FU -point for some non-empty affine open subset U ⊆ X0. After shrinking U, we may assume that U ⊆ U0. Thus q′ U = qU , and q′ is isotropic over FU . Hence q is isotropic over FU . U0 such that ai = bic2 discrete valuation ring T of characteristic not two. Let S equal X, or a non-empty connected open proper subset W ⊂ X, or a non-empty connected proper subset V ⊂ X consisting of a union of irreducible components of X. In these three cases, let L respectively equal F or FW Proposition 4.4. Let bX be a projective normal curve with closed fiber X over a complete or FP , where in the third case we consider the model bY of F obtained by contracting V and where the point P ∈ bY is the image of V . Let q be a quadratic form over L. (a) If dim(q) 6= 2 and qFQ is isotropic for each point Q ∈ S, then q is isotropic. (b) If q is a regular quadratic form, then iW(q) ∈ {min(iW(qFQ)), min(iW(qFQ)) − 1}, where the minimum is taken over all Q ∈ S. The second case occurs precisely when q is Witt equivalent to an anisotropic binary Pfister form that becomes isotropic over each FQ. 28 (c) The kernel of the map π : W(L) →QQ∈S W(FQ) is equal to the kernel XS(L, µ2) of the local-global map H 1(L, µ2) →QQ∈S H 1(FQ, µ2) in Galois cohomology. Proof. We begin with the observation that if qFQ is isotropic for each point Q ∈ S, then there is a finite set P of closed points of S such that qFU is isotropic for each connected component U of S r P. To see this, note that for each irreducible component S0 of S, the form qFη is isotropic, where η is the generic point of S0. Hence q is isotropic over FU for some non-empty open subset U ⊆ S0, by Lemma 4.3. We may now take U to be the collection of these sets U (one for each irreducible component of S), and take P to be the complement in S of the union of the sets U ∈ U. This proves the observation. We now prove part (a). By the above observation, there are sets P and U as above such that qFξ is isotropic for each ξ ∈ P ∪ U. Consider Example 4.2 in the situation of Proposition 3.7, 3.9, or 3.10, if S is equal to X, W , or V respectively. By Theorem 4.1(d) in the context of this example, q is isotropic over L. For part (b), take the Witt decomposition q = qa ⊥ ih, where qa is anisotropic, h is a hyperbolic plane, and i ≥ 0. The assertion is trivial if q is itself hyperbolic, and so we may assume that qa is a non-trivial form. If qa remains anisotropic over FQ0 for some Q0 ∈ S, then iW(q) = i = iW(qQ0), and iW(qQ) ≥ i for all other Q ∈ S. Thus iW(q) = min(iW(qFQ)) in this case. The other case is that qa becomes isotropic over each FQ. Then by the above claim, qa becomes isotropic over Fξ for each ξ ∈ P ∪ U as above. So by Theorem 4.1(e) in the context of Example 4.2, qa is an anisotropic binary Pfister form that becomes isotropic (or equivalently, hyperbolic) over each Fξ, and hence over each FQ. Thus q is Witt equivalent to such a form and iW(q) = i = iW(qFQ) − 1 for all Q ∈ S. For part (c), observe that by Theorem 4.1(a,b) in the context of Example 4.2, the kernel of the map πP : W(L) →Qξ W(Fξ) is equal to the kernel XP(L, µ2) of the local-global map H 1(L, µ2) →Qξ H 1(Fξ, µ2), where ξ ranges over the elements of P ∪ W in each product. So it suffices to show that ker(π) = S ker(πP) and XS(L, µ2) = S XP(L, µ2), where in each all P. It therefore remains to show that ker(π) ⊆S ker(πP) and XS(L, µ2) ⊆S XP(L, µ2). case P ranges over the non-empty sets of closed points of S. For any choice of P (and hence of W), FU ⊂ FQ for all Q ∈ U ∈ W. Thus ker(πP) ⊆ ker(π) and XP(L, µ2) ⊆ XS(L, µ2) for We begin with the first of these inclusions. By Witt decomposition, every non-trivial class in ker(π) is represented by a non-trivial anisotropic form q. Such a q becomes hyperbolic and hence isotropic over each FQ. So by the observation at the beginning of the proof, q becomes isotropic over Fξ for every ξ ∈ P ∪ W, for some choice of P. Also, since q is not hyperbolic over F but becomes hyperbolic over each FQ, it follows from part (b) above that q is a binary form. Thus the isotropic forms qFξ are also binary and hence hyperbolic, and so the class of q lies in ker(πP) as desired. To prove the second inclusion, note that a non-trivial element of XS(L, µ2) is given by a quadratic field extension of L, of the form L[a1/2] for some non-square a ∈ L×. By definition of XS(L, µ2), the element a is a square in Lη = Fη for every generic point η of S. For each η, choose an irreducible connected open neighborhood U0 ⊂ S. By Corollary 3.8(b), U0. Thus b is a square in Fη; i.e. the µ2-torsor given by a = bc2 for some b ∈ F × and c ∈ bR× 29 F [b1/2] over F has an Fη-point. By [HHK15, Proposition 5.8], this torsor has an FU -point for some open neighborhood U ⊆ U0 of η; i.e. b and hence a is a square in FU . Let P be the complement of the union of the sets U as η varies. Then the given element of XS(L, µ2) lies in XP(L, µ2). In the above result, the first case (where S = X and L = F ) was previously shown in Theorem 9.3, Corollary 9.5, and Theorem 9.6 of [HHK15]; but here a uniform argument proves all three cases. Analogous local-global assertions have been also proven with respect to discrete valuations on F rather than with respect to points on X; see [CPS12, Theo- rem 3.1] and [HHK15, Theorem 9.11]. By combining Proposition 4.4 with the strategy used in [HHK15, Proposition 9.10], we obtain a local-global principle for isotropy over FP : Proposition 4.5. Let bX be a normal projective T -curve, let P be a closed point of bX, and let q be a quadratic form on FP of dimension 6= 2. Assume char(k) 6= 2. Then q is isotropic over FP if and only if it is isotropic over (FP )v for every discrete valuation v on FP . Even more is true: q is isotropic over FP provided that it is isotropic over (FP )v for each discrete valuation v on FP whose restriction to F is induced by a codimension one point on a regular projective model of F over T . Proof. The forward direction of the first assertion is trivial. For the reverse direction, con- sider a quadratic form q on FP that is isotropic on (FP )v for every discrete valuation v on By resolution of singularities for surfaces ([Abh69], [Lip75]) and Weierstrass Preparation FP . We may assume that q is a diagonal form ha1, . . . , ani, with ai ∈ bRP . ([HHK13], Corollary 3.7), there is a birational projective morphism π : bX ′ → bX such that bX ′ is regular, and such that on the pullback π′ : bX ′ P → Spec(bRP ) of π with respect to Spec(bRP ) → bX, the support of q is a normal crossing divisor at every point of π′−1(P ) ⊂ bX ′ P (which we may identify with V := π−1(P ) ⊂ bX ′; here the support of q is defined to be the union of the supports of the divisors of the elements ai.) By the third case of Proposition 4.4(a), in order to show that q is isotropic over FP it suffices to show that q is isotropic over FQ for every point Q ∈ V . First note that by [HHK15, Proposition 7.5], for any Q ∈ V and any discrete valuation v on FQ, the restriction of v to F is a (non-trivial) discrete valuation. Since F ⊆ FP ⊆ FQ, it follows that the same holds for the restriction of v to FP . Consider a closed point Q of V . By the condition on the support of q at Q, there exists a generating set {x, y} for the maximal ideal of bRQ whose support contains that of q in Spec(bRQ). Let v be the y-adic valuation on FQ, and let v0 = vFP . Thus q is isotropic over the completion (FP )v0, by the previous paragraph and by the hypothesis of this direction of the proposition. Hence q is also isotropic over (FQ)v, which contains (FP )v0. Since (FQ)v has residue characteristic unequal to two, it follows from [HHK15, Lemma 9.9] that q is isotopic over FQ. The other case is that Q is a generic point of V . Thus FQ is a complete discretely valued field, say with valuation v. Again, the restriction v0 of v to FP is a discrete valuation such that q is isotropic over (FP )v0. Hence q is also isotropic over FQ, which contains (FP )v0. This completes the proof of the reverse implication. 30 For the last assertion, note that we may assume in the above argument that the model to check that the valuations used in the above argument are induced by codimension one If Q is a generic point of V , then this condition is trivial, since Q is itself a codimension on that model. So consider the case that Q is a closed point of V , and take the valuation v0 = vFP considered in the argument above. By the above condition on branches, the hypothesis of [HHK13, Theorem 3.1(c)] is automatically satisfied; and so there exist b ∈ F bX ′ has the property that distinct branches of the closed fiber X ′ at any closed point must lie on distinct irreducible components of X ′. With respect to this choice of model bX ′, it suffices points on bX ′. one point on bX ′, and the restriction of v0 to F is the valuation associated to that point and c ∈ bR× Q such that y = bc. Here b = yc−1 ∈ bRQ, and {x, b} is a generating set for the maximal ideal of bRQ. So there is a unique irreducible component D of the zero locus of b on bX ′ that passes through Q. The y-adic valuation v on FQ is equal to the b-adic valuation on bX ′, as desired. FQ, and the valuation v0 on FP thus restricts to the b-adic valuation on F . That is, v0F is the discrete valuation associated to the generic point of D, which is of codimension one on Lemma 4.6. Let E be the fraction field of a two-dimensional Noetherian complete local domain R. Then E is isomorphic to a finite separable extension of a field of the form FP . Moreover if R is regular or equicharacteristic, then E is itself of the form FP . Proof. If R is regular, then by [Coh46, Theorem 15] it is of the form T [[x]] for some complete discrete valuation ring T . Thus E = FP with respect to a point on the projective T -line. In the general case, by [Coh46, Theorem 16], R is a finite extension of a two-dimensional regular complete local domain having residue field k. So by the previous paragraph, E is a finite extension of a field of the form FP . This extension is automatically separable if char(E) = 0; and by [GaOr08, Théorème 7.1], it can be chosen to be separable if char(E) > 0. Hence E is isomorphic to a finite separable extension of FP . Finally, if R is equicharacteristic, then by the above it is a finite generically separable extension of some T [[x]], where T = k[[t]] since it is equicharacteristic. So E is a finite separable extension of k((t, x)), and thus of the form FP by [HHK13], Lemma 3.8. Proposition 4.5 and Lemma 4.6 then yield the following strengthening of [Hu12, Theo- rem 1.2]): Corollary 4.7. Let E be the fraction field of a regular or equicharacteristic two-dimensional complete local ring whose residue field k has characteristic unequal to two. Then a quadratic form q over E of dimension 6= 2 is isotropic if and only if it becomes isotropic over Ev for every discrete valuation v on E. Corollary 4.8. In the situation of Proposition 3.7 (resp. 3.9 or 3.10), with char(K) 6= 2, let L be the field F (resp. FW or FP ) and let S be the set X (resp. W or V ). (a) Then u(L) ≤ max ξ∈P∪W u(Fξ) ≤ sup Q∈S u(FQ). 31 (b) If the residue field k of T has characteristic unequal to two, then u(L) = max ξ∈P∪W u(Fξ) = u(FQ) = sup v∈Ω u(Lv), where Ω is the set of discrete valuations on L whose restriction sup Q∈S to F is a discrete valuation that is induced by a codimension one point on a regular model of F . (c) If k is perfect of characteristic two, and char(K) = 0, then the four quantities u(L), maxξ∈P∪W u(Fξ), supQ∈S u(FQ), supv∈Ω u(Lv) are each less than or equal to 8. Proof. For ξ ∈ P ∪ W, the field Fξ is not quadratically closed, since the integrally closed ring bRξ is not. Thus u(Fξ) ≥ 2 by [Lam05], Chapter XI, Example 6.2(1). If q is a quadratic form over L having dimension greater than maxξ∈P∪W u(Fξ), then qξ is isotropic for all ξ ∈ P ∪ W. Hence q is isotropic over L by Theorem 4.1(d) in the context of Example 4.2, using that dim(q) > 2. This shows that u(L) ≤ maxξ∈P∪W u(Fξ). Next, if U ∈ W and q is a quadratic form over FU of dimension greater than supQ∈S u(FQ), then qFQ is isotropic for every point Q of U. Hence q is isotropic over FU by Proposition 4.4(a) for FU , using that dim(q) > 2. Thus maxξ∈P∪W u(Fξ) ≤ supQ∈S u(FQ). This proves part (a). Next, we show part (b), assuming that char(k) 6= 2. By part (a), it suffices to show the two inequalities supQ∈S u(FQ) ≤ supv∈Ω u(Lv) ≤ u(L). Proposition 4.4(a) to FQ for every Q ∈ S at which π is not an isomorphism, we see that For the first of these inequalities, we may consider just closed points Q. Let π : bY → bX be a birational projective morphism such that bY is smooth, and let Σ = π−1(S). Applying supQ∈S u(FQ) ≤ supQ∈Σ u(FQ). So after replacing bX by bY , we may assume that bX is regular. Next, since char(k) 6= 2, there is a split cover ω : bX ′ → bX, say with function field extension F ′/F and closed fiber X ′, such that for each Q′ ∈ S′ := ω−1(S) and each a ∈ F ′ Q′, there exist b ∈ F ′ and c ∈ F ′× Q′ such that a = bc2. (See Corollaries 3.3(c) and 3.7 of [HHK13], the latter applied to the set of non-unibranched points of S.) By Proposition 5.1 of [HHK15], the set of Q′, for Q′ ∈ X ′, is the same as the set of isomorphism classes isomorphism classes of fields F ′ of fields FQ, for Q ∈ X. Also, since bX is regular, the analogous assertion is true for the fields Fv, by Proposition 7.6 of [HHK15]. So we may replace bX by bX ′, and therefore assume that bX satisfies the above factorization condition on elements a ∈ FQ. Now let q be a quadratic form over FQ for some Q ∈ S, and assume that n := dim(q) > u(Lv) for all v ∈ Ω. To prove the first inequality we wish to show that q is isotropic over FQ. We may assume that q = ha1, . . . , ani with ai ∈ FQ. By the above condition, we may write ai = bic2 Q . Replacing q by the FQ-equivalent form hb1, . . . , bni, we may assume that q is defined over F . Now let w be any discrete valuation on FQ whose restriction to F is a discrete valuation induced by a codimension one point on a regular projective model of F . Thus v := wL ∈ Ω, and Lv is contained in (FQ)w. But q is isotropic over Lv by the dimension assumption on q, since v ∈ Ω and q is a quadratic form over Lv. Thus q is isotropic over each (FQ)w. By Proposition 4.5, it follows that q is isotropic over FQ. This completes the proof of the first inequality. i with bi ∈ F and c ∈ F × For the second inequality, let v be a discrete valuation on L. Since char(k) 6= 2, the residue field κv of Lv also has characteristic unequal to two. So u(Lv) = 2u(κv) for each 32 v ∈ Ω by Springer's theorem [Spr55, Proposition 2], and 2u(κv) ≤ u(L) by the first part of [HHK09, Lemma 4.9]. Hence u(Lv) ≤ u(L), concluding the proof of part (b). For part (c), we first show under the given hypotheses that u(FQ) ≤ 8 for all Q ∈ S. Let q be a quadratic form of dimension 9 over FQ; we wish to show that q is isotropic. Since a normal crossing at Q, [PS14, Proposition 4.6] asserts that q is isotropic. More generally, We may assume that each ai is non-zero. Let D be the union of the supports of the divisors char(L) = char(K) = 0, we may assume that q is a diagonal form ha1, . . . , a9i, with ai ∈ bRP . (a1), . . . , (a9), (2) on Spec(bRQ). In the special case that bX is regular at Q and D has at most let π : bX ′ → bX be a blow-up such that bX ′ is regular and D has only normal crossings on Spec(bRQ) × bX bX ′, the corresponding blow-up of Spec(bRQ). Then q is isotropic over FQ′ for every Q′ ∈ π−1(Q), by the special case just shown. By Proposition 4.4(a), q is isotropic over FQ, as desired. Thus u(FQ) ≤ 8. So by part (a), u(L) ≤ max ξ∈P∪W u(FQ) ≤ 8 in this case. To complete the proof u(Fξ) ≤ sup Q∈S of part (c), it suffices to show that u(Lv) ≤ 8 for each v ∈ Ω. If the residue characteristic of v is zero (and thus not two), then u(Lv) ≤ u(L) by the same argument as at the end of the proof of part (b); and so u(Lv) ≤ 8. Otherwise, v is the valuation associated to the generic point of a component of the special fiber of some model of L, and u(Lv) ≤ 8 as in the first part of the proof of [PS14, Theorem 4.7]. Remark 4.9. (a) Corollary 4.8(b) remains valid if one takes the supremum over all the discrete valuations on L, since the above argument that u(Lv) ≤ u(L) does not use that v ∈ Ω. (b) Corollary 4.8(b) is related to Theorem 4.9 in [Hu15]. The proof in [Hu15] used that the function field there was assumed to be purely transcendental. Note that char(k) 6= 2 in that assertion, by a standing hypothesis there. (c) Part (c) of the corollary extends Theorem 4.7 of [PS14], which asserted that u(F ) ≤ 8 under these hypotheses. The next result generalizes an assertion given in [HHK13, Theorem 4.1] concerning the value of u(FU ) for an open subset U of X. Here, as in [HHK09] and [HHK13], given a field E, us(E) denotes the smallest n such that u(L) ≤ 2in for every finitely generated field extension L/E of transcendence degree i ≤ 1. The proof is as for [HHK13, Theorem 4.1], but using Corollary 3.8 in place of the less general [HHK13, Corollary 3.3(a)]. As in that result, we need to assume that the residue field k of the complete discrete valuation ring T has characteristic unequal to two (not just that its fraction field K has characteristic 6= 2). (a) Then u(FU ) ≤ 4us(k). nected open subset of the closed fiber X. Assume that char(k) 6= 2. Proposition 4.10. Let bX be a normal projective T -curve, and let U be a non-empty con- (b) Let eX be the normalization of X, and let eQ ∈ eX be a closed point lying over some point Q ∈ U. Then u(FU ) ≥ 4u(κ(eQ)). 33 See also [Cuo13, Corollary 3.7]. Theorem 4.11. Let E be one of the following: (i) the fraction field of a two-dimensional Noetherian complete local domain R that is regular or equicharacteristic; or (ii) a finite separable extension of the fraction field of the t-adic completion of T [x], where T is a complete discrete valuation ring with uniformizer t. Assume that the residue field k of R (resp. T ) does not have characteristic two. (a) Then u(E) ≤ 4us(k). (b) If u(k) = us(k), and if u(k′) = u(k) for every finite extension k′ of k, then u(E) = 4u(k). Proof. In case (i), E is of the form FP by Lemma 4.6. The assertion then follows in this case from [HHK13], Theorem 4.1, by choosing a finite set of points P on the closed fiber X of the In case (ii), E is a finite separable extension of FU , where U = A1 T . By Proposition 3.5(b), E is isomorphic to a field F ′ k, viewed as an open subset of the closed fiber of P1 U ′ for some finite extension F ′ of F , where F is the fraction field of T [x], and for some non-empty model bX, such that P contains P and the points where distinct components of X meet. connected open subset U ′ ⊂ X ′. (Here X ′ is the closed fiber of a projective normal model bX ′ that u(κ(eQ)) = u(k) by hypothesis, for any closed point Q ∈ U ′. of F ′.) By Proposition 4.10, u(E) = u(F ′ U ′) ≤ 4us(k), proving part (a). For part (b), u(E) ≤ 4us(k) = 4u(k) by part (a); and the reverse inequality follows from Proposition 4.10(b), using Example 4.12. Theorem 4.11 applies in particular to the broad class of fields k that satisfy Leep's An(2) property. Recall that k is called an An(2) field if for every r > 0, every system of r homogeneous forms of degree two over k in more than r · 2n variables has a non-trivial zero over a finite extension of k having odd degree. (see [Lee13, Section 2]). Every Cn field is an An(2) field ([Sha72], Lemma IV.3.7), but not conversely. Although p-adic fields are not Cn for any n, they are A2(2) field for all primes p, including p = 2 (see [Lee13], Corollary 2.7). Moreover if k is an An(2) field then so is every finite extension of k ([Lee13], Theorem 2.5); and the fields k(z) and k((z)) are An+1(2) fields ([Lee13], Theorem 2.3). Since u(k) ≤ 2n for an An(2) field ([Lee13], Theorem 2.2), it follows from the above properties that us(k) ≤ 2n, and thus u(E) ≤ 2n+2 in the notation of the corollary, provided char(k) 6= 2. In the special case that u(k′) = 2n for every finite extension k′/k, it follows from [Lee13, Lemma 3.2] that u(k) = us(k) = 2n. Thus the hypotheses of Theorem 4.11(b) hold. As a consequence, we obtain Theorem 1.3: Proof of Theorem 1.3. By Example 4.12, the fields k are respectively A0(2), A1(2), A2(2), A3(2), as are their finite extensions; and moreover the hypotheses of Theorem 4.11(b) hold (using also [Lee13, Theorem 3.4] in the last two cases). We conclude by that theorem. As pointed out to us by David Leep, the case of the theorem for u(cid:0)k((x, t))(cid:1), where k = Qp or Qp(z) or Qp((z)), can be deduced directly from results in [Lee13]. Namely, if k is an An(2)-field of characteristic unequal to two, then k((t))(x) is an An+2(2) field by [Lee13, 34 Theorem 2.3] and so u(cid:0)k((t))(x)(cid:1) ≤ 2n+2. If u(k) = 2n (as in the case of k = Qp or Qp(z) or Qp((z))), it then follows that u(cid:0)k((x, t))(cid:1) = 2n+2 by [Lee13, Proposition 5.1]. Theorem 4.11 also provides explicit values for the u-invariant in mixed characteristic, when the residue characteristic is odd: the maximal unramified extension of Zp. p [[x]] or p [x]). Let E be the fraction field of a finite extension S of R; and if R = Zp[[x]] or the Corollary 4.13. Let p be an odd prime, and Zur p Let R be Zp[[x]] or the p-adic completion of Zp[x] (resp. the p-adic completion of Zur of Zur p-adic completion of Zur p [[x]], assume that S is regular. Then u(E) = 8 (resp. 4). Proof. In the first two cases, let T = Zp and apply Theorem 4.11, using that the hypotheses of part (b) hold with us(Fp) = 2, by [HHK09, Theorem 4.10]. In the other cases, let T be the p-adic completion of Zur p ; this is the ring of Witt vectors of Fp. Again the hypotheses of Theorem 4.11(b) hold, this time with us(Fp) = 1. In the case of mixed characteristic with residue characteristic two, we obtain the following, by combining [PS14] with Theorem 4.11: Corollary 4.14. Let k be a complete discretely valued field of characteristic zero whose residue field κ is perfect of characteristic two. If E is a finite extension of k((x, t)) or of the fraction field of k[x][[t]], then u(E) ≤ 16. Proof. Every finite extension λ of κ is perfect, so u(λ) ≤ 2 by [MMW91, Corollary 1]. Thus u(ℓ) ≤ 4 for every finite extension ℓ of k, by a theorem of Springer ([Spr55, Proposition 2]). By [PS14, Theorem 4], u(F ) ≤ 8 for every finitely generated extension F of k of transcen- dence degree one. Thus us(k) ≤ 4. We conclude by Theorem 4.11, where separability holds since char(k) = 0. 4.2 Applications to central simple algebras Finally, we turn to applications of our results to central simple algebras, especially concerning the period and index of elements of the Brauer group of fields of the sort considered in Section 3, and their finite extensions. In particular, for a finite separable extension E of a field of the form FP or FU , we find an integer d such that ind(α) divides per(α)d for α ∈ Br(E). See Theorems 4.21 and 4.22, as well as the associated corollaries, for the precise statements, which strengthen and extend results in [HHK13, Section 4]. For example, for two- dimensional p-adic cases, we obtain a sharp bound for the period-index bound d regardless of the period of α; see Theorems 4.23 and 1.2. As in Section 4.1, we begin with an abstract local-global result (Theorem 4.15) that ap- plies to diamonds that satisfy patching. But unlike the analogous Theorem 4.1, Theorem 4.15 does not require a factorization hypothesis. This makes it more applicable, permitting its use in conjunction with Corollary 3.12, which in turn makes possible the period-index appli- cations mentioned above for finite separable extensions of fields FP and FU . Theorem 4.15 also yields local-global results about the value of the period-index bound for the fields FP and FU ; see Example 4.16 and Corollary 4.17. 35 Azumaya algebra over E is the same as a product of central simple Fv-algebras. Below we use that if E is a product of fields Fv, then Br(E) = Q Br(Fv) because an and each Fi a finite direct product of fields. Write Fi =Qv∈Vi Theorem 4.15. Suppose that F• = (F ≤ F1, F2 ≤ F0) is a diamond of rings with F a field Fv where each Fv is a field. Assume moreover that patching holds for F•. Then: (a) We have a short exact sequence 0 → Br(F ) → Br(F1) × Br(F2) → Br(F0), where the map on the right is given by taking the difference of the restrictions of the two Brauer classes. (b) For a class α ∈ Br(F ), we have ind(α) = lcm{ind(αv) v ∈ V1 ∪ V2}. Proof. For any field L, the natural map GLn(L) → PGLn(L) is surjective, by Hilbert's Theorem 90 and the long exact cohomology sequence associated to the short exact sequence of algebraic groups 1 → Gm → GLn → PGLn → 1. Now factorization for GLn holds for F•, since F• has the patching property (see Theorem 2.8(a)). The above surjectivity then implies that factorization for PGLn also holds for F•. Theorem 2.13 then in turn implies Recall that H 1(F, PGLn) classifies isomorphism classes of central simple F -algebras of degree n ([KMRT98, p. 396]). So if A is a central simple F -algebra such that AFv is split for that the map of pointed sets H 1(F, PGLn) →Qv H 1(Fv, PGLn) has trivial kernel. each v, then A itself is split. Thus the homomorphism Br(F ) →Qv Br(Fv) is injective. Now, suppose that we have classes αi ∈ Br(Fi) = Qv∈Vi algebras Av over Fv of degree n such that the class of Ai =Qv∈Vi Av in Br(Fi) =Qv∈Vi Br(Fv) for i = 1, 2 such that (α1)F0 = (α2)F0. We wish to show that there is an α ∈ Br(F ) with αFi = αi. Choose a positive integer n that is divisible by ind(αv) for each v. We may then choose central simple Br(Fv) is αi. Since (α1)F0 = (α2)F0, the algebras (A1)F0 and (A2)F0 are Brauer equivalent, and thus isomorphic, being of the same degree. Using patching for central simple algebras (see ∼= Ai, compatibly. This Example 2.7), there is a central simple F -algebra A such that AFi gives exactness of the given sequence, proving part (a). We now turn to part (b). By [Pie82, Proposition 13.4(iv)], ind(αFv) divides ind(α). It thus suffices to show that if each ind(αFv) divides an integer i then so does ind(α). Choose a central simple algebra A with Brauer class α and with some degree n > i. Let SBi be the i-th generalized Severi-Brauer variety, parametrizing ni-dimensional right ideals of A (see [VdB88, p. 334], [See99, Theorem 3.6], or [HHK09, p. 255]). For any field extension L/F , the group PGL1(A)(L) acts transitively on the L-points of SBi, via [KMRT98, Proposition 1.12, Definition 1.9] (see also [HHK09, p. 255]). Also, ind(AL) divides i if and only if SBi(L) 6= ∅, by [KMRT98, Proposition 1.17]. In particular, SBi(F1) and SBi(F2) are non-empty. We claim that factorization holds for PGL1(A)(F•). Assuming this for the moment, it follows from Theorem 2.13 that SBi(F ) 6= ∅. Therefore ind(α) divides i, as desired. To complete the proof of part (b), it remains to prove the above claim. By Theorem 2.13, it suffices to prove that the map H 1(F, PGL1(A)) →Qv H 1(Fv, PGL1(A)) has trivial kernel. Let β be in the kernel of this map. Now for any field E, the cohomology set H 1(E, PGL1(A)) parametrizes the set of isomorphism classes of central simple E-algebras of degree n, with the 36 trivial element corresponding to the class of A (see [KMRT98, Proposition 29.1 and p. 396]). Let B be a central simple F -algebra B of degree n whose isomorphism class corresponds to β. Thus B ⊗ Aop induces the trivial element in Br(Fv) for all v; and hence B ⊗ Aop ∈ Br(F ) is itself trivial by part (a). Equivalently, β is trivial in H 1(F, PGL1(A)). This proves the claim and hence the result. Example 4.16. Under the hypotheses of Proposition 3.7 (resp. Proposition 3.9 or Proposi- tion 3.10 or Corollary 3.12), the conclusions of Theorem 4.15 hold, since its hypotheses hold by those propositions. In particular, a central simple algebra A over F (resp. over FW or FP or E) is split if and only if it is split over each Fξ for ξ ∈ P ∪ W. Moreover the index of A is the least common multiple of the indices of the algebras AFξ. See also [HH10, Theorem 7.2], [HHK09, Theorem 5.1], and [RS13, Theorem 2] for related results. Corollary 4.17. Let L be the field F (resp. FW or FP or E) in the situation of Proposi- tion 3.7 (resp. Proposition 3.9 or Proposition 3.10 or Corollary 3.12). Let d be a positive integer and let α ∈ Br(L). For ξ ∈ P ∪ W, let αξ be the image of α in Br(Fξ). If ind(αξ) divides per(αξ)d for all ξ ∈ P ∪ W, then ind(α) divides per(α)d. Proof. Since per(αξ) divides per(α) for each ξ ∈ P ∪ W, we have that ind(αξ) divides per(α)d for each ξ. Since ind(α) = lcm(cid:0)ind(αξ)(cid:1)ξ by Example 4.16, it follows that ind(α) divides per(α)d. We now turn to results that provide period-index bounds for finite separable extensions of fields of the form FP and FW , in terms of such bounds for k. Even in the special case of the fields FP and FW themselves, the results strengthen [HHK09, Corollary 5.10] and [HHK13, Theorem 4.6] by improving the exponent on the period and also considering more general open subsets. First we prove some lemmas. Proof. Let β ∈ nBr(F ) be as above. Thus m := per(β) divides n. By [Sal97], Proposition 1.2, of degree m and hence have index dividing m. Thus γ1, γ2 have periods dividing m and ind(β) ind(β0) ind(γ1) ind(γ2) per(β0)dm2 md+2 = per(β)d+2. residue field k and fraction field F . Let n, d be positive integers such that µn ⊂ k and such that ind α(per α)d for every n-torsion Brauer class α ∈ nBr(k). Let β ∈ nBr(F ) be a Brauer Lemma 4.18. Suppose that bR is a excellent complete regular local ring of dimension 2 with class ramified only along a regular sequence for bR. Then ind β (per β)d+2. we may write β = β0 + γ1 + γ2, where β0 ∈ Br(bR), where γi are the classes of cyclic algebras hence the same is true of β0. By [Mil80, Corollary IV.2.13], we may identify Br(bR) = Br(k) via specialization. Moreover the index of the class β0 ∈ Br(bR) ⊂ Br(F ) divides that of of k and of bR, and hence of étale splittings of associated central simple algebras. Thus Lemma 4.19. Suppose bR is a 2-dimensional excellent ring with fraction field F , t ∈ bR, and bR is complete with respect to the t-adic topology. Suppose that bR/tbR ∼= k[U] is the α ∈ mBr(k[U]). If β ∈ mBr(F ) is ramified only at the support of tbR, then ind β (per β)d+1. coordinate ring of a regular affine k-curve U with char(k) 6 m, and that ind α (per α)d for all its image in Br(k), since specialization induces a natural bijection between étale extensions 37 Proof. The ramification of β defines a finite connected cover V → U of curves of degree n := per(β), together with a generator σ of its cyclic Galois group Cn. Applying [Sal08, the cover V → U is actually unramified and hence étale. By [AGV72, VII.5.5], specialization Let L be the fraction field of S, and consider the cyclic algebra C = (L/F, σ, t). By [Sal99, Lemma 10.2], C and β define the same cyclic cover V → U and the same Galois generator σ. Theorem 1.1] at each closed point of U, and using that β is ramified only at tbR, it follows that induces an equivalence of categories between the étale covers of U and of Spec(bR); and hence there is an étale algebra S/bR that lifts V → U and which is Galois with generator σ lifting σ. Thus the Brauer class β − [C] ∈ Br(F ) is unramified over bR; i.e. it lies in Br(bR), and is represented by an Azumaya algebra B over bR. Now per([C]) ind([C]) n = per(β), and so By reducing modulo (t), the bR-algebra B induces an Azumaya algebra B0 over k[U], hence SBi associated to B over bR. Its fiber (SBi)0 modulo (t) is the i-th generalized Severi-Brauer variety associated to B0 over k[U]. Since the index of B0 over k(U) is i, there is a k(U)-point on (SBi)0, by the key property of generalized Severi-Brauer varieties (recalled in the proof of Theorem 4.15 above). Now (SBi)0 is smooth and projective over k[U], and U is a regular curve; so the valuative criterion for properness implies that this k(U)-point extends to a k[U]- in turn a class in the Brauer group of k(U). We claim that ind(B) divides i := ind(B0), over F and k(U) respectively. To see this, consider the i-th generalized Severi-Brauer variety per(B) = per(β − [C]) per(β). generic point of the image of this section is an F -point of SBi. Thus SBi(F ) is non-empty, and so the index of [B] ∈ Br(F ) divides i, proving the claim. point, viz. a section of (SBi)0 → U = Spec(k[U]) = Spec(bR/tbR). By Lemma 4.5 of [HHK09] and the comment after that, this section over U lifts to a section of SBi → Spec(bR). The Now per(B0) per(B), since Br(bR) → Br(k[U]) is a group homomorphism taking [B] to [B0]. Thus per(B0) per(β) = n m. The above claim and the hypothesis on Br(k[U]) then yield that ind(B) ind(B0) (per B0)d (per B)d. But β = [B] + [C]. Hence we have that ind(β) ind(B) ind(C) (per B)d per(β) (per β)d+1, as asserted. Lemma 4.20. For a (general) field L and an integer n, the following are equivalent: 1. For every finite field extension L′/L, and α ∈ nBr(L′), we have ind(α) per(α)d. 2. For every prime q dividing n, every finite field extension L′/L and every Brauer class α ∈ Br(L′) of period q, we have ind(α)qd. Moreover if char(L) does not divide n then the same assertion holds with respect to the class of finite separable extensions L′/L in conditions 1 and 2. Proof. The forward implication is trivial. For (2) ⇒ (1), by considering primary parts we may assume that per(α) is a prime power qr. The implication is then given by [PS14, Lemma 1.1]. In the case that char(L) does not divide n, the corresponding assertion for sep- arable extensions L′/L holds because the proof of [PS14, Lemma 1.1] involves only extensions of q-power degree. 38 Following [Lie11] and [HHK09], we define the Brauer dimension of a field k away from a prime p as follows: The value is 0 if the absolute Galois group of k is a pro-p group (e.g. if k is separably closed). Otherwise, it is the infimum of the positive integers d such that for every finite generated field extension E/k of transcendence degree i ≤ 1, every α ∈ Br(E) of period prime to p satisfies ind α (per α)d+i−1. (We note that the term "Brauer dimension" was used in somewhat different senses in the manuscripts [ABGV11] and [PS14].) Theorem 4.21. Let T be a complete discrete valuation ring whose residue field k has Brauer dimension d away from p := char(k). Let bX be a normal projective T -curve with function field F and closed fiber X. Let ξ be either a closed point P ∈ X or a connected Zariski open subset W ⊂ X, and let E be a finite separable extension of Fξ. Then ind(α) per(α)d+1 for all α in Br(E) of period not divisible by p. Proof. By hypothesis, per(α) is not divisible by char(k) and hence also not by the charac- teristic of K, the fraction field of T . By Lemma 4.20, we may assume that per(α) is a prime number q 6= char(K), char(k). Since the degree [K(ζq) : K] is prime to q, we may also assume that K, k each contain µq. Namely, let K ′ = K(ζq), k′ = k(ζq), and E′ := E(ζq). Then the period of the induced element α′ ∈ Br(E′) is equal to the period of α ∈ Br(E), by [Pie82, Proposition 14.4.b(v)]. Also, the index of α (which is a power of q) divides [E′ : E] ind(α′) by [Pie82, Proposition 13.4(v)]; and hence it divides the index of α′ since [E′ : E] is relatively prime to q. So we may henceforth assume that K, k each contain µq. In the case that ξ = W , we may assume that W is affine, since FW = FW ′ for some connected affine open set W ′ on a normal model of T (see Remark 3.2(b)). In both cases Observe that the assertion holds in the special case that π is an isomorphism. Namely, if ξ = P , then ind α(per α)d+1 for all α ∈ qBr(E) by Lemma 4.18, since ind γ(per γ)d−1 for all γ ∈ qBr(k eP ) by the assumption on the Brauer dimension of k. (Here k eP is the residue ξ = P, W , let S be the integral closure of bRξ in E, let D be the ramification divisor of α on Spec(S), and let eξ be the inverse image of ξ under Spec(S) → Spec(bRξ). Applying Corollary 3.12, we obtain a birational projective morphism π : bV → Spec(S) and a non- empty set P of closed points of V := π−1(eξ) satisfying the four conditions there. field at eP .) Similarly, if ξ = W , then ind α(per α)d+1 for α ∈ qBr(E) by Lemma 4.19. fiber U ⊂ Spec(bRU ) for each U ∈ W. So we may now assume that π is not an isomorphism, and therefore that the patching assertion in the last part of Corollary 3.12 holds. Let D′, W, B be as in that result. By properties (ii) and (iii) of Corollary 3.12, αU ∈ Br(FU ) is unramified away from the closed Now ind αQ(per αQ)d+1 in Br(FQ) for each Q ∈ P by Lemma 4.18, and therefore ind αQ(per α)d+1, since per αQ per α. Similarly, ind αU (per α)d+1 in Br(FU ) for all U ∈ W by Lemma 4.19. But the index of α ∈ Br(E) is the least common multiple of the indices of all the induced elements αQ, αU , for Q ∈ P and U ∈ W, by Example 4.16 in the context of Corollary 3.12. So the desired conclusion follows. Above, we restricted attention to elements of the Brauer group whose period is prime to the residue characteristic p. But in [PS14], a result was shown about elements whose 39 period is equal to p. Namely, suppose that char(K) = 0 and char(k) = p > 0. If F is a finitely generated K-algebra of transcendence degree one, and if α ∈ Br(F ) has period p, then ind(α) divides p2n+2, where n is the p-rank of the residue field k of T . (See [PS14, Theorem 3.6]. Recall that the p-rank, or imperfect exponent, of a field k of characteristic p is the integer n such that [k : kp] = pn.) Combining the ideas there with the ideas above, we obtain the following: Theorem 4.22. In the situation of Theorem 4.21, suppose that T has mixed characteristic (0, p). If the period of α ∈ Br(E) is a power of p, then ind(α) divides (per α)2n+2, where n is the p-rank of k. Proof. Case I : per(α) = p. As in the proof of Theorem 4.21, we may assume that the fraction field K (though not the residue field k) contains a primitive p-th root of unity. As in the proof of Theorem 4.21, finite set P ⊂ V . we obtain a birational projective morphism π : bV → Spec(S) and an associated non-empty In the case that π is not an isomorphism, we proceed as in the proof of Theorem 4.21 but use [PS14, Proposition 3.5 and Theorem 2.4] instead of using Lemmas 4.18 and 4.19. Namely, as before we obtain a set W consisting of the components of V r P. If U ∈ W, then U is irreducible and we may consider its generic point η. The p-rank of the field Fη is n + 1. Applying [PS14, Theorem 2.4] to Fη, we thus obtain that ind(αFη) divides p2n+2. By [HHK15, Proposition 5.8] and [KMRT98, Proposition 1.17], after shrinking U (and correspondingly enlarging P), we have that ind(αFU ) = ind(αFη), which divides p2n+2. Meanwhile, by [PS14, Proposition 3.5], ind(αFP ) divides p2n+2 for every P ∈ P. As in the proof of Theorem 4.21, we conclude via Example 4.16. In the case that π is an isomorphism, we similarly use those two results in [PS14] instead of Lemmas 4.18 and 4.19. In the case that ξ = W , we apply [PS14, Theorem 2.4] at each generic point of W , and as above obtain a finite collection of points and open sets that partition W . We then conclude via Example 4.16 as in the above case. Case II : General case. Let E′ be a finite extension of E. Recall that the p-rank of E′ is also equal to n, i.e. [E′ : E′p] = [E : Ep], because [E′ : E] = [E′p : Ep] via the Frobenius isomorphism. Thus Case I applies to elements of Br(E′) having period p. The result now follows from Lemma 4.20 applied to the field E and the integer per(α). Theorem 4.23. Let E be one of the following: (i) the fraction field of a two-dimensional Noetherian complete local domain R; or (ii) a finite separable extension of the fraction field of the t-adic completion of T [x], where T is a complete discrete valuation ring with uniformizer t. Assume that the residue field k of R (resp. T ) has Brauer dimension d away from p := char(k), and has p-rank n. Let α ∈ Br(E). Then ind(α) per(α)d+1 if p does not divide per(α); and ind(α) per(α)max(d+1,2n+2) with no restriction on per(α) if char(E) = 0. 40 Proof. In case (i), Lemma 4.6 says that E is a finite separable extension of a field of the form FP . In case (ii), E is a finite separable extension of FU , where U = A1 T . So in both cases, E is of the form considered in Theorems 4.21 and 4.22. k ⊂ P1 Those two theorems therefore yield this result if the period of α is either prime to p or a power of p. The general case follows by decomposing α into its primary parts. For example, taking T = k[[t]], this theorem applies to finite separable extensions of the fraction field of k[x][[t]], as well as of k((x, t)). This strengthens [HHK13, Corollary 4.7]. Corollary 4.24. In the situation of Theorem 4.23, suppose that k is finite (resp. algebraically closed). If char(E) does not divide per(α) then ind(α) divides per(α)2. Moreover ind(α) = per(α) in the algebraically closed case if char(k) does not divide per(α). Proof. If k is algebraically closed, then d = 0 and n = 0 in the notation of Theorem 4.23. Since the period always divides the index, the assertion in this case follows from the theorem. If k is finite, then d = 1 by Wedderburn's Theorem and [Rei75, Theorem 32.19]; and n = 0 since k is perfect. So again the result follows from the theorem. In particular, in the p-adic case this yields Corollary 1.2 and the assertion after it. See also [Hu13, Theorem 3.4] for a related result in the local case. References [Abh69] In 1969 Shreeram S. Abhyankar. Resolution of singularities of algebraic surfaces. Algebraic Geometry (Internat. Colloq., Tata Inst. Fund. Res., Bombay, 1968), pp. 1– 11, Oxford Univ. Press, London. [AGV72] Michael Artin, Alexander Grothendieck, and Jean-Louis Verdier, eds. Théorie des Topos et Cohomologie Étale des Schémas, Tome 2. SGA 4, Séminaire de Géométrie Al- gébrique. Lecture Notes in Mathematics, vol. 270. Springer-Verlag, Berlin, Heidelberg, New York, 1972. [ABGV11] Asher Auel, Eric Brussel, Skip Garibaldi, and Uzi Vishne. Open problems on central simple algebras. Transform. Groups 16 (2011), 219–264. [BT13] [BLR90] Oren Ben-Bassat and Michael Temkin. Berkovich spaces and tubular descent. Adv. Math. 234 (2013), 217–238. Siegfried Bosch, Werner Lütkebohmert, and Michel Raynaud. Néron models. Ergebnisse der Mathematik, vol. 21. Springer-Verlag, Berlin and Heidelberg, 1990. [Bou72] Nicolas Bourbaki. Commutative Algebra. Hermann and Addison-Wesley, 1972. [Coh46] Irvin Sol Cohen. On the structure and ideal theory of complete local rings. Trans. Amer. Math. Soc. 59, (1946), 54–106. 41 [CPS12] [Cuo13] [deJ04] Jean-Louis Colliot-Thélène, R. Parimala, and V. Suresh. Patching and local-global principles for homogeneous spaces over function fields of p-adic curves. Comment. Math. Helv. 87 (2012), 1011–1033. Doan Trung Cuong. Fibers of flat morphisms and Weierstrass Preparation Theorem. J. Algebra 411 (2014), 337–355. Aise Johan de Jong. The period-index problem for the Brauer group of an algebraic surface. Duke Math. J. 123 (2004), 71–94. [GaOr08] Ofer Gabber and Fabrice Orgogozo. Sur la p-dimension des corps. Invent. Math. 174 (2008), 47–80. [Har84] [HH10] David Harbater. Convergent arithmetic power series. Amer. J. Math. 106 (1984), 801–846. David Harbater and Julia Hartmann. Patching over fields. Israel Journal of Math. 176 (2010), 61–107. [HHK09] David Harbater, Julia Hartmann, and Daniel Krashen. Applications of patching to quadratic forms and central simple algebras. Invent. Math. 178 (2009), 231–263. [HHK13] David Harbater, Julia Hartmann, and Daniel Krashen. Weierstrass preparation and algebraic invariants. Mathematische Annalen 356 (2013), 1405–1424. [HHK14] David Harbater, Julia Hartmann, and Daniel Krashen. Local-global principles for Galois cohomology. Commentarii Mathematici Helvetici 89 (2014), 215–253. [HHK15] David Harbater, Julia Hartmann, and Daniel Krashen. Local-global principles for torsors over arithmetic curves. American Journal of Mathematics, 137 (2015), 1559– 1612. [Hu12] [Hu13] [Hu15] Yong Hu. Local-global principle for quadratic forms over fraction fields of two-dimen- sional Henselian domains. Ann. Inst. Fourier (Grenoble) 62 (2012), 2131–2143 (2013). Yong Hu. Division algebras and quadratic forms over fraction fields of two-dimensional henselian domains. Algebra and Number Theory 7 (2013), 1919–1952. Yong Hu. The Pythagoras number and the u-invariant of Laurent series fields in several variables. J. Algebra 426 (2015), 243-258. [KMRT98] Max-Albert Knus, Alexander S. Merkurjev, Markus Rost, and Jean-Pierre Tignol. The Book of Involutions. American Mathematical Society, Providence, RI, 1998. [Kra10] Daniel Krashen. Field patching, factorization and local-global principles. In: Quadratic Forms, Linear Algebraic Groups, and Cohomology (J.-L. Colliot-Thélène, S. Garibaldi, R. Sujatha, V. Suresh, eds.). Developments in Mathematics, vol. 18, Springer-Verlag, New York, 2010, 57–82. 42 [Lam05] Tsit-Yuen Lam. Introduction to quadratic forms over fields. American Mathematical Society, Providence, RI, 2005. [Lan70] Serge Lang. Algebraic Number Theory. Addison-Wesley, Reading, Massachusetts, 1970. [Lee13] [Lie11] [Lip75] David Leep. The u-invariant of p-adic function fields. J. reine angew. Math. 679 (2013), 65–73. Max Lieblich. Period and index in the Brauer group of an arithmetic surface. With an appendix by Daniel Krashen. J. Reine Angew. Math. 659 (2011), 1–41. Joseph Lipman. Introduction to resolution of singularities. In Algebraic geometry (Proc. Sympos. Pure Math., Vol. 29, Humboldt State Univ., Arcata, Calif., 1974), pages 187– 230. Amer. Math. Soc., Providence, R.I., 1975. [MMW91] Pasquale Mammone, Remo Moresi, and Adrian R. Wadsworth. u-invariants of fields of characteristic 2. Math. Z. 208 (1991), 335–347. [Mat80] Hideyuki Matsumura. Commutative algebra. Second edition. Mathematics Lecture Note Series, vol. 56. Benjamin/Cummings Publishing Co., Inc., Reading, Mass., 1980. [Mil80] [PN10] [PS10] [PS14] [Pie82] [RS13] [Rei75] [Sal97] [Sal99] James S. Milne, Étale cohomology. Princeton Mathematical Series, vol. 33. Princeton University Press, Princeton, N.J., 1980. Elad Paran and Danny Neftin. Patching and admissibility over two-dimensional com- plete local domains. Algebra and Number Theory 4 (2010), 743–762. R. Parimala and V. Suresh. The u-invariant of the function fields of p-adic curves. Ann. of Math. (2) 172 (2010), 1391–1405. R. Parimala and V. Suresh. Period-index and u-invariant questions for function fields over complete discretely valued fields. Invent. Math., 197 (2014), 215–235. Richard S. Pierce. Associative algebras. Graduate Text in Math., vol. 88. Springer- Verlag, New York, 1982. B. Reddy and V. Suresh. Admissibility of groups over function fields of p-adic curves. Adv. Math. 237 (2013), 316–330. Irving Reiner. Maximal Orders. Academic Press, London, New York and San Francisco, 1975. David J. Saltman. Division algebras over p-adic curves. J. Ramanujan Math. Soc. 12 (1997), 25–47. David J. Saltman. Lectures on division algebras, volume 94 of CBMS Regional Confer- ence Series in Mathematics. American Mathematical Society, Providence, RI, 1999. [Sal08] David Saltman. Division algebras over surfaces. J. Algebra 320 (2008), 1543–1585. 43 [See99] [Sha72] [Spr55] [VdB88] George F. Seelinger. Brauer-Severi schemes of finitely generated algebras. Israel J. of Math. 111 (1999), 321–337. Stephen S. Shatz. Profinite groups, arithmetic, and geometry. Annals of Mathematics Studies, No. 67. Princeton University Press, Princeton, 1972. Tonny Albert Springer. Quadratic forms over fields with a discrete valuation. I. Equiv- alence classes of definite forms. Nederl. Akad. Wetensch. Proc. Ser. A. 58 = Indag. Math. 17 (1955), 352–362. Michel Van den Bergh. The Brauer-Severi scheme of the trace ring of generic matrices. In: Perspectives in ring theory (Antwerp, 1987), pp. 333–338, NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., vol. 233, Kluwer Acad. Publ., Dordrecht, 1988. Author Information: David Harbater Department of Mathematics, University of Pennsylvania, Philadelphia, PA 19104-6395, USA email: [email protected] Julia Hartmann Department of Mathematics, University of Pennsylvania, Philadelphia, PA 19104-6395, USA email: [email protected] Daniel Krashen Department of Mathematics, University of Georgia, Athens, GA 30602, USA email: [email protected] The first author was supported in part by NSF grants DMS-0901164 and DMS-1265290, and NSA grant H98230-14-1-0145. The second author was supported by the German Excellence Initiative via RWTH Aachen University and by the German National Science Foundation (DFG). The third author was supported in part by NSF grant DMS-1007462 and NSF CAREER Grant DMS-1151252. 44
0810.2154	4	0810	2010-03-02T08:07:55	Iterative building of Barabanov norms and computation of the joint spectral radius for matrix sets	[ "math.RA", "math.NA" ]	The problem of construction of Barabanov norms for analysis of properties of the joint (generalized) spectral radius of matrix sets has been discussed in a number of publications. The method of Barabanov norms was the key instrument in disproving the Lagarias-Wang Finiteness Conjecture. The related constructions were essentially based on the study of the geometrical properties of the unit balls of some specific Barabanov norms. In this context the situation when one fails to find among current publications any detailed analysis of the geometrical properties of the unit balls of Barabanov norms looks a bit paradoxical. Partially this is explained by the fact that Barabanov norms are defined nonconstructively, by an implicit procedure. So, even in simplest cases it is very difficult to visualize the shape of their unit balls. The present work may be treated as the first step to make up this deficiency. In the paper two iteration procedure are considered that allow to build numerically Barabanov norms for the irreducible matrix sets and simultaneously to compute the joint spectral radius of these sets.	math.RA	math	ITERATIVE BUILDING OF BARABANOV NORMS AND COMPUTATION OF THE JOINT SPECTRAL RADIUS FOR MATRIX SETS Victor Kozyakin Institute for Information Transmission Problems Russian Academy of Sciences Bolshoj Karetny lane 19, Moscow 127994 GSP-4, Russia Abstract. The problem of construction of Barabanov norms for analysis of properties of the joint (generalized) spectral radius of matrix sets has been dis- cussed in a number of publications. In [18,21] the method of Barabanov norms was the key instrument in disproving the Lagarias-Wang Finiteness Conjecture. The related constructions were essentially based on the study of the geomet- rical properties of the unit balls of some specific Barabanov norms. In this context the situation when one fails to find among current publications any detailed analysis of the geometrical properties of the unit balls of Barabanov norms looks a bit paradoxical. Partially this is explained by the fact that Bara- banov norms are defined nonconstructively, by an implicit procedure. So, even in simplest cases it is very difficult to visualize the shape of their unit balls. The present work may be treated as the first step to make up this deficiency. In the paper an iteration procedure is considered that allows to build numer- ically Barabanov norms for the irreducible matrix sets and simultaneously to compute the joint spectral radius of these sets. 1. Introduction. Let A = {A1, . . . , Ar} be a set of real m × m matrices. As usual, for n ≥ 1, let us denote by A n the set of all n-products of matrices from A ; A 0 = I. For each n ≥ 1, define the quantity ¯ρn(A ) := max Aij ∈A ρ(Ain ··· Ai2 Ai1 ), where maximum is taken over all possible products of n matrices from the set A , and ρ(·) denotes the spectral radius of a matrix, that is the maximal magnitude of its eigenvalues. Clearly, if n > r then some matrices in the product Ain ··· Ai2 Ai1 will occur several times. The limit ¯ρ(A ) := lim sup n→∞ (¯ρn(A ))1/n is called the generalized spectral radius of the matrix set A [9, 11]. Similarly, given a norm k · k in Rm, for each n ≥ 1, define the quantity ρn(A ) := max Aij ∈A kAin ··· Ai2 Ai1k, 2000 Mathematics Subject Classification. Primary: 15A18; 15A60; Secondary: 65F15. Key words and phrases. Infinite matrix products, generalized spectral radius, joint spectral radius, extremal norms, Barabanov norms, irreducibility, numerical algorithms. 1 2 VICTOR KOZYAKIN where kAk, for a matrix A, is the matrix norm generated by the vector norm k · k in Rm, that is kAk = supkxk=1 kAxk. Then the limit ρ(A ) := lim sup n→∞ (ρn(A ))1/n (1) does not depend on the choice of the norm k·k and is called the joint spectral radius of the matrix set A [33]. When r = 1 this definition coincides with the famous Gelfand formula for the spectral radius of a matrix [13] as in this case A = {A} is a singleton matrix set and ρn(A ) = kAnk. For matrix sets A consisting of a finite amount of matrices, as is our case, the quantities ¯ρ(A ) and ρ(A ) coincide with each other [5] and their common value is denoted as ρ(A ) := ¯ρ(A ) = ρ(A ), while the quantities ¯ρn(A ) and ρn(A ) form lower and upper bounds, respectively, for the joint/generalized spectral radius: ¯ρn(A ) ≤ ¯ρ(A ) = ρ(A ) ≤ ρn(A ), ∀ n ≥ 0. This last formula may serve as a basis for a posteriori estimating the accuracy of computation of ρ(A ). The first algorithms of a kind in the context of control theory problems have been suggested in [6], for linear inclusions in [2], and for problems of wavelet theory in [8 -- 10]. Later the computational efficiency of these algorithms was essentially improved in [14, 26]. Unfortunately, the common feature of all such algorithms is that they do not provide any bounds for the number of computational steps required to get desired accuracy of approximation of ρ(A ). Recently, in [19] explicit computable estimates for the rate of convergence of the quantities kAnk1/n to ρ(A) end their extension to the case of the joint spectral radius were obtained. Probably, these results will help to make more constructive the problem of evaluating of ρ(A ) by the generalized Gelfand formula (1) . Some works suggest different formulas to compute ρ(A ). So, in [7] it is shown that ρ(A ) = lim sup n→∞ max Aij ∈A tr(Ain ··· Ai2 Ai1 )1/n , where, as usual, tr(·) denotes the trace of a matrix. Given a norm k · k in Rm, denote kA k := max A∈A kAk. Then the spectral radius of the matrix set A can be defined by the equality ρ(A ) = inf k·k kA k, (2) where infimum is taken over all norms in Rd [12, 33], and therefore ρ(A ) ≤ kA k for any norm k·k in Rm. For irreducible matrix sets,1 the infimum in (2) is attained, and for such matrix sets there are norms k · k in Rm, called extremal norms, for which (3) kA k ≤ ρ(A ). 1A matrix set A is called irreducible if the matrices from A have no common invariant sub- spaces except {0} and Rm. In [23 -- 25] such a matrix set was called quasi-controllable. ITERATIVE BUILDING OF BARABANOV NORMS 3 In analysis of the joint spectral radius ideas suggested by N.E. Barabanov [2 -- 4] play an important role. These ideas have got further development in a variety of publications among which we would like to distinguish [35]. Theorem 1.1 (N.E. Barabanov). Let the matrix set A = {A1, . . . , Ar} be irre- ducible. Then the quantity ρ is the joint (generalized) spectral radius of the set A iff there is a norm k · k in Rm such that ρkxk ≡ max i kAixk. (4) Throughout the paper, a norm satisfying (4) will be called a Barabanov norm corresponding to the matrix set A . Note that Barabanov norms are not unique. Similarly, as is shown in [31, Thm 3.3] and [32], the value of ρ equals to ρ(A ) if and only if for some central-symmetric convex body2 S the following equality holds ρS = conv r [i=1 AiS! , (5) where conv(·) denotes the convex hull of a set and ρS := {ρx : x ∈ S}. As is noted in [31], the relation (5) was proved by A.N. Dranishnikov and S.V. Konyagin, so it is natural to call the central-symmetric set S the Dranishnikov-Konyagin-Protasov set. The set S can be treated as the unit ball of some norm k · k in Rd (recently this norm is usually called the Protasov norm). Note that Barabanov and Protasov norms are the extremal norms, that is they satisfy the inequality (3). In [28, 29, 36] it is shown that Barabanov and Protasov norms are dual to each other. Remark that formulas (3), (4) and (5) define the joint or generalized spectral radius for a matrix set in an apparently computationally nonconstructive manner. In spite of that, namely such formulas underlie quite a number of theoretical con- structions (see, e.g., [1, 18, 21, 27, 35, 36]) and algorithms [30] for computation of ρ(A ). Different approaches for constructing Barabanov norms to analyze properties of the joint (generalized) spectral radius are discussed, e.g., in [15, 17] and [34, Sect. 6.6]. In [18, 21] the method of Barabanov norms was the key instrument in disproving the Lagarias-Wang Finiteness Conjecture. The related constructions were essentially based on the study of the geometrical properties of the unit balls of some specific Barabanov norms. In [16, 17] the method of extremal polytope norms was the key tool in investigation of the finiteness properties of pairs of 2 × 2 matrices. In this context the situation when one fails to find among current publications any detailed analysis of the geometrical properties of the unit balls of Barabanov norms looks a bit paradoxical. Partially this is explained by the fact that Barabanov norms are defined nonconstructively, by an implicit procedure. So, even in simplest cases it is very difficult to visualize the shape of their unit balls. The present work may be treated as the first step to make up this deficiency. In the paper, an iteration procedure is considered that allows to build numerically Barabanov norms for the irreducible matrix sets and simultaneously to compute the joint spectral radius of these sets. A similar iteration procedure is also discussed in [22]. The structure of the paper is as follows. In Introduction we give basic definitions and present the motivation of the work. In Section 2 the main iteration procedure is 2The set is called body if it contains at least one interior point. 4 VICTOR KOZYAKIN introduced and Main Theorem stating convergence of this procedure is formulated. The iteration procedure under consideration is called the max-relaxation procedure since in it the next approximation to the Barabanov norm is constructed as the maximum of the current approximation and some auxiliary norm. In Section 3 proof Main Theorem is given. In Section 4, to build simplest examples, the max-relaxation scheme is adapted for computations with 2× 2 matrices. Results of numerical tests are illustrated by two examples. At last, in concluding Section 5 we discuss some shortcomings of the proposed approach and formulate further unresolved problems. 2. Max-relaxation iteration scheme. Given r, m ≥ 1, let A = {A1, . . . , Ar} be an irreducible set of m × m real matrices. Throughout the paper, a continuous function γ(t, s) defined for t, s > 0 and satisfying γ(t, t) = t, min{t, s} < γ(t, s) < max{t, s} for t 6= s, will be called an averaging function. Examples for averaging functions are: γ(t, s) = t + s 2 , γ(t, s) = √ts, γ(t, s) = 2ts t + s . Given some averaging function γ(·,·), a norm k · k0 in Rm, and a vector e ∈ Rm n, n = 1, 2, . . ., such that kek0 = 1, construct recursively the norms k · kn and k · k◦ in accordance with the following rules: MR1: if the norm k · kn has been already defined compute the quantities n , ρ+ n ); maxi kAixkn maxi kAixkn , , γn = γ(ρ− ρ+ n = max x6=0 ρ− n = min x6=0 kxkn kxkn MR2: define the norms k · kn+1 and k · k◦ kxkn+1 = maxnkxkn, γ−1 kxk◦ n+1 = kxkn+1/kekn+1. n+1: n max i kAixkno , (6) (7) (8) Remark that the number of operations needed to perform one step of algorithm MR1-MR2 is of order rm2ν(ε), where ν(ε) is the number of operations needed to compute, for an arbitrary vector x ∈ Rm, the value of the norm kxk with a relative In general, the value ν(ε) is of order ε−m. So, the total number of accuracy ε. operations needed to perform n steps of algorithm MR1-MR2 has the same rate of growth as nrm2ε−m. Remark also, that the procedure (6) of calculation of ρ± n resembles the technique of iterative approximation of the joint spectral radius suggested in [14]. Main Theorem. For any irreducible matrix set A , nonzero vector e ∈ Rm, initial norm k · k0, and any averaging function γ(t, s), the sequences {ρ± n} constructed by the iteration procedure MR1, MR2 converge to ρ(A ), and the sequence of norms k·kn converges uniformly on each bounded set to some Barabanov norm k·k∗ of the matrix set A . Moreover, the sequence {ρ− n} is nonincreasing, and n } is nondecreasing, the sequence {ρ+ for all n = 1, 2, . . . , which provides an a posteriori estimate for the computational error of ρ(A ). n ≤ ρ(A ) ≤ ρ+ ρ− n ITERATIVE BUILDING OF BARABANOV NORMS 5 3. Proof of Main Theorem. Let us suppose that we managed to prove the fol- lowing assertions: A1: the sequences {ρ+ A2: the limits of the sequences {ρ+ ρ = lim n→∞ n} and {ρ− n} and {ρ− ρ+ n = lim n→∞ n} are convergent; ρ− n ; n} coincide: n converge pointwise to a limit k · k∗. A3: the norms k · k◦ Then the function k · k∗ will be a semi-norm in Rm. Moreover, by (8) each norm n meets the normalization condition kek◦ n→∞kek◦ kek∗ = lim n = 1, and hence n = 1, k · k◦ which implies kxk∗ 6≡ 0. Note also that due to (8) the norms k·k◦ only by numerical factors. Therefore, the quantities ρ± n differ from k ·kn ρ+ n = max x6=0 maxi kAixk◦ n kxk◦ n , ρ− n = min x6=0 n can be defined as maxi kAixk◦ n . kxk◦ n (9) Then, passing to the limit in (9), one can conclude that the semi-norm kxk∗ satisfies the Barabanov condition ρkxk∗ = max i kAixk∗. But as shown in [21, Thm. 3], any semi-norm kxk∗ 6≡ 0 satisfying the Barabanov condition for an irreducible matrix set is a Barabanov norm. Thus, under assumptions A1, A2 and A3, the iteration procedure (6) -- (8) allows to build a Barabanov norm and to find the joint spectral radius of the matrix set A . So, to complete the proof of Main Theorem we need to justify assertions A1, A2 and A3 which will be done in Sections 3.1 -- 3.6. In Section 3.1 we establish convergence of the sequence of norms {k·k◦ n} to some limit which allows to prove in Lemma 3.2 that Assertion A3 is a corollary of Assertions A1 and A2. Section 3.2 demonstrates that the quantities {ρ− n} form the family of lower bounds for the joint spectral radius ρ of the matrix set A , while the quantities {ρ+ n} form the family of upper bounds for ρ. In Section 3.3 we prove that the sequences {ρ± n} are bounded and monotone which implies the existence of the limits ρ− = limn→∞ ρ− n n . At last, in Sections 3.4 and 3.5 we prove that ρ− = ρ+ which and ρ+ = limn→∞ ρ+ allows to justify in Section 3.6 the validity of Assertions A1 and A2 and thus to finalize the proof of Main Theorem. 3.1. Convergence of the sequence of norms {k · k◦ k · k′ and k · k′′ in Rm define the quantities n}. Given a pair of norms e−(k · k′,k · k′′) = min x6=0 e+(k · k′,k · k′′) = max x6=0 (10) kxk′ kxk′′ . kxk′ kxk′′ , Since all norms in Rm are equivalent to each other, the quantities e−(k·k′,k·k′′) and e+(k · k′,k · k′′) are correctly defined and 0 < e−(k · k′,k · k′′) ≤ e+(k · k′,k · k′′) < ∞. Therefore the quantity ecc(k · k′,k · k′′) = e+(k · k′,k · k′′) e−(k · k′,k · k′′) ≥ 1, (11) 6 VICTOR KOZYAKIN which is called the eccentricity of the norm k · k′ with respect to the norm k · k′′ (see, e.g., [36]), is also correctly defined. Lemma 3.1. Let k · k∗ be a Barabanov norm for the matrix set A . Then ecc(k · k◦ n,k · k∗) = ecc(k · kn,k · k∗), ∀ n, (12) and the sequence of the numbers ecc(k · kn,k · k∗) is nonincreasing. Proof. Note first that by (8) each norm k · k◦ n differs from the corresponding norm k · kn only by a numerical factor. From this, by the definition (10), (11) of the eccentricity of one norm with respect to another, the equality (12) follows. Denote by ρ the joint spectral radius of the matrix set A . Then, by definitions of the function e+(·) and of the Barabanov norm k · k∗, from (6), (7) we obtain: kxkn+1 = maxnkxkn, γ−1 i kAixkno ≤ n max ≤ e+(k · kn,k · k∗) maxnkxk∗, γ−1 n max i kAixk∗o = Therefore = e+(k · kn,k · k∗) max(cid:8)kxk∗, γ−1 n ρkxk∗(cid:9) . (13) Similarly, by definitions of the function e−(·) and of the Barabanov norm k · k∗, kxkn+1 = maxnkxkn, γ−1 e+(k · kn+1,k · k∗) ≤ e+(k · kn,k · k∗) max(cid:8)1, γ−1 i kAixkno ≥ n ρ(cid:9) . n max from (6), (7) we obtain: ≥ e−(k · kn,k · k∗) maxnkxk∗, γ−1 n max i kAixk∗o = = e−(k · kn,k · k∗) max(cid:8)kxk∗, γ−1 n ρkxk∗(cid:9) . Therefore (14) e+(k · kn+1,k · k∗) e−(k · kn+1,k · k∗) ≤ n ρ(cid:9) . = ecc(k · kn,k · k∗). e−(k · kn+1,k · k∗) ≥ e−(k · kn,k · k∗) max(cid:8)1, γ−1 By dividing termwise the inequality (13) on (14) we get e+(k · kn,k · k∗) e−(k · kn,k · k∗) Denote by Nloc(Rm) the topological space of norms in Rm with the topology of ecc(k · kn+1,k · k∗) = Hence, the sequence {ecc(k · kn,k · k∗)} is nonincreasing. uniform convergence on bounded subsets of Rm. Corollary 1. The sequence of norms {k · k◦ Proof. For each n and any x 6= 0, by the definition (10) of the functions e+(·) and e−(·) the following relations hold n} is compact in Nloc(Rm). e−(k · kn,k · k∗) ≤ kxkn e−(k · kn,k · k∗) ≤ kekn kxk∗ ≤ e+(k · kn,k · k∗), kek∗ ≤ e+(k · kn,k · k∗), and then from which 1 ecc(k · kn,k · k∗) kxk∗ kek∗ kekn ≤ kxkn ≤ ecc(k · kn,k · k∗)kxk∗ kek∗ kekn. ITERATIVE BUILDING OF BARABANOV NORMS 7 Since here by construction the norms {k · k◦ kek◦ n ≡ 1, and by Lemma 3.1 ecc(k · k◦ n} satisfy the normalization condition n,k · k∗) ≤ ecc(k · k◦ 0,k · k∗), we have 1 ecc(k · k0,k · k∗) kxk∗ kek∗ ≤ kxkn ≤ ecc(k · k0,k · k∗)kxk∗ kek∗ . Therefore the norms k ·kn, n ≥ 1, are equicontinuous and uniformly bounded on each bounded subset of Rm. Moreover, their values are also uniformly separated from zero on each bounded subset of Rm separated from zero. From here by the Arzela-Ascoli theorem the statement of the corollary follows. Corollary 2. If at least one of subsequences of norms from {k · k◦ Nloc(Rm) to some Barabanov norm then the whole sequence {k · k◦ in Nloc(Rm) to the same Barabanov norm. n} which converges in Nloc(Rm) to Proof. Let {k · k◦ some Barabanov norm k · k∗. Then by definition of the eccentricity of one norm with respect to another nk} be a subsequence of {k · k◦ n} converges in n} also converges nk ,k · k∗) → 1 Here by Lemma 3.1 the eccentricities ecc(k · k◦ then the following stronger relation holds ecc(k · k◦ as k → ∞. n,k · k∗) are nonincreasing in n, and Note now that by the definition (10), (11) of the eccentricity of one norm with ecc(k · k◦ n,k · k∗) → 1 as n → ∞. (15) respect to another n 1 ecc(k · k◦ n,k · k∗) ≤ kxk◦ kxk∗ ≤ ecc(k · k◦ n,k · k∗), from which by (15) it follows that the sequence of norms {k·k◦ Nloc(Rm) to the norm k · k∗. Lemma 3.2. Assertion A3 is a corollary of Assertions A1 and A2. Proof. By Corollary 1 the sequence of norms {k · k◦ nk} that converges in space Nloc(Rm) to some norm k · k∗. Then, passing to the limit in (9) as n = nk → ∞, we get by Assertions A1 and A2: ∀ x 6= 0, n} has a subsequence {k · k◦ n} converges in space maxi kAixk∗ ρ = , kxk∗ which means that k · k∗ is a Barabanov norm for the matrix set A . This and Corollary 2 then imply that the sequence {k · k◦ n} converges in space Nloc(Rm) to the Barabanov norm k · k∗. Assertion A3 is proved. So, in view of Lemma 3.2 to prove that the iteration procedure (6) -- (8) is con- vergent it suffices to verify only that Assertions A1 and A2 hold. 3.2. Relations between ρ± mate the spectral radius of a matrix set. n and ρ. The following lemma provides a way to esti- Lemma 3.3. Let α, β be numbers such that in some norm k · k the inequalities αkxk ≤ max Ai∈A kAixk ≤ βkxk, hold. Then α ≤ ρ ≤ β, where ρ is the joint spectral radius of the matrix set A . 8 VICTOR KOZYAKIN Proof. Let k · k∗ be some Barabanov norm for the matrix set A . Since all norms in Rm are equivalent, there are constants σ− > 0 and σ+ < ∞ such that σ−kxk∗ ≤ kxk ≤ σ+kxk∗. Consider, for each k = 1, 2, . . ., the functions ∆k(x) = max 1≤i1,i2,...,ik≤r kAik . . . Ai2 Ai1 xk. Then, as is easy to see, Similarly consider, for each k = 1, 2, . . ., the functions αkkxk ≤ ∆k(x) ≤ βkkxk. ∆∗ k(x) = 1≤i1,i2,...,ik≤r kAik . . . Ai2 Ai1 xk∗. max (16) (17) For these functions, by definition of Barabanov norms the following identity hold which is stronger than (17). Now, note that (16) and the definition of the functions ∆k(x) and ∆∗ k(x) imply ∆∗ k(x) ≡ ρkkxk∗, (18) σ−∆∗ k(x) ≤ ∆k(x) ≤ σ+∆∗ k(x). Then, by (17), (18), σ− σ+ αk ≤ ρk ≤ σ+ σ− βk, ∀ k, So, Lemma 3.3 and the definition (6) of ρ± n imply that the quantities {ρ− from which the required estimates α ≤ ρ ≤ β follow. n } form the family of lower bounds for the joint spectral radius ρ of the matrix set A , while the quantities {ρ+ n} form the family of upper bounds for ρ. This allows to estimate a posteriori errors of computation of the joint spectral radius with the help of the iteration procedure (6) -- (8). 3.3. Convergence of the sequences {ρ± By definition, i (cid:26)max(cid:26)kAixkn, γ−1 = max(cid:26)max n }. Estimate the value of maxi kAixkn+1. j kAiAjxkn(cid:27)(cid:27) = i kAiAjxkn(cid:27) . max i kAixkn+1 = max n max i kAixkn, γ−1 n max j max Here by the definition (6) of the quantities ρ± equalities can be estimated as follows: n the right-hand part of the chain of ρ− n max(cid:26)kxkn, γ−1 n max j kAjxkn(cid:27) ≤ i kAixkn, γ−1 n max j ≤ max(cid:26)max max i kAiAjxkn(cid:27) ≤ n max(cid:26)kxkn, γ−1 ρ+ n max j kAjxkn(cid:27) . Therefore, by definition of the norm kxkn+1, ρ− n kxkn+1 ≤ max i kAixkn+1 ≤ ρ+ nkxkn+1, ITERATIVE BUILDING OF BARABANOV NORMS 9 maxi kAixkn+1 ρ− n ≤ ≤ ρ+ n , ∀ x 6= 0, n+1 ≤ ρ+ n+1 ≤ ρ+ n . from which and then, kxkn+1 n ≤ ρ− ρ− So, the following lemma holds. Lemma 3.4. The sequence {ρ− sequence {ρ+ each member of the sequence {ρ− n} is bounded from above by each member of the n} and is nondecreasing. The sequence {ρ+ n} is bounded from below by n} and is nonincreasing. In view of Lemma 3.4 there are the limits ρ− = lim n→∞ ρ− n , ρ+ = lim n→∞ ρ+ n , γ = lim n→∞ γn = lim n→∞ where γ(ρ− n , ρ+ n ), which means that Assertion A1 holds. Hence, to prove that the iteration procedure (6) -- (8) is convergent it remains only to justify Assertion A2: ρ− = ρ+. ρ− ≤ γ ≤ ρ+, To prove that ρ− = ρ+, below it will be supposed the contrary, which will lead us to a contradiction. 3.4. Transition to a new sequence of norms. To simplify further constructions we will switch over to a new sequence of norms for which the quantities ρ± n will be independent of n. By Corollary 1 the sequence of the norms k · k◦ Hence, there is a subsequence of indices {nk} such that the the norms k · k◦ k · knk /keknk converge to some norm k · k• kek• 0 = 1. Then, passing to the limit in (9), by Lemma 3.4 we obtain: n is compact in space Nloc(Rm). nk = 0 satisfying the normalization condition ρ+ = max x6=0 maxi kAixk• 0 kxk• 0 , ρ− = min x6=0 maxi kAixk• 0 kxk• 0 , γ = γ(ρ−, ρ+). Now by induction the following statement can be easily proved. Lemma 3.5. For each n = 0, 1, 2, . . ., the sequence of the norms k · knk+n/keknk converges to some norm k · k• n. Moreover, for each n = 0, 1, 2, . . ., we have the equalities (19) (20) (21) 3.5. Sets ωn. Define, for each n = 0, 1, 2, . . ., the set max x6=0 and the recurrent relations kxk• maxi kAixk• n = ρ−, kxk• n maxi kAixk• n = ρ+, min x6=0 n kxk• n+1 = maxnkxk• ωn =nx ∈ Rm : ρ−kxk• maxi kAixk• n kxk• n n, γ−1 max no . i kAixk• no . i kAixk• n = max By (19) ωn is the set on which the quantity attains its minimum. Lemma 3.6. If x ∈ ωn then kxk• n+1 = kxk• n. 10 VICTOR KOZYAKIN Proof. The statement of the lemma is obvious for x = 0. So, suppose that x ∈ ωn, x 6= 0. In this case (21) and the inequalities ρ− ≤ ρ+ imply i kAixk• max n = ρ−kxk• n ≤ γkxk• n or, what is the same, kxk• n ≥ γ−1 max i kAixk• n. From here by the definition (20) of the norm k · k• n+1 we get the required equality: n, γ−1 max i kAixk• n+1 = maxnkxk• no = kxk• kxk• The lemma is proved. Lemma 3.7. If ρ− < ρ+ then ωn+1 ⊆ ωn for each n = 0, 1, 2, . . .. Proof. Let x ∈ ωn+1. If x = 0 then clearly x ∈ ωn, so suppose that x 6= 0. By definitions of the set ωn+1 and of the norm k·k• n the following equalities take place: n. i kAixk• max n+1 = max i (cid:26)max(cid:26)kAixk• = max(cid:26)max i kAixk• n, γ−1 max n, γ−1 max j kAjAixk• i,j kAjAixk• n(cid:27)(cid:27) = n(cid:27) = Let here = ρ−kxk• n+1 = ρ− maxnkxk• i kAixk• n ≤ γ−1 max n. kxk• n, γ−1 max no . i kAixk• (22) (23) Then from (22) it follows that n, γ−1 max i kAixk• i,j kAjAixk• n(cid:27) = ρ−kxk• max(cid:26)max But by the conditions of the lemma ρ− < ρ+. Then γ = γ(ρ−, ρ+) > ρ−, and the right-hand part of the above equalities is strictly less than maxi kAixk• n. A contra- diction, since the left-hand part of the same equalities is no less than maxi kAixk• n. The above contradiction is caused by the assumption (23), and therefore it is proved that the condition x 6= 0 ∈ ωn+1 implies the strict inequality n+1 = γ−1ρ− max i kAixk• n. kxk• n > γ−1 max i kAixk• n. In this case from (22) it follows that max(cid:26)max i kAixk• n, γ−1 max i,j kAjAixk• n(cid:27) = ρ−kxk• n. Let us show that the equality (24) implies i kAixk• max n = ρ−kxk• n. (24) (25) Indeed, supposing the contrary, by definition of the quantity ρ−, there should be valid the strict inequality maxi kAixk• n > ρ−kxk• n. Then the left-hand part of the equality (24) should be strictly greater than ρ−kxk• n, that is greater than the right- hand part of the same equality, which is impossible. This last contradiction shows that the equality (25) holds as soon as x 6= 0 ∈ ωn+1, which means by (19) that x ∈ ωn. ITERATIVE BUILDING OF BARABANOV NORMS 11 Corollary 3. If ρ− < ρ+ then ω := ∩n≥0ωn 6= 0 and kxk• n = . . . , 0 = kxk• 1 = ··· = kxk• ∀ x 6= 0 ∈ ω. (26) Proof. By Lemma 3.7 {ωn} is the family of embedded closed conic3 sets. Then the intersection ω of these sets is also a closed conic set such that ω 6= {0}. By definition of the set ω, if x ∈ ω then x ∈ ωn for every integer n ≥ 0. Hence, by Lemma 3.6 kxk• 3.6. Completion of the proof of Assertion A2. By Corollary 3 there is a non- zero vector g on which all the norms k · k• n, from which the equalities (26) follow. n take the same values: n+1 = kxk• kgk• 0 = kgk• 1 = ··· = kgk• n = ··· . Then, due to uniform boundedness of the eccentricities of the norms k · k• n with respect to some Barabanov norm k·k∗ (this fact can be proved by verbatim repetition of the analogous proof for the norms k ·kn), the norms k ·k• n form a family which is uniformly bounded and equicontinuous with respect to the Barabanov norm k · k∗: ∃ µ± ∈ (0,∞) : µ−kxk∗ ≤ kxk• n ≤ µ+kxk∗, n = 0, 1, 2, . . . . n} is compact in Hence, by the Arzela-Ascoli theorem the family of norms {k · k• Nloc(Rm). n it follows also that i kAixk• From the definition (20) of the norms k · k• kxk• Then the norms k · k• n are monotone increasing in n and bounded (with respect to the Barabanov norm k · k∗) and therefore they pointwise converge to some norm k · k•. Moreover, since the family of norms {k · k• n} is equicontinuous with respect to the Barabanov norm k · k∗, the norms k · k• n converge to the norm k · k• in space Nloc(Rm). Now, passing to the limit in the relations n+1 = maxnkxk• no ≥ kxk• n, γ−1 max n. kxk• n+1 = maxnkxk• n, γ−1 max i kAixk• no ≥ γ−1 max i kAixk• n, which follow from (20), we obtain kxk• ≥ γ−1 max i kAixk•. From here max x6=0 maxi kAixk• kxk• ≤ γ. (27) On the other hand, passing to the limit in the first relation of (19), we obtain max x6=0 maxi kAixk• kxk• = ρ+. (28) Relations (27) and (28) imply the inequality ρ+ ≤ γ which contradicts the as- sumption ρ− < ρ+ because by definition of the function γ(·,·) the condition ρ− < ρ+ implies the inequality γ = γ(ρ−, ρ+) < ρ+. The obtained contradiction completes the proof of the equality ρ− = ρ+ as well as of the convergence of the iteration procedure (6) -- (8). 3A set X is called conic if together with each its point x it contains the ray {tx : t ≥ 0}. 12 VICTOR KOZYAKIN 4. Computational scheme for 2 × 2 matrices. Let A = {A1, . . . , Ar} be a set of real 2 × 2 matrices Ai =  a(i) 11 a(i) 21 a(i) 12 a(i) 22   . Let (r, ϕ) be the polar coordinates in R2. Then, for a vector x ∈ R2 with Cartesian coordinates x = {x1, x2}, we have x = {r cos ϕ, r sin ϕ} and r = r(x) =qx2 Define, for an arbitrary norm k · k, the function 1 + x2 2, ϕ = ϕ(x) = arctan (x2/x1) . R(ϕ) = k{cos ϕ, sin ϕ}k. Then the norm kxk of the vector x with polar coordinates (r, ϕ) is determined by the equality (29) and the unit sphere in the norm k · k is determined as the geometrical locus of the vectors x polar coordinates of which satisfy the relations kxk = rR(ϕ), rR(ϕ) ≡ 1 or r = 1 R(ϕ) , see Fig. 1. x = (r, ϕ) kxk = rR(ϕ) 1/R(ϕ) kxk = 1 ⇔ rR(ϕ) = 1 Figure 1. Definition of the function R(ϕ). Now, let Rn(ϕ) be the function defining in the polar coordinates the graph of the unit sphere kxkn = 1 of the norm k · kn determined by the iteration procedure (6) -- (8). Rewrite the relations (6) -- (8) in terms of the functions Rn(ϕ). To do it we should express the quantities kAixkn, i = 0, 2, . . . , r, in terms of the functions Rn(ϕ). By (29) kAixkn = r(Aix)Rn(ϕ(Aix)). ITERATIVE BUILDING OF BARABANOV NORMS 13 Here by definition of the matrix Ai r(Aix) = rHi(ϕ), where Hi(ϕ) =r(cid:16)a(i) 11 cos ϕ + a(i) 12 sin ϕ(cid:17)2 +(cid:16)a(i) 21 cos ϕ + a(i) 22 sin ϕ(cid:17)2 . Similarly, by definition of the matrix Ai where ϕ(Aix) = Φi(ϕ), Φi(ϕ) = arctan a(i) 21 cos ϕ + a(i) a(i) 11 cos ϕ + a(i) 22 sin ϕ 12 sin ϕ! . From the obtained relations it follows that the first two equalities in (6) take the form ρ+ n = max ϕ max i Hi(ϕ)Rn(Φi(ϕ)) Rn(ϕ) , ρ− n = min ϕ max i Hi(ϕ)Rn(Φi(ϕ)) Rn(ϕ) , or, what is the same, where ρ+ n = max ϕ R∗ n(ϕ) Rn(ϕ) , ρ− n = min ϕ R∗ n(ϕ) Rn(ϕ) , R∗ n(ϕ) = max i {Hi(ϕ)Rn(Φi(ϕ))} . The relations (7) take the form or, equivalently, n R∗ rRn+1(ϕ) = max(cid:8)rRn(ϕ), rγ−1 Rn+1(ϕ) = max(cid:8)Rn(ϕ), γ−1 n R∗ n(ϕ)(cid:9) n(ϕ)(cid:9) and the normalization condition (8) takes the form (30) (31) (32) rR◦ n+1(ϕ) = rRn+1(ϕ) reRn+1(ϕe) , where (re, ϕe) are polar coordinates of the vector e. Taking in place of e the vector with polar coordinates (1, 0) the normalization condition can be rewritten in the form R◦ n+1(ϕ) = Rn+1(ϕ) Rn+1(0) . (33) So, the max-relaxation iteration scheme can be represented as follows. Given an averaging function γ(·,·), set R0(ϕ) ≡ 1, and build recursively the 2π -- periodic functions Rn(ϕ) and R◦ n(ϕ), n = 1, 2, . . ., in accordance with the following rules: MR1: regarding the function Rn(ϕ) already known compute the numerical val- ues ρ+ n and ρ− n by formulas (30), (31) and set γn = γ(ρ− n , ρ+ n ); MR2: define Rn+1(ϕ) by (32) and R◦ norm k · k◦ of the vector x. n+1 as kxk◦ n+1 = rR◦ n+1(ϕ) by (33), and then determine the n+1(ϕ), where (r, ϕ) are the polar coordinates 14 VICTOR KOZYAKIN 6 4 2 0 −2 −4 −6 −6 kA1xk∗ = ρ kA2xk∗ = ρ kxk∗ = 1 −4 −2 0 2 4 6 2 1.5 1 0.5 0 −0.5 −1 −1.5 −2 −2 kA1xk∗ = ρ kA2xk∗ = ρ kA3xk∗ = ρ kxk∗ = 1 −1 0 1 2 Figure 2. Examples of computation of Barabanov norms for 2 × 2 matrices. Example 1. Consider the family A = {A1, A2} of 2 × 2 matrices A1 =(cid:18) 1 0 0 1 (cid:19) . 1 1 (cid:19) , A2 =(cid:18) 1 1 The functions Φi(ϕ), Hi(ϕ), Rn(ϕ), R∗ n(ϕ) were chosen to be piecewise linear with 3000 nodes uniformly distributed over the interval [−π, π]. It was needed 13 iter- ations of algorithm MR1-MR2 with the averaging function γ(t, s) = t+s imple- 2 mented in MATLAB to compute the joint spectral radius ρ(A ) with the absolute accuracy 10−3. The computed value of the joint spectral radius is ρ(A ) = 1.617. The computed unit sphere of the Barabanov norm k · k∗ after the 13th iteration of algorithm MR1-MR2 is shown on Fig. 2 on the left. As is seen from Fig. 2, in Example 1 the sets kA1xk = ρ and kA2xk = ρ have exactly 4 intersection points. This was theoretically proved in [18, 21] for the case when one of the matrices A1, A2 is lower triangle and the other is upper triangle, and their entries are nonnegative. In [18, 21] this fact was one of key points in disproving the Finiteness Conjecture. We do not know whether this fact is valid in a general case or not, but numerical tests based on algorithm MR1-MR2 with several dozens pairs of matrices A1, A2 testify for this fact. Example 2. Consider the family A = {A1, A2, A3} of 2 × 2 matrices 0 1 (cid:19) , A2 =(cid:18) 0.8 −0.6 Here the functions Φi(ϕ), Hi(ϕ), Rn(ϕ), R∗ n(ϕ) were also chosen to be piecewise lin- ear with 3000 nodes uniformly distributed over the interval [−π, π]. It was needed 31 iterations of algorithm MR1-MR2 with the averaging function γ(t, s) = t+s im- 2 plemented in MATLAB to compute the joint spectral radius ρ(A ) with the absolute accuracy 10−3. The computed value of the joint spectral radius is ρ(A ) = 1.347. The computed unit sphere of the Barabanov norm k · k∗ after the 31d iteration of algorithm MR1-MR2 is shown on Fig. 2 on the right. The MATLAB code for this Example can be found in [20]. A1 =(cid:18) 1 1 0.6 0.8 (cid:19) , A3 =(cid:18) 1 −0.4 1.3 (cid:19) . 0 ITERATIVE BUILDING OF BARABANOV NORMS 15 5. Concluding remarks. The max-relaxation algorithm suggested in the paper allows to calculate the joint spectral radius of a finite matrix family (of arbitrary matrix size and arbitrary amount of matrices in the set) with any required accuracy and to evaluate a posteriori the computational error. Still, this algorithm gives rise to a set of open problems. Problem. While the quantities {ρ± n } provide a posteriori bounds for the accuracy of approximation of ρ(A ) the question about the accuracy of approximation of the Barabanov norm k · k∗ by the norms k · k◦ n is open. It seems, the difficulty in resolving this problem is caused by the fact that in gen- eral the Barabanov norms for a matrix family are determined ambiguously. Namely to overcome this difficulty we preferred to consider relaxation algorithm instead of direct one. Moreover, as can be shown, both theoretically and numerically, if to set kxkn+1 = γ−1 n maxi kAixkn in (7) then the obtained direct computational analog of algorithm MR1-MR2 may turn out to be non-convergent. Problem. The question about the rate of convergence of the sequences {ρ+ {ρ− n} to the joint spectral radius is also open. Remark also that in this paper mainly the algorithm for building of Barabanov norms rather than its computational details was studied. The numerical aspects of implementation of this algorithm require additional analysis. n} and Problem. An estimation of the computational cost of the max-relaxation algorithm is required. The dependance of the algorithm on parameters r, m and the choice of the averaging function γ(t, s) is acute, too. Acknowledgements. This work was supported by the Russian Foundation for Basic Research, project no. 10-01-00175. REFERENCES [1] N. Barabanov, Lyapunov exponent and joint spectral radius: Some known and new results, in: Proceedings of the 44th IEEE Conference on Decision and Control and European Control Conference 2005, Seville, Spain, December 12 -- 15, pp. 2332 -- 2337, 2005. 3 [2] N. E. Barabanov, On the Lyapunov exponent of discrete inclusions. I, Avtomat. i Telemekh., pp. 40 -- 46, in Russian, translation in Automat. Remote Control 49 (1988), no. 2, part 1, 152 -- 157. 2, 3 [3] N. E. Barabanov, On the Lyapunov exponent of discrete inclusions. II, Avtomat. i Telemekh., pp. 24 -- 29, in Russian, translation in Automat. Remote Control 49 (1988), no. 3, part 1, 283 -- 287. 3 [4] N. E. Barabanov, On the Lyapunov exponent of discrete inclusions. III, Avtomat. i Telemekh., pp. 17 -- 24, in Russian, translation in Automat. Remote Control 49 (1988), no. 5, part 1, 558 -- 565. 3 [5] M. A. Berger and Y. Wang, Bounded semigroups of matrices, Linear Algebra Appl., 166 (1992), 21 -- 27. 2 [6] R. K. Brayton and C. H. Tong, Constructive stability and asymptotic stability of dynamical systems, IEEE Trans. Circuits Syst., 27 (1980), 1121 -- 1130. 2 [7] Q. Chen and X. Zhou, Characterization of joint spectral radius via trace, Linear Algebra Appl., 315 (2000), 175 -- 188. 2 [8] D. Colella and C. Heil, The characterization of continuous, four-coefficient scaling functions and wavelets, IEEE Trans. Inform. Theory, 38 (1992), 876 -- 881. 2 [9] I. Daubechies and J. C. Lagarias, Sets of matrices all infinite products of which converge, Linear Algebra Appl., 161 (1992), 227 -- 263. 1, 2 [10] I. Daubechies and J. C. Lagarias, Two-scale difference equations. II. Local regularity, infinite products of matrices and fractals, SIAM J. Math. Anal., 23 (1992), 1031 -- 1079. 2 16 VICTOR KOZYAKIN [11] I. Daubechies and J. C. Lagarias, Corrigendum/addendum to: "Sets of matrices all infinite products of which converge", Linear Algebra Appl., 327 (2001), 69 -- 83. 1 [12] L. Elsner, The generalized spectral-radius theorem: an analytic-geometric proof, Linear Alge- bra Appl., 220 (1995), 151 -- 159. 2 [13] I. Gelfand, Normierte Ringe, Rec. Math. [Mat. Sbornik] N. S., 9 (51) (1941), 3 -- 24. 2 [14] G. Gripenberg, Computing the joint spectral radius, Linear Algebra Appl., 234 (1996), 43 -- 60. 2, 4 [15] N. Guglielmi and M. Zennaro, On the asymptotic properties of a family of matrices, Linear Algebra Appl., 322 (2001), 169 -- 192. 3 [16] N. Guglielmi and M. Zennaro, Polytope norms and related algorithms for the computation of the joint spectral radius, in: Proceedings of the 44th IEEE Conference on Decision and Control and European Control Conference 2005, Seville, Spain, December 12 -- 15, pp. 3007 -- 3012, 2005. 3 [17] N. Guglielmi and M. Zennaro, An algorithm for finding extremal polytope norms of matrix families, Linear Algebra Appl., 428 (2008), 2265 -- 2282. 3 [18] V. Kozyakin, A dynamical systems construction of a counterexample to the finiteness con- jecture, in: Proceedings of the 44th IEEE Conference on Decision and Control and European Control Conference 2005, Seville, Spain, December 12 -- 15, pp. 2338 -- 2343, 2005. 1, 3, 14 [19] V. Kozyakin, On accuracy of approximation of the spectral radius by the Gelfand for- mula, Linear Algebra Appl., 431 (2009), 2134 -- 2141, doi:10.1016/j.laa.2009.07.008, URL http://dx.doi.org/10.1016/j.laa.2009.07.008. 2 [20] V. Kozyakin, Max-Relaxation iteration procedure for building of Barabanov norms: Conver- gence and examples, ArXiv.org e-Print archive, 2010, arXiv:1002.3251. 14 [21] V. S. Kozyakin, Structure of extremal trajectories of discrete linear systems and the finiteness conjecture, Automat. Remote Control, 68 (2007), 174 -- 209, doi:10.1134/S0005117906040171. 1, 3, 5, 14 [22] V. S. Kozyakin, On the computational aspects of the theory of joint spectral radius, Dokl. Akad. Nauk, 427 (2009), 160 -- 164, doi:10.1134/S1064562409040097 , in Russian, translation in Doklady Mathematics 80 (2009), no. 1, 487 -- 491. 3 [23] V. S. Kozyakin and A. V. Pokrovskiı, The role of controllability-type properties in the study of the stability of desynchronized dynamical systems, Dokl. Akad. Nauk, 324 (1992), 60 -- 64, in Russian, translation in Soviet Phys. Dokl. 37 (1992), no. 5, 213 -- 215. 2 [24] V. S. Kozyakin and A. V. Pokrovskii, "Estimates of Amplitudes of Transient Regimes in Quasi-Controllable Discrete Systems," CADSEM Report 96 -- 005, Deakin University, Geelong, Australia, 1996, arXiv:0908.4138. 2 [25] V. S. Kozyakin and A. V. Pokrovskii, Quasi-controllability and estimation of the amplitudes of transient regimes in discrete systems, Izv., Ross. Akad. Estestv. Nauk, Mat. Mat. Model. Inform. Upr., 1 (1997), 128 -- 150, in Russian. 2 [26] M. Maesumi, An efficient lower bound for the generalized spectral radius of a set of matrices, Linear Algebra Appl., 240 (1996), 1 -- 7. 2 [27] P. A. Parrilo and A. Jadbabaie, Approximation of the joint spectral radius using sum of squares, Linear Algebra Appl., 428 (2008), 2385 -- 2402, doi:10.1016/j.laa.2007.12.027, arXiv:0712.2887. 3 [28] E. Plischke and F. Wirth, Duality results for the joint spectral radius and transient behavior, Linear Algebra Appl., 428 (2008), 2368 -- 2384, doi:10.1016/j.laa.2007.12.009. 3 [29] E. Plischke, F. Wirth and N. Barabanov, Duality results for the joint spectral radius and transient behavior, in: Proceedings of the 44th IEEE Conference on Decision and Control and European Control Conference 2005, Seville, Spain, December 12 -- 15, pp. 2344 -- 2349, 2005. 3 [30] V. Protasov, The geometric approach for computing the joint spectral radius, in: Proceedings of the 44th IEEE Conference on Decision and Control and European Control Conference 2005, Seville, Spain, December 12 -- 15, pp. 3001 -- 3006, 2005. 3 [31] V. Yu. Protasov, The joint spectral radius and invariant sets of linear operators, Fundam. Prikl. Mat., 2 (1996), 205 -- 231, in Russian. 3 [32] V. Yu. Protasov, A generalization of the joint spectral radius: the geometrical approach, Facta Univ. Ser. Math. Inform., pp. 19 -- 23. 3 [33] G.-C. Rota and G. Strang, A note on the joint spectral radius, Indag. Math., 22 (1960), 379 -- 381. 2 ITERATIVE BUILDING OF BARABANOV NORMS 17 [34] J. Theys, "Joint Spectral Radius: theory and approximations," Ph.D. thesis, Facult´e des sciences appliqu´ees, D´epartement d'ing´enierie math´ematique, Center for Systems Engineering and Applied Mechanics, Universit´e Catholique de Louvain, 2005. 3 [35] F. Wirth, The generalized spectral radius and extremal norms, Linear Algebra Appl., 342 (2002), 17 -- 40. 3 [36] F. Wirth, On the structure of the set of extremal norms of a linear inclusion, in: Proceedings of the 44th IEEE Conference on Decision and Control, and the European Control Conference 2005 Seville, Spain, December 12 -- 15, pp. 3019 -- 3024, 2005. 3, 6 E-mail address: [email protected]
1801.05910	1	1801	2018-01-18T02:32:01	Dual Third-order Jacobsthal Quaternions	[ "math.RA" ]	In 2016, Y\"uce and Torunbalc\i\ Ayd\i n \cite{Yuc-Tor} defined dual Fibonacci quaternions. In this paper, we defined the dual third-order Jacobsthal quaternions and dual third-order Jacobsthal-Lucas quaternions. Also, we investigated the relations between the dual third-order Jacobsthal quaternions and third-order Jacobsthal numbers. Furthermore, we gave some their quadratic properties, the summations, the Binet's formulas and Cassini-like identities for these quaternions.	math.RA	math	DUAL THIRD-ORDER JACOBSTHAL QUATERNIONS GAMALIEL CERDA-MORALES Instituto de Matem´aticas, Pontificia Universidad Cat´olica de Valpara´ıso, Blanco Viel 596, E-mails: [email protected] / [email protected] Valpara´ıso, Chile. Abstract. In 2016, Yuce and Torunbalcı Aydın [17] defined dual Fibonacci quaternions. In this paper, we defined the dual third-order Jacobsthal quater- nions and dual third-order Jacobsthal-Lucas quaternions. Also, we investi- gated the relations between the dual third-order Jacobsthal quaternions and third-order Jacobsthal numbers. Furthermore, we gave some their quadratic properties, the summations, the Binet's formulas and Cassini-like identities for these quaternions. Mathematical subject classification: Primary: 11R52; Secondary: 11B37, 20G20. Key words: Thir-order Jacobsthal number, third-order Jacobsthal-Lucas number, third-order Jacobsthal quaternions, third-order Jacobsthal-Lucas quaternions, dual quaternion. 1. Introduction The real quaternions are a number system which extends to the complex num- bers. They are first described by Irish mathematician William Rowan Hamilton in 1843. In 1963, Horadam [8] defined the n-th Fibonacci quaternion which can be represented as (1.1) QF = {Qn = Fn +iFn+1 +jFn+2 +kFn+3 : Fn is n-th Fibonacci number}, where i2 = j2 = k2 = ijk = −1. In 1969, Iyer [13, 14] derived many relations for the Fibonacci quaternions. In 1977, Iakin [11, 12] introduced higher order quaternions and gave some identities for these quaternions. Furthermore, Horadam [9] extend to quaternions to the complex Fibonacci numbers defined by Harman [6]. In 2012, Halıcı [5] gave generating functions and Binet's formulas for Fibonacci and Lucas quaternions. In 2006, Majernik [15] defined a new type of quaternions, the so-called dual quaternions in the form QN = {a + bi + cj + dk : a, b, c, d ∈ R}, with the following multiplication schema for the quaternion units (1.2) i2 = j2 = k2 = 0, ij = −ji = jk = −kj = ik = −ki = 0. 1 2 G. CERDA-MORALES In 2009, Ata and Yaylı [1] defined dual quaternions with dual numbers coefficient as follows: (1.3) QD = {A + Bi + Cj + Dk : A, B, C, D ∈ D, i2 = j2 = k2 = ijk = −1}, where D = R[ε] = {a + bε : a, b ∈ R, ε2 = 0, ε 6= 0}. It is clear that QN and QD are different sets. In 2014, Nurkan and Guven [16] defined dual Fibonacci quaternions as follows: (1.4) DF = {Qn = bFn + ibFn+1 + jbFn+2 + kbFn+3 : bFn = Fn + εFn+1}, where i2 = j2 = k2 = ijk = −1 and bFn is the n-th dual Fibonacci number. In 2016, Yuce and Torunbalcı Aydın [17] defined dual Fibonacci quaternions as follows: (1.5) NF = {Qn = Fn +iFn+1 +jFn+2 +kFn+3 : Fn is n-th Fibonacci number}, where i2 = j2 = k2 = 0, ij = −ji = jk = −kj = ik = −ki = 0. In the other hand, the Jacobsthal numbers have many interesting properties and applications in many fields of science (see, e.g., [2]). The Jacobsthal numbers Jn are defined by the recurrence relation (1.6) J0 = 0, J1 = 1, Jn+1 = Jn + 2Jn−1, n ≥ 1. Another important sequence is the Jacobsthal-Lucas sequence. This sequence is defined by the recurrence relation jn+1 = jn + 2jn−1, n ≥ 1 and j0 = 2, j1 = 1. (see, [10]). In [4] the Jacobsthal recurrence relation (1.6) is extended to higher order re- currence relations and the basic list of identities provided by A. F. Horadam [10] is expanded and extended to several identities for some of the higher order n }n≥0, and third-order cases. In particular, third-order Jacobsthal numbers, {J (3) Jacobsthal-Lucas numbers, {j(3) n+2 + J (3) n }n≥0, are defined by 0 = 0, J (3) n+3 = J (3) J (3) n+1 + 2J (3) 1 = J (3) n , J (3) (1.7) 2 = 1, n ≥ 0, and (1.8) n+3 = j(3) j(3) n+2 + j(3) n+1 + 2j(3) n , j(3) 0 = 2, j(3) 1 = 1, j(3) 2 = 5, n ≥ 0, respectively. The following properties given for third-order Jacobsthal numbers and third- order Jacobsthal-Lucas numbers play important roles in this paper (for more, see [4]). (1.9) 3J (3) n + j(3) n = 2n+1, DUAL THIRD-ORDER JACOBSTHAL QUATERNIONS 3 , , , 1 J (3) n+2 − 4J (3) n = 2j(3) n−3, j(3) n − 4J (3) if n 6≡ 1 (mod 3) j(3) n+1 + j(3) n = 3J (3) n+2, 2 if n ≡ 0 (mod 3) −3 if n ≡ 1 (mod 3) 1 if n ≡ 2 (mod 3) n − 3J (3) j(3) n =( −2 if n ≡ 1 (mod 3) n = n+2 = n − J (3) j(3) (cid:16)j(3) n−3(cid:17)2 k =( J (3) n (cid:17)2 n (cid:17)2 − 9(cid:16)J (3) (cid:16)j(3) if n ≡ 0 (mod 3) 1 −1 if n ≡ 1 (mod 3) 0 if n ≡ 2 (mod 3) if n 6≡ 0 (mod 3) J (3) n+1 − 1 if n ≡ 0 (mod 3) + 3J (3) n j(3) n = 4n, = 2n+2j(3) nXk=0 n−3. J (3) n+1 (1.10) (1.11) (1.12) (1.13) (1.14) (1.15) (1.16) and (1.17) (1.18) and (1.19) Using standard techniques for solving recurrence relations, the auxiliary equa- tion, and its roots are given by x3 − x2 − x − 2 = 0; x = 2, and x = −1 ± i√3 2 . Note that the latter two are the complex conjugate cube roots of unity. Call them ω1 and ω2, respectively. Thus the Binet formulas can be written as J (3) n = 2 7 2n − 3 + 2i√3 21 ωn 1 − 3 − 2i√3 21 ωn 2 = j(3) n = 8 7 2n + 3 + 2i√3 7 ωn 1 + 3 − 2i√3 7 ωn 2 = 1 n (cid:17) 7(cid:16)2n+1 − V (3) n (cid:17) , 7(cid:16)2n+3 + 3V (3) 1 respectively. Here V (3) n is the sequence defined by (1.20) V (3) n = 3 + 2i√3 3 ωn 1 + 3 − 2i√3 3 ωn 2 = 2 if n ≡ 0 (mod 3) −3 if n ≡ 1 (mod 3) if n ≡ 2 (mod 3) 1 . 4 G. CERDA-MORALES Recently in [3], we have defined a new type of quaternions with the third-order Jacobsthal and third-order Jacobsthal-Lucas number components as J Q(3) n = J (3) n + J (3) n+1i + J (3) n+2j + J (3) n+3k and jQ(3) n = j(3) n + j(3) n+1i + j(3) n+2j + j(3) n+3k, respectively, where i2 = j2 = k2 = ijk = −1, and we studied the properties of these quaternions. Also, we derived the generating functions and many other identities for the third-order Jacobsthal and third-order Jacobsthal-Lucas quater- nions. In this paper, we define the dual third-order Jacobsthal quaternions and dual third-order Jacobsthal-Lucas quaternions as follows: (1.21) and (1.22) J N (3) m = J (3) m + J (3) m+1i + J (3) m+2j + J (3) m+3k (m ≥ 0) jN (3) m = j(3) m + j(3) m+1i + j(3) m+2j + j(3) m+3k (m ≥ 0), respectively. Here i2 = j2 = k2 = 0, ij = −ji = jk = −kj = ik = −ki = 0. Also, we investigated the relations between the dual third-order Jacobsthal quaternions and third-order Jacobsthal numbers. Furthermore, we gave some their quadratic properties, the Binet's formulas, d'Ocagne and Cassini-like identities for these quaternions. 2. Dual Third-order Jacobsthal Quaternions We can define dual third-order Jacobsthal quaternions by using third-order Jacobsthal numbers. The n-th third-order Jacobsthal number J (3) is defined by Eq. (1.7). Then, we can define the dual third-order Jacobsthal quaternions as follows: n (2.1) NJ = {J N (3) m = J (3) m + J (3) m+1i + J (3) m+2j + J (3) m+3k}, where J (3) m is the m-th third-order Jacobsthal number and {i, j, k} as in Eq. (1.2). Also, we can define the dual third-order Jacobsthal-Lucas quaternion as follows: (2.2) Nj = {jN (3) m = j(3) m + j(3) m+1i + j(3) m+2j + j(3) m+3k}, where j(3) m is the m-th third-order Jacobsthal-Lucas number. DUAL THIRD-ORDER JACOBSTHAL QUATERNIONS 5 Then, the addition and subtraction of the dual third-order Jacobsthal and dual J N (3) m ± jN (3) third-order Jacobsthal-Lucas quaternions is defined by m+2j + J (3) m+2j + j(3) m+1 ± j(3) m+1i + J (3) m+1i + j(3) m ) + (J (3) m = (J (3) ± (j(3) = (J (3) + (J (3) m + J (3) m + j(3) m ± j(3) m+3 ± j(3) m+3)k (2.3) m+3k) m+3k) m+1)i + (J (3) m+2 ± j(3) m+2)j and the multiplication of the dual third-order Jacobsthal and dual third-order Jacobsthal-Lucas quaternions is defined by J N (3) m jN (3) m (2.4) = J (3) = (J (3) m+1i + J (3) m + J (3) m j(3) + (J (3) m + (J (3) m j(3) m j(3) m+3 + J (3) m+2j + J (3) m+1 + J (3) m+3j(3) m )k. m+3k)(j(3) m + j(3) m j(3) m+1i + j(3) m+2 + J (3) m+2j + j(3) m+2j(3) m )j m+3k) m+1j(3) m )i + (J (3) Now, the scalar and the vector part of the J N (3) m which is the m-th term of the dual third-order Jacobsthal sequence {J N (3) m }m≥0 are denoted by (2.5) (S J N (3) m , V J N (3) m ) = (J (3) m , J (3) m+1i + J (3) m+2j + J (3) m+3k). Thus, the dual third-order Jacobsthal J N (3) relation (2.4) is defined by m is given by S J N (3) m + V J N (3) m . Then, (2.6) J N (3) m jN (3) m = S J N (3) m S (3) m jN + S J N (3) m V (3) m jN + S (3) m jN V J N . (3) m The conjugate of dual third-order Jacobsthal quaternion J N (3) J N defined as (3) m and it is J N m+2j − J (3) m+1i − J (3) m − J (3) m = J (3) m+3k. The norm of J N (3) m is denoted by m is (3) (2.7) kJ N (3) m k2 = J N (3) m J N (3) m = J N (3) m J N (3) Then, we give the following theorem using statements (2.1) and (2.2). m (cid:17)2 m =(cid:16)J (3) . m and J N (3) Theorem 2.1. Let J (3) sthal sequence {J (3) {J N (3) m }m≥0, respectively. m be the m-th terms of the third-order Jacob- m }m≥0 and the dual third-order Jacobsthal quaternion sequence In this case, for m ≥ 0 we can give the following relations: (2.8) (2.9) 2J N (3) m + J N (3) m+1 + J N (3) m+2 = J N (3) m+3, J N (3) m − J N (3) m+1i − J N (3) m+2j − J N (3) m+3k = J (3) m , 6 (2.10) G. CERDA-MORALES kJ N (3) m k2 + kJ N (3) m+1k2 + kJ N (3) m+2k2 = 1 7  3 · 22(m+1)(1 + 4i + 8j + 16k) −2m+2U N (3) m m (i + 2j + 4k) −2m+3U (3) +2(1 − i − j + 2k) 7(cid:16)V (3) ,  m+2(cid:17). where U N (3) m = U (3) m + U (3) m+1i + U (3) m+2j + U (3) m+3k and U (3) m = 1 m+1 + 3V (3) Proof. (2.8): By the equations J N (3) we get m = J (3) m + J (3) m+1i + J (3) m+2j + J (3) m+3k and (1.7), 2J N (3) m + J N (3) = (2J (3) + (J (3) + (J (3) m+2 m+1 + J N (3) m + 2J (3) m+1 + J (3) m+2 + J (3) m + J (3) m+2 + J (3) = (2J (3) + (2J (3) = J (3) m+3k) m+1i + 2J (3) m+2i + J (3) m+3i + J (3) m+1 + J (3) m+2j + 2J (3) m+3j + J (3) m+4j + J (3) m+2) + (2J (3) m+4k) m+5k) m+1 + J (3) m+3 + J (3) m+4)j + (2J (3) m+3 + J (3) m+3 + J (3) m+4i + J (3) m+5j + J (3) m+6k = J N (3) m+3. m+2 + J (3) m+3)i m+4 + J (3) m+5)k) (2.9): By using J N (3) m in the Eq. (1.7) and conditions (1.2), we get J N (3) m − J N (3) m+1i − J N (3) m+2j − J N (3) m + J (3) m+3k = J (3) − (J (3) − (J (3) − (J (3) = J (3) m . m+1 + J (3) m+2 + J (3) m+3 + J (3) m+1i + J (3) m+2j + J (3) m+3k m+3j + J (3) m+4j + J (3) m+5j + J (3) m+2i + J (3) m+3i + J (3) m+4i + J (3) m+4k)i m+5k)j m+6k)k (2.10): By using Eqs. (2.4) and (1.18), we get (2.11) m (cid:17)2 (cid:16)J N (3) m (cid:17)2 =(cid:16)J (3) + 2J (3) m J (3) m+1i + 2J (3) m J (3) m+2j + 2J (3) m J (3) m+3k, DUAL THIRD-ORDER JACOBSTHAL QUATERNIONS 7 and (2.12) where U (3) obtain = = = 1 1 1 m = 1 m+1 + 3V (3) m + 2V (3) m+1 + 4V (3) m+2(cid:17)2(cid:19) +(cid:16)2m+3 − V (3) m+2) + 14(cid:17) m (cid:17)2 +(cid:16)J (3) (cid:16)J (3) m+1(cid:17)2 +(cid:16)J (3) m+2(cid:17)2 49(cid:18)(cid:16)2m+1 − V (3) m (cid:17)2 +(cid:16)2m+2 − V (3) m+1(cid:17)2 49(cid:16)21 · 22(m+1) − 2m+2(V (3) m + 2(cid:17) , 7(cid:16)3 · 22(m+1) − 2m+2U (3) 7(cid:16)V (3) m (cid:17)2 +(cid:16)J N (3) m+1(cid:17)2 (cid:16)J N (3) m (cid:17)2 =(cid:16)J (3) + 2(cid:16)J (3) + 2(cid:16)J (3) + 2(cid:16)J (3) 7 3 · 22(m+1)(1 + 4i + 8j + 16k) − 2m+2U N (3) m (i + 2j + 4k) + 2(1 − i − j + 2k) ! , m+2(cid:17). Finally, from the Eqs. (2.11) and (2.12), we +(cid:16)J N (3) m+2(cid:17)2 +(cid:16)J (3) m+1(cid:17)2 +(cid:16)J (3) m+2(cid:17)2 m+1 + J (3) m+2 + J (3) m+3 + J (3) m+3(cid:17) i m+4(cid:17) j m+5(cid:17) k m+1J (3) m+1J (3) m+1J (3) m+2 + J (3) m+3 + J (3) m+4 + J (3) m+2J (3) m+2J (3) m+2J (3) m J (3) m J (3) m J (3) = 1 m −2m+3U (3) m+1i + U (3) where U N (3) m = U (3) m + U (3) m+2j + U (3) m+3k. (cid:3) Theorem 2.2. Let J N (3) m and jN (3) Jacobsthal quaternion sequence {J N (3) Lucas quaternion sequence {jN (3) m be the m-th terms of the dual third-order m }m≥0 and the dual third-order Jacobsthal- m }m≥0, respectively. The following relations are satisfied (2.13) (2.14) (2.15) m (cid:17)2 (cid:16)jN (3) jN (3) m+3 − 3J N (3) m+3 = 2jN (3) m , jN (3) m+1 + jN (3) m = 3J N (3) m+2, + 3J N (3) m+3jN (3) m+3 = 4m+3(1 + 4i + 8j + 16k). 8 G. CERDA-MORALES Proof. (2.13): From identities between third-order Jacobsthal number and third- order Jacobsthal-Lucas number (1.10) and (2.3), it follows that jN (3) m+3 − 3J N (3) m+3 = j(3) m+4i + j(3) m+5j + j(3) m+3 + j(3) − 3(J (3) = (j(3) + (j(3) m+3 + J (3) m+3 − 3J (3) m+5 − 3J (3) m + 2j(3) = 2j(3) m+4i + J (3) m+3) + (j(3) m+5)j + (j(3) m+6k m+5j + J (3) m+4 − 3J (3) m+4)i m+6 − 3J (3) m+3k m+2j + 2j(3) m+1i + 2j(3) m+6k) m+6)k = 2jN (3) m . The proof of (2.14) is similar to (2.13), using the identity (1.13). (2.15): Now, using Eqs. (2.4), (2.11) and (1.15), we get m (cid:17)2 (cid:16)jN (3) +3J N (3) m+3jN (3) m+3 m j(3) m+3k m+1i + 2j(3) m+3j(3) m+5j(3) m j(3) m+4 + J (3) m+3)j + 3(J (3) m+2j + 2j(3) m+4j(3) m+3)i m+3j(3) m+6 + J (3) m+6j(3) m+3)k m (cid:17)2 =(cid:16)j(3) + 3J (3) + 3(J (3) m+3j(3) m+3j(3) m (cid:17)2 =(cid:16)j(3) + 2j(3) m j(3) m+3 + 3(J (3) m+5 + J (3) + 3J (3) m+3j(3) m+1 + 3(J (3) m+2 + 3(J (3) m+3 + 3(J (3) m+3 + (2j(3) + (2j(3) m j(3) m j(3) m j(3) m+3j(3) m+3j(3) m+3j(3) + (2j(3) = 4m+3(1 + 4i + 8j + 16k). m+4 + J (3) m+5 + J (3) m+6 + J (3) m+4j(3) m+5j(3) m+6j(3) m+3))i m+3))j m+3))k Theorem 2.3. Let J N (3) quaternion sequence {J N (3) m be the m-th term of the dual third-order Jacobsthal m }m≥0. Then, we have the following identity (cid:3) (2.16) mXs=0 J N (3) s = J N (3) m+1 − 1 21(cid:16)7(1 + i + 4j + 7k) − 4V N (3) m+1 + V N (3) m (cid:17) , where V N (3) m = V (3) m + V (3) m+1i + V (3) m+2j + V (3) m+3k. DUAL THIRD-ORDER JACOBSTHAL QUATERNIONS 9 Proof. Since (2.17) mXs=0 (see [4]), we get s = J (3) J (3) m+1 − 1 21(cid:16)7 − 4V (3) m+1 + V (3) m (cid:17) =( J (3) m+1 if m 6≡ 0 (mod 3) J (3) m+1 − 1 if m ≡ 0 (mod 3) J N (3) s = J (3) s + i mXs=0 J (3) s = J (3) m+1 − mXs=0 +(cid:18)J (3) +(cid:18)J (3) +(cid:18)J (3) m+2 − m+3 − m+4 − 1 1 J (3) s + j J (3) s + k m+1 + V (3) m+2 + V (3) m+3Xs=3 m+2Xs=2 m+1Xs=1 m (cid:17) 21(cid:16)7 − 4V (3) m+1(cid:17)(cid:19) i 21(cid:16)7 − 4V (3) m+2(cid:17)(cid:19) j 21(cid:16)28 − 4V (3) m+3(cid:17)(cid:19) k 21(cid:16)49 − 4V (3) 21(cid:16)7(1 + i + 4j + 7k) − 4V N (3) m+3 + V (3) m+4 + V (3) 1 1 1 = J N (3) m+1 − m + V (3) where V N (3) m = V (3) m+1i + V (3) m+2j + V (3) m+3k. m+1 + V N (3) m (cid:17) , (cid:3) Theorem 2.4. Let J N (3) m and jN (3) Jacobsthal quaternion sequence {J N (3) Lucas quaternion sequence {jN (3) jN (3) m J N (3) m − jN m be the m-th terms of the dual third-order m }m≥0 and the dual third-order Jacobsthal- m }m≥0, respectively. Then, we have (3) m J N (3) m ), m = 2(J (3) m J N (3) m jN (3) jN (3) m J N (3) m + jN (3) m J N (3) m = 2j(3) m − j(3) m J N (3) m . (2.18) (2.19) Proof. (2.18): By the Eqs. (2.4) and (2.6), we get jN (3) m J N (3) m − jN (3) m J N (3) m = (j(3) − (j(3) = 2J (3) m + j(3) m − j(3) m (j(3) m jN (3) = 2(J (3) m+1i + j(3) m+1i − j(3) m+2j + j(3) m+2j − j(3) m+3k)(J (3) m+3k)(J (3) m+3k) − 2j(3) m − J (3) m + J (3) m (J (3) m+1i + j(3) m − j(3) m+2j + j(3) m J N (3) m ). m+1i − J (3) m+1i + J (3) m+2j − J (3) m+2j + J (3) m+3k) m+3k) m+1i + J (3) m+2j + J (3) m+3k) 10 (2.19): G. CERDA-MORALES (3) m jN (3) m J N (3) = (j(3) + (j(3) = j(3) m + jN (3) m J N m + j(3) m+1i + j(3) m − j(3) m+1i − j(3) m J (3) m + (j(3) m J (3) m J (3) m+3 + j(3) m J (3) m − (j(3) m J (3) m+3 + j(3) m J N (3) m . m+2j + j(3) m+2j − j(3) m+1 + j(3) m+3J (3) m )k m J (3) m+3J (3) m+1 + j(3) m )k + (j(3) + j(3) − (j(3) = 2j(3) m+3k)(J (3) m+3k)(J (3) m + J (3) m − J (3) m J (3) m+1i + J (3) m+1i − J (3) m+2 + j(3) m+2j + J (3) m+2j − J (3) m+2J (3) m )j m+1J (3) m )i + (j(3) m+3k) m+3k) m+1J (3) m )i − (j(3) m J (3) m+2 + j(3) m+2J (3) m )j Theorem 2.5 (Binet's Formulas). Let J N (3) third-order Jacobsthal quaternion sequence {J N (3) Jacobsthal-Lucas quaternion sequence {jN (3) m be m-th terms of the dual m }m≥0 and the dual third-order m }m≥0, respectively. For m ≥ 0, the m and jN (3) Binet's formulas for these quaternions are as follows: (cid:3) (2.20) and (2.21) J N (3) m = = jN (3) m = 8 7 1 3 + 2i√3 2 7 1 21 2mα − 7(cid:16)2m+1α − V N (3) m (cid:17) ω1ωm 1 − 2mα + ω1ωm 1 + 3 + 2i√3 7 3 − 2i√3 21 ω2ωm 2 3 − 2i√3 7 ω2ωm 2 = 7(cid:16)2m+3α + 3V N (3) m (cid:17) , m = 2 − 3i + j + 2k if m ≡ 0 (mod 3) −3 + i + 2j − 3k if m ≡ 1 (mod 3) 1 + 2i − 3j + k if m ≡ 2 (mod 3) , respectively, where V (3) m is the sequence defined by (2.22) V N (3) α = 1 + 2i + 4j + 8k and ω1,2 = 1 + ω1,2i + ω2 1,2j + k. DUAL THIRD-ORDER JACOBSTHAL QUATERNIONS 11 Proof. Repeated use of (1.18) in (2.1) enables one to write for α = 1 + 2i + 4j + 8k and ω1,2 = 1 + ω1,2i + ω2 1,2j + k (2.23) J N (3) m = J (3) 2m+1 − 2m+2 − 7 m + J (3) 2 7 2n − + 2 + 2 + 2 α2m+1 − 1 7 7 7 2m+3 − = = m+1i + J (3) 3 + 2i√3 m+3k m+2j + J (3) 3 − 2i√3 ωn 2 21 ωn 1 − 3 + 2i√3 21 3 + 2i√3 21 3 + 2i√3 21 ωm+1 1 ωm+2 1 ωm+3 1 21 3 + 2i√3 21 ω1ωm 1 − 21 − − 3 − 2i√3 3 − 2i√3 3 − 2i√3 − 3 − 2i√3 21 21 21 ωm+1 2 ! i 2 ! j 2 ! k ωm+3 ωm+2 ω2ωm 2 and similarly making use of (1.19) in (2.2) yields (2.24) = jN (3) m = j(3) m+1i + j(3) m+2j + j(3) m+3k m + j(3) 1 7 α2m+3 + 3 + 2i√3 7 ω1ωm 1 + 3 − 2i√3 7 ω2ωm 2 . The formulas in (2.23) and (2.24) are called as Binet's formulas for the dual third- order Jacobsthal and dual third-order Jacobsthal-Lucas quaternions, respectively. Using notation in (2.22), we obtain the results (2.20) and (2.21). (cid:3) Theorem 2.6 (D'Ocagne-like Identity). Let J N (3) dual third-order Jacobsthal quaternion sequence {J N (3) n ≥ m ≥ 0, the d'Ocagne identities for J N (3) 7 α(2n+1U N (3) m+1 − J N (3) (2.25) J N (3) n+1J N (3) n J N (3) m = m is as follows: 1 m be the m-th terms of the m }m≥0. In this case, for m+1 − 2m+1U N (3) n+1) +(1 − i − j + 2k)U (3) n−m ! , (2.26) (cid:16)J N (3) m+1(cid:17)2 − J N (3) m+2J N (3) m = 1 7 2m+1α(2U N (3) +(1 − i − j + 2k) m+1 − U N (3) m+2) ! , where U N (3) m+1 = 1 7 (2V N (3) m − V N (3) m+1) and α = 1 + 2i + 4j + 8k. 12 G. CERDA-MORALES Proof. (2.25): Using Eqs. (2.20) and (2.22), we get (2.27) J N (3) n J N (3) m+1 − J N (3) n+1J N (3) m = = = 1 1 m ) ! n )(2m+2α − V N (3) m+1) n+1)(2m+1α − V N (3) 49 (2n+1α − V N (3) −(2n+2α − V N (3) 49 −2n+1αV N (3) m+1 − 2m+2V N (3) m + 2m+1V N (3) 7 (1 + 2i + 4j + 8k)(2n+1U N (3) +(1 − i − j + 2k)U (3) n α + V N (3) n V N (3) m+1 n+1α − V N (3) n+1V N (3) m+1 − 2m+1U N (3) n+1) +2n+2αV N (3) m ! ! , n−m 1 m+1 = 1 where U N (3) n = m + 1 in Eq. (2.27), we obtain for m ≥ 0, (2.28) m − V N (3) 7 (2V N (3) m+1) and V N (3) (cid:16)J N (3) m+1(cid:17)2 −J N (3) m+2J N (3) m = 1 7 (1 + 2i + 4j + 8k)2m+1(2U N (3) +(1 − i − j + 2k) m+1 − U N (3) m+2) ! . (cid:3) m as in (2.22). In particular, if We will give an example in which we check in a particular case the Cassini-like identity for dual third-order Jacobsthal quaternions. Example 2.7. Let J N (3) 1 , J N (3) 2 cobsthal quaternions such that J N (3) J N (3) 2 = 1 + 2i + 5j + 9k and J N (3) 0 , J N (3) and J N (3) 3 0 = i + j + 2k, J N (3) be the dual third-order Ja- 1 = 1 + i + 2j + 5k, 3 = 2 + 5i + 9j + 18k. In this case, (cid:16)J N (3) 1 (cid:17)2 − J N (3) 2 J N (3) 0 = (1 + i + 2j + 5k)2 − (1 + 2i + 5j + 9k)(i + j + 2k) = (1 + 2i + 4j + 10k) − (i + j + 2k) = 1 + i + 3j + 8k = 1 7 2(1 + 2i + 4j + 8k)(2U N (3) +(1 − i − j + 2k) 1 − U N (3) 2 ) ! and (cid:16)J N (3) 2 (cid:17)2 − J N (3) 3 J N (3) 1 = (1 + 2i + 5j + 9k)2 − (2 + 5i + 9j + 18k)(1 + i + 2j + 5k) = (1 + 4i + 10j + 18k) − (2 + 7i + 13j + 28k) = −1 − 3i − 3j − 10k = 1 7 4(1 + 2i + 4j + 8k)(2U N (3) +(1 − i − j + 2k) 2 − U N (3) 3 ) ! DUAL THIRD-ORDER JACOBSTHAL QUATERNIONS 13 3. Conclusions There are two differences between the dual third-order Jacobsthal and the dual coefficient third-order Jacobsthal quaternions. The first one is as follows: the dual coefficient third-order Jacobsthal quaternionic units are i2 = j2 = k2 = ijk = −1 whereas the dual third-order Jacobsthal quaternionic units are i2 = j2 = k2 = 0, ij = −ji = jk = −kj = ik = −ki = 0. The second one is as follows: the elements of the dual coefficient third-order Jacobsthal quaternion are J (3) m+1 (ε2 = 0, ε 6= 0) whereas the elements of the dual third-order Jacobsthal quaternions are m-th third-order Jacobsthal number J (3) m . m + εJ (3) References [1] E. Ata and Y. Yaylı, Dual quaternions and dual projective spaces, Chaos Solitons and Fractals, 40(3), (2009) 1255 -- 1263. [2] P. Barry, Triangle geometry and Jacobsthal numbers, Irish Math. Soc. Bull. 51 (2003), 45 -- 57. [3] G. Cerda-Morales, Identities for Third Order Jacobsthal Quaternions, Advances in Applied Clifford Algebras 27(2) (2017), 1043 -- 1053. [4] Ch.K. Cook and M.R. Bacon, Some identities for Jacobsthal and Jacobsthal-Lucas num- bers satisfying higher order recurrence relations, Annales Mathematicae et Informaticae 41 (2013), 27 -- 39. [5] S. Halıcı On Fibonacci quaternions, Adv. Appl. Clifford Algebr. 22(2), (2012), 321 -- 327. [6] C.J. Harman, Complex Fibonacci numbers, The Fibonacci Quarterly. 19 (1), (1981), 82 -- 86. [7] A.F. Horadam, A generalized Fibonacci sequence, The American Mathematical Monthly. 68 (5), (1961), 455 -- 459. [8] A.F. Horadam, Complex Fibonacci numbers and Fibonacci quaternions, American Math. Monthly. 70 (3), (1963), 289 -- 291. [9] A.F. Horadam, Quaternion recurrence relations, Ulam Quarterly. 2 (2), (1993), 23 -- 33. [10] A. F. Horadam, Jacobsthal representation numbers, Fibonacci Quarterly 34 (1996), 40 -- 54. [11] A.L. Iakin, Generalized quaternions of higher order, The Fibonacci Quarterly. 15 (4), (1977), 343 -- 346. [12] A.L. Iakin, Generalized quaternions with quaternion components, The Fibonacci Quarterly. 15 (4), (1977), 350 -- 352. [13] M.R. Iyer, A note on Fibonacci quaternions, The Fibonacci Quarterly. 7 (3), (1969), 225 -- 229. [14] M.R. Iyer, Some results on Fibonacci quaternions, The Fibonacci Quarterly. 7 (3), (1969), 201 -- 210. [15] V. Majernik, Quaternion formulation of the Galilean space-time transformation, Acta Phy. Slovaca., 56 (1), (2006), 9 -- 14. [16] K.S. Nurkan and I.A. Guven, Dual Fibonacci quaternions. Adv. Appl. Clifford Algebras. 25(2), (2015), 403 -- 414. [17] S. Yuce and F. Torunbalcı Aydın, A new aspect of dual Fibonacci quaternions, Adv. Appl. Clifford Algebras. 26(2), (2016), 873 -- 884.
1606.06701	1	1606	2016-06-21T18:46:42	On non-commutative rank and tensor rank	[ "math.RA" ]	We study the relationship between the commutative and the non-commutative rank of a linear matrix. We give examples that show that the ratio of the two ranks comes arbitrarily close to 2. Such examples can be used for giving lower bounds for the border rank of a given tensor. Landsberg used such techniques to give nontrivial equations for the tensors of border rank at most $2m-3$ in $K^m\otimes K^m\otimes K^m$ if $m$ is even. He also gave such equations for tensors of border rank at most $2m-5$ in $K^m\otimes K^m\otimes K^m$ if $m$ is odd. Using concavity of tensor blow-ups we show non-trivial equations for tensors of border rank $2m-4$ in $K^m \otimes K^m \otimes K^m$ for odd $m$ for any field $K$ of characteristic 0. We also give another proof of the regularity lemma by Ivanyos, Qiao and Subrahmanyam.	math.RA	math	ON NON-COMMUTATIVE RANK AND TENSOR RANK HARM DERKSEN AND VISU MAKAM Abstract. We study the relationship between the commutative and the non-commutative rank of a linear matrix. We give examples that show that the ratio of the two ranks comes arbitrarily close to 2. Such examples can be used for giving lower bounds for the border rank of a given tensor. Landsberg used such techniques to give nontrivial equations for the tensors of border rank at most 2m − 3 in K m ⊗ K m ⊗ K m if m is even. He also gave such equations for tensors of border rank at most 2m − 5 in K m ⊗ K m ⊗ K m if m is odd. Using concavity of tensor blow-ups we show non-trivial equations for tensors of border rank 2m− 4 in K m ⊗ K m ⊗ K m for odd m for any field K of characteristic 0. We also give another proof of the regularity lemma by Ivanyos, Qiao and Subrahmanyam. 1. Introduction 1.1. Linear matrices. We fix an infinite field K. A linear matrix (or matrix pencil) A over K is a matrix whose coefficients are linear expressions in variables t1, t2, . . . , tm. The commu- tative rank crk(A) is defined as the rank over the commutative function field K(t1, t2, . . . , tm). On the other hand, one can take t1, t2, . . . , tm to be independent non-commuting variables and compute the rank over the free skew field K (< t1, t2, . . . , tm >) . This is called the non- commutative rank, and is denoted ncrk(A). See [11] for details. Over a skew field, the rank of a matrix is defined as the (left) row rank, which is equal to the (right) column rank. In particular, adding left multiplied rows to other rows and right multiplied columns to other columns does not affect the rank. Note also that a square matrix is invertible over a skew field if and only if it is of full rank. We refer to [12, 15] for several equivalent definitions of non-commutative rank. The following well-known example shows that the commutative and non-commutative rank of a linear matrix may differ. Example 1.1. This example is based on skew symmetric matrices. Let 6 1 0 2 n u J 1 2 ] . A R h t a m [ 1 v 1 0 7 6 0 . 6 0 6 1 : v i X r a A =  0 1 −1 0 0 t1 t2 0 1 −1 0 −t1 −t2 0 . 0 [t2, t1] , 0 0 It is easy to see that crk(A) = 2. However, over the free skew field K (< t1, t2 >) , we can do row and column transformations to transform A to which is clearly of full rank since [t2, t1] = t2t1−t1t2 is non-zero over the skew field K (< t1, t2 >) . The first author was supported by NSF grant DMS-1302032 and the second author was supported by NSF grant DMS-1361789. 1  d 0 0 c 0 −b c 0 −a 0 −a 0 d 0 0 −b  and  0 −b −d 0 0 −c −d 0 0 0 −d b a c b 0  are linear matrices that have commutative rank 3, and non-commutative rank 4. Here a, b, c, d are variables. There are very few examples known where there is a discrepancy between commutative and non-commutative rank. One well known family is based on skew symmetric matrices for odd m (see Example 1.1). In [7, Section 4], a more interesting family is given based on the Cayley-Hamilton theorem. 1.2. Linear subspaces of matrices. Linear matrices can also be studied from the point of view of linear subspaces and their tensor blow-ups. We denote by Matp,q the space of p × q matrices over the field K. Definition 1.3. We define the rank of a linear subspace X ⊆ Matp,q to be the maximal rank among its members, and denote it by rk(X ). The set of matrices in X having this maximal rank is Zariski open. Since the underlying field K is infinite, we can relate the commutative rank of a linear matrix to the rank of a linear subspace (see [11, Lemma 3.1]) as follows: Lemma 1.4 ([11]). Let A = X0 + t1X1 + t2X2 + · · · + tmXm be a linear matrix and let X = span(X0, X1, X2, . . . , Xm). Then crk(A) = rk(X ). Example 1.2. In [9], several low rank examples can be found. For example, they show that It turns out that non-commutative rank can be understood from the perspective of linear subspaces. In order to do this, we require the notion of tensor blow-ups for linear subspaces. Definition 1.5. Let X be a linear subspace of Matk,n. We define its (p, q) tensor blow-up X {p,q} to be X ⊗ Matp,q =nXi Xi ⊗ Ti(cid:12)(cid:12)(cid:12)Xi ∈ X , Ti ∈ Matp,qo viewed as a linear subspace of Matkp,nq. We will write X {d} = X {d,d}. Given X ∈ X having rank r = rk(X ), observe that X ⊗ I ∈ X {d} has rank rd. Hence rk(X {d}) is at least d · rk(X ). Example 1.6. Let X denote the linear subspace of skew symmetric 3 × 3 matrices. The rank of this subspace is 2. Let X1, X2, X3 be any basis of X . Domokos showed in [8] that 1 0(cid:19) + X3 ⊗(cid:18)0 0 0 1(cid:19) ∈ X {2} has full rank 6, which is larger than X1 ⊗(cid:18)1 0 0 0(cid:19) + X2 ⊗(cid:18)0 1 2 · 2 = 4. The above example shows rk(X {d}) could be larger than d · rk(X ). However, Ivanyos, Qiao and Subrahmanyam showed that there is a very strong restriction on the possible ranks of tensor blow-ups. They proved the following regularity lemma ([15, Lemma 11 and Remark 10]). 2 Proposition 1.7 ([15], Regularity Lemma). If X is a linear subspace of matrices, then rk(X {d}) is a multiple of d. In [15], this is proved by giving an algorithm that takes a matrix of rank ≥ rd + 1 in X {d} and produces another matrix in X {d} of rank ≥ (r + 1)d. Analyzing their algorithm (see [15, 16]), they show that it runs in polynomial time. We give another proof of the regularity lemma using Amitsur's universal division algebras. While our proof is less constructive than the original proof, it is conceptually more satisfying. The following characterization of non-commutative rank in terms of ranks of tensor blow- ups appears in [15]. Lemma 1.8 ([15]). Let A = X0 + t1X1 + t2X2 + · · · + tmXm be a linear matrix and let X = span(X0, X1, X2, . . . , Xm). Then: ncrk(A) = max rk(X {d}) rk(X {d}) . d d = lim d→∞ d We observe that we can define non-commutative ranks for linear subspaces of matrices, and do so. Definition 1.9. For a linear subspace of matrices X , define rk(X {d}) = lim d→∞ rk(X {d}) . d ncrk(X ) = max rk(X {d}) and lim d→∞ d d rk(X {d}) d The existence of max d follows from the following partial in- creasing property of ranks of blow-ups (see [15, Corollary 12]), along with the aforementioned regularity lemma. Lemma 1.10 ([15]). Let X be a linear subspace of matrices, Then for d ≥ n, weakly increasing. rk(X {d}) d is Observe that ncrk(X ) ≥ rk(X ), since rk(X {d}) is at least d · rk(X ). On the other hand, it is shown in [11] that ncrk(X ) ≤ 2 rk(X ). In fact, modifying their argument, one can show that the ratio must be < 2. Proposition 1.11. For any linear subspace X , we have ncrk(X ) < 2 rk(X ). The authors of [15] comment that the statement of Lemma 1.10 is perhaps true for d < n as well, but are unable to prove it. Making use of the proposition above, we are able to show: Proposition 1.12. Let X be a linear subspace of matrices, Then for d ≥ n is weakly increasing. 2 − 1, rk(X {d}) d We also show that the increasing property need not hold for small values of d. We combine a construction of Bergman in [4] of a rational identity satisfied by 3 × 3 matrices but not by 2×2 matrices, with a construction of Hrubes and Wigderson in [14] to give a counterexample. Proposition 1.13. There exists a linear subspace X such that rk(X {2}) 2 rk(X {3}) . 3 > 3 In fact, using an existential result in [4], we can show: Theorem 1.14. For any n, m ∈ Z>0 such that n ∤ m, there is a linear subspace X such that rk(X {n}) n rk(X {m}) . > m 1.3. Ratio of non-commutative rank to commutative rank. We have seen in Propo- sition 1.11 that the ratio of non-commutative rank to commutative rank is bounded by 2. In [11], Fortin and Reutanauer comment that the bound of 2 is perhaps not sharp, and suggest that 3/2 might be the right bound based on the examples available. In this paper, we give a family of examples for which the ratio comes arbitrarily close to 2, thus showing that the sharpest possible bound is actually 2! We have seen above that linear matrices can be studied through the linear subspaces of matrices they define, and we give our examples in the language of linear subspaces of matrices. Given an element in v ∈ K n, we can define a map Lv : Vi K n → Vi+1 K n given by x 7→ v ∧ x. This gives a linear map L : K n → Hom(Vi K n,Vi+1 K n). The image is a linear Theorem 1.15. Let X (p, 2p+1) denote the image of L : K 2p+1 → Hom(Vp K 2p+1,Vp+1 K 2p+1). subspace. We have ncrk(X (p, 2p + 1)) rk(X (p, 2p + 1)) = 2p + 1 p + 1 . The linear subspaces in the theorem above provide a family of examples for which the ratio approaches 2. In fact, these linear subspaces give rise to more examples which have a discrepancy between the commutative and non-commutative rank. Corollary 1.16. Let X (i, n) denote the image of the linear map L : K n → (Vi K n,Vi+1 K n). Then (1) ncrk(X (i, n)) is full; (2) If i 6= 0, n − 1, then rk(X (i, n)) is not full. 1.4. Applications to tensor rank. Let char K = 0. A simple dimension counting ar- gument shows that there is an open dense subset containing tensors of border rank ≥ n3/(3n − 2). However, the polynomial equations that are satisfied by tensors of border rank ≤ m get very complicated as m becomes large. In [21], Landsberg gives non-trivial equations for tensors in K m ⊗ K m ⊗ K m of border rank 2m− 3 when m is even, and 2m− 5 when m is odd. We show that Landsberg's methods are essentially an investigation of ranks of tensor blow- ups for the linear subspaces in Theorem 1.15. In [7], we showed a concavity property for the ranks of tensor blow-ups. For odd m, using the concavity property, we give non-trivial equations for tensors of border rank 2m − 4 in K m ⊗ K m ⊗ K m (see Theorem 6.1). We use our equations to prove that certain explicit tensors have border rank ≥ 2m − 3 (see Proposition 6.3). 2. Tensor blow-ups and Universal division algebras 2.1. Tensor blow-ups. Let A = X0 + t1X1 + t2X2 + · · · + tmXm be an p × q linear matrix. The (i, j)th entry of A is a linear function in the indeterminates tk's with coefficients in K. 4 In fact if ck ∈ K is the (i, j)th entry of Xk, then the (i, j)th entry of A is given by Ai,j = c0+c1t1+· · ·+cmtm. Suppose S1, . . . , Sm are d×d matrices, then X0⊗I+X1⊗S1+· · ·+Xm⊗Sm is a p × q block matrix and the size of each block is d × d. Moreover, the (i, j)th block is c0I + c1S1 + · · · + cmSm. In effect X0 ⊗ I + X1 ⊗ S1 + · · · + Xm ⊗ Sm is simply the block matrix obtained by substituting Sk for tk in the linear matrix A. Hence, we make the following definition. Definition 2.1. Let A = X0 + t1X1 + · · · + tmXm be a linear matrix. For any m-tuple of matrices S = (S1, S2, . . . , Sm), we define A(S) = X0 ⊗ I + X1 ⊗ S1 + · · · + Xm ⊗ Sm. j,k1 ≤ j, k ≤ d, i ∈ Z>0} be a collection of j,k]. j,k}] denote the j,k for 1 ≤ j, k ≤ d, i ∈ Z>0. Observe that for each i, the 2.2. The ring of generic matrices. Let {ti independent commuting indeterminates. For each i ∈ Z>0, define the d× d matrix Ti = [ti By a generic matrix, we will refer to a matrix of indeterminates. Let K[{ti polynomial ring in the variables ti generic matrix Ti lies in Matd,d(K[{ti Definition 2.2. The ring of generic matrices Rd ⊆ Matd,d(K[{ti algebra generated by {Tii ∈ Z>0}. Lemma 2.3. Let A = X0+t1X1+· · ·+tmXm be a linear matrix, and let X = span(X1, X2, . . . , Xm). Then, we have j,k}]) is defined as the sub- j,k}]). rk(X {d}) = rk(X0 ⊗ I + X1 ⊗ T1 + · · · + Xm ⊗ Tm), where Ti is a generic matrix for i = 1, 2, . . . , m. Proof. We first show rk(X {d}) ≤ rk(X0⊗I +X1⊗T1+· · ·+Xm⊗Tm). For S = (S0, S1, . . . , Sm) in a non-empty Zariski open subset of Matm+1 d,d , we have rk(X0⊗S0+X1⊗S1+· · ·+Xm⊗Sm) = rk(X {d}) = r, since K is infinite. There is an S in this Zariski open subset for which S0 is invertible. For such an S, observe that rk(X0 ⊗ I + X1 ⊗ S−1 0 Sm) = r. The corresponding r × r minor in X0 ⊗ I + X1 ⊗ T1 + · · · + Xm ⊗ Tm must be a non-zero polynomial. The other inequality, i.e., rk(X {d}) ≥ rk(X0⊗I +X1⊗T1+· · ·+Xm⊗Tm) is straightforward. 0 S1 + · · · + Xm ⊗ S−1 (cid:3) Example 2.4. Let A = X0 + t1X1 + t2X2 = T1, T2, we have: 0 1 −1 0 −t1 −t2 t1 t2 0  . Then for generic matrices  0 I −I 0 −T1 −T2 T1 T2 0  −→ 5 0 I −I 0 0 0 0 0 [T2, T1]  . A(T1, T2) = X0 ⊗ I + X1 ⊗ T1 + X2 ⊗ T2 = 0 I −I 0 −T1 −T2 T1 T2 0  . As observed in the introduction, we can do row and column transformations to transform Hence the rk A(T1, T2) = 2d + rk([T1, T2]). If the Ti are generic matrices of size 1 × 1, then [T1, T2] = 0, and if T1, T2 are generic matrices of size d × d for d ≥ 2, then [T1, T2] is invertible, and hence of full rank. Thus for T1, T2 generic matrices of size d × d, we have rk A(T1, T2) =(2 if d = 1, 3d if d ≥ 2. In particular, observe that rk A(T1, T2) is always a multiple of d. Using Lemma 2.3, one sees that the regularity lemma is satisfied in for the linear subspace of 3× 3 skew symmetric matrices. 2.3. Universal division algebras. Observe, as in Example 2.4, that for generic d × d matrices, the expression [T1, T2] was either identically zero, or invertible depending upon the value of d. This is a special case of a surprising general phenomenon, namely that any non-zero non-commutative rational expression in some d× d generic matrices must in fact be invertible! This follows from the fact that Amitsur's universal division algebras are division algebras. We describe these universal division algebras. Recall the ring of generic matrices Rd ⊆ Matd,d(K[{ti j,k}]). Let Zd denote the center of Rd, and let its field of fractions be Qd. The following result is due to Amitsur (see [1, 2, 3]). One can also find it in standard texts (for example [6, Section 7.2]). Theorem 2.5 (Amitsur). UD(d) := Qd⊗Zd Rd is a division algebra and is called a universal division algebra of degree d. Proof. Posner proved that the central quotient of a prime PI-ring is a simple algebra (see [24]). The ring Rd satisfies a polynomial identity, namely the Amitsur-Levitzki polynomial. Amitsur showed (see [1, Theorem 4]) that Rd is in fact an integral domain, and in particular a prime ring. Hence its central quotient UD(d) is a simple algebra. By the Wedderburn- Artin theorem (see [18, Section 3.13]), it must be a matrix algebra over a division algebra, i.e., UD(d) ∼= Matr,r(D) for some division algebra D and r ∈ Z>0. Further, since Rd is an integral domain, UD(d) has no nilpotents. Hence UD(d) ∼= Mat1,1(D) ∼= D. (cid:3) Note that UD(d) ⊆ Matd,d(K({ti as we mentioned in the introduction. Proof of Theorem 1.7. Let X0, X1, X2, . . . , Xm span the linear subspace X ⊆ Matp,q, and set A = X0 + t1X1 + · · · + tmXm. Then by Lemma 2.3, we have j,k})). We now give another proof of the regularity lemma, rk(X {d}) = rk(X0 ⊗ I + X1 ⊗ T1 + · · · + Xm ⊗ Tm). A(T1, T2, . . . , Tm) = X0⊗I + X1⊗T1 +· · ·+ Xm⊗Tm can be viewed as a p×q block matrix whose blocks are linear expressions in the generic matrices Ti, and in particular elements of UD(d), a division algebra. By row and column operations in UD(d) ⊆ Matd,d(K({ti j,k})), we can make the transformation: (X0 ⊗ I + X1 ⊗ T1 + · · · + Xm ⊗ Tm) −→ 6 I I . . . 0 0 0   Since each I denotes a d × d identity matrix, it is clear that rk(A(T1, . . . , Tm)) = rk(X {d}) (cid:3) is a multiple of d. 3. Combinatorics of ranks of tensor blow-ups 3.1. Weakly increasing property of blow-ups. We modify the argument used by Fortin and Reutenauer (see [11]) to prove Proposition 1.11. Proof of Proposition 1.11. Let r be the smallest non-negative integer such that we have a linear subspace X ⊆ Matp,q of rank r for some p, q, such that ncrk(X ) = 2r. We have r > 1 since rk(X ) = 1 implies ncrk(X ) = 1 (see [11, Remark 1] and [7, Lemma 2.9]). We use a result of Flanders (see [10, Lemma 1]) to see that X is equivalent to a subspace C B(cid:19) with C of size r × r (see also [11, Corollary 2]). Since ncrk(X ) = 2r, of the form(cid:18)A 0 we must have ncrk(A) ≥ r, since we must have at least 2r linearly independent rows. But A has only r columns, and hence ncrk(A) = r. A similar argument considering columns shows that ncrk(B) = r. We have rk(A), rk(B) ≥ r/2 because the ratio is at most 2. We cannot have rk(A) = r/2 or rk(B) = r/2 as that would violate the minimality of r. Thus rk(A), rk(B) > r/2. However, this means that rk(X ) ≥ rk(A) + rk(B) > r. (cid:3) To prove Proposition 1.12, we improve the proof of Proposition 1.10 given in [15] by making use of Proposition 1.11. Proof of Proposition 1.12. Suppose rk(X {d})/d = r. Choose a basis X1, . . . , Xm of X . There exist T1, . . . , Tm ∈ Matd,d such thatPi Xi ⊗ Ti ∈ X {d} has rank rd. Choose a1, . . . , am ∈ K such thatPi aiXi ∈ X has rank equal to rk(X ). Then let eTi ∈ Matd+1,d+1 be given by  . eTi = i=1 Xi ⊗ Ti) + rk(Pm i=1 Xi ⊗eTi) ≥ rk(Pm i=1 aiXi) = rd + rk(X ) Then it is easy to see that rk(Pm Furthermore, we have 0 ... 0 ai 0 . . . 0 Ti rk(X ) > 1 2 ncrk(X ) ≥ 1 2r. In the above, the first inequality follows from Proposition 1.11, and the second follows from the Definition 1.9. Hence, we have rk(X {d+1}) ≥ rk(Pi Xi ⊗ eTi) > rd + 1 2r. 2 − 1 ≥ r Since d ≥ n 2r ≥ (r − 1)(d + 1). Now, by the regularity lemma (Proposition 1.7) we must have rk(X {d+1})/(d + 1) ≥ r = rk(X {d})/d. (cid:3) The following result follows from the concavity of tensor blow-ups (see [7, Proposition 2.10]). 2 − 1, we have rk(X {d+1}) > rd + 1 7 Proposition 3.1. Let n ≥ 2 and suppose d ≥ n − 1, and assume rk(X {d+1})/(d + 1) = r. Then rk(X {d})/d ≥ r. Combining Proposition 1.12 with the above proposition, we get Corollary 3.2. Let n ≥ 2. Then rk(X {d})/d is constant for d ≥ n − 1. 3.2. Rational identities. In [4, 5], Bergman proved a number of remarkable results on rational relations and rational identities in division rings. In particular, he came up with an explicit construction of a rational expression which is an identity on 3 × 3 matrices, but invertible on general 2 × 2 matrices. We introduce some notation. Let Y ′ denote the commutator [X, Y ], and let δ(Y ) denote (Y 2)′[(Y −1)′]−1. In [4], Bergman proves the following result (see also [6, Theorem 7.4.3]). Theorem 3.3 ([4]). Let n = 2 or 3. For X, Y ∈ Matn,n(K), we have: ψ = δ(Y ′)δ(Y ′′)[(δ(Y ′′)−1)′][(δ(Y ′′′)−1)′] =(1 0 if n = 3, if n = 2. Corollary 3.4. The rational expression ψ − 1 is an identity for 3 × 3 matrices, but is invertible for general choices of 2 × 2 matrices. Bergman also showed the existence of such rational functions more generally. Let E(d) be the set of rational expressions that can be evaluated on generic d × d matrices. Theorem 3.5 ([4]). Assume n, m ∈ Z>0. Then E(n) ⊆ E(m) if and only if n m. 3.3. Non-commutative arithmetic circuits with division. A non-commutative arith- metic circuit is a directed acyclic graph, whose vertices are called gates. Gates of in-degree 0 are elements of K or variables ti. The other allowed gates are inverse, addition and multi- plication gates of in-degrees 1, 2 and 2 respectively. The edges going into an multiplication gate are labelled left and right to indicate the order of multiplication. A formula is a cir- cuit, where every node has out-degree at most 1. The number of gates in a circuit is called It is easy to observe that each output gate its size. Let Φ be a circuit in m variables. p,p. In the process of evaluation, if the input of an inverse gate of a circuit Φ computes a rational expression. We denote by bΦ(T ) the evaluation of Φ at is not invertible, then bΦ(T ) is undefined. Φ is called a correct circuit if bΦ(T ) is defined for T = (T1, T2, . . . , Tm) ∈ Matm some T . For further details, we refer to [14]. In [14], Hrubes and Wigderson reduce non-commutative rational identity testing to de- ciding the invertibility of linear matrices. A deterministic algorithm over Q for deciding the invertibility of linear matrices was given by Garg, Gurvits, Oliviera and Wigderson in [12] by analyzing an algorithm of Gurvits in [13]. In [7], we give bounds for the size of matrices required to detect invertibility, and this gives another proof that over Q, invertibility of lin- ear matrices can be decided in polynomial time. Moreover, the bounds in [7] immediately show that invertibility of a linear matrix can be decided in randomized polynomial time over arbitrary characteristic. In [16], Ivanyos, Qiao and Subrahmanyam use the bounds in [7] to give a deterministic algorithm that works over arbitrary characteristic. Given a non-commutative formula of size n, Hrubes and Wigderson construct a family of linear matrices Au for each gate u of the formula. We refer to [14, Theorem 2.5] for details. 8 We are content to remark that these matrices can be constructed explicitly in time which is polynomial in n. We recall [14, Propostion 7.1]. Proposition 3.6 ([14]). Let R be a ring which contains K in its center. For a formula Φ, and a1, a2, . . . , am ∈ R, the following are equivalent: (1) bΦ(a1, a2, . . . , am) is defined. (2) For every gate u, the Au(a1, a2, . . . , am) is invertible. Now, we can put Bergman's results together with Hrubes and Wigderson's results to give a proof of Proposition 1.13. Proof of Proposition 1.13. Let Φ be the non-commutative formula that computes the ratio- nal expression (ψ−1)−1. By the construction of Hrubes and Wigderson mentioned above, we where the Ti are generic 2 × 2 matrices by Theorem 3.3. Thus, the Au(T ) is invertible for all u. On the other hand, if the Ti are generic 3 × 3 matrices, then once again by Theorem 3.3, t1X1 +· · · + tmXm and let X = span(X0, X1, . . . , Xm). Then, using Lemma 2.3, we conclude have linear matrices Au for each gate u. Observe thatbΦ(T ) is defined for T = (T1, T2, . . . , Tm) bΦ(T ) is not defined. Thus, for some u, Au is not invertible. For this u, write Au = X0 + rk(X {2}) rk(X {3}) > 2 3 . (cid:3) For the general case, we use Theorem 3.5. Proof of Theorem 1.14. If n ∤ m, then there exists r ∈ E(n) such that r /∈ E(m). Let Φ be the non-commutative formula that computes r. The argument in the proof of Proposition 1.13 applied to Φ gives the required conclusion. (cid:3) 4. Ratio of non-commutative and commutative ranks We assume char K = 0 for this section. For v ∈ K n, define Lv :Vp(K n) →Vp+1(K n) by w 7→ v ∧ w. Let e1, e2, . . . , en be a basis for K n. Note that a basis forVp(K n) is given by {ei1 ∧ ei2 ∧ · · · ∧ eip1 ≤ i1 < i2 < · · · < ip ≤ n}. Let A(p, n) denote the linear matrix given by t1Le1 + t2Le2 + · · · + tnLen. Lemma 4.1. For a particular choice of basis, the linear matrix A(p, n) has the form  tnI A(p − 1, n − 1) A(p, n − 1) 0  Proof. Let A = {(ei1 ∧ ei2 ∧ · · · ∧ eip−1) ∧ en 1 ≤ i1 < · · · < ip−1 ≤ n − 1}, and B = {ei1 ∧ ei2 ∧ · · · ∧ eip 1 ≤ i1 < · · · < ip ≤ n − 1}. 9 C = {(ei1 ∧ ei2 ∧ · · · ∧ eip) ∧ en 1 ≤ i1 < · · · < ip ≤ n − 1}, and D = {ei1 ∧ ei2 ∧ · · · ∧ eip+1 1 ≤ i1 < · · · < ip+1 ≤ n − 1}. Then clearly A ∪ B is a basis forVp(K n). Similarly, let Then C ∪ D is a basis forVp+1(K n). It is easy to see that there Len : B → C is a bijection. Now, order the basis elements forVp(K n) by taking the basis vectors from B first, and then from A. ForVp+1(K n), order the basis vectors by taking the basis vectors from C first, and then from D. Within the basis vectors of C, we order them in the same order as the vectors from B via the aformentioned bijection given by Le1. (cid:3) Remark 4.2. The description in [20, Section 4] is the same as the one above. See also [22, 23, Section 2]. Corollary 4.3. If A(p, n) has full column rank, then so does A(p − 1, n − 1). Similarly, if A(p, n) has full row rank, then so does A(p, n − 1). Proof. Assume without loss of generality that v = en. By the above choice of basis Len = Corollary 4.4. For any non-zero v ∈ K n, rk(Lv) =(cid:0)n−1 p (cid:1). (cid:20) I 0 0 0 (cid:21). Hence p (cid:19). rk(Lv) = B = C =(cid:18)n − 1 Corollary 4.5. We have crk(A(p, n)) = rk(X (p, n)) =(cid:0)n−1 p (cid:1). p+1(cid:0)2p p(cid:1) =(cid:0)2p+1 Proof. This follows from Lemma 1.4. (cid:3) (cid:3) 2p+1 Observe that in order to prove Theorem 1.15, it suffices to prove that ncrk(A(p, 2p + 1)) = p (cid:1). Further note that dimVp(K 2p+1) = dimVp+1(K 2p+1) =(cid:0)2p+1 words, we want to show that the non-commutative rank of A(p, 2p + 1) is full. We will use a result of Landsberg in [21]. Proposition 4.6 ([21]). Let e1, e2, . . . , e2p+1 be a basis for C2p+1. For −p ≤ r ≤ p, let Sr be the (p + 1) × (p + 1) matrix such that p (cid:1). In other Sr(j, k) =(1 0 if j = k + r, otherwise. . Then, we have that Le1 ⊗ S−p + Le2 ⊗ S−p+1 + · · · + Le2p+1 ⊗ Sp is invertible. The set {S−p, S−p+1, . . . , Sp} is a basis for the space of Toeplitz matrices of size (p + 1) × (p + 1). Any other basis of the Toeplitz matrices would work as well. Corollary 4.7. The linear matrix A(p, 2p + 1) has full non-commutative rank. Proof of Theorem 1.15. Observe that the linear subspace associated to A(p, 2p+1) is X (p, 2p+ 1). Hence Corollary 4.5 and Corollary 4.7 give us the required conclusion. (cid:3) 10 Proof of Corollary 1.16. To prove (1), consider A(i, n). If i < n/2, then let k = n − 2i − 1. The linear matrix A(i+k, n+k) has full column rank by Corollary 4.7, since 2(i+k)+1 = n+k. By repeated application of Corollary 4.3, we conclude that A(i, n) has full column rank. Since i < n/2, the matrix A(i, n) has more rows than columns, and hence has full non-commutative rank. If i ≥ n/2, then we observe that A(i, 2i + 1) has full non-commutative rank. Once again by repeated application of Corollary 4.3, we conclude that A(i, n) has full row rank. Since A(i, n) has more columns than rows, it has full non-commutative rank. Finally, observe that the linear subspace defined by A(i, n) is the linear subspace X (i, n). To prove (2), use Corollary 4.5. (cid:3) 5. Lower bounds on border rank Definition 5.1. For a tensor T ∈ K a1 ⊗ K a2 ⊗ · · · ⊗ K al, we define its tensor rank trk(T ) as the smallest m such that T can be written as a sum of m pure tensors. Let Zm denote the set of tensors with tensor rank ≤ m. The set Zm need not be a Zariski closed subset, and we can consider its Zariski closure Zm. Definition 5.2. For a tensor T ∈ K a1 ⊗ K a2 ⊗ · · · ⊗ K al, we define its border rank brk(T ) as the smallest m such that T ∈ Zm. Tensor rank and border rank have been studied extensively, especially with respect to the matrix multiplication tensor. See [19] for details. K a ⊗ K b ⊗ K c, we can write T = Pi si ⊗ Xi, with si ∈ K a and Xi ∈ K b ⊗ K c. Let We consider tensor product spaces with three tensor factors. Given a tensor in T ∈ L : K a → Matp,q be a linear map, and denote the image by XL. We identify K b ⊗ K c with Matb,c and identify Matp,q ⊗ Matb,c with Matpb,qc. This gives the following map. ψL : K a ⊗ K b ⊗ K c −→ Matpb,qc Xi si ⊗ Xi 7−→Xi L(si) ⊗ Xi. Lemma 5.3. For a tensor T ∈ K a ⊗ K b ⊗ K c we have rk(ψL(T )) ≤ brk(T ) rk(XL). Proof. Let T = s ⊗ b ⊗ c be a tensor of rank 1. Then ψL(T ) = L(a) ⊗ (b ⊗ c), and hence rk(ψL(T )) ≤ rk(L(a)) ≤ rk(XL). Therefore, if we take a tensor T ∈ K a ⊗ K b ⊗ K c of rank r, then rk(ψL(T )) ≤ r rk(XL) = D. Observe that the (D + 1) × (D + 1) minors of ψL(T ) are polynomial equations that vanish all tensors of rank ≤ r, i.e they vanish on Zr. Hence these equations vanish on Zr as well. Hence if brk(T ) = r, we must have rk(ψL(T )) ≤ D = r rk(XL) = brk(T ) rk(XL). (cid:3) rk(ψL(T )) rk(XL) Remark 5.4. In particular, ψL(T ) ∈ X {p,q} ). Hence in order to get a good lower bound, it would be useful for the blow-up to have large rank, which in turn is only possible if XL has a large ratio of non-commutative rank to commutative rank. , and hence rk(ψL(T )) ≤ rk(X {b,c} is a lower bound for brk(T ). Further, observe that L L 11 Corollary 5.5. Let D = r rk(XL). Then the (D+1)×(D+1) minors of ψL(T ) give equations that are satisfied by all tensors of border rank ≤ r. Landsberg's technique (see [21]) for obtaining lower bounds for border rank is the same as the one we describe above. For any r, the above corollary gives polynomials that are satisfied by all tensors having border rank ≤ r. It follows that if these polynomials do not vanish on a tensor T , then we must have brk(T ) > r, providing a possible method for showing lower bounds for border rank. However, this method is only useful if these polynomials are non- trivial, i.e., not identically zero. The non-triviality of these equations essentially depends on the rank of the blow-up X {b,c} Lemma 5.6. One of the d × d minors of ψL(T ) is a non-trivial polynomial if and only if rk(X {b,c} Proof. Suppose rk(X {b,c} that rk(ψL(T1)) = rk(X {b,c} and hence that d × d minor is a non-trivial polynomial. , there exists T1 ∈ K a ⊗ K b ⊗ K c such ) ≥ d. Hence there is a d × d minor in ψL(T1) that is non-zero, ) ≥ d. Since im(ψL) = X {b,c} The converse follows immediately since the underlying field K is infinite. ) ≥ d. (cid:3) L . L L L L We discuss a few results that can help in finding lower bounds for the ranks of blow-ups. Lemma 5.7. For a linear subspace X , if X {d} has full rank, then X {kd} has full rank for any k ∈ Z>0. Proof. If rk(X {d}) is full, then we have somePi Xi ⊗ Si is invertible for some Xi ∈ X and Si ∈ Matd,d. Clearly, (Pi Xi ⊗ Si) ⊗ Ik is also invertible, where Ik ∈ Matk,k denotes the identity matrix. ThusPi Xi ⊗ (Si ⊗ Ik) ∈ X {kd} is invertible. In view of Corollary 5.5 and Lemma 5.6, it would be useful to show lower bounds on the rank of blow-ups. For this, the concavity properties of blow-ups that we showed in [7] will be very useful. Proposition 5.8 ([7]). For a linear subspace of matrices X , let r(p, q) = rk(X {p,q}). Then we have: (cid:3) (1) r(p, q + 1) ≥ r(p, q); (2) r(p + 1, q) ≥ r(p, q); (3) r(p, q + 1) ≥ 1 (4) r(p + 1, q) ≥ 1 In particular, this shows that that r(p, q) is increasing and concave down in either variable 2 (r(p, q) + r(p, q + 2)); 2 (r(p, q) + r(p + 2, q)). independently. 6. Border rank of tensors in K m ⊗ K m ⊗ K m for odd m Let m = 2p + 1 be a positive odd integer. Let L : K m → Hom(Vp K m,Vp+1 K m) be the linear map defined in Theorem 1.15. For, this L, we define ψL as in the previous section, i.e., ψL : K m ⊗ K m ⊗ K m −→ Mat(2p+1 Xi si ⊗ Xi 7→Xi 12 p )m p )m,(2p+1 L(si) ⊗ Xi. (D + 1) × (D + 1) minors of ψL gives a non-trivial equation for tensors in K m ⊗ K m ⊗ K m of border rank ≤ 2m − 4. p(cid:1)(2m − 4). Then at least one of the p(cid:1) by Corollary 4.5. Hence, by Corollary 5.5 and Lemma 5.6, Theorem 6.1. Let ψL be as above, and let D = (cid:0)2p Proof. Observe that rk(XL) =(cid:0)2p it suffices to show that rk(X {m} ) ≥ D + 1. By Proposition 4.6, we know that X {p+1} we have X {2p+2} properties from Proposition 5.8. Let M = dimVp K m = dimVp+1 K m =(cid:0)2p+1 has full rank as well. To find lower bounds on rk(X {2p+1} has full rank and as a consequence of Lemma 5.7, ), we use the p (cid:1), and let r(p, q) = rk(X {p,q}). Then we have r(p + 1, p + 1) = (p + 1)M, and r(2p + 2, 2p + 2) = (2p + 2)M by the above discussion. We have L L L L Further, by concavity in the second variable, we have r(p + 1, 2p + 1) ≥ r(p + 1, p + 1) ≥ (p + 1)M. r(2p + 2, 2p + 1) ≥ (2p + 1)r(2p + 2, 2p + 2) + r(2p + 2, 0) 2p + 2 (2p + 1)(2p + 2)M ≥ = (2p + 1)M. 2p + 2 Now, by concavity in the first variable, we have r(2p + 1, 2p + 1) ≥ ≥ = pr(2p + 2, 2p + 1) + r(p + 1, 2p + 1) p + 1 p(2p + 1)M + (p + 1)M p + 1 2p2 + 2p + 1 p + 1 M. Hence, we have ) L rk(X {2p+1} (cid:0)2p p(cid:1) (2p2 + 2p + 1)(cid:0)2p+1 p (cid:1) (p + 1)(cid:0)2p p(cid:1) (2p2 + 2p + 1)(2p + 1) (p + 1)(p + 1) ≥ = > 4p − 2 = 2m − 4 ) >(cid:0)2p p(cid:1)(2m − 4) as required. Thus rk(X {m} L Recall that the non-commutative rank is at most twice the commutative rank. Hence (cid:3) L ) rk(X {m} crk(XL) ≤ m · ncrk(XL) crk(XL) 13 < 2m. This alone shows that there is very little room for improvement for the lower bounds we obtain using this method. Remark 6.2. For m = 5 i.e., p = 2, Landsberg shows that in fact X {m} has full rank, thus giving non-trivial equations for tensors of border rank 8. Experimental evidence shows that in fact this is true for p = 3 and 4 as well, suggesting that it is perhaps true for all p, which would give non-trivial equations for tensors of border rank 2m − 2. L mPi=1 mPi=1 p )m,(2p+1 Proposition 6.3. The tensor T = ≥ 2m − 3. In K m ⊗ K m ⊗ K m, Landsberg gives explicit tensors having border rank ≥ 2m − 2 (resp. 2m− 4) when m is even (resp. odd) (see [21]). For m odd, we can give explicit tensors whose border rank is ≥ 2m − 3. Let m = 2p + 1 be odd, and let Sr ∈ Matp+1,p+1 for −p ≤ r ≤ p be as in Proposition 4.6. obtained from Qr by removing the last column and last row of Qr. Identifying Mat2p+1,2p+1 with K m ⊗ K m, we can consider the tensor T = e1, e2, . . . , em is the standard basis for K m. For each r, consider Qr = Sr ⊕ Sr ∈ Mat2p+2,2p+2, and let eQr ∈ Mat2p+1,2p+1 be the matrix ei ⊗ eQi−p−1 ∈ K m ⊗ K m ⊗ K m, where ei ⊗ eQi−p−1 ∈ K m ⊗ K m ⊗ K m has border rank Proof. Let L : K m → Hom(Vp K m,Vp+1 K m) be the linear map defined in Theorem 1.15. We p (cid:1) rows from A. Hence, we have ψL(T ) is obtained by removing(cid:0)2p+1 p (cid:1) columns and(cid:0)2p+1 p (cid:19) rk(ψL(T )) ≥ rk(A) − 2(cid:18)2p + 1 p (cid:19) p (cid:19) − 2(cid:18)2p + 1 = (m + 1)(cid:18)2p + 1 p (cid:19)(2p). =(cid:18)2p + 1 have ψL(T ) = Mat(2p+1 L(ei)⊗ Qi−p−1 ∈ p )(m+1) has full rank by Proposition 4.6 and Lemma 5.7. Observe that L(ei)⊗eQi−p−1 ∈ Mat(2p+1 p )m. Observe that A = mPi=1 mPi=1 p )(m+1),(2p+1 Thus, we have (cid:3) rk(XL) brk(T ) ≥ (cid:0)2p+1 p (cid:1)(2p) = (cid:0)2p+1 p (cid:1)(2p) (cid:0)2p p(cid:1) > 2m − 4. Hence brk(T ) ≥ 2m − 3 as required. 14 References [1] S. A. Amitsur, The T -ideals of the free ring, J. London Math. Soc. 30 (1955), 470-475. [2] S. A. Amitsur, On central division algebras, Israel J. Math. 12 (1972), 408-420. [3] S. A. Amitsur, The generic division rings, Israel J. Math. 17 (1974), 241-247. [4] G. Bergman, Rational relations and rational identities in division rings. I, Journal of Algebra 43, (1976), 252-266. [5] G. Bergman, Rational relations and rational identities in division rings. II, Journal of Algebra 43, (1976), 267-297. [6] P. M. Cohn Skew fields. Theory of general division rings, Encyclopedia of Mathematics and its Appli- cations 57, Cambridge University Press, Cambridge, 1995. [7] H. Derksen and V. Makam Polynomial degree bounds for matrix semi-invariants, arXiv:1512.03393 [math.RT], 2015. [8] M. Domokos, Relative invariants of 3 × 3 matrix triples, Linear and Multilinear Algebra 47 (2000), [9] D. Eisenbud and J. Harris, Vector spaces of matrices of low rank, Adv. in Math. 70, (1988), 135-155. 175-190. [10] H. Flanders, On spaces of linear transformations with bounded rank, J. London Math. Soc. 37 (1962), 10-16. [11] M. Fortin and C. Reutenauer, Commutative/non-commutative rank of linear matrices and subspaces of matrices of low rank, S´em. Lothar. Combin. 52:B52f, 2004. [12] A. Garg, L. Gurvits, R. Oliveira and A. Widgerson, A deterministic polynomial time algorithm for non-commutative rational identity testing, arXiv:1511.03730, 2015. [13] L. Gurvits, Classical complexity and quantum entanglement, J. Comp. Syst. Sci. 69 (2004), 448-484. [14] P. Hrubes and A. Wigderson, Non-commutative arithmetic circuits with division, ITCS'14, Princeton, NJ, USA, 2014. [15] G. Ivanyos, Y. Qiao and K. V. Subrahmanyam, Non-commutative Edmonds' problem and matrix semi- invariants arXiv:1508.00690 [cs.DS], 2015. [16] G. Ivanyos, Y. Qiao and K. V. Subrahmanyam, Constructive noncommutative rank computation in deterministic polynomial time over fields of arbitrary characteristics, tt arXiv:1512.03531 [cs.CC], 2015. [17] N. Jacobson, PI-algebras, Lecture Notes in Mathematics, Vol. 441, Springer-Verlag, Berlin-New York, 1975. [18] N. Jacobson, Basic algebra. II, Second Edition, W. H. Freeman and Company, New York, 1989. [19] J. M. Landsberg, Tensors: Geometry and Applications, Graduate Studies in Mathematics 128, Ameri- can Math. Soc., Providence, RI, 2012. [20] J. M. Landsberg, New lower bounds for the rank of matrix multiplication, SIAM Journal on Comput- ing 43 (2014), 144-149. 3 least 2m − 2, J. Pure Appl. Algebra 219 (2015), 3677-3684. Linear Algebra Appl. 438 (2013), 4500-4509 [21] J. M. Landsberg, Nontriviality of equations and explicit tensors in Cm ⊗ Cm ⊗ Cm of border rank at [22] A. Massarenti and E. Ravioli, The rank of n × n matrix multiplication is at least 3n2 − 2√2n 2 − 3n, [23] A. Massarenti and E. Ravioli, Corrigendum to "The rank of n × n matrix multiplication is at least 2 − 3n" [Linear Algebra Appl. 438 (11) (2013) 4500–4509], Linear Algebra Appl. 445 [24] E. C. Posner, Prime rings satisfying a polynomial identity, Proc. Amer. Math. Soc. 11 (1960), 180-184. 3 3n2 − 2√2n (2014), 369-371. 15
1202.5819	1	1202	2012-02-27T03:18:56	On the exponent of spinor groups	[ "math.RA", "math.RT" ]	In this paper, we show that all the exponents of degree greater than 2 of spinor groups divide the Dynkin index 2.	math.RA	math	ON THE EXPONENT OF SPINOR GROUPS SANGHOON BAEK 1. Introduction Let G be a split simple simply connected group of rank n over a field F . Fix a maximal split torus T of G and a Borel subgroup B containing T . We denote by W the Weyl group of G with respect to T . Let Λ be the weight lattice of G (hence, T ∗ = Λ). We denote by ω1, · · · , ωn the fundamental weights of Λ. We let IK := Ker(Z[Λ] → Z) and ICH := Ker(S ∗(Λ) → Z) be the augmentation ideals, where Z[Λ] → Z (respectively, S ∗(Λ) → Z) is the map from the group ring Z[Λ] (respectively, the symmetric algebra) of Λ to the ring of integers by sending eλ to 1 (respectively, any element of positive degree to 0). For any i ≥ 0, we consider the ring homomorphism φ(i) : Z[Λ] → Z[Λ]/I i+1 K → S ∗(Λ)/I i+1 CH → Si(Λ), where the first and the last maps are projections and the middle map sends ePn j=1 aj ωj to Qn j=1(1 − ωj)−aj . The ith-exponent of G (denoted by τi), as intro- duced in [1], is the gcd of all nonnegative integers Ni satisfying Ni · (I W CH)(i) ⊆ φ(i)(I W K ), K := hZ[Λ]W ∩ IKi (respectively, I W CH := hS ∗(Λ)W ∩ ICHi) denotes the where I W CH )(i) = W -invariant augmentation ideal of Z[Λ] (respectively, S ∗(Λ)) and (I W I W CH ∩ Si(Λ). Informally, these numbers τi measure how far is the ring S ∗(Λ)W from being a polynomial ring in basic invariants. For any i ≤ 4, it was shown that the ith-exponent of G divides the Dynkin in- dex in [1] and this integer was used to estimate the torsion of the Grothendieck gamma filtration and the Chow groups of E/B, where E/B denotes the twisted form of the variety of Borel subgroups G/B for a G-torsor E. In this paper, we show that all the remaining exponents of spinor groups divide the Dynkin index 2. Acknowledgments. The work has been partially supported from the Fields Institute and from Zainoulline's NSERC Discovery grant 385795-2010. 1 2 S. BAEK 2. Exponent Let G be Spin2n+1 (n ≥ 3) or Spin2n (n ≥ 4). The fundamental weights are defined by ω1 = e1, ω2 = e1 + e2, · · · , ωn−1 = e1 + · · · + en−1, ωn = ω1 = e1, ω2 = e1 + e2, · · · , ωn−1 = e1 + · · · + en−1 − en 2 , ωn = e1 + · · · + en 2 e1 + · · · + en , , 2 respectively, where the canonical basis of Rn is denoted by ei (1 ≤ i ≤ n). For 1 ≤ i ≤ n, let (1) q2i := e2i 1 + · · · + e2i n be the basic invariants of the group G, i.e., be algebraically independent homo- geneous generators of S ∗(Λ)W as a Q-algebra (see [2, §3.5 and §3.12]), together with (2) if G = Spin2n. q′ n := e1 · · · en For any λ ∈ Λ, we denote by W (λ) the W -orbit of λ. For any finite set A of weights, we denote −A the set of opposite weights. The Weyl groups of Spin2n+1 and Spin2n are (Z/2Z)n ⋊Sn and (Z/2Z)n−1 ⋊ Sn, respectively. Hence, by the action of these Weyl groups, one has the following decomposition of W -orbits: if G = Spin2n+1 (respectively, G = Spin2n), then for any 1 ≤ k ≤ n − 1 (respectively, 1 ≤ k ≤ n − 2) (3) W (ωk) = W+(ωk) ∪ −W+(ωk), where W+(ωk) = {ei1 ± · · · ± eik }i1<···<ik. If n is even, then the W -orbits of the last two fundamental weights of Spin2n are given by (4) W (ωn−1) = W+(ωn−1) ∪ −W+(ωn−1) and W (ωn) = W+(ωn) ∪ −W+(ωn), where W+(ωn−1) (respectively, W+(ωn)) is the subset of W (ωn−1) (respectively, W (ωn)) containing elements of the positive sign of e1. For any λ = Pn Pn j=1 ajωm the following lemma: j . For example, λ(0) = Pn j=1 ajωj ∈ Λ and any integer m ≥ 0, we set λ(m) = j=1 aj and λ(1) = λ. We shall need Lemma 2.1. (i) If G is Spin2n+1 (respectively, Spin2n), then for any odd integer p, any nonnegative integers m1, · · · , mp and, any 1 ≤ k ≤ n − 1 (respectively, any 1 ≤ k ≤ n − 2), we have Xλ∈W (ωk) λ(m1) · · · λ(mp) = 0. (ii) If G is Spin2n with odd n, then for any even integer p and any nonneg- ative integers m1, · · · , mp, we have Xλ∈W (ωn) λ(m1) · · · λ(mp) = Xλ∈W (ωn−1) λ(m1) · · · λ(mp). ON THE EXPONENT OF SPINOR GROUPS 3 (iii) If G is Spin2n, then for any odd integer p < n and any nonnegative integers m1, · · · , mp, we have Xλ∈W (ωn) λ(m1) · · · λ(mp) = Xλ∈W (ωn−1) λ(m1) · · · λ(mp) = 0. Proof. (i) It follows from (3) that Xλ∈W (ωk) λ(m1) · · · λ(mp) = Xλ∈W+(ωk) = Xλ∈W+(ωk) = 0. λ(m1) · · · λ(mp) + Xλ∈−W+(ωk) λ(m1) · · · λ(mp) − Xλ∈W+(ωk) λ(m1) · · · λ(mp) λ(m1) · · · λ(mp) (ii) If G is Spin2n with odd n, then we have W (ωn) = −W (ωn−1). Hence, the result immediately follows from the assumption that p is even. (iii) If n is even, then the result follows from (4) by the same argument In general, note that for any λi1, · · · , λip ∈ W+(ω1) as in the proof of (i). the term λi1(m1) · · · λip(mp)/2p (respectively, -λi1(m1) · · · λip(mp)/2p) appears 2n−2 times (respectively, 2n−2) in both sums in (iii). (cid:3) Let p be an even integer and q ≥ 2 an integer. For any nonnegative integers m1, · · · , mp, we define Λ(p, q)(m1, · · · , mp) := X λj1(m1) · · · λjp(mp), where the sum ranges over all different λi1, · · · , λiq ∈ W+(ω1) and all λi1, · · · , λip ∈ {λi1, · · · , λiq} such that the numbers of λi1, · · · , λiq appearing in λi1, · · · , λip are all nonnegative even solutions of x1 + · · · + xq = p. If p < 2q, then we set Λ(p, q)(m1, · · · , mp) = 0. Given m1, · · · , mp, we simply write Λ(p, q) for Λ(p, q)(m1, · · · , mp). For instance, Λ(4, 2) is the sum of λj1(m1)λj2(m2)λj3(m3)λj4(m4) for all j1, j2, j3, j4 ∈ {i, j} and all 1 ≤ i 6= j ≤ n such that two i's and two j's appear in j1, j2, j3, j4. Example 2.2. We observe that (5) (x1 + x2)(x′ 1 + x′ 2) + (x1 − x2)(x′ 1 − x′ 2) = 2(x1x′ 1 + x2x′ 2). If G is Spin2n+1 or Spin2n, then by (3) and (5) we have XW+(ω2) λ(m1)λ(m2) = 2(n − 1) XW+(ω1) λ(m1)λ(m2) for any nonnegative integers m1 and m2 as we have (n − 1) choices of such pairs in the left hand side of (5) from W+(ω2), which implies that XW (ω2) λ(m1)λ(m2) = 2(n − 1) XW (ω1) λ(m1) · · · λ(m2), 4 S. BAEK (cf. [1, Lemma 5.1(ii)]). For any even p ≥ 4, we apply the same argument with the expansion of (x1 + x2) · · · (x(p) 2 ). Then, we have 2 ) + (x1 − x2) · · · (x(p) 1 − x(p) 1 + x(p) XW+(ω2) λ(m1) · · · λ(mp) = 2(n − 1) XW+(ω1) λ(m1) · · · λ(mp) + 2Λ(p, 2), which implies that XW (ω2) λ(m1) · · · λ(mp) = 2(n − 1) XW (ω1) λ(m1) · · · λ(mp) + 22Λ(p, 2). We generalize Example 2.2 to any ωk as follows. Lemma 2.3. If G is Spin2n+1 (respectively, Spin2n), then for any 1 ≤ k ≤ n − 1 (respectively, 1 ≤ k ≤ n − 2), any even p, and any nonnegative integers m1, · · · mp we have k Xj=2 2k(cid:18)n − j k − j(cid:19)Λ(p, j). λ(m1) · · · λ(mp)+ λ(m1) · · · λ(mp) = 2k−1(cid:18)n − 1 k − 1(cid:19) XW (ω1) XW (ωk) Proof. For any λ ∈ W (ω1), there are 2k(cid:0)n−1 ing λ in W (ωk), thus we have the term 2k−1(cid:0)n−1 PW (ωk) λ(m1) · · · λ(mp). of PW (ωk) λ(m1) · · · λ(mp), where λi1, · · · , λip ∈ W (ω1), then by the action of Weyl group this term vanishes in PW (ωk) λ(m1) · · · λ(mp). Hence, the remain- ing terms in PW (ωk) λ(m1) · · · λ(mp) are a linear combination of Λ(p, j) for all 2 ≤ j ≤ k such that p ≥ 2k. As each term Λ(p, j) appears 2k(cid:0)n−j k−j(cid:1) times in PW (ωk) λ(m1) · · · λ(mp), the result follows. k−1(cid:1) choices of the element contain- k−1(cid:1)PW (ω1) λ(m1) · · · λ(mp) in If an element λ ∈ W (ω1) appears odd times in a term λi1(m1) · · · λip(mp) For any λ ∈ Λ, we denote by ρ(λ) the sum of all elements eµ ∈ Z[Λ] over all elements µ of W (λ). Let i! · φ(i)(eλ) = λi + Si for any i ≥ 1, where Si is the (cid:3) sum of remaining terms in i! · φ(i)(eλ) and λ = P ajωj, aj ∈ Z. Hence, for any fundamental weight ωk we have (6) i! · φ(i)(ρ(ωk)) = XW (ωk) λi + XW (ωk) Si. We view i! · φ(i)(eλ) as a polynomial in variables λ, λ(m1), · · · , λ(mj) for some nonnegative integers m1, · · · , mj. Let Ti be the sum of monomials in Si whose degrees are even. If G is Spin2n+1 (respectively, Spin2n), then by Lemma 2.1(i) the equation (6) reduces to (7) i! · φ(i)(ρ(ωk)) = XW (ωk) λi + XW (ωk) Ti. for any 1 ≤ k ≤ n − 1 (respectively 1 ≤ k ≤ n − 2). ON THE EXPONENT OF SPINOR GROUPS 5 Given p and q, we define Ω(p, q) := X Λ(p, q)(m1, · · · , mp), where the sum ranges over all m1, · · · , mp which appear in all monomials of Ti. Example 2.4. (i) If G is Spin2n+1 or Spin2n and i = 4, then by (7) and Lemma 2.3 we have 4!φ(4)(ρ(ω1)) = XW (ω1) 4!φ(4)(ρ(ω2)) = XW (ω2) = XW (ω2) T4, λ4 + XW (ω1) λ4 + XW (ω2) λ4 + 2(n − 1) XW (ω1) T4 T4, which implies that 4!(φ(4)(ρ(ω2)) − 2(n − 1)φ(4)(ρ(ω1))) = XW (ω2) λ4 − 2(n − 1) XW (ω1) λ4. By Lemma 2.3, the right-hand side of the above equation is equal to 4Λ(4, 2) = 4 · 4! 2!2! Xi<j e2 i e2 j . Hence, we have φ(4)(ρ(ω2)) − 2(n − 1)φ(4)(ρ(ω1)) = Xi<j i e2 e2 j . (ii) If G is Spin2n+1 (n ≥ 4) or Spin2n (n ≥ 5) and i = 6, then by (7) and Lemma 2.3 we have 6!φ(6)(ρ(ω1)) = XW (ω1) 6!φ(6)(ρ(ω2)) = XW (ω2) 6!φ(6)(ρ(ω3)) = XW (ω3) which implies that T6, λ6 + XW (ω1) λ6 + 2(n − 1) XW (ω1) 2 (cid:19) XW (ω1) λ6 + 4(cid:18)n − 1 T6 + 4Ω(4, 2), T6 + 8(n − 2)Ω(4, 2), φ(6)(ρ(ω3)) − 2(n − 2)φ(6)(ρ(ω2)) + 2(n − 1)(n − 2)φ(6)(ρ(ω1)) = Xi<j<k e2 i e2 j e2 k. Lemma 2.5. (i) If G is Spin2n, then we have XW (ωn) λn − XW (ωn−1) λn = n!e1 · · · en. 6 S. BAEK (ii) If G is Spin2n, then for any 1 ≤ p ≤ n − 1 and any nonnegative integers m1, · · · , mp we have XW (ωn) λ(m1) · · · λ(mp) = XW (ωn−1) λ(m1) · · · λ(mp). Proof. (i) First, assume that n ≥ 4 is even. We show that XW+(ωn) λn − XW+(ωn−1) λn = (n!/2)e1 · · · en. As W+(ωn) = W+(ωn−1) = 2n−2, we have (n!/2n)2n−2e1 · · · en − (−(n!/2n)2n−2e1 · · · en) = (n!/2)e1 · · · en in PW (ωn) λn − PW (ωn−1) λn. If one of the exponents i1, · · · , in in ei1 1 · · · ein n (except the case i1 = · · · = in = 1) is odd, then from the orbits W+(ωn) and W+(ωn−1) this monomial vanishes in each sum of PW+(ωn) λn −PW+(ωn−1) λn. Otherwise, the terms 2n−2Pn mn = 1 are in both PW+(ωn) λn and PW+(ωn−1) λn. Now, we assume that n ≥ 4 is odd. As W (ωn) = W (ωn−1) = 2n−1, we j , Λ(n, 2) · · · , Λ(n, n/2) with m1 = · · · = j=1 en have (n!/2n)2n−1e1 · · · en − (−(n!/2n)2n−1e1 · · · en) = n!e1 · · · en in PW (ωn) λn −PW (ωn−1) λn. By the same argument, if one of the exponents i1, · · · , in in ei1 monomial vanishes in each sum of PW (ωn) λn −PW (ωn−1) λn. This completes n (except the case i1 = · · · = in = 1) is odd, then this the proof of (i). 1 · · · ein (ii) By Lemma 2.1(ii)(iii), it is enough to consider the case where both n and tion of Weyl group, any term λi1(m1) · · · λip(mp), where an element λ ∈ W (ω1) p are even. For any p and any n ≥ p+2, we have 2n−2(PW+(ω1) λ(m1) · · · λ(mp)) in both PW+(ωn) λ(m1) · · · λ(mp) and PW+(ωn−1) λ(m1) · · · λ(mp). By the ac- appears odd times in either PW+(ωn) λ(m1) · · · λ(mp) − 2n−2(PW+(ω1) λ(m1) · · · λ(mp)) orPW+(ωn−1) λ(m1) · · · λ(mp)−2n−2(PW+(ω1) λ(m1) · · · λ(mp)), van- ishes. As each term of Λ(p, 2), · · · , Λ(p, p/2) appears in bothPW+(ωn) λ(m1) · · · λ(mp) and PW+(ωn−1) λ(m1) · · · λ(mp), this completes the proof. (cid:3) Theorem 2.6. If G is Spin2n+1 (respectively, Spin2n), then for any i ≥ 3 and any n ≥ [i/2] + 1 (respectively, n ≥ [i/2] + 2) the exponent τi divides the Dynkin index τ2 = 2. Proof. As B2 = C2 and D3 = A3, we have 1 = τ3 2 by [1, Theorem 5.4]. If G is Spin2n for any n ≥ 4, then by Lemma 2.5(i)(ii) we have n = φ(n)(ρ(ωn)) − φ(n)(ρ(ωn−1)), q′ ON THE EXPONENT OF SPINOR GROUPS 7 which implies that the invariant q′ φ(n). As there are no invariants of odd degree except q′ n is in the ideal generated by the image of n, we have τ2i+1 τ2i for all i ≥ 1. Therefore, it suffices to show that τ2i τ2 for any i ≥ 2. By Lemma 2.3 together with the same argument as in Example 2.4 we have (8) φ(2i)(ρ(ωi)) + i−1 Xj=1 ajφ(2i)(ρ(ωi−j)) = Xj1<···<ji j1 · · · e2 e2 ji, where the integers a1, · · · , ai−1 satisfy i−2 Xj=k (cid:16) 2j+1(cid:18)n − 1 − k j − k (cid:19)aj+1(cid:17) + 2i(cid:18)n − 1 − k i − 1 − k(cid:19) = 0, for 0 ≤ k ≤ i − 2. Let pi be the right-hand side of (8). Then this equation implies that pi is in the image of φ(2i). We show that the invariant q2i is in the ideal φ(2i)(I W K ) for any i ≥ 2. We proceed by induction on i. As q2 = φ(2)(ρ(ω1)), the case i = 2 is obvious. By Newton's identities we have (9) (−1)i−1q2i = ipi − i−1 Xj=1 (−1)j−1pi−1−jq2j with p0 = 1. By the induction hypothesis, the sum of (9) is in φ(2i)(I W Hence, q2i is in φ(2i)(I W K ). (cid:3) K ). For any nonnegative integer n, we denote by v2(n) the 2-adic valuation of n. For a smooth projective variety X over F , we denote by Γ∗K(X) the gamma filtration on the Grothendieck ring K(X). We let cCH : S ∗(Λ) → CH(G/B) be the characteristic map. Corollary 2.7. Let G be Spin2n (respectively, Spin2n+1). If 2m(i)(ker cCH)(i) ⊆ CH)(i) for some nonnegative integer m(i), then for any i ≥ 3 and any (I W n ≥ [i/2]+2 (respectively, n ≥ [i/2]+1) the torsion of ΓiK(G/B)/Γi+1K(G/B) is annihilated by 2g(i), where g(i) = 1 + m(i) + v2((i − 1)!). Remark 2.8. It is shown that m(3) = 0 and m(4) = 1 in [1, Lemma 6.4]. Proof. The proof of [1, Theorem 6.5] still works with Theorem 2.6. (cid:3) Corollary 2.9. Let G be Spin2n (respectively, Spin2n+1). If 2m(i)(ker cCH)(i) ⊆ CH)(i) for some nonnegative integer m(i), then for any G-torsor E, any i ≥ 3 (I W and any n ≥ [i/2] + 2 (respectively, n ≥ [i/2] + 1) the torsion of CHi(E/B) is annihilated by 2t(i), where t(i) = 1 +Pi Proof. By [3, Theorem 2.2(2)], we have j=3 g(j) + v2((i − 1)!). ΓiK(G/B)/Γi+1K(G/B) ≃ ΓiK(E/B)/Γi+1K(E/B). 8 S. BAEK As the torsion of CHi(E/B) is annihilated by (i − 1)! i Yj=1 e(ΓiK(E/B)/Γi+1K(E/B)), where e(ΓiK(E/B)/Γi+1K(E/B)) denotes the finite exponent of its torsion subgroup (see [1, p.149]), the result follows from Corollary 2.7. (cid:3) References [1] S. Baek, E. Neher, K. Zainoulline, Basic polynomial invariants, fundamental represen- tations and the Chern class map, Doc. Math. 17 (2012), 135 -- 150. [2] J. Humphreys, Reflection groups and Coxeter groups. Cambridge studies in Advanced Math. 29, Cambridge Univ. Press (1990). [3] I. A. Panin, On the algebraic K-theory of twisted flag varieties, K-Theory 8 (1994), no. 6, 541 -- 585. Department of Mathematics and Statistics, University of Ottawa, Canada E-mail address: [email protected]

Subsets and Splits

No community queries yet

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